CECW-EH-Y Department of the Army EM 1110-2-

U.S. Army Corps of Engineers 1415

Engineer Washington, DC 20314-1000
Manual 5 March 1993
1110-2-1415
Engineering and Design

HYDROLOGIC FREQUENCY ANALYSIS

Distribution Restriction Statement

Approved for public release; distribution is unlimited.
 DEPARTMENT OF THE ARMY EM 1110-2-1415
U.S. Army Corps of Engineers
Washington. D.C. 20314-1000

CECW-EH-Y

Engineer Manual
No. 1110-2-1415 5 March 1993

Engineering and Design

HYDROLOGIC FREQUENCY ANAL YSJS

1. Purpose. This manual provides guidance and procedures for frequency analysis of: flood flows, low
flows, precipitation. water surface elevation, and flood damage.

2. Applicabilit}'. This manual applies to major subordinate commands. districts, and laboratories ha,·ing
responsibility for the design of civil works projects.

3. General. Frequency estimates of hydrologic, climatic and economic data are required for the
planning, design and evaluation of flood control and navigation projects. The text illustrates many of
the statistical techniques appropriate for hydrologic problems by example. The basic theory is usually
not provided, but references are provided for those who wish to research the techniques in more detail.

FOR THE COMMANDER:

WILLIAM D. BROWN
Colonel, Corps of Engineers
Chief of Staff
 DEPARTMENT OF THE ARMY EM 1110-2-1415
U.S. Army Corps of Engineers
Washington, D.C. 20314-1000

CECW-EH-Y

Engineer Manual
No. 1110-2-1415 5 March 1993

Engineering and Design

HYDROLOGIC FREQUENCY ANALYSIS

Table of Contents

Subject Paragraph Page

CHAPTER INTRODUCTION
Purpose and Scope . . . . . . . . . . . . . . . . . . . . I- I 1-1
References ......................... . 1-2 1-1
Definitions ......................... . 1-3 1-1
Need for Hydrologic Frequency Estimates .. . 1-4 1-1
Need for Professional Judgement ........ . 1-5 1-2

CHAPTER 2 FREQUENCY ANALYSIS

Definition .......................... . 2-1 2-1
Duration Curves ..................... . 2-2 2-3
Selection of Data for Frequency Analysis .. . 2-3 2-6
Graphical Frequency Analysis ........... . 2-4 2-10
Analytical Frequency Analysis .......... . 2-5 2-12

CHAPTER 3 FLOOD FREQUENCY ANALYSIS

Introduction ........................ . 3-1 3-1
Log-Pearson Type III Distribution ....... . 3-2 3-1
Weighted Skew Coefficient ............. . 3-3 3-6
Expected Probability . . . . . . . . . . . . . . . . . .. 3-4 3-7
Risk ............................. . 3-5 3-10
Conditional Probability Adjustment ...... . 3-6 3-11
Two-Station Comparison ............... . 3-7 3-13
Flood Volumes ...................... . 3-8 3-18
Effects of Flood Control Works on
Flood Frequencies ................... . 3-9 3-22
Effects of Urbanization ............... . 3-10 3-28

CHAPTER 4 LOW-FLOW FREQUENCY ANALYSIS

Uses ............................. . 4-1 4-1
Interpretation ....................... . 4-2 4-1
Application Problems ................. . 4-3 4-1
EM 1110-2-1415
5 Mar 93

Subject Paragraph Page

CHAPTER 5 PRECIPITATION FREQUENCY ANALYSIS

General Procedures ................... . 5-1 5-1
Available Regional Information ......... . 5-2 5-2
Derivation of Flood-Frequency Relations
from Precipitation .................. . 5-3 5-2

CHAPTER 6 STAGE(ELEVATION)-FREQUENCY ANALYSIS

Uses ............................. . 6-1 6-1
Stage Data .......................... . 6-2 6-1
Frequency Distribution ................ . 6-3 6-2
Expected Probability .................. . 6-4 6-2

CHAPTER 7 DAMAGE-FREQUENCY RELATIONSHIPS

Introduction ........................ . 7-1 7-1
Computation of Expected Annual Damage .. 7-2 7-1
Equivalent Annual Damage ............. . 7-3 7-3

CHAPTER 8 STATISTICAL RELIABILITY CRITERIA

Objective .......................... . 8-1 8-1
Reliability of Frequency Statistics ........ . 8-2 8-1
Reliability of Frequency Curves ......... . 8-3 8-1

CHAPTER 9 REGRESSION ANALYSIS AND APPLICATION TO REGIONAL

STUDIES
Nature and Application ................ . 9-1 9-1
Calculation of Regression Equations ...... . 9-2 9-1
The CorrelationCoefficient and Standard
Error ............................ . 9-3 9-3
Simple Linear Regression Example . . . . . . . . 9-4 9-4
Factors Responsible of Nondetennination .. . 9-5 9-7
Multiple Linear Regression Example ...... . 9-6 9-8
Partial Correlation .................... . 9-7 9-8
Verification of Regression Results ....... . 9-8 9-10
Regression by Graphical Techniques ...... . 9-9 9-10
Practical Guidelines . . . . . . . . . . . . . . . . . . . 9-10 9-10
Regional Frequency Analysis ........... . 9-11 9-11

CHAPTER 10 ANALYSIS OF MIXED POPULATIONS

Definition . . . . . . . . . . . . . . . . . . . . . . . . . .. 10-1 10-1
Procedure .......................... . 10-2 10-1
Cautions ........................... . 10-3 10-2

CHAPTER II FREQUENCY OF COINCIDENT FLOWS

Introduction ........................ . 11-1 11-1
A Procedure for Coincident Frequency Analysis 11-2 11-1

11
 EM 1110-2-1415
5 Mar 93

Subject Paragraph Page

CHAPTER 12 STOCHASTIC HYDROLOGY

Introduction ........................ . 12-1 12-1
Applications ........................ . 12-2 12-1
Basic Procedure ..................... . 12-3 12-1
Monthly Streamflow Model ............. . 12-4 12-3
Data Fill In ......................... . 12-5 12-6
Application In Areas of Limited Data ..... . 12-6 12-6
Daily Streamflow Model ............... . 12-7 12-6
Reliability .......................... . 12-8 12-7

APPENDIX A SELECTED BIBLIOGRAPHY

References and Textbooks .............. . A A-I
Computer Programs .................. . B A-4

APPENDIX B GLOSSARY ........................ . B-1

APPENDIX C COMPUTATION PROCEDURE FOR EXTREME

VALUE (GUMBEL) DISTRIBUTION ..... C-1

APPENDIX D HISTORIC DATA ................... . D-1

APPENDIX E EXAMPLES OF RELIABILITY TESTS FOR THE

MEAN AND STANDARD DEVIATION ... E-1

APPENDIX F STATISTICAL TABLES ............... . F-1

Median Plotting Positions .............. . F-2
Deviates for Pearson Type III Distribution .. F-4
Normal Distribution .................. . F-6
Percentage Points of the One-Tailed
t-Distribution ...................... . F-7
Values of Chi-Square Distribution ....... . F-8
Values of the F Distribution ............ . F-9
Deviates for the Expected Probability
Adjustment ........................ . F-10
Percentages for the Expected Probability
Adjustment ........................ . F-11
Confidence Limit Deviates for Normal
Distribution ....................... . F-12
Mean-Square Error of Station Skew
Coefficient ........................ . F-16
Outlier Test K Values (10 Percent
Significance Level) .................. . F-17
Binomial Risk Tables ................. . F-18

111
 EM 1110-2-1415
5 Mar 93

List of Figures

Title Number Page

Histogram and Probability Density Function ............. . 2-la 2-2

Sample and Theoretical Cumulative Distribution Functions .. . 2-lb 2-2
Daily Flow-Duration Curve .......................... . 2-2 2-3
Daily Flow-Duration Curves for Each Month ............. . 2-3 2-5
Illustration of Chronologie Sequence and Arrayed Flood Peaks . 2-4 2-9
Example of Graphical Frequency Analysis ............... . 2-5 2-12
Partial Duration Frequency Curve, Log-Log Paper ......... . 2-6a 2-14
Partial Duration Frequency Curve, Probability Paper ....... . 2-6b 2-14
Annual Frequency Curve ............................ . 3-1 3-4
Confidence Limit Curves based on the Non-central t-Distribution 3-2 3-8
Cumulative Probability Distribution of Exceedances per 100 Years 3-3 3-9
Two-Station Comparison Computations ................. . 3-4 3-16
Observed and Two-Station Comparison Frequency Curves ... . 3-5 3-17
Coordination of Flood-Volume Statistics ................ . 3-6 3-20
Flood-Volume Frequency Curves ...................... . 3-7 3-21
Flood- Volume Frequency Relations .................... . 3-Sa 3-23
Storage Requirement Determination .................... . 3-Sb 3-23
Daily Reservoir Elevation-Duration Curve ............... . 3-9 3-25
Seasonal Variation of Elevation-Duration Relations ........ . 3-10 3-25
Example With-Project versus Without-Project Peak Flow Relations 3-11 3-27
Example Without-Project and With-Project Frequency Curves . 3-12 3-27
Typical Effect of Urbanization on Flood Frequency Curves .. . 3-13 3-29
Low-Flow Frequency Curves ......................... . 4-1 4-3
Frequency Curve, Annual Precipitation ................. . 5-1 5-1
Flow-Frequency Curve, Unregulated and Regulated Conditions 6-1 6-1
Rating Curve for Present Conditions ................... . 6-2 6-3
Derived Stage-Frequency Curves, Unregulated and
Regulated Conditions .............................. . 6-3 6-3
Maximum Reservoir Elevation-Frequency Curve .......... . 6-4 6-4
Schematic for Computation of Expected Annual Damage .... . 7-1 7-2
Frequency Curve with Confidence Limit Curves .......... . 8-1 8-2
Computation of Simple Linear Regression Coefficients ..... . 9-1 9-5
Illustration of Simple Regression ...................... . 9-2 9-6
Example Multiple Linear Regression Analysis ............ . 9-3 9-9
Regression Analysis for Regional Frequency Computations .. . 9-4 9-14
Regional Analysis Computations for Mapping Errors ....... . 9-5 9-15
Regional Map of Regression Errors .................... . 9-6 9-15
Annual Peaks and Sequential Computed Skew by Year ...... . 9-7 9-18
Nonhurricane, Hurricane and Combined Flood Frequency Curves 10-1 10-3
Illustration of Water-Surface Profiles in Coincident
Frequency Analysis ............................... . 11-1 11-2
Data Estimation from Regression Line .................. . 12-la 12-2
Data Estimation with Addition of Random Errors ......... . 12-lb 12-2

iv
 EM 1110-2-1415
5 Mar 93

List of Tables

Title Number Page

Daily Flow-Duration Data and Interpolated Values ........ . 2-1 2-4

Annual Peaks, Sequential and Arrayed with Plotting Positions . 2-2 2-11
Partial Duration Peaks, Arrayed with Plotting Positions ..... . 2-3 2-13
Computed Frequency Curve and Statistics ............... . 3-1 3-3
High-Flow Volume-Duration Data ..................... . 3-2 3-18
Low-Flow Volume-Duration Data ..................... . 4-1 4-2

v
 EM I I 10-2-1415
5 Mar 93

ENGINEERING AND DESIGN

HYDROLOGIC FREQUENCY Al\OALYSIS
CHAPTER 1

INTRODUCTION

I -1. Purpose and Scope. This manual provides guidance in applying statistical principles
to the analysis of hydrologic data for Corps of Engineers Civil Works activities. The text
illustrates, by example, many of the statistical techniques appropriate for hydrologic
problems. The basic theory is usually not provided, but references are provided for those
who wish to research the techniques in more detail.
1-2. References. The techniques described herein are taken principally from "Guidelines
for Determining Flood Flow Frequency" (46) 1, "Statistical Methods in Hydrology" (1 ), and
"Hydrologic Frequency Analysis" (41 ). References cited in the text and a selected
bibliography of literature pertaining to frequency analysis techniques are contained in
Appendix A.

1-3. Definitions. Appendix B contains a list of definitions of terms common to

hydrologic frequency analysis and symbols used in this manual.

1-4. Need for Hydrologic Frequency Estimates.

a. Applications. Frequency estimates of hydrologic, climatic and economic data are

required for the planning, design and evaluation of water management plans. These plans
may consist of combinations of structural measures such as reservoirs, levees, channels,
pumping plants, hydroelectric power plants, etc., and nonstructural measures such as flood
proofing, zoning, insurance programs, water use priorities, etc. The data to be analyzed
could be streamflows, precipitation amounts, sediment loads, river stages, lake stages,
storm surge levels, flood damage, water demands, etc. The probability estimates from
these data are used in evaluating the economic, social and environmental effects of the
proposed management action.

b. Objective. The objective of frequency analysis in a hydrologic context is to infer

the probability that various size events will be exceeded or not exceeded from a given
sample of recorded events. Two basic problems exist for most hydrologic applications.
First the sample is usually small, by statistical standards, resulting in uncertainty as to the
true probability. And secondly, a single theoretical frequency distribution does not always
fit a particular data-type equally well in all applications. This manual provides guidance
in fitting frequency distributions and construction of confidence limits. Techniques are
presented which can possibly reduce the errors caused by small sample sizes. Also, some
types of data are noted which usually do not fit any theoretical distributions.

c. General Guidance. Frequency analysis should not be done without consideration

of the primary application of the results. The application will have a bearing on the type
of analysis (annual series or partial duration series), number of stations to be included,

1
Numbered references refer to Appendix A, Selected Bibliography.

1-1
EM 1110-2 1415
S Mar 93

whether regulated frequency curves will be needed, etc. A frequency study should be
well coordinated with the hydrologist, the planner and the economist.

1-5. Need for Professional Judgment. It is not possible to define a set of procedures that
can be rigidly applied to each frequency determination. There may be applications where
more complex joint or conditional frequency methods, that were considered beyond the
scope of this guidance, will be required. Statistical analyses alone will not resolve all
frequency problems. The user of these techniques must insure proper application and
interpretation has been made. The judgment of a professional experienced in hydrologic
analysis should always be used in concert with these techniques.

1-2
 EM 1110-2-1415
5 Mar 93

CHAPTER 2
FREQUENCY ANALYSIS

2-1. Definition.

a. Frequency. Many of the statistical techniques that are applied to hydrologic data
(to enable inferences to be made about particular attributes of the data) can be labeled
with the term "frequency analysis" techniques. The term "frequency" usually connotes a
count (number) of events of a certain magnitude. To have a perspective of the importance
of the count, the total number of events (sample size} must also be known. Sometimes the
number of events within a specified time is used to give meaning to the count, e.g., two
daily flows were this low in 43 years. The probability of a certain magnitude event
recurring again in the future, if the variable describing the events is continuous, (as are
most hydrologic variables), is near zero. Therefore, it is necessary to establish class
intervals (arbitrary subdivisions of the range) and define the frequency as the number of
events that occur within a class interval. A pictorial display of the frequencies within
each class interval is called a histogram (also known as a frequency polygon).

b. Relative Frequency. Another means of representing the frequency is to compute

the relative frequency. The relative frequency is simply the number of events in the class
interval divided by the total number of events:

f. • n./N (2-1)
1 1

where:

f. • relative frequency of events in class interval i

1

n. • number of events in interval i

1

N • total number of events

A graph of the relative frequency values is called a frequency distribution or histogram,

Figure 2-la. As the number of observations approaches infinity and the class interval size
approaches zero, the enveloping line of the frequency distribution will approach a smooth
curve. This curve is termed the probability density function (Figure 2-la).

c. Cumulative Frequency. In hydrologic studies, the probability of some magnitude

being exceeded (or not exceeded) is usually the primary interest. Presentation of the data
in this form is accomplished by accumulating the probability (area) under the probability
density function. This curve is termed the cumulative distribution function. In most
statistical texts, the area is accumulated from the smallest event to the largest. The
accumulated area then represents non-exceedance probability or percentage (Figure 2-1 b).
It is more common in hydrologic studies to accumulate the area from the largest event to
the smallest. Area accumulated in this manner represents exceedance probability or
percentage.

2-1
EM I I 10-2 1415
5 Mar 93

Potomac RiYer at Point of Roc:k1, MD

80
I I I I
~ -
~ '" -
1- -
se
~
-· .L ~
-
-
..• ~ -
-
•c: ~
I
..."'
liil S2 I-- -
-
..
~
0 ~

J
~ -
E ~ -
z
:I
• I--
1- / -
-
- -
.. ~
I--·
~

~
1\ 1,m
-
-
-
-
~ -
0
.. ..... ....
Lopritbm of Annual Pew
1!1.2 1!1.8
•

Figure 2-1 a. Histogram and Probability Density Function.

Potomac RiYer at Point of Roc:ka, MD
s

•c:u o.e
.,•
••u
c3 o.e
c:'
0

...z
~
...
0

o ....
i
A
e
A.
O.R

Lopritbm of Annual Pew

Figure 2-1 b. Sample and Theoretical Cumulative Distribution Functions.

2-2
 EM 111 0-2- 1415
5 Mar 93

2-2. Duration Curves.

a. Computation. The computation of flow-duration curves was probably the first

attempt to analyze hydrologic data by statistical techniques. The events for flow-duration
curves are usually mean daily flow values. One of the first steps in preparation of a
duration curve is dividing the range of the data into class intervals. Table 2-1 shows the
class intervals of daily flows input into the computer program ST A TS (58) for a duration
analysis of Fishkill Creek at Beacon, New York. The flows tend to be grouped near the
low end with very few large flows. Therefore, the relative frequency curve is skewed to
the right. It has been found that making the logarithmic transform reduces the skewness
of the curve. The class intervals in Table 2-1 are based on a logarithmic distribution of
the flows. Plotting the data in Table 2-1 on log-probability paper, Figure 2-2, provides a
plot that is easily read at the extremities of the data. The daily flow-duration curve
cannot be considered a frequency curve in the true sense, because the daily flow on a
particular day is highly correlated with the flow on the preceding day. For this reason,
the abscissa is labeled "percent of time."

Pienkill Creek •t a ..can, NY 1946-1988

/
v
183 /
~
•
'"u.
3
... ~
v
v
D
18 2
II.

...
:I
-4
c•

181 /
v
c~
v
aa.aa aa.s as ae 68 18 1 .1 • 81

Figure 2-2. Daily Flow-Duration Curve.

2-3
 EM 1110-2 1415
S Mar 93

Table 2-1. Daily Flow-Duration Data and Interpolated Values.

-DURATION DATA- FISHKILL CR AT BEACON, NY - DAILY FLOWS

***************************************************************************************
* LOWER NUMBER PERCENT • LOWER NUMBER PERCENT *
* CLASS CLASS IN ACCUM EQUAL OR * CLASS CLASS IN ACCUM EQUAL OR *
* NUMBER LIHIT CLASS NUMBER EXCEED * NUMBER LIHIT CLASS NUMBER EXCEED *
* FLOW,CFS * FLOW,CFS *
·------------------------------------------·------------------------------------------·
* 1 1.00 5 8766 100.00 * 16 100.00 888 5606 63.95 *
* 2 2.00 4 8761 99.94 • 17 150.00 783 4718 53.82 *
• 3 3.00 8 8757 99.90 • 18 200.00 1246 3935 44.89 *
* 4 4.00 22 8749 99.81 • 19 300.00 822 2689 30.68 *
• s 5.00 37 8727 99.56 • 20 400.00 484 1867 21.30 •
* 6 6.00 156 8690 99.13 • 21 500.00 340 1383 15.78 *
* 7
8
8.00
10.00
95
254
8624 98.38
97.30
• 22
23
600.00
800.00
465
239
1043
578
11.90
6.59
*
* 8529 * *
* 9 15.00 261 8275 94.40 • 24 1000.00 251 339 3.87 *
* 10 20.00 423 8014 91.42 • 25 1500.00 47 88 1. 00 *
* 11 30.00 405 7591 86.60 * 26 2000.00 32 41 0.47 *
* 12 40.00 359 7186 81.98 * 27 3000.00 6 9 0.10 *
* 13 50.00 332 6827 77.88 • 28 4000.00 0 3 0.03 *
* 14 60.00 480 6495 74.09 * 29 6000.00 3 3 0.03 *
* 15 80.00 409 6015 68.62 * 30 7000.00 0 0 0.00 *
***************************************************************************************

-INTERPOLATED DURATION CURVE- FISHKILL CR AT BEACON, NY - DAILY FLOWS

*****************************************************************
• PERCENT INTERPOLATED • PERCENT INTERPOLATED •
• EQUAL OR HAGNITUOE •EQUAL OR HAGNITUOE •
* EXCEED FLOW,CFS • EXCEED FLOW,CFS •
·-------------------------------·-------------------------------·
* 0.01 6970.0 • 60.00 118.0 •
• 0.05 3480.0 • 70.00 74.5 •
• 0.10 3020.0 • 80.00 44.7 •
* 0.20 2530.0 • 85.00 33.4 *
* 0.50 1960.0 • 90.00 22.7 •
* 1.00 1500.0 • 95.00 14.0 *
* 2.00 1230.0 • 98.00 8.8 *
* 5.00 903.0 * 99.00 6.3 *
* 10.00 658.0 * 99.50 5.2 *
* 15.00 518.0 * 99.80 4.0 *
* 20.00 420.0 * 99.90 2.9 *
* 30.00 306.0 • 99.95 1.9 •
* loO.OO 230.0 • 99.99 1.5 *
* 50.00 171.0 * 100.00 1.1 *
*****************************************************************

Output from HEC computer program STATS.

b. l.lm. Duration curves are useful in assessing the general low flow characteristics
of a stream. If the lower end drops rapidly to the probability scale, the stream has a low
ground-water storage and, therefore, a low or no sustained flow. The overall slope of the
flow-duration curve is an indication of the flow variability in the stream. Specific uses
that have been made of duration curves are: 1) assessing the hydropower potential of
run-of-river plants; 2) determining minimum flow release; 3) water quality studies; 4)
sediment yield studies; and 5) comparing yield potential of basins. It must be
remembered that the chronology of the flows is lost in the assembly of data for duration
curves. For some studies, the low-flow sequence, or persistence, may be more important
(see Chapter 4).

2-4
 EM 1110-2-1415
5 Mar 93

c. Monthlv Curves. Occasionally the distribution of the flows during particular

seasons of the year is of interest. Figure 2-3 illustrates a way of presenting
daily-flow-duration curves that were computed from the daily flows during each month.
d. Stage-Duration. Stage-duration curves are often used to establish vertical
navigation clearances for bridges. If there have been no changes in the discharge versus
stage relationship (rating curve), then the stages may be used instead of flows to compute
a stage-duration curve. But, if there have been significant changes to the rating curve
(because of major levee construction, for instance) then the stage-duration curve should
be derived from the flow-duration curve and the latest rating curve. The log
transformation is !lQ1 recommended for stages.

FISHKILL CREEK
10000

/- MAX

en
""
u
....z
1000
---
.......-- 1%
5%
10%
30%
a
w
a
w
w
u

--
X
w
X
0
50% w
._....
.J ::E
100 70%
"">
._""
90%
.....J 0

0""' 95%
zw
99% u
a:
10 '. ·.· MIN w
Q.

MONTH

Figure 2-3. Daily Flow-Duration Curves for Each Month ..

2-5
 EM 1110-2-1415
5 Mar 93

e. Future Probabilities. A duration curve is usually based on a fairly large sample

size. For instance, Figure 2-2 is based on 8766 daily values (Table 2- I). Even though the
observed data can be used to make inferences about future probabilities, conclusions
drawn from information at the extremities can be misleading. The data indicate there is a
zero percent chance of exceeding 6970 cfs, however, it is known that there is a finite
probability of experiencing a larger flow. And similarly, there is some chance of
experiencing a lower flow than the recorded 1.1 cfs. Therefore, some other means is
needed for computing the probabilities of infrequent future events. Section 2-4 describes
the procedure for assigning probabilities to independent events.

2-3. Selection of Data for Frequency Analysis.

a. Selection Based on APplication of Results. The primary question to be asked

before selection of data for a frequency study is: how will the frequency estimates be
used? If the frequency curve is to be used for estimating damage that is related to the
peak flow in a stream, maximum peak flows should be selected from the record. If the
damage is best related to a longer duration of flow, the mean flow for several days'
duration may be appropriate. For example, a reservoir's behavior may be related to the
3-day or 10-day rain flood volume or to the seasonal snowmelt volume. Occasionally, it is
necessary to select a related variable in lieu of the one desired. For example, where
mean-daily flow records are more complete than the records of peak flows, it may be
more desirable to derive a frequency curve of mean-daily flows and then, from the
computed curve, derive a peak-flow curve by means of an empirical relation between
mean daily flows and peak flows. All reasonably independent values should be selected,
but only the annual maximum events should be selected when the application of analytical
procedures discussed in Chapter 3 is contemplated.

b. Uniformity of Data.

(I) General Considerations. Data selected for a frequency study must measure the
same aspect of each event (such as peak flow, mean-daily flow, or flood volume for a
specified duration), and each event must result from a uniform set of hydrologic and
operational factors. For example, it would be improper to combine items from old records
that are reported as peak flows but are in fact only daily readings, with newer records
where the instantaneous peak was actually measured. Similarly, care should be exercised
when there has been significant change in upstream storage regulation during the period
of record to avoid combining unlike events into a single series. In such a case, the entire
record should be adjusted to a uniform condition (see Sections 2-3f and 3-9). Data should
always be screened for errors. Errors have been noted in published reports of annual
flood peaks. And, errors have been found in the computer files of annual flood peaks.
The transfer of data to either paper or a computer file always increases the probability
that errors have been accidentally introduced.

(2) Mixed Populations. Hydrologic factors and relationships during a general winter
rain flood are usually quite different from those during a spring snowmelt flood or during
a local summer cloudburst flood. Where two or more types of floods are distinct and do
not occur predominately in mutual combinations, they should not be combined into a
single series for frequency analysis. It is usually more reliable in such cases to segregate
the data in accordance with type and to combine only the final curves, if necessary. In
the Sierra Nevada region of California and Nevada, frequency studies are made separately
for rain floods which occur principally during the months of November through March,

2-6
 EM 1110-2-1415
5 Mar 93

and for snowmelt floods, which occur during the months of April through July. Flows for
each of these two seasons are segregated strictly by cause - those predominantly caused by
snowmelt and those predominantly caused by rain. In desert regions, summer
thunderstorms should be excluded from frequency studies of winter rain flood or spring
snowmelt floods and should be considered separately. Along the Atlantic and Gulf Coasts,
it is often desirable to segregate hurricane floods from non hurricane events. Chapter I 0
describes how to combine the separate frequency curves into one relation.
c. Location Differences. Where data recorded at two different locations are to be
combined for construction of a single frequency curve, the data should be adjusted as
necessary to a single location, usually the location of the longer record. The differences in
drainage area, precipitation and, where appropriate, channel characteristics between the
two locations must be taken into account. When the stream-gage location is different
from the project location, the frequency curve can be constructed for the stream-gage
location and subsequently adjusted to the project location.
d. Estimating Missjng Events. Occasionally a runoff record may be interrupted by a
period of one or more years. If the interruption is caused by destruction of the gaging
station by a large flood, failure to fill in the record for that flood would result in a biased
data set and should be avoided. However, if the cause of the interruption is known to be
independent of flow magnitude, the record should be treated as a broken record as
discussed in Section 3-2b. In cases where no runoff records are available on the stream
concerned, it is usually best to estimate the frequency curve as a whole using regional
generalizations, discussed in Chapter 9, instead of attempting to estimate a complete series
of individual events. Where a longer or more complete record at a nearby station exists, it
can be used to extend the effective length of record at a location by adjusting frequency
statistics (Section 3-7) or estimating missing events through correlation (Chapter 12).
e. Climatic Cycles. Some hydrologic records suggest regular cyclic variations in
precipitation and runoff potential, and many attempts have been made to demonstrate that
precipitation or streamflows evidence variations that are in phase with various cycles,
particularly the well established 11-year sunspot cycle. There is no doubt that
long-duration cycles or irregular climatic changes are associated with general changes of
land masses and seas and with local changes in lakes and swamps. Also, large areas that
have been known to be fertile in the past are now arid deserts, and large temperate regions
have been covered with glaciers one or more times. Although the existence of climatic
changes is not questioned, their effect is ordinarily neglected, because the long-term
climatic changes have generally insignificant effect during the period concerned in water
development projections, and short-term climatic changes tend to be self-compensating.
For these reasons, and because of the difficulty in differentiating between stochastic
(random) and systematic changes, the effect of natural cycles or trends during the analysis
period is usually neglected in hydrologic frequency studies.
f. Effect of Basin Development on Frequency Relations.
(1) Hydrologic frequency estimates are often used for some purpose relating to
planning, design or operation of water resources control measures (structural and
nonstructural). The anticipated effects of these measures in changing the rate and volume
of flow is assessed by comparing the without project frequency curve with the
corresponding with project frequency curve. Also, projects that have existed in the past
have affected the rates and volumes of flows, and the recorded values must be adjusted to
reflect uniform conditions in order that the frequency analysis will conform to the basic

2-7
 EM 1110-2-1415
5 Mar 93

assumption of homogeneity. In order to meet the assumptions associated with analytical

frequency analysis techniques, the flows must be essentially unregulated by manmade
storage or diversion structures. Consequently, wherever practicable, recorded runoff
values should be adjusted to natural (unimpaired) conditions before an analytical
frequency analysis is made. In cases where the impairment results from a multitude of
relatively small improvements that have not changed appreciably during the period of
record, it is possible that analytical frequency analysis techniques can be applied. The
adjustment to natural conditions may be unnecessary and, because of the amount of work
involved, not cost effective.
(2) One approach to determining a frequency curve of regulated or modified
runoff consists of routing all of the observed flood events under conditions of proposed or
anticipated development. Then a relationship is developed between the modified and the
natural flows, deriving an average or dependable relationship. A frequency curve of
modified flows is derived from this relationship and the frequency curve of natural flows.
In order to determine frequencies of runoff for extreme floods, routings of multiples of
the largest floods of record or multiples of a large hypothetical flood can be used.
Techniques of estimating project effects are outlined in Chapter 3-09d.
g. Annual Series Versus Partial Duratjon Series. There are two basic types of
frequency curves used to estimate flood damage. A curve of annual maximum events is
ordinarily used when the primary interest lies in the larger events or when the second
largest event in any year is of little concern in the analysis. The partial-duration curve
represents the frequency of all independent events of interest, regardless of whether two
or more occurred in the same year. This type of curve is sometimes used in economic
analysis, where there is considerable damage associated with the second largest and third
largest floods that occurred in some of the years. Caution must be exercised in selecting
events because they must be both hydrologically and economically independent. The
selected series type should be established early in the study in coordination with the
planner and/or economist. The time interval between flood events must be sufficient for
recovery from the earlier flood. Otherwise damage from the later flood would not be as
large as computed. When both the frequency curve of annual floods and the
partial-duration curve are used, care must be exercised to assure that the two are
consistent. A graphic demonstration of the relation between a chronologie record, an
annual-event curve and a partial-duration curve is shown on Figure 2-4.
h. Presentation of Data and Results of Frequency Analysis. When frequency curves
are presented for technical review, adequate information should be included to permit an
independent review of the data, assumptions and analysis procedures. The text should
indicate clearly the scope of the studies and include a brief description of the procedure
used, including appropriate references. A summary of the basic data consisting of a
chronological tabulation of values used and indicating sources of data and any adjustments
should be included. The frequency data should also be presented in graphical form,
ordinarily on probability paper, along with the adopted frequency curves. Confidence
limit curves should also be included for analytically-derived frequency curves to illustrate
the relative value of the frequency relationships. A map of the gage locations and tables
of the adopted statistics should also be included.

2-8
 EM 1110-2-1415
5 Mar 93

Chronolocical Sequence of Flooda, 1~6~-1968

10
~
... 9
~
c: 8
••
:I 7
0
6
.s= 5
• "'
0
r;: 3
•.. 2, lit
.ac
g., I I I d I I I
Ill
0
II I I I I I I I -I
1959 1960 1961 1962 1963 196"' 1965 1966 1967 1968
Year

Array of Annual Flood Pe&kl, 1P46-1~

(Annual Seri•)
ol! 10
... 9
~
; 8
a r
=0

.s
6
5

r;:~ "'
3
1 2 I I
•
g., 1 1111111111
0 11111111111111
1 10 15 20 2"'
Arrayed Sequence Number

Array of Flood Pe&kl above 1500 cla, 1~6-1968

(Partial-Duration Seri•)
10
ol! 9
u
~ 8
c:
=7
:I

.
_g 6
...
.s 5
0
r;:
"3
1 2
•
g., 1 I 11111111111111111 II II I II I Ill II Ill I
0
I Ill 11111111111111 IIIII II lllln-lrrr
5 10 15 20 25 30 40 45 50
Arrayed Sequence Number

Figure 2-4. Illustration of Chronologie Sequence and Arrayed Flood Peaks.

2-9
EM 1110-2-1415
5 Mar 93

2-4. Graphical Frequency Analysis.

a. Advamages and Limitations. Every set of frequency data should be plotted

graphically, even though the frequency curves are obtained analytically. It is important to
visually compare the observed data with the derived curve. The graphical method of
frequency-curve determination can be used for any type of frequency study, but
analytical methods ha\•,• certain advantages when they are applicable. The principal
advantages of graphical methods are that they are generally applicable, that the derived
curve can be easily visualized, and that the observed data can be readily compared with
the computed results. However, graphical methods of frequency analysis are generally less
consistent than analytical methods as different individuals would draw different curves.
Also, graphical procedures do not provide means for evaluating the reiiability of the
estimates. Comparison of the adopted curve with plotted points is not an index of
reliability, but it is often erroneously assumed to be, thus implying a much greater
reliability than is actually attained. For these reasons, graphical methods should be
limited to those data types where analytical methods are known not to be generally
applicable. That is, where frequency curves are too irregular to compute analytically, for
example, stream or reservoir stages and regulated flows. Graphical procedures should
always be to visually check the analytical computations.

b. Selection and Arrangement of Peak Flow Data. General principles in the selection
of frequency data are discussed in Section 2-3. Data used in the construction of
frequency curves of peak flow consist of the maximum instantaneous flow for each year
of record (for annual-event curve) or all of the independent events that exceed a selected
base value (for partial-duration curve). This base value must be smaller than any flood
flow that is of importance in the analysis, and should also be low enough so that the total
number of floods in excess of the base equals or exceeds the number of years of record.
Table 2-2 is a tabulation of the annual peak flow data with dates of occurrence, the data
arrayed in the order of magnitude, and the corresponding plotting positions.

c. Plotting Formulas. Median plotting positions are tabulated in Table F- I. In

ordinary hydrologic frequency work, N is taken as the number of years of record rather
than the number of events, so that percent chance exceedance can be thought of as the
number of events per hundred years. For arrays larger than 100, the plotting position, P 1,
of the largest event is obtained by use of the following equation:

(2-2a)

The plotting position for the smallest event (P") is the complement (1-P 1) of this value,
and all the other plotting positions are interpolated linearly between these two. The
median plotting positions can be approximated by

Pm • lOO(m -.3)/(N +.4) (2-2b)

where m is the order number of the event.

For partial-duration curves, particularly where there are more events than years (N),
plotting positions that indicate more than one event per year can also be obtained using

2-10
 EM 1110-2-1415
5 Mar 93

Table 2-2. Annual Peaks, Sequential and Arrayed with Plotting Positions.

-PLOTTING POSITIONS-FISHKILL CREEK AT BEACON, N.Y.

****************************************************************
* ..... EVENTS ANALYZED ...... * ........... ORDERED EVENTS .......... *
• • WATER MEDIAN •
* l'f:lN DAY YEAR FL~.CFS • RANK YEAR FL~,CFS PLOT POS *
·--------------------------·-----------------------------------·
• 3 5 1945 2290. • 1 1955 8800. 2.87 •
* 12 27 1945 1470. * 2 1956 8280. 6.97 •
• 3 15 1947 2220. • 3 1961 4340. 11.07 •
• 3 18 1948 2970. • 4 1968 3630. 15.16 •
• 1 1 1949 3020. • 5 1953 3220. 19.26 •
• 3 9 1950 1210. • 6 1952 3170. 23.36 •
• 4 1 1951 2490. • 7 1962 3060. 27.46 •
• 3 12 1952 3170. * 8 1949 3020. 31.56 •
* 1 25 1953 3220. • 9 1948 2970. 35.66 •
• 9 13 1954 1760. • 10 1958 2500. 39.75 •
• 8 20 1955 8800. • 11 1951 2490. 43.85 •
* 10 16 1955 8280. * 12 1945 2290. 47.95 •
• 4 10 1957 1310. • 13 1947 2220. 52.05 •
* 12 21 1957 2500. * 14 1960 2140. 56.15 *
* 2 11 1959 1960. * 15 1959 1960. 60.25 *
* 4 6 1960 2140. * 16 1963 1780. 64.34 *
* 2 26 1961 4340. * 17 1954 1760. 68.44 •
* 3 13 1962 3060. * 18 1967 1580. 72.54 •
* 3 28 1963 1780. * 19 1946 1470. 76.64 *
* 1 26 1964 1380. • 20 1964 1380. 80.74 •
* 2 9 1965 980. • 21 1957 1310. 84.84 •
• 2 15 1966 1040 . * 22 1950 1210. 88.93 •
* 3 30 1967 1580. * 23 1966 1040. 93.03 *
• 3 19 1968 3630. • 24 1965 980. 97.13 *
****************************************************************

Output from HEC computer program HECWRC.

Equation 2-2b. This is simply an approximate method used in the absence of knowledge
of the total number of events in the complete set of which the partial-duration data
constitute a subset.

d. Plotting Grids. The plotting grid recommended for annual flood flow events is
the logarithmic normal grid developed by Allen Hazen (ref 13) and designed such that a
logarithmic normal frequency distribution will be represented by a straight line, Figure
2-5. The plotting grid used for stage frequencies is often the arithmetic normal grid. The
plotting grids may contain a horizontal scale of exceedance probability, exceedance
frequency, or percent chance exceedance. Percent chance exceedance (or nonexceedance)
is the recommended terminology.

e. Examole Plotting of Annual-Event Frequency Curve. Figure 2-5 shows the

plotting of a frequency curve of the annual peak flows tabulated in Table 2-2. A smooth
curve should be drawn through the plotted points. Unless computed by analytical
frequency procedures, the frequency curve should be drawn as close to a straight line as
possible on the chosen probability graph paper. The data plotted on Figure 2-5 shows a
tendency to curve upward, therefore, a slightly curved line was drawn as a best fit line.

f. Example Plotting of Partial-Duration Curve. The partial-duration curve

corresponding to the partial-duration data in Table 2-3 is shown of Figure 2-6a. This
curve has been drawn through the plotted points, except that it was made to conform with
the annual-event curve in the upper portion of the curve. The annual-event curve was

2-11
EM 1110-2-1415
5 Mar 93

developed in accordance with the procedures described in Chapter 3. When

partial-duration data must include more events than there are years of record (see
Subparagraph b) it will be necessary to use logarithmic paper for plotting purposes, as on
Figure 2-6a, in order to plot exceedance frequencies greater than 100 percent. Otherwise,
the curve can be plotted on probability grid, as illustrated on Figure 2-6b.

2-5. Analytical Freauency Analysis.

a. General Procedures and Common Distributions. The fitting of data by an

analytical procedure consists of selecting a theoretical frequency distribution, estimating
the parameters of the distribution from the data by some fitting technique, and then
evaluating the distribution function at various points of interest. Some theoretical
distributions that have been used in hydrologic frequency analysis are the normal
(Gaussian), log normal, exponential, two-parameter gamma, three-parameter gamma,
Pearson type III, log-Pearson type III, extreme value (Gumbel) and log Gumbel. Chapter
3 describes the fitting of the log-Pearson type III to annual flood peaks and Appendix C
describes fitting the extreme value (Gumbel) distribution.

v
...u•
...
J
0
-4
1&. ./
v
113 L
v
.:t

••
A.
-4
•c
::J
c
~

88.98 98.9 99 98 &e 18 1 .1 .81

Pe~ent C~nce Exceedence

Figure 2-5. Example of Graphical Frequency Analysis

2-12
 EM 1110-2-1415
5 Mar 93

Table 2-3. Partial Duration Peaks, Arrayed with Plotting Positions.

FISHKILL CREEK AT BEACON, NY -- PEAKS ABOVE 1500 CFS

••••••••••••••••••••••••••••••••••••• •••••••••••••••••••••••••••••••••••••
* ........... ORDERED EVENTS .......... * *·· ......... ORDERED EVENTS .......... *
• WATER MEDIAN * • WATER MEDIAN *
• RANK YEAR FLOW,CFS PLOT POS • • RANK YEAR FLOW,CFS PLOT POS *
·-----------------------------------·
• 1 1955 8800. 2.87 • ·-----------------------------------·
• 26 1952 1820 . 105.33
• 2 1956 8280. 6.97 • • 27 19~5 1780 . 109.43
• 3 1961 4340. 11.07 • • 28 1963 1780 . 113.52 *
4 1968 3630. 15.16 • • 29 1956 1770 . 117.62
• 5 1953 3220. 19.26 • • 30 195~ 1760 . 121.72 *
• 6 1952 3170. 23.36 * • 31 1952 1730. 125.82
* 7 1962 3060. 27.~6 * • 32 1968 1720. 129.92
• 8 1949 3020. 31.56 • • 33 1955 1660. 134.02 *
• 9 19~8 2970. 35.66 • • 34 1958 1650 . 138.11
• 10 19~8 2750. 39.75 • • 35 1958 1650. 142.21
• 11 1949 2700. 43.85 * 36 1953 1630. 146.31 *
• 12 1958 2500. 47.95 • • 37 1960 1610 . 150.41 *
• 13 1951 2490. 52.05 • • 38 1956 1600. 154.51 *
• 14 1952 2460. 56.15 * • 39 1958 1590. 158.61 *
* 15 1945 2290. 60.25 * • 40 1958 1580. 162.70 *
* 16 1953 2290. 64.34 • • 41 1967 1580. 166.80
• 17 1958 2290. ·68.4~ • • 42 1951 1560. 170.90 *
• 18 1953 2280. 72.5~ • • 43 1959 1560. 175.00 *
• 19 1948 2220. 76.64 • • 44 1955 1550. 179. 10 *
• 20 1951 2210. 80.74 • • 45 1951 1540. 183.20 *
• 21 1960 2140. 84.84 * • 46 1968 1530. 187.30 *
• 22 1953 2080. 88.93 • • 47 1960 1520. 191.39
• 23 1959 1960. 93.03 • • 48 1958 1520. 195.49 *
* 24 1959 1920. 97.13 * • 49 1952 1520. 199.59 *
• 25 1958 1900. 101.23 * * 50 1948 1510. 203.69 *
* * * 51 1963 1510. 207.79 *
************************************* *************************************

b. Advantages. Determining the frequency distribution of data by the use of

analytical techniques has several advantages. The use of an established procedure for
fitting a selected distribution would result in consistent frequency estimates from the same
data set by different persons. Error distributions have been developed for some of the
theoretical distributions that enable computing the degree of reliability of the frequency
estimates (see Chapter 8). Another advantage is that it is possible to regionalize the
parameter estimates which allows making frequency estimates at ungaged locations (see
Chapter 9).
c. Disadvantages. The theoretical fitting of some data can result in very poor
frequency estimates. For example, stage-frequency curves of annual maximum stages are
shaped by the channel and valley characteristics, backwater conditions, etc. Another
example is the flow-frequency curve below a reservoir. The shape of this frequency
curve would depend not only on the inflow but the capacities, operation criteria, etc.
Therefore, graphical techniques must be used where analytical techniques provide poor
frequency estimates.

2-13
EM 1110-2-1415
5 Mar 93

l"i.,kiU c.-..k et . .ecan HY 1..a•1MB

u4 •

" ~
'r--.
f'

.•u
~
~
D

..;
L
D c

~~
""•• D
~
~I'll
Q.

~~

Figure 2-6a. Partial Duration Frequency Curve, Log-Log Paper

..•
u ll
ll
..
L
;
D 1e 4

""•
•
IL /
/c

,_ ~~ ~
I""
~
D

... ...

118.88 88.8 88 1 .1 • e1

Figure 2-6b. Partial Duration Frequency Curve, Probability Paper.

2-14
 EM 1110-2-1415
5 Mar 93

CHAPTER 3
FLOOD FREQUENCY ANALYSIS

3-1. Introduction.
The procedures that federal agencies are to follow when computing a frequency
curve of annual flood peaks have been published in Guidelines for Determining Flood
Flow Frequency, Bulletin 17B (46). As stated in Bulletin 17B, "Flood events ... do not fit
any one specific known statistical distribution." Therefore, it must be recognized that
occasionally, the recommended techniques may not provide a reasonable fit to the data.
When it is necessary to use a procedure that departs from Bulletin 17B, the procedure
should be fundamentally sound and the steps of the procedure documented in the report
along with the frequency curves.

This report contains most aspects of Bulletin 17B, but in an abbreviated form.
Various aspects of the procedures are described in an attempt to clarify the computational
steps. The intent herein is to provide guidance for use with Bulletin 17B. The step by
step procedures to compute a flood peak frequency curve are contained in Appendix 12 of
Bulletin 17B and are not repeated herein.
3-2. Log-Pearson Type III Distribution.
a. General. The analytical frequency procedure recommended for annual maximum
streamflows is the logarithmic Pearson type III distribution. This distribution requires
three parameters for complete mathematical specification. The parameters are: the mean,
or first moment, (estimated by the sample mean, X); the variance, or second moment,
(estimated by the sample variance, S2); the skew, or third moment, (estimated by the
sample skew, G). Since the distribution is a logarithmic distribution, all parameters are
estimated from logarithms of the observations, rather than from the observations
themselves. The Pearson type III distribution is particularly useful for hydrologic
investigations because the third parameter, the skew, permits the fitting of non-normal
samples to the distribution. When the skew is zero the log-Pearson type III distribution
becomes a two-parameter distribution that is identical to the logarithmic normal (often
called log-normal) distribution.
b. Fitting the Distribution.

(1) The log-Pearson type III distribution is fitted to a data set by calculating the
sample mean, variance, and skew from the following equations:

}:X
X • -- (3-1)
N

2 Lx 2 L(X-X) 2
s = = (3-2a)
N-1 N-1

3-1
EM 1110-2-1415
5 Mar 93

. (3-2b)
N-1

N(L (X-X)3 )
G • = (3-3a)
(N-l){N-2)53 (N-I)(N-2)53

N2 <I x3> - 3N{l: x><I x2) + 2(L X) 3

- N{N-1 )(N-2)53
(3-3b)

in which:

X = mean logarithm
X == logarithm of the magnitude of the annual event
N • number of events in the data set
52 = unbiased estimate of the variance of logarithms
X - X-X, the deviation of the logarithm of asingle event from the mean
logarithm
G • unbiased estimate of the skew coefficient of logarithms

The precision of the computed values is more sensitive to the number of significant digits
when Equations 3-2b and 3-3b are used.
(2) In terms of the frequency curve itself, the mean represents the general
magnitude or average ordinate of the curve, the square root of the variance (the standard
deviation, 5) represents the slope of the curve, and the skew represents the degree of
curvature. Computation of the unadjusted frequency curve is accomplished by computing
values for the logarithms of the streamflow corresponding to selected values of percent
chance exceedance. A reasonable set of values and the results are shown in Table 3-1.
The number of values needed to define the curve depends on the degree of curvature (i.e.,
the skew). For a skew value of zero, only two points would be needed, while for larger
skew values all of the values in the table would ordinarily be needed.
(3) The logarithms of the event magnitudes corresponding to each of the selected
percent chance exceedance values are computed by the following equation:

log Q = X+ K5 (3-4)

where X and 5 are defined as in Equations 3-1 and 3-2 and where

3-2
 EM 1110-2-1415
5 Mar 93

log Q = logarithm of the flow (or other variable) corresponding to a specified value
of percent chance exceedance

K = Pearson type III deviate that is a function of the percent chance exceedance
and the skew coefficient.

c. Examole Computation.

( 1) As shown i.!l. the following example, Equation 3-4 is solved by using the
computed values of X and S and obtaining from Appendix V -3 the value of K
corresponding to the adopted skew, G, and...!}le selected percent chance exceedance (P).
An example computation for P-= 1.0, where X, S and G are taken from Table 3-1, is:

log Q = 3.3684 + 2.8236 (.2456)

4.0619
11::

Q • 11500 cfs
(2) It has been shown (36) that a frequency curve computed in this manner is biased
in relation to average future expectation because of uncertainty as to the true mean and
standard deviation. The effect of this bias for the normal distribution can be eliminated
by an adjustment termed the expected probability adjustment that accounts for the actual
sample size. This adjustment is discussed in more detail in Section 3-4. Table 3-1 and
Figure 3-1 shows the derived frequency curve along with the expected probability
adjusted curves and the 5 and 95 percent confidence limit curves.

Table 3-l. Computed Frequency Curve and Statistics.

-FREQUENCY CURVE- 01-3735 FISHKILL CREEK AT BEACON, NEW YORK
**************************************************************••
* ........ FLOW,CFS ........ * PERCENT * ... CONFIDENCE LIMITS .... •
* EXPECTED • CHANCE * •
* COMPUTED PROBABILITY * EXCEEDANCE * 0.05 LIMIT 0.95 LIMIT *
•·------------------------·------------·------------------------·
19200. 28300. * 0.2 * 39100. 12300 . *
* 14500. 19000. * 0.5 * 26900. 9740. *
* 11500. 14100. * 1.0 * 20100. 8080. •
• 9110. 10500. * 2.0 * 14800. 6640 . •
• 7100. 7820 . • 4.0 • 10800 . 5380. •
• 4960. 5210. * 10.0 • 6850. 3950. *
• 3650. 3740. • 20.0 • 4710 . 2990 . *
* 2190. 2190. • 50.0 • 2650. 1790. •
• 1440. 1420. • 80.0 • 1760. 1110 . •
• 1200. 1170 . * 90.0 • 1490. 884. •
• 1040. 1010. • 95.0 • 1320 . 746 . *
• 841. 791 . • 99.0 • 1100 . 568. •
*1111111111111111111111 I 1111111111111111111111111 I 1111111111111*
• FREQUENCY CURVE STATISTICS • STATISTICS BASED ON •
·--------------------------------·-~---------------------------·
• MEAN LOGARITHM 3.3684 * HISTORIC EVENTS 0 *
• STANDARD DEVIATION 0.2456 * HIGH OUTLIERS 0 •
• COMPUTED SKEW 0.7300 • LOW OUTLIERS 0 •
• GENERALIZED SKEW 0.6000 • ZERO OR MISSING 0 *
* ADOPTED SKEW 0.7000 * SYSTEMATIC EVENTS 24 *
****************************************************************

3-3
 EM 1110-2-1415
5 Mar 93

Fiankill C~••k et Beecon, NY 1946-1968

0 Obaerved Annual Peake

--. -.
Computed Frequency Curve
With Expected Probability Adjuatment
.,<+>

~
Confidence Limit Curvea

•u
v
/
~
18 4

v
3
....D
1&.
.:t

••
/
0..
....
•:J A!f!"'
c 18 3
c
q
~-

99.99 99.9 99 98 68 18 1 .1 . 81

Figure 3-1. Annual Frequency Curve.

d. Broken Record. A broken record results when one or more years of annual peaks
are missing for any reason not related to the flood magnitude. In other words, the missing
events were caused by a random occurrence. The gage may have been temporarily
discontinued for budgetary or other reasons. The different segments of the record are
added together and analyzed as one record, unless the different parts of the record are
considered non-homogeneous. If a portion of the record is missing because the gage was
destroyed by a flood or the flood was too low to record, then the observed events should
be analyzed as an incomplete record.
e. Incomplete Record. An incomplete record can result when some of the peak flow
events were either too high or too low. Different analysis procedures are recommended
for missing high events and for missing low events. Missing high events may result from
the gage being out of operation or the stage exceeding the rating table. In these cases,
every effort should be made to obtain an estimate of the missing events. Missing low
floods usually result when the flood height is below the minimum reporting level or the
bottom of a crest stage gage. In these cases, the record should be analyzed using the
conditional probability adjustment described in Appendix 5 of Bulletin 17B and Section
3-6 of this report.

3-4
 EM 1110-2-1415
5 Mar 93

f. Zero-flood years. Some of the gaging stations in add regions record no flow for
the entire year. A zero flood peak precludes the normal statistical analysis because the
logarithm of zero is minus infinity. In this case the record should be analyzed using the
conditional probability adjustment described in Appendix 5 of Bulletin 17B and Section
3-6 of this report.

g. Outliers.

(I) Guidance. The Bulletin 17B (46) defines outliers as "data points which depart
significantly from the trend of the remaining data." The sequence of steps for testing for
high and low outliers is dependent upon the skew coefficient and the treatment of high
outliers differs from that of low outliers. When the computed (station) skew coefficient is
greater than +0.4, the high-outlier test is applied first and the adjustment for any high
outliers and/or historic information is made before testing for low outliers. When the
skew coefficient is less than -0.4, the low-outlier test is applied first and the adjustment
for any low outlier(s) is made before testing for high outliers and adjusting for any
historic information. When the skew coefficient is between -0.4 and +0.4, both the high-
and low-outlier tests are made to the systematic record (minus any zero flood events)
before any adjustments are made.

(2) Eauation. The following equation is used to screen for outliers:

(3-5)

where:

X 0 = outlier threshold in log units

X = mean logarithm (may have been adjusted for high or low outliers, and/or
historical information depending on skew coefficient)

S • standard deviation (may be adjusted value)

KN = K value from Appendix 4 of Bulletin I7B or Appendix F, Table 11 of this
report. Use plus value for high-outlier threshold and minus value for
low-outlier threshold
N = Sample size (may be historic period (H) if historically adjusted statistics are
used)

(3) High Outliers. Flood peaks that are above the upper threshold are treated as high
outliers. The one or more values that are determined to be high outliers are weighted by
the historical adjustment equations. Therefore, for any flood peak(s) to be weighted as
high outlier(s), either historical information must be available or the probable occurrence
of the event(s) estimated based on flood information at nearby sites. If it is not possible
to obtain any information that weights the high outlier(s) over a longer period than that of
the systematic record, then the outlier(s) should be retained as part of the systematic
record.

3-5
EM 1110-2-1415
5 Mar 93

(4) Low Outliers. Flood peaks that are below the low threshold value are treated as
low outliers. Low outliers are deleted from the record and the frequency curve computed
by the conditional probability adjustment (Section 3-6). If there are one or more values
very near, but above the threshold value, it may be desirable to test the sensitivity of the
results by considering the value(s) as low outlier(s).

h. Historic Events and Historical Information.

( 1) Definitions. Historic events are large flood peaks that occurred outside of the
systematic record. Historical information is knowledge that some flood peak, either
systematic or historic, was the largest event over a period longer than that of the
systematic record. It is historical information that allows a high outlier to be weighted
over a longer period than that of the systematic record.

(2) Eauations. The adjustment equations are applied to historic events and high
outliers at the same time. It is important that the lowest historic peak be a fairly large
peak, because every peak in the systematic record that is equal to or larger than the lowest
historic peak must be treated as a high outlier. Also a basic assumption in the adjusting
equations is that no peaks higher than the lowest historic event or high outlier occurred
during the unobserved part of the historical period. Appendix D in this manual is a
reprint of Appendix 6 from Bulletin 17B and contains the equations for adjusting for
historic events and/or historical information.

3-3. Weighted Skew Coefficient.

a. General. It can be demonstrated, either through the theory of sampling

distributions or by sampling experiments, that the skew coefficient computed from a small
sample is highly unreliable. That is, the skew coefficient computed from a small sample
may depart significantly from the true skew coefficient of the population from which the
sample was drawn. Consequently, the skew coefficient must be compared with other
representative data. A more reliable estimate of the skew coefficient of annual flood
peaks can be obtained by studying the skew characteristics of all available streamflow
records in a fairly large region and weighting the computed skew coefficient with a
generalized skew coefficient. (Chapter 9 provides guidelines for determining generalized
skew coefficients.)

b. Weighting Equation. Bulletin 17B recommends the following weighting equation:

MSEG(G) + MSEG(G)
G\J = (3-6)
MSEG + MSEG
where:
G\1 = weighted skew coefficient
G = computed (station) skew
G = generalized skew
MSEG = mean-square error of generalized skew
MSEG = mean-square error of computed (station) skew

3-6
 EM 1110-2-1415
5 Mar 93

c. Mean Square Error.

(I) The mean-square error of the computed skew coefficient for log-Pearson type
III random variables has been obtained by sampling experiments. Equation 6 in Bulletin
17B provides an approximate value for the mean-square error of the computed (station)
skew coefficient:

MSE ""' 10<A·B[log10(N/10)]} (3-7a)

G

(3-7b)

A = -0.33 + 0.08 IGI if IGI ~ 0.90

= -0.52 + 0.30 IGI if IGI > 0.90

B = 0.94 - 0.26 IGI if IGI ~ .50

= 0.55 if IGI > 1.50

where:

IGI = absolute value of the computed skew

N = record length in years

Appendix F-1 0 provides a table of mean-square error for several record lengths and skew
coefficients based on Equation 3-7a.

(2) The mean-square error (MSE) for the generalized skew will be dependent on the
accuracy of the method used to develop generalized skew relations. For an isoline map,
the MSE would be the average of the squared differences between the computed (station)
skew coefficients and the isoline values. For a prediction equation, the square of the
standard error of estimate would approximate the MSE. And, if an arithmetic mean of
the stations in a region were adopted, the square of the standard deviation (variance)
would approximate the MSE.

3-4. Expected Probability.

a. The computation of a frequency curve by the use of the sample statistics, as an

estimate of the distribution parameters, provides an estimate of the true frequency curve.
(Chapter 8 discusses the reliability and the distribution of the computed statistics.) The
fact of not knowing the location of the true frequency curve is termed uncertainty. For
the normal distribution, the sampling errors for the mean are defined by the t distribution
and the sampling errors for the variance are defined by the chi-squared distribution.
These two error distributions are combined in the formation of the non-central t
distribution. The non-central t-distribution can be used to construct curves that, with a
specified confidence (probability), encompass the true frequency curve. Figure 3-2 shows

3-7
EM 1110-2-1415
5 Mar 93

the confidence limit curves around a frequency curve that has the following assumed
statistics: N= 10, X•O., S= 1.0.

b. If one wished to design a flood protection work that would be exceeded, on the
average, only one time every 100 years (one percent chance exceedance), the usual design
would be based on the normal standard deviate of 2.326. Notice that there is a 0.5 percent
chance that this design level may come from a "true" curve that would average 22
exceedances per 100 years. On the other side of the curve, instead of the expected one
exceedance, there is a 99.5 percent chance that the "true" curve would indicate 0.004
exceedances. Note the large number of exceedances possible on the left side of the curve.
This relationship is highly skewed towards the large exceedances because the bound on the
right side is zero exceedance. A graph of the number of possible "true" exceedances
versus the probability that the true curve exceeds this value, Figure 3-3, provides a curve
with an area equal to the average (expected) number of exceedances.

c. The design of many projects with a target of 1 exceedance per 100 years at each
project and assuming N= I 0 for each project, would actually result in an average of 2.69
exceedances (see Apper. . ix F-8).

NORMAL DISTRIBUTION, SAMPLE SIZE•10

4 .50
....
~
u
z
UJ

-
c
1-
< ~
u
.75 :
>
w ~
c
l .90
c !
a:
< .950 :
c .975
z
< .990
1-
en .995

91.19 99.9 99 90 50 10 0 1 .01

PERCENT CHANCE EXCEEDANCE

Figure 3-2. Confidence Limit Curves based on the Non-central t Distribution.

3-8
 EM 1110-2-1415
5 Mar 93
Z5
I I I I
TARGET Of 1 EXCEEDA~CE PER 100 EVE~TS
- AND IASEC ON SAMPlE SIZE OF 10 EVENTS

zo
en
cr
...:
w
>-

-
0
0 15
-

\
cr
w
a..
en
w
(.,)
z
c 10
c
w
w
(.,)
><
w
\ AREA UNDER CURVE IS EXPECTED NUMBER
OF EXCEEDANCES PER 100 EVENTS

5
~-....... 2.69 (Appendix V·8l

---
-
- ~
r--..,
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
PROBABILITY OF EXCEEDING
Figure 3-3. Cumulative Probability Distribution of Exceedances per 100 Years.

d. There are two methods that can be used to correct (expected probability
adjustment) for this bias. The first method, as described above, entails plotting the curve
at the "expected" number of exceedances rather that at the target value, drawing the new
curve and then reading the adjusted design level. Appendix F-8 provides the percentages
for the expected probability adjustment.

e. The second method is more direct because an adjusted deviate (K value) is used in
Equation 3-4 that makes the expected probability adjustment for a given percent chance
exceedance. Appendix F-7 contains the deviates for the expected probability adjustment.
These values may be derived from the t-distribution by the following equation:

KP,N = t [(N+l)/N]YI (3-8)

P,N·1

where:

P = exceedance probability (percent chance exceedance divided by 100)

N = sample size

K expected probability adjusted deviate

= Student's t-statistic from one-tailed distribution

3-9
EM 1110-2-1415
5 Mar 93

f. For a sample size of 10 and a 1% percent chance exceedance, the expected

probability adjusted deviate is 2.959 as compared to the value of 2.326 used to derive the
computed frequency curve.
g. As mentioned in the first paragraph, the non-central t distribution, and
consequently the expected probability adjustment, is based on the normal distribution.
The expected probability adjustment values in Appendices F-7 and F-8 are considered
applicable to Pearson type III distributions with small skew coefficients. The phrase
"small skew coefficients" is usually interpreted as being between -0.5 to +0.5. Note also
that the uncertainty in the skew coefficient is not considered. In other words, the skew
coefficient is treated as if it were the population skew coefficient.
h. The expected probability adjustment can be applied to frequency curves derived
by other than analytical procedures if the equivalent worth (in years) of the procedure can
be computed or estimated.
3-5. :flli.k.
a. Definition. The term risk is usually defined as the possibility of suffering loss or
injury. In a hydrologic context, risk is defined as "the probability that one or more events
will exceed a given flood magnitude within a specified period of years" (46). Note that
this narrower definition includes a time specification and assumes that the annual
exceedance frequency is exactly known. Uncertainty is .D.Q1 taken into account in this
definition of risk. Risk then enables a probabilistic statement to be made about the
chances of a particular location being flooded when it is occupied for a specified number
of consecutive years. The percent chance of the location being flooded in any given year
is assumed to be known.
b. Binomjal Distribution. The computation of risk is accomplished by the equation
for the binomial distribution:

N! (3-9)
I!(N-I)!

where:
R1 • risk (probability) of experiencing exactly I flood events
N = number of years (trials)
I .. number of flood events (successes)
P • exceedance probability, percent chance exceedance divided by 100, of the
annual event (probability of success)
(The terms in parentheses are those usually used in statistical texts)

When I equals zero (no floods), Equation 3-9 reduces to:

3-10
 EM 1110-2-1415
5 Mar 93

and the probability of experiencing one or more floods is easily computed by taking the
complement of the probability of no floods:

D
n(1 or more)
,. ,I -,1 ,I _ 'P, 1\N

c. Aoplication.

( 1) Risk is an important concept to convey to those who are or will be protected by

flood control works. The knowledge of risk alerts those occupying the flood plain to the
fact that even with the protection works, there could be a significant probability of being
flooded during their lifetime. As an example, if one were to build a new house with the
ground floor at the 1% chance flood level, there is a fair (about one in four) chance that
the house will be flooded before the payments are completed, over the 30-year mortgage
life. Using Equation 3-1 Ob:

R<1 or more) "" 1-(1-.01)30

= 1-.9930

= 1-.74
.. .26 or 26% chance

(2) Appendix F-12 provides a table for risk as a function of percent chance
exceedance, period length and number of exceedances. This table could also be used to
check the validity of a derived frequency curve. As an example, if a frequency curve is
determined such that 3 observed events have exceeded the derived 1% chance exceedance
level during the 50 years of record, then there would be reason to question the derived
frequency curve. From Appendix F-12, the probability of this occurring is 0.0122 or
about 1%. It is possible for the situation to occur, but the probability of occurring is very
low. This computation just raises questions about the validity of the derived curve and
indicates that other validation checks may be warranted before adopting the derived
curve.
3-6. Conditional Probability Adjustment. The conditional probability adjustment is made
when flood peaks have either been deleted or are not available below a specified
truncation level. This adjustment will be applied when there are zero flood years, an
incomplete record or low outliers. As stated in Appendix 5 of Bulletin 17B, this
procedure is not appropriate when 25 percent or more of the events are truncated. The
computation steps in the conditional probability adjustment are as follows:
1. Compute the estimated probability (P) that an annual peak will exceed the
truncation level:

P = N/n (3-lla)

3-11
EM 1110-21415
5 Mar 93

where N is the number of peaks above the truncation level and n is the total number
of years of record. If the statistics reflect the adjustments for historic information,
then the appropriate equation is

H- WL
p .. (3-11 b)
H

where H is the length of historic period, W is the systematic record weight and L is
the number of peaks truncated.
2. The computed frequency curve is actually a conditional frequency curve. Given
that the flow exceeds the truncation level, the exceedance frequency for that flow
can be estimated. The conditional exceedance frequencies are converted to annual
frequencies by the probability computed in Step 1:.

(3-12)

where P is the annual percent chance exceedance and Pd is the conditional percent
chance exceedance.
3. Interpolate either graphically or mathematically to obtain the discharge values
(QP) for I, 10 and 50 percent chance exceedances.

4. Estimate log-Pearson type III statistics that will fit the upper portion of the
adjusted curve with the following equations:

log (Q 1/Q 10 )
G5 • -2.50 + 3.12 - - - - - (3-13)

log (Q 1/Q 50 )
ss .. (3-14)

(3-15)

where G 5 , S5 and X are the synthetic skew coefficient, standard deviation and
mean, respectively; Q1, Q1D and Q50 the discharges determined in Step 3; and K 1 and
K 50 are the Pearson Type III deviates for percent change exceedances of I and 50
and skew coefficient G 5 •

3-12
 EM 1110-2-1415
5 Mar 93

5. Combine the synthetic skew coefficient with the generalized skew by use of
Equation 3-6 to obtain the weighted skew.
6. Develop the computed frequency curve with the synthetic statistics and compare
it with the plotted observed flood peaks.
3-7. Two-Station Comparison.

a. Purpose.

(I) In most cases of frequency studies of runoff or precipitation there are locations
in the region where records have been obtained over a long period. The additional period
of record at such a nearby station is useful for extending the record at a short record
station provided there is reasonable correlation between recorded values at the two
locations.

(2) It is possible, by regression or other techniques, to estimate from concurrent

records at nearby locations the magnitude of individual missing events at a station.
However, the use of regression analysis produces estimates with a smaller variance than
that exhibited by recorded data. While this may not be a serious problem if only one or
two events must be estimated to "fill in" or complete an otherwise unbroken record of
several years, it can be a significant problem if it becomes necessary to estimate more than
a few events. Consequently, in frequency studies, missing events should not be freely
estimated by regression analysis.

(3) The procedure for adjusting the statistics at a short-record station involves three
steps: (1) computing the degree of correlation between the two stations, (2) using the
computed degree of correlation and the statistics of the longer record station to compute
an adjusted set of statistics for the shorter-record station, and (3) computing an equivalent
"length of record" that approximately reflects the "worth" of the adjusted statistics of the
short-record station. The longer record station selected for the adjustment procedure
should be in a hydrologically similar area and, if possible, have a drainage area size
similar to that of the short-record station.

b. Computation of forrelation. The degree of correlation is reflected in the

correlation coefficient R as computed through use of the following equation:

(3-16)

where:

R2 = the determination coefficient

Y = the value at the short-record station

X = the concurrent value at the long-record station

N = the number of years of concurrent record

3-13
EM 1110-2 1415
5 Mar 93

For most studies involving streamflow values, it is appropriate to use the logarithms of the
values in the equations in this section.
c. Adjustment of Mean. The following equation is used to adjust the mean of a
short-record station on the basis of a nearby longer-record station:

(3-17)

where:
Y • the adjusted mean at the short-record station
Y 1 == the mean for the concurrent record at the short-record station
X 3 == the mean for the complete record at the longer-record station
X1 • the mean for the concurrent record at the longer-record station
R = the correlation coefficient
the standard deviation for the concurrent record at the short-record station
the standard deviation for the concurrent record at the longer-record
station

All of the above parameters may be derived from the logarithms of the data where
appropriate, e.g., for annual flood peaks. The criterion for determining if the variance of
the adjusted mean will likely be less than the variance of the concurrent record is:

R 2 > l/(N 1 - 2) (3-18)

where N 1 equals the number of years of concurrent record. If R 2 is less than the
criterion, Equation 3-17 should not be applied. In this case just use the computed mean at
the short-record station or check another nearby long-record station. See Appendix 7 of
Bulletin 17B for procedures to compare the variance of the adjusted mean against the
variance of the entire short-record period.
d. Adjustment of Standard Deviation. The following equation can be used to adjust
the standard deviation:

Sy2 = S 2 + (S 2_ S 2) R 2(S 2/S ~ (3-19)

Y1 X X1 Y1 X1

(approximate)

3-14
 EM 1110-2-1415
5 Mar 93

where:

Sy = the adjusted standard deviation at the short-record station

Sy
1
= the standard deviation for the period of concurrent record at the short-record
station

Sx = the standard deviation for the complete record at the base station

Sy
1
= the standard deviation for the period of concurrent record at the base station

R2 = the determination coefficient

All of the above parameters may be derived from the logarithms of the data where
appropriate, e.g., for annual flood peaks. This equation provides approximate results
compared to Equation 3- I 9 in Appendix 7 of Bulletin 17B, but in most cases the
difference in the results does not justify the additional computations.

e. Adjustment of Skew coefficient. There is no equation to adjust the skew

coefficient that is comparable to the above equations. When adjusting the statistics of
annual flood peaks either a weighted or a generalized skew coefficient may be used
depending on the record length.

f. Equivalent Record Length. The final step in adjusting the statistics is the
computation of the "equivalent record length" which is defined as the period of time
which would be required to establish unadjusted statistics that are as reliable (in a
statistical sense) as the adjusted values. Thus, the equivalent length of record is an
indirect indication of the reliability of the adjusted values of Y and Sy. The equivalent
record length for the adjusted mean is computed from the following equation:

Ny
Ny"" _ _ _ _ _ _ _ _ _._ _ _ _ _ _ _ __
(3-20)
2 2
I - [(Nx- Ny )/NxJ [R - (l - R )/(Ny - 3)]
1 1

where:

Ny • the equivalent length of record of the mean at the short-record station

N y1 = the number of years of concurrent record at the two stations

Nx = the number of years of record at the longer-record station

R = the adjusted correlation coefficient

Figure 3-4 shows the data and computations for a two-station comparison for a short
record station with 21 events and a long record station with 60 systematic events. It can
be seen that the adjustment of the frequency statistics provides an increased reliability in
the mean equivalent to having an additional 17 years of record at the short-record station.

3-15
EM 1110-2-1415
5 Mar 93

Year

River
flow

-1915- -----------
12600
River
--
Chattooge hllulell

-- -- r-~otal-~~~::~~~--=~~~::
Systeutic Events 60 21 39
STATISTICS OF DATA
long ~ecord

--------
Not
;r-- ort
ord

----
21
1917
1918
14000
5900 Historic Period 71 0 0
------0
1919 16000
1920 8200 log lleen 3.8866 3.9310 3.8650 3.4796
1921
1922
4100
6200 Standard Deviation 0.2322 0.2678 0.2140
---
0.2609
1923
1924
5300
9200 CCM~PUted 0.5015 0.1588 0. 7340
--
·0. 0811
1925 3900 Skew Generalized . - . .
1926 6200 Adopted . - . ·0.1000
1927 3600
1928 20100
1929 11400 CQMPUTATIONS FOR TWO-STATION COMPARISON:
1940 29000•
1941 7530 Slope: b • 0.790495
1942 6870 Correlation Coefficient: R •0.811351 CSee Figure 9·01>
1943 6870
1944 3840 Fr~ equations 19, 21, and 22:
1945 2930
1946 6650 '1 • 3.4796 • 0.790495 (3.8866 • 3.9310)
1947 6440 ,, • 3.4445
1948 12400
1949 13900
1950 4740 s,2 • ((0.2322)2 - (0.2678) 2 Jc0.8114) 2 c0.2609/0.2678> 2
1951 5220 .. (0.2609)2
1952 13400
1953 ~020 s, • 0.23~
1954 6230
1955 5820 1 --~~-71"11'1'""
II --.....---oj,---.:2..:. • 38
1956 5820 ,, • - !!.;_1 (0.6583 - - 0.6583,
1957 5820 I I 21 " l
1958 5620
1959 5620
1960 6440 ii • o.o IISEG • 0.302 G • ·0.0811 MSEG • 0.142
1961 7310
1962 9660 G (0.302)(·0.0811) + (0.0142)(0.0)
1963 5420 w• o.3o2. o.142
1964 9880
1965 2noo 7440 Gw. -0.055 -~ -0.1
1966 13400 5140
1967 15400 2800
1968 5620 3100 FREQUEIICY CURVE, TALLULAH RIVER NEAR CLAYTON, GA
1969 14700 2470
1970 3480 2010 •••••• FLOW,CFS •••••••• PERCE liT ..COIIFIDENCE LIMITS ••
1971 3290 976 EXPECTED CHANCE
19n 7440 2160 COMPUTED PIIIOIAIILITY EXCEEDANCE .05 LIMIT .95 LIMIT
1973 19600 8500
1974 6400 4660 12700 14200 .2 18800 9580
1975 6340 2410 10900 11900 .5 15600 8410
1976 18500 6530 9590 10300 1.0 13400 7540
1977 13000 3580 8350 8800 2.0 11300 6680
1978 7850 4090 6760 6990 5.0 8810 5550
1979 14800 6240 5590 5710 10.0 7040 4680
1980 10900 2880 4430 4480 20.0 5370 3780
1981 4120 1600 2810 2810 50.0 3260 2420
1982 5000 1960 1760 1740 80.o 2060 1450
1983 7910 3260 1370 1340 90.0 1640 1080
1984 4810 2000 1110 1070 95.0 1360 847
1985 4740 1010 745 686 99.0 957 524
• Histor-Ic inforution, peak Iargen since 1915.

Figure 3-4. Two-Station Comparison Computations.

3-16
 EM 1110-2-1415
5 Mar 93
(Figure 9-1 shows the computations for b and R and Figure 9-2 shows the Tallulah River
annual peaks plotted against the Chattooga River peaks.) Figure 3-5 shows the resulting
unadjusted and adjusted frequency curves based on the computed and adjusted statistics
in Figure 3-4. Although N is actually the equivalent years of record for the mean, the
value is used as an estimateyequivalent record length in the computation of confidence
limits and the expected probability adjustment.

g. Summary of Steps. The procedure for computing and adjusting frequency

statistics using a longer-record station can be summarized as follows:

(I) Arrange the streamflow data by pairs in order of chronological sequence.

(2) Compute Y 1 and Sy for the entire record at the short-record station.
1
(3) Compute X and Sx for the entire record at the longer-record station.

(4) Compute X 1 and Sx for the portion of the longer-record station which is
concurrent with the shott-record station.

(5) Compute the correlation coefficient using Equation 3-16.

(6) Com put£' Y and Sy for the short-record station using Equations 3-17 and 3-18.

T•llul.n Riu.r n.er Cl•~ton, QA

- --- Expected Probability Curve from Data

-Expected Probability Curve from Two-Station Compariaon

.,
.,
...•u 184
., l.,...- ~

3' I" ~

...IL0 .,.. --~

.;t ~
••
Q. ~ ~

... ~
fi""
•:::Jc 18 3 ~
c
G:
,

99.99 99.9 99 1 .1 .81

Figure 3-5. Observed and Two-Station Comparison Frequency Curves.

3-17
EM 1110-2-1415
5 Mar 93

(7) Calculate the equivalent length of record of the mean for the short-record station
using Equation 3-20.
(8) Compute the frequency curve using adjusted values of Y and S in Equation 3-4
and K values from Appendix F-2 corresponding to the adopted skew coefficient.

(9) Compute the expected probability adjustment and the confidence limits.

3-8. Flood Volumes.

a. Nature and Purpose. Flood volume frequency studies involve frequency analysis
of maximum runoff within each of a set of specified durations. Flood volume-duration
data normally obtained from the USGS WATSTORE files consists of data for I, 3, 7, 15,
30, 60, 90, 120, and 183 days. These same values are the default values in the HEC
computer program ST ATS (Table 3-2). Runoff volumes are expressed as average flows in
order that peak flows and volumes can be readily compared and coordinated. Whenever it
is necessary to consider flows separately for a portion of the water year such as the rain
season or snowmelt season, the same durations (up to the 30-day or 90-day values) are
selected from flows during that season only. Flood volume-duration curves are used
primarily for reservoir design and operation studies, and should generally be developed in
the design of reservoirs having flood control as a major function.

Tabl~ 3-2, High FIQw Vglym~-.JlyratiQn Jlata

- VOLUME-DURATION DATA - FISHKILL CR AT BEACON, NY - DAILY FLOWS
-----------------------------------------------------------------------------------------------
HIGHEST MEAN VALUE FOR DURATION, FLOW,CFS
YEAR 1 3 7 15 30 60 90 120 183
-----------------------------------------------------------------------------------------------
19~5 2080.0 1936.7 1714.3 1398.7 1106.8 752.3 742.2 649.4 559.2
1946 1360.0 1180.3 923.0 837.3 657.8 605.3 476.2 451.5 379.9
191o7 1800.0 1616.7 1159.1 820.5 687.1 611.9 5S8.5 485.8 396.4
191o8 2660.0 21o30.0 2322.9 1641.7 1llo5 .1 862.0 706.2 638.1 512.7
1949 2900.0 2346.7 1715.7 1358.9 888.9 680.7 586.8 522.4 422.4
1950 1050.0 909.7 746.9 639.7 588.1 455.9 423.0 387.2 335.1
1951 2160.0 1886.7 1741o.3 1248.1 872.9 832.1 781.2 689.8 568.9
1952 2870.0 2266.7 1557.6 1186.5 1032.8 925.1 854.1 732.6 692.9
1953 2850.0 2233.3 1644.3 1317.2 1145.5 994.6 831.1 794.4 654.5
1954 1520.0 1096.7 811.7 620.4 482.9 397.0 405.7 372.7 348.1
1955 6970.0 4536.7 2546.1 1360.0 759.2 608.0 494.0 463.1 478.7
1956 6760.0 5456.7 3354.3 1959.7 1572.8 1080.9 767.7 635.8 641.7
1957 1230.0 1117.3 1037.7 758.9 524.2 408.8 363.3 373.4 324.4
1958 2130.0 1916.7 1587.1 1354.5 1128.1 972.0 848.2 777.8 654.1
1959 1670.0 986.7 782.1 586.6 517.6 486.7 437.5 398.8 346.2
1960 2080.0 1770.0 1374.3 1046.9 712.3 605.5 530.5 515.1 468.4
1961 3440.0 2966.7 2155.7 1590.2 1152.3 845.2 759.5 656.2 491.4
1962 2570.0 2070.0 15~7.7 1105.0 857.7 600.9 461.3 429.4 325.0
1963 1730.0 1616.7 1309.0 1216.0 900.8 569.1 438.0 370.8 305.9
1961o 1300.0 1106.7 945.3 737.8 541.2 514.8 486.6 450.1 368.3
1965 900.0 826.3 652.6 455.7 375.8 303.3 275.7 235.0 175.0
1966 930.0 774.7 693.3 546.5 445.7 352.5 296.2 272.5 209.0
1967 1520.0 1416.7 1247.1 1023.5 906.8 701.3 581.4 521.1 436.8
1968 3500.0 2810.0 1934.3 1328.5 878.7 611.7 609.5 567.3 460.3
-----------------------------------------------------------------------------------------------
Note - Data based on water year of October I of preceeding year through September 30
of given year.

3-18
 EM 1110-2-1415
5 Mar 93

b. Data for Comprehensive Series. Data to be used for a comprehensive flood

volume-duration frequency study should be selected from nearly complete water year
records. Unless overriding reasons exist, the durations in Table 3-2 should be used in
order to assure consistency among various studies for comparison purposes. Maximum
flood events should be selected only for those years when recorder gages existed or when
the maximum events can be estimated by other means. Where a minor portion of a water
year's record is missing, the longer-duration flood volumes for that year can often be
estimated adequately. If upstream regulation or diversion is known to have an effect, care
should be exercised to assure that the period selected is the one when flows would have
been maximum under the specified (usually natural) conditions.

c. Statistics for Comprehensive Series.

( 1) The probability distribution recommended for flood volume-duration frequency

computations is the log-Pearson type III distribution; the same as that used for annual
flood peaks. In practice, only the first two moments, mean and standard deviations are
based on station data. As discussed in Section 3-3, the skew coefficient should not be
based solely on the station record, but should be weighted with information from regional
studies. To insure that the frequency curves for each duration are consistent, and
especially to prevent the curves from crossing, it is desirable to coordinate the variation in
standard deviation and skew with that of the mean. This can be done graphically as
shown in Figure ~-6. For a given skew coefficient, there is a maximum and minimum
allowable slope for the standard deviation-versus-mean relation which prevents the curves
from crossing within the established limits. For instance, to keep the curves from crossing
within 99.99 and .01 percent chance exceedances with a skew of 0., the slope must not
exceed .269, nor be less than -.269, respectively. The value of this slope constraint is
found by stating that the value of one curve (XA for curve A) must equal or exceed the
value for a second curve (X 8 for curve B) at the desired exceedance frequency. Each of
these values can be found by substitution into Equation 3-4 (the K for zero skew and
99.99 percent chance exceedance is -3.719):

XA > Xa
XA + (-3.719) SA > X 8 + (-3.719) S8

3.719 (SB- SA) > XB- XA

(SB - SA)
> 0.269
(XB- XA)
where:
XA == Value of frequency curve- A at 99.99 percent chance exceedance
X 8 == Value of frequency curve Bat 99.99 percent chance exceedance
XA = Mean of frequency curve A
X8 = Mean of frequency curve B
SA = Standard deviation of frequency curve A
S8 = Standard deviation of frequency curve B

3-19
EM 1110-2-1415
5 Mar 93

0.30

~
> It
<

~
0.26 . - - - - - - - - - 4 -
i ..,~-~.-
"'g-
-l I
0
c:
.2 ,..
s> 0.20 <

t! ~
C)
<')
"'0

jIll
(i.j 0.15

0.10
2.4 2.8 2.8 3.0 3.2 3.4

1.0

•
/
s-
Ill
t:ll
0.5

v
- /.
0
....J
0

v
1 0.0
~
8
0
3:
G)
l}lj / •
~.5

v
-1.0
/
2.4 2.8 2.8 3.0 3.2 3.-4

Mean of Logartthns (X)

Figure 3-6. Coordination of Flood- Volume Statistics.

3-20
 EM 1110-2-1415
5 Mar 93

(2) When the skew changes between durations, it is probably easiest to adopt
smoothed relations for the standard deviation and skew and input the statistics into a
computer program that computes the ordinates. The curves can then be inspected for
consistency.
(3) If the statistics for the peak flows have been computed according to the
procedures in Bulletin 178, the smoothing relations should be forced through those points.
The procedure for computing a least-squares line through a given intersection can be
found in texts describing regression analyses.
d. Frequency Curves for Comorehensive Series.
( 1) General Procedure. Frequency curves of flood volumes are computed
analytically using general principles and methods of Chapters 2 and 3. They should also
be shown graphically and compared with the data on which they are based. This is a
general check on the analytic work and will ordinarily reveal any inconsistency in data
and methodology. The computed frequency curves and the observed data should be
plotted on a single sheet for comparison purposes, Figure 3-7.

(f)
~
u PEAK
1-0AY
.z.
-""
0
1-

a:
10~ 3-DAY

7-DAY
:::>
0 15-DAY
a:
0 30-0AY
~

3: 60-0AY
0
-1 10 3
u...
z
""
UJ
::E

99 90 50 10 •1 . 01
PERCENT CHANCE EXCEEDANCE

Figure 3-7. Flood-Volume Frequency Curves.

3-21
EM 1110-2-1415
5 Mar 93

(2) Interpolation Between Fixed Durations. The runoff volume for any specified
frequency can be determined for any duration between 1-day and 365-days by drawing a
curve on logarithmic paper relating mean discharge (or volume) to duration for that
specified frequency (see Figure 3-8a). When runoff volumes for durations shorter than 24
hours are important, special frequency studies should be made. These could be done in
the same manner as for the longer durations, using skew coefficients interpolated in some
reasonable manner between those used for peak and 1-day flows.
e. Applications of Flood Volume-Duration Frequencies.
(I) Volume-duration Curves. The use of flood volume-duration frequencies in
solving reservoir planning, design, and operation problems usually involves the
construction of volume-duration curves for specified frequencies. These are drawn first
on logarithmic paper for interpolation purposes, as illustrated on Figure 3-8a. The mean
discharge values are multiplied by appropriate durations to obtain volumes and are then
replotted on an arithmetic grid as shown on the Figure 3-8b. A straight line on this grid
represents a constant rate of flow. The straight line represents a uniform flow of 1,500
cfs, and the maximum departure from the 2% chance exceedance curve demonstrates that
a reservoir capacity of 16,000 cfs-days (31,700 acre-feet) is required to control the
indicated runoff volumes by a constant release of 1,500 cfs. The curve also indicates that
a duration of about 8 days is critical for this project release rate and associated
flood-control storage space.
(2) Apolication to a Sjngle Reservoir. In the case of a single flood-control reservoir
located immediately upstream of a single damage center. the volume frequency problems
are relatively simple. A series of volume-duration curves, similar to that shown on Figure
3-8, corresponding to selected exceedance frequencies should first be drawn. The project
release rate should be determined, giving due consideration to possible channel
deterioration, encroachment into the flood plain, and operational contingencies. This
procedure can be used not only as an approximate aid in selecting a reservoir capacity, but
also as an aid in drawing filling-frequency curves.
(3) Application to a Reservoir System. In solving complex reservoir problems,
representative hydrographs at all locations can be patterned after one or more past floods.
The ordinates of these hydrographs can be adjusted so that their volumes for the critical
durations will equal corresponding magnitudes at each location for the selected frequency.
A design or operation scheme based on regulation of such a set of hydrographs would be
reasonably well balanced. Some aspects of this problem are described in Section 3-9g.
3-9. Effects of Flood Control Works on Flood Frequencies.
a. Nature of the Problem. Flood control reservoirs are designed to substantially
affect the frequency of flood flows (or flood stages) at various downstream locations.
Many land use changes such as urbanization, forest clearing, etc. can also have significant
effects on downstream flood flows (see Section 3-1 0). Channel improvements (intended to
reduce stages) and levee improvements (intended to confine flows) at specified locations
can substantially affect downstream flows by eliminating some of the natural storage
effects. Levees can also create backwater conditions that affect river stages for a
considerable distance upstream. The degree to which flows and stages are modified by
various flood control works or land use changes can depend on the timing, areal
distribution and magnitude of rainfall (and snowmelt, if pertinent) causing the flood.
Accordingly, the studies should include evaluations of the effects on representative flood

3-22
 EM 1110-2-1415
5 Mar 93

FISHKILL CREEK AT BEACON NY 1945 1968 -

._,._ --+ ·- +-j -;- '--
I

___
j_

~ ~+ 1-
-. l -- _j
, --r- r--.!'t-- I i
I i I
:
r- 1-r-- ~--~ I
..."'u I I

~~
II
~ ~ZI
I
,;
. i
0
...
...J sol
t-- --- - t "·~ j--0--

......
z
:E I

I
I I I
!
I I I

I l 10
i
I
ii
DURATION. DAYS

Figure 3-Sa. Flood- Volume Frequency Relations.

10000 FISHKILL CREEK AT BEACON NY 1945 1968 -

/
v
/
// \) .. i.

,."' / ~
.c.t.
60000
/ ,.,t./ i-1-'i.~

"'>-...
0 ~
l,4v"
v. .,.
I

...u
"'
'
/

•oooo r:.<:)
..; //
;F
:E
::l •4...";;
...J
0
>
/
. . . ;.,..;t-
//: .. fc,."","' "-"'
STDIAGE IEOUIIEMEMT COMPUTATION
21000 1200 • 16000 CfS·DATS_

1/)-
20000
' ' 01 31700 AC · F T
'

0 '
10 zo ]0 50 60
DURATION. DAYS

Figure 3-Sb. Storage Requirement Determination.

3-23
EM 1110-2-1415
5 Mar 93

events. with careful consideration given to the effects of different temporal and areal
distributions.
b. Terminology.
(I) Natural Conditions. Natural conditions in the drainage basin are defined as
hydrologic conditions that would prevail if no regulatory works or other works of man
were constructed. Natural conditions, however, include the effects of natural lakes,
swamp areas, etc.
(2) Present Conditions. Present or base conditions are defined as the conditions that
exist as of the date of the study or some specified time.
(3) Without- Project Conditions. Without-project conditions are defined as the
conditions that would exist without the projects under consideration, but with all existing
projects and may include future projects whose construction is imminent.
(4) With-Project Conditions. With-project conditions are defined as the conditions
that will exist after the projects under consideration are completed.
c. Reservoir-Level Frequency Computation.
(I) Factors to be Considered. Factors affecting the frequency of reservoir levels
include historical inflow rates and anticipated future inflow rates estimated by
volume-frequency studies, the storage-elevation curves, and the plan of reservoir
regulation including location and size of reservoir outlets and spillway. A true frequency
curve of annual maxima or minima can only be computed when the reservoir completely
fills every year. Otherwise, the events would not be independent. If there is dependence
between annual events, the ordinate should be labeled "percent of years exceeded" for
maximum events and "percent of years not exceeded" for minimum events.
(2) Comoutation and Presentation of Results. A frequency curve of annual
maximum reservoir elevations (or stages) is ordinarily constructed graphically. using
procedures outlined in Section 2-4. Observed elevations (or stages) are used to the extent
that these are available, if the reservoir operation will remain the same in the future.
Historical and/or large hypothetical floods may also be routed through the reservoir using
future operating plans. A typical frequency curve is illustrated on Figure 6-4.
Elevation-duration curves are constructed from historical operation data or from routings
of historical runoff in accordance with procedures discussed in Section 2-2, Figure 3-9.
Such curves may be constructed for the entire period of record or for a selected wet
period or dry period. For many purposes, particularly recreation uses, the seasonal
variation of reservoir elevation (stages) is important. In this case a set of frequency or
duration curves for each month of the year may be valuable. One format for presenting
this information is illustrated on Figure 3-10.

3-24
 EM 1110-2-1415
5 Mar 93

TOP OF liiDIJCII)
SUIQIA&CE

TOP OF FL.OOD
CONTROl. POOL
------
'

1030
z0
...
....c
....>-1 ' ___ _ j _____ ..
1020 ; --- --- i

.... 1

I
I
TOP OF VATU
SUPPLY POOl.

......... .. 10
PERCENT OF TIME EXCEEDED
10 10

Figure 3-9. Daily Reservoir Elevation-Duration Curve.

CLARKS I-LL LAKE. 1962-1985

837
/ i"-.
338
)v
a3l5

LL ,, "" 1\

- "' "-
334

333
............Vj' \,
....... Percent of

v/ ~ ~
~ .... Daya Exoeedlld

"'
t: ~

e ~~
~ c:--~
~ ..,....., ~ A
a 0.1'1.

-r--
331 + 1.0'4
c(
~ / -...::::: ...........
~
~ rj/
>
,.,.,. --- --
w 330
...........
al ~ ...;
~~
328
v ~ ......._-.......,

""
0 5.0'4
321 ........... 1:>. 10.0'1.
~ ~
~
3'L7 ~ X 20.0'1.

328 ~
"' v 50.0'1.

Figure 3-10. Seasonal Variation of Elevation-Duration Relations.

3-25
 EM 1110-2-1415
5 Mar 93

d. Effects of Reservoirs on Flows at Downstream Points.

( 1) Routing for Period of Record. The frequency of reservoir outflows or of flows
at a downstream location can be obtained from routings of the period-of -record runoff by
the following methods:
(a) Determine the annual maximum flow at each location of interest and construct a
frequency curve of the regulated flows by graphical techniques (Section 2-4).
(b) Construct a graph of with-project versus without-project flows at the location of
interest and draw a curve relating the two quantities as illustrated on Figure 3-11. The
points should be balanced in the direction transverse to the curve, but factors such as
flood volume of the events and reliability of regulation must be considered in drawing the
curve. This curve can be used in conjunction with a frequency curve of without-project
flows to construct a frequency curve of with-project flows as illustrated on Figure 3-12.
This latter procedure assures consistency in the analysis and gives a graphical presentation
of the variability of the regulated events for a given unregulated flow.
(2) Use of Hyoothetical- Flood Routings. Usually recorded values of flows are not
large enough to define the upper end of the regulated frequency curve. In such cases, it is
usually possible to use one or more large hypothetical floods (whose frequency can be
estimated from the frequency curve of unregulated flows) to establish the corresponding
magnitude of regulated flows. These floods can be multiples of the largest observed
floods or of floods computed from rainfall; but it is best not to multiply any one flood by
a factor greater than two or three. The floods are best selected or adjusted to represent
about equal severity in terms of runoff frequency of peak and volumes for various
durations. The routings should be made under reasonably conservative assumptions as to
initial reservoir stages.

(3) Incidental Control by Water Supoly Space. In constructing fre :-mency curves of
regulated flows, it must be recognized that reservoir operation for purr es other than
flood control will frequently provide incidental regulation of floods. H-wever, the
availability of such space cannot usually be depended upon, and its value is considerably
diminished for this reason. Consequently, the effects of such space on the reduction of
floods should be estimated very conservatively.
(4) Allowance for Operational Contingencies. In constructing frequency curves of
regulated flows, it should be recognized that actual operation is rarely perfect and that
releases will frequently be curtailed or diminished because of unforeseen operation
contingencies. Also, where flood forecasts are involved in the reservoir operation, it must
be recognized that these are subject to considerable uncertainty and that some allowance
for uncertainty will be made during operation. In accounting for these factors, it will be
found that the actual control of floods is somewhat less than could be expected if full
release capacities and downstream channel capacities were utilized efficiently and if all
forecasts were exact.
(5) Runoff from Unregulated Areas. In estimating the frequency of runoff at a
location that is a considerable distance downstream from one or more reservoir projects, it
must be recognized that none of the runoff from the intermediate areas between the
reservoir(s) and the damage center will be regulated. This factor can be accounted for by
constructing a frequency curve of the runoff from the intermediate area, and using this
curve as an indicator of the lower limit for the curve of regulated flows. Streamflow

3-26
 EM 1110-2-1415
5 Mar 93

c Pe&k Flow Even~•

Derived Relationship /

- - - -- Equal Flow Line .. '

/

~ •u "
...-"Spillway Duign Flood
116
,-
......
0
Standard Project Flood
/
.X.

•,•
.-[7
a. . ~
.•
...• u6 ""
"
.. "'"

~
&
•
a:
" ~~
" all ~"'""
. ap
" c
"
" ""
u"'

Figure 3-11. Example With-Project versus Without-Project Peak Flow Relations.

(Only ~op 20 even~• ahown) ..a 1

+-4-+-~~-4---~~~~-+,v~~~~
~•
u
aP-e" lj
j
......
0

.X. 1.16
•
a.•
...
•c
J
c
~

88.98 98.8 89 81 lSI 11 1 .1 .11

Figure 3-12. Example Without-Project and With-Project Frequency Curves.

3-27
 EM 1110-2-1415
5 Mar 93

routing and combining of both the flows from the unregulated area and those from the
regulated area is the best procedure for deriving the regulated frequency curve.
e. Effects of Channel. Levee and Floodway Improvements. The effect of channel,
levee and floodway improvements on river stages at the project location and on river
discharges downstream from the project location can generaJiy be evaluated by routing
several typical floods through the reaches of the improvement and the upstream reaches
affected by backwater. The stages or discharges thus derived can be plotted against
corresponding without- project values, and a smooth curve drawn. This curve could be
used in conjunction with a frequency curve of without-project values to construct a
frequency curve of with-project values as discussed in Paragraph 3-09d( 1)b.
Corresponding stages upstream from the selected control point can be estimated from
water-surface profile computations.

f. Changes in Stage-Discharge Relationships. Changes in stage-discharge relations

due to channel improvements, levee construction or flow obstructions can best be
evaluated by computing theoretical water surface profiles for each of a number of
discharges. The resulting relationships for modified conditions can be used to modify
routing criteria to enable evaluation of the downstream effects of these changes.
g. Effects of Multiple Reservoir Systems.
(I) Representative events. When more than one reservoir exists above a damage
center, the problem of evaluating reservoir stages and downstream flows under project
conditions becomes increasingly complex. Whenever practicable, it is best to make
complete routings of five to ten historic flood events and a large event that has been
developed from a hypothetical rainfall pattern. If necessary, it is possible to supplement
these events by using multiples of the flow values. Care muse be exercised in selecting
events that have representative flood volumes, timings, and areal distributions. Also,
there should be a balance of events caused by particular climatic factors, i.e. snowmelt,
tropical storm, thunderstorm, etc. Furthermore, the flood-volume-duration characteristics
of the hypothetical events should be similar to the recorded events (see Section 3-8).
Hypothetical events must be used with caution, however, because certain characteristics of
atypical floods may be responsible for critical flooding conditions. Accordingly, such
studies should be supplemented by a critical examination of the potential effects of atypical
floods.

(2) Computer Program. It is generaJiy impossible to make all of the flood routings
necessary to evaluate the effect of a reservoir system by hand computations. Computer
programs have been developed to route floods through a reservoir system with complex
operational criteria (55).

3-10. Effects of Urbanization.

a. General Effects. Urbanization has tw9 major effects on the watershed which
influence the runoff characteristics. First, there is a substantial increase in the impervious
area, which results in more water entering the stream system as direct runoff. Second, the
drainage system coJiecting the runoff is generally more efficient and tends to concentrate
the water faster in the downstream portion of the channel system. It is important to keep
these two effects in mind when considering the changes in the flood peak frequency curve
caused by increasing urbanization.
3-28
 EM 1110-2-1415
5 Mar 93

b. Effect on Frequency Relations. A general statement can be made about the

effects of urbanization on flood-peak frequency relations. The usual effect on the
frequency relation is to cause a significant increase in the magnitude of the more frequent
events, but a lesser increase in the less frequent events. This results in an increase in the
mean of the annual flood peaks. a decrease in the standard deviation and an unpredictable
effect on the skew coefficient (see Figure 3- 13 ). The resulting frequency relation may
not fit any of the standard theoretical distributions. Graphical techniques should be
applied if a good fit is not possible by an analytical distribution.

c. Other Considerations. The actual effect of urbanization at a specific location is

dependent on many factors. Some of the factors that must be considered are basin slope,
basin shape, previous land use and ground cover, number of depressional areas drained,
magnitude and nature of urban development and the typical flood source (snowmelt,
thunderstorm, hurricane, or frontal storm).lt is possible for urbanization to cause a
decrease in the flood peaks at a particular site. For instance, consider an area downstream
of two tributary areas of such size and shape that the large floods are caused by the
addition of the nearly coincident peaks from the two tributaries. Urbanization in one of
the tributary areas will likely cause the contribution from this area to arrive downstream
earlier. This change in the timing of the peaks would result in lower downstream peaks.
Of course. when both areas have become equally urbanized, the flood peaks may coincide
again. The construction of bridges or other encroachments can reduce the flood peak
downstream. but causes backwater flooding upstream.

~ •u ~~
183
..
3
~/ "'

~
....u..0
~
Ill
• ' 4
/

Q.
II
c
~ """
.... . ~f.·' ..,
II
J
c: 182 , ~·~·
c:
<I

99.99 99.9 99 90 50 10 1 .1 . 01

Figure 3-13. Typical Effect of Urbanization on Flood Frequency Curves.

3-29
 EM 1110-2-1415
5 Mar 93

d. Adjustment of a Series of Nonstationary Peak Discharges. When the annual peak

discharges have been recorded at the outlet of a basin which has been undergoing
progressive urbanization during the period of record, the peak discharges are
nonstationary because of the varying basin condition. It is generally necessary to adjust
the discharges to a stationary series representative of existing conditions. One approach
to adjusting the peaks to a stationary series is as follows:
(I) Develop and calibrate a rainfall-runoff model for existing basin conditions and
for conditions at several other points in time during the period of record.
(2) Develop a hypothetical storm for the basin using generalized rainfall criteria,
such as that contained in Weather Bureau Technical Paper 40 (14). Select the
magnitude of the storm, e.g., a 25-year recurrence interval, to be used. The
recurrence interval is arbitrary as it is not assumed in this approach that runoff
frequency is equal to rainfall frequency. The purpose of adopting a specific
magnitude is to establish a base storm to which ratios can be applied for subsequent
steps in the analysis.
(3) Apply several ratios (say 5 to 8) to the hypothetical storm developed in the
previous step such that the resulting calculated peak discharges at the gage will cover
the range desired for frequency analysis. Input the balanced storms to the
rainfall-runoff model for each of the basin conditions selected in step (I), and
determine peak discharges at the gaged location.
(4) From the results of step (3 ), plot curves representing peak discharge versus storm
ratio for each basin condition (or point in time).
(5) Use the curves developed in step (4) to adjust the observed annual peak
discharges. For example, an observed annual peak discharge that occurred in 1975 is
adjusted by entering the "1975" curve (or interpolating) with that discharge, locating
the frequency of that event, and reading the magnitude of the adjusted peak from
the base-condition curve for the same frequency. The adjusted peak thus obtained is
assumed to be the peak discharge that would have occurred for the catchment area
and development at the base condition. It is not necessary to adjust to natural
conditions. A stationary series could be developed for one or more points in time.
(6) A conventional frequency analysis can be performed on the adjusted peak
discharges determined in the preceding step. If the data represent natural conditions,
Bulletin 17B procedures would be applicable. If the basin conditions represent
significant urbanization, graphical analysis may be appropriate.
e. Development of Freauency Curves at Ungaged Sjtes. There are several
approaches that can be taken to develop frequency curves at ungaged sites that have been
subject to urbanization. In order of increasing difficulty, they are: 1) application of
simple transfer procedures (e.g., Q • CIA); 2) application of available region-specific
criteria, e.g., USGS regression equations; 3) application of rainfall-runoff models to
hypothetical storm events; 4) application of simple and detailed rainfall-runoff models
with observed storm events and 5) complete period-of -record simulation. As approaches
(3) and (4) are often applied, the computational steps are presented in some detail.

3-30
 EM 1110-2-1415
5 Mar 93

(I) Hypothetical Storm Approach

(a) Develop peak-discharge frequency curve for specific land use conditions
from available gaged data and/or regional relationships.
(b) Develop balanced storms of various frequencies using data from generalized
criteria, a nearby gage or the equivalent.

(c) Develop rainfall-runoff model for the specific watershed with the adopted
land-use conditions. Calibrate runoff and routing parameters by reproducing
observed hydrographs occurring under natural conditions.

(d) Input balanced storms (from b) to rainfall-runoff model (from c).

Determine exceedance probabilities to associate with balanced storms from
adopted specific land-use conditions peak discharge frequency curve (from a)
with computed peak discharges.

(e) Modify parameters of rainfall-runoff model to reflect future urban runoff

characteristics. Input balanced storms to the urban- conditions model.
(f) Plot results assuming frequency of each event is the same for both the
adopted land use and the future urban conditions.
{2) "Simple" and Detailed Simulation of Historic Events

(a) Simulate all major historic events with a relatively simple model to establish
the ranking of events and an approximate peak discharge for each. The
approximate peaks could be developed by using a multiple linear regression
approach, by using a very simple rainfall-runoff model, or by any other
approach that will capture the hydrologic response of the basin.

(b) Perform a conventional frequency analysis of the approximate peaks

obtained in step a.
(c) Make detailed simulations of selected events and correlate the more precise
peaks with the approximate peaks.

(d) Use the relationship developed in step c to determine the desired frequency
curve. The same approach can be followed for both existing and future
conditions.

3-31
 EM 1110-2-1415
5 Mar 93

CHAPTER 4
LOW- FLOW FREQUENCY ANALYSIS

4-1. ~. Low-flow frequency analyses are used to evaluate the ability of a stream to
meet specified flow requirements at a particular location. The analysis can provide an
indication of the adequacy of the natural flow to meet a given demand with a stated
probability of experiencing a shortage. Additional analyses can indicate the amount of
storage that would be required to meet a given demand, again with a stated probability of
being deficient. The design of hydroelectric power plants, determination of minimum
flow requirements for water quality and/or fish and wildlife, and design of water storage
projects can benefit from low-flow frequency analysis.
4-2. Interpretation.
a. Analytical frequency techniques are usually not applicable to low-flow data
because most theoretical frequency distributions cannot satisfactorily fit the recorded data.
It is recommended that graphical techniques be used and that known geologic and
hydrologic conditions be kept in mind when developing the relationships. As the low
values are the major interest, the data are arranged with the smallest value first. The
probability scale is usually labeled "percent chance nonexceedance."
b. Annual low flows are usually computed for several durations (in days) with the
flow rate expressed as the mean flow for the period. For example, the USGS WATSTORE
output provides the mean flow values for daily durations of I, 3, 7, 14, 30, 60, 90, 120
and 183 days. The default values for the HEC program STATS are the same with the
exception of using a 15-day duration instead of 14 (Table 4-1 ). Often a climatic year
from April 1 to March 31 is specified to provide a definite separation of the seasonal
low-flow periods. Figure 4-1 is a plot of the data in Table 4-1.
4-3. Aoolication Problems.
a. Basin Develooment. The effects of any basin developments on low flows are
usually quite significant. For example, a relatively moderate diversion can be neglected
when evaluating flood flow relations, but it would reduce, or even eliminate, low flows.
Accordingly, one of the most important aspects of low flows concerns the evaluation of
past and future effects of basin developments.
b. Multi- Year Events. In regions of water scarcity and where a high degree of
development has been attained, projects that entail carryover of water for several years
are often planned. In such projects it is desirable to analyze low-flow volume frequencies
for periods ranging from 1-1/2 to 8-1/2 years or more. Because the number of
independent low-flow periods of these lengths, in even the longest historical records, is
very small and because the concept of multi- annual periods is somewhat inconsistent with
the basic concept of an "annual event;" there is no truly satisfactory way for computing
the percent chance nonexceedance for low-flow periods that are more than 1 year in
length. One procedure described in reference (37) has been used with long sequences of
synthetically generated streamflows to derive estimates of drought frequency. Although

4-1
EM 1110-2 1415
5 Mar 93

Table 4-1. Low-Flow Volume-Duration Data.

- VOLUME-DURATION DATA - FISHKILL CR AI BEACON, NY - DAILY FLOWS
-----------------------------------------------------------------------------------------------
LOWEST MEAN VALUE FOR DURATION, FLOW,CFS
YEAR 1 3 7 15 30 60 90 120 183
-----------------------------------------------------------------------------------------------
1945 92.0 104.0 115.1 127.7 H3.0 179.5 220.7 254.1 305.3
1946 9.4 12.8 17.6 21.3 28.5 49.8 62.1 58.7 75.4
1947 9.4 12.8 17.3 19.0 21.2 32.1 41.0 62.0 137.0
1948 8.3 10.2 15.7 15.7 18.9 21.6 27.4 33.5 78.1
1949 7.1 8.2 9.0 9.1 10.0 11.3 12.3 14.3 21.2
1950 22.0 22.0 23.9 27.0 32.6 37.0 43.1 51.1 119.0
1951 20.0 33.3 40.9 45.5 58.4 73.2 84.0 88.7 116.4
1952 34.0 39.7 43.0 44.0 46.2 64.6 100.1 97.9 135.3
1953 4,4 4.8 4.9 7. 3 10.4 11.0 15.2 25.3 49.9
1954 8.4 9.5 12.3 14.6 16.7 22.9 39.8 99.8 160.8
1955 6.1 6.3 7.0 11.2 14.7 37.2 67.6 148.4 247.9
1956 19.0 21.3 23.0 26.8 29.5 53.5 59.5 71.3 98.2
1957 3.7 5.1 5.8 6.4 8.9 9.8 12.6 15.5 25.6
1958 12.0 13.3 17.4 19.2 23.8 29.6 41.5 50.9 118.5
1959 17.0 17.3 20.6 25.1 39.0 49.8 53.2 60.4 111.0
1960 48.0 48.3 53.3 63.9 77.5 122.1 136.3 149.1 213.9
1961 17.0 17.0 19.7 22.9 27.9 32.1 31.9 37.2 57.2
1962 5.9 6.6 7.0 7.8 9.9 15.0 16.7 20.6 41.9
1963 19.0 19.0 19.6 21.6 27.4 32.1 38.8 58.9 70.5
1964 1.1 1.4 1.8 2.5 4.2 7.1 9.0 11.5 19.1
1965 5. 7 6.7 9.9 11.7 12.1 13.7 15.8 20.6 26.1
1966 4.0 4.5 4.7 4.8 5.0 6.4 13.2 21.8 64.8
1967 43.0 44.0 49.3 58.1 62.0 92.4 122.4 147.2 188.6

lfot.a - Data baaed on Climatic Year of April 1 of &ivan year throuah March 31 of next year.

the results obtained through the use of this procedure seem reasonable, it is impossible to
verify the accuracy of the frequency estimates.
c. Regionalization. Regionalization of low-flow events is usually not very
successful. The variations in geologic conditions such as depth to ground water, size of
ground water basin, permeability of the aquifer, etc., are not easily quantifiable to enable
translation into probable low-flow rates. It may be possible to estimate low-flow rates on
a per unit area basis for a given exceedance frequency if the study area is relatively
homogeneous with respect to geology, topography, and climate. If information is needed
at several ungaged sites, the procedures described by Riggs (28) should be reviewed for
applicability.

4-2
 EM 1110-2-1415
5 Mar 93

z
0 102~~----~----~--~~~~~~~--~~---------r--------~----4-----~
....c
1-4

a:
::1
0
a:
0
Y..
z 113-DAY
0 6
...J 10~~----~----~--------~--------_,~~~~~~~~~120-DAY--~----~
Y.. 0
z 110-DAY
c 110-DAY
w
% 30-DAY
1!5-DAY
7-DA'I'
3-DA'I'
0
s-ou
11!1.11!1 111.11 II 10 50 tO •t . 01
PERCENT CHANCE NONEXCEEDANCE

Figure 4-1. Low-Flow Frequency Curves.

4-3
 EM 1110-2-1415
5 Mar 93

CHAPTER 5
PRECIPITATION FREQUENCY ANALYSIS

5-1. General Procedures. The computation of frequency curves of station

precipitation can be done by procedures similar to those for streamflow analysis described
in the preceding sections. Both graphical and analytical methods may be used. In
precipitation studies, however, instantaneous peak intensities are ordinarily not analyzed
since they are virtually impossible to measure and are of little practical value. Cumulative
precipitation amounts for specified durations are commonly analyzed, mostly for durations
of less than 3 or 4 days. The National Weather Service has traditionally used the
Fisher-Tippett Type I frequency distribution with Gumbel's fitting procedure. The
logarithmic normal, Pearson Type III and log- Pearson Type III (Figure 5-l) distributions,
have also been used with success. Station precipitation alone is not adequate for most
hydrologic studies, and some method of evaluating the frequency of simultaneous or
near-simultaneous precipitation over an area is necessary. Procedures for obtaining
depth-area frequency curves are usually available from National Weather Service
publications (references are given in subsequent paragraphs).

D.ui• WSW, CA 18~-1986

•• ~
.r:.
u
c: /
/
·-I

..
c:

~
0
...,
•-4

...,• 181
•-4

~
~
•-4
u
•
I.
c
/
-4

c
•J
¢
c:

99.99 99.9 99 1 .1 • 91

Figure 5-1. Frequency Curve, Annual Precipitation.

5-1
EM 1110-2-1415
5 Mar 93

5-2. Available Regional Information. Where practical, use should be made of previous
precipitation-frequency-duration studies that have incorporated regional information. For
durations of 5 to 60 minutes in an area generally east of 105th meridian, see Hydro-35 (9).
For durations of 2 to 24 hours in the same area see Technical Paper 40 (10). Because of
the orographic effect, individual reports have been prepared for each of the 11 western
states (24). These reports have maps for 6- and 24-hour durations with extrapolation
procedures to obtain durations Jess than 6 hours. Longer duration events (2- to 10-days)
are presented in references (21), (22) and (23).
5-3. Derivation of Flood-Freauency Relations from Precipitation.
a. Application. Precipitation-frequency relations are often used to derive
flood-frequency relations where inadequate flow data are available or where existing (or
proposed) watershed changes have modified (or will modify) the rainfall-runoff
relationships. Guidelines for developing runoff frequencies from precipitation
frequencies are presented in references (10) and (44). Flood-frequency curves developed
by rainfall-runoff procedures often have tess variance (lower standard deviation) than
those developed from annual flood peaks. This results because not all the possible loss
rates for a given magnitude of precipitation are modeled. If extensive use wiJI be made of
frequency curves derived by rainfall-runoff modeling, an appropriate ratio adjustment for
the standard deviation should be developed for the region.
b. Calibration. Reference (44) describes the procedures involved in calibrating a
HEC-1 model to a flow-frequency curve based either on gaged data from a portion of the
basin or on regional flood-frequency relations. The coefficients from the calibrated
model must be consistent with those from nearby basins that have also been modeled. It
must be remembered that a frequency curve computed from observed flood peaks is based
on a relatively small sample. It is possible that the flow-frequency curve derived from
precipitation-frequency data is more representative of the population flow-frequency
curve than the one computed from the statistics of the observed flood peaks. But, there
are also errors in calibrating the model and establishing loss rates approximate with the
different frequency events. Therefore, the derivation of frequency relations by
rainfall-runoff modeling requires careful checking for consistency at every step.
c. Partial ouration. The precipitation-frequency relations presented in the National
Weather Service publications represent all the events above a given magnitude; therefore,
these relations are from a partial-duration series. The resulting flood frequency relations
must be adjusted if an annual peak flood frequency relationship is desired. Or, more
typically, the partial-duration series precipitation estimates are adjusted to represent
annual series estimates prior to use.

5-2
 EM 1110-2-1415
5 Mar 93

CHAPTER 6
STAGE (ELEVATION) - FREQUENCY ANALYSIS

6-1. l.llli. Maximum stage-frequency relations are often required to evaluate inundation
damage. Inundation can result from a flooding river, storm surges along a lake or ocean,
wind driven waves (runup), a filling reservoir, or combinations of any of these. Minimum
elevation-frequency curves are used to evaluate the recreation benefits at a lake or
reservoir, to locate a water supply intake, to evaluate minimum depths available for
navigation purposes, etc. (Stages are referenced to an arbitrary datum; whereas, elevations
are generally referenced to mean sea level.)
6-2. Stage Data.

a. The USGS WATSTORE Peak Flow File has, in addition to annual peak flows,
maximum annual stages at most sites. Also, some sites located near estuaries have only
stage information because the flow is affected by varying backwater conditions.
b. River stages can be very sensitive to changes in the river channel and floodway.
Therefore, the construction of levees, bridges, or channel modifications can result in stage
data that is non-homogeneous with respect to time. For riverine situations, it is usually
recommended that the flow-frequency curve (Figure 6-1) and a rating curve (stage versus

1···-..---~----P------.--------P-------~------~--~----~

t
i .· .......·/
v
.....

]~

. . . . ·. . . vv
11 6'-ti---~----t------+--------~~----~-,~--+---_,~--ti

• • • • • • UnrecuJated Computed Frequency Curve

- Reculated Frequency Curve Derived by Graphical Procedure•

Figure 6-1. Flow Frequency Curve, Unregulated and Regulated Conditions.

6-1
EM 1110-2-1415
5 Mar 93

flow, Figure 6-2) be used to derive a stage-frequency curve (Figure 6-3). A

stage-frequency curve derived by indirect methods may not always represent the true
relation if the site is subject to occasional backwater situations. Backwater conditions can
be caused by an ice jam, a debris flow, a downstream reservoir, a high tide, a storm surge,
or a downstream river. A coincident frequency analysis may be necessary to obtain an
accurate estimate of the stage-frequency relationship (see Chapter 11 ).
c. Usually the annual extreme value is used to develop an annual series, but a
seasonal series or a partial-duration series could be developed if needed. Caution must be
used in selecting independent events. Independent events are not easily determined if the
events are elevations of a large lake or reservoir; in fact even the annual events may be
significantly correlated.
6-3. Frequency Distribution. Stage (elevation) data are usually not normally distributed
(not a straight line on probability paper). Therefore, an analytical analysis should not be
made without observing the fit to the plotted points (see Chapter 2). Usually, an
arithmetic-probability plot is appropriate for stage or elevation data, but there may be
situations where a logarithmic or some other appropriate transformation will make the plot
more nearly linear. When drawing the curve, known constraints must be kept in mind.
As an example, the bottom elevation, bankfull stage, levee heights, etc., would be
important for a riverine site. The minimum pool, top of conservation pool, top of flood
control pool, spillway elevation, operation criteria, etc., all influence the
elevation-frequency relation for a reservoir, Figure 6-4. These constraints usually make
these frequency relations very non-linear. Extrapolation of stage (elevation) frequency
relations must be done very cautiously. Again, any constraints acting on the relations
must be used as a guide in drawing the curves. Historical information can be incorporated
into a graphical analysis of stage (elevation) data by use of the procedures in Appendix 6
of Bulletin 17B (ref 46). The statistical tests (Appendix 4, ref 46) to screen for outliers
should IlQ1 be applied unless the stage (elevation) data can be shown to nearly fit a normal
distribution.
6-4. Expected Probability. The expected probability adjustment should not be made to
frequency relations derived by graphical methods. The median plotting position formula
corrects for the bias caused by small sample sizes. The expected probability adjustment
should be made when an analytical method is used directly to derive the stage (elevation)
frequency relation. The expected probability adjustment should be made to the
flow-frequency curve when the stage (elevation) frequency relation is derived indirectly.

6-2
 EM 1110-2-1415
5 Mar 93

~
~
... i!
•• I""
IL

•
~
Cl
...Ill• i!
/
1
I 0 Obeerved Peak Flow• and Stare•, 19-'9-85

•
I -

seeel
Ratinr Curve for Gace Location, Nov 12, 19-'8 to Pruent

111111
I I I
a••••• 2151111
I
311111

Figure 6-2. Rating Curve for Present Conditions.

3

.. · /
3
.. v
...
, ~

...
.
....
: i!

I
IL

•
.
Cl
•
.... 2oe
...
/
Ill

~
...E . ..
•" 2

.....
..
/
~

.
.
.. .
2
.....
... v
Derived Stare-Frequenc:y Curve, Unrerulated Condition•
- - Derived Stare-Frequenc:y Curve, Rerulated Condition•
1
I I I I I I
SUI. SISI SUI. II IISI Sll 151 11 1 .1 .11
P•~•nt Ch.nce Exceea.nce

Figure 6-3. Derived Stage-Frequency Curves, Unregulated and Regulated Conditions

6-3
EM 1110-2-1415
5 Mar 93

!XANPl[ ~!SE~VOIA • ANNUAL NAXINUN [L[VATIONS

IDI D
-- I I' I
I
10S6.S 1- TOP OF DAM

IDS0
-- '
I' I 1 I
---
TOP OF INDUCED
'
I 1047. s 1- SWCHARCE

-
;

!
!
oys 104/o. s 1- TOP OF FLOOD
I CONTllOL POOL
IU 0
--
z"
-- ! ~f. I
I
......c
0
1030
i i
- l
Cl
I

> i
IU
..J I
IU I I
1020

1 '

IDS ~
~ I
1010.0 ~ TOP OF IIATER

I~ SUPPLY POOL

---
1000
..... •••• " 10 so
PERCENT CHANCE EXCEEDANCE
ID I .I .DI
971.0 C....•TOP OF liiACTIV!

Figure 6-4. Maximum Reservoir Elevation-Frequency Curve.

6-4
 EM 1110-2-1415
S Mar 93

CHAPTER 7
DAMAGE-FREQUENCY RELATIONSHIPS

7-1. Introduction.

a. There are three methods that may be used to compute average annual damage and
are herein termed the historic method, the simulation method and the frequency method.
If 50 years of damage information were available for an area that has remained in
essentially the same land use with a reasonably constant level of economic activity,
historic damage could be scaled to the present to account for price differences (inflation)
and the average simply computed. This approach is termed the historic method and is the
most direct but is seldom used because sufficient data usually do not exist and the land use
and economic activity of an area are usually changing.

b. A hydrologic simulation model could be developed, or the historic record used,

along with damage functions to generate a time trace of simulated damage. The average
of the time trace of damage would be the average annual damage. This would be termed
the "simulation" method. The simulation method has the advantage of permitting the use
of complex damage functions that can consider more than a single parameter and thus
enable a more accurate computation of damage. The disadvantage of this method is that
the future floods are assumed to exactly duplicate the historic floods and no consideration
given to the possibility of larger floods.

c. The most widely used approach within the Corps of Engineers is the frequency
technique. This technique is described in detail in Section 7-2. This technique addresses
the disadvantages of the previous two methods, and yet is fairly easily applied.
Experience in the development and application of damage functions is essential to
computation of reasonable estimates. Care should be taken to assure the rating curve is
not looped so that discharge is a unique function of stage. Otherwise more complex
functions that correctly relate stage and discharge should be developed and applied.
Damage functions in agricultural areas are often a function of the season and the duration
of flooding. Sensitivity analysis may be useful in determining the reliability of the
computed expected annual damage considering the uncertainties involved.

7-2. Comoutation of Expected Annual Damage.

a. Figure 7-1 shows a schematic of the application of the three basic damage
evaluation functions used to compute the expected value of the annual damage. The term
"expected" is used rather than "average" because a frequency curve is used to represent the
distribution of future flood events and the expected value of damage is computed by the
summation of probability weighted estimates of damage.

b. The steps involved in determining the reduction in annual damage due to project
measures are:

( 1) Develop the basic relationships (stage-damage, stage-discharge, and

discharge-exceedance frequency functions) for each index location for existing
conditions.

7-1
EM 1110-2-1415
5 Mar 93

Rating Cw"ve

0.18 0.50 0.10y

Exceedanc:e Probabity
ArttiYnellc or L

Discharge. cfs
Scale

Stage-FreqJeney Qxve ..---.::D.;;;;ama~"""-'R...;.;el;.;.;a::.;;tion~---.

0.118 0.110
Damage,$
Exceedanee Probabllty

Clrve

0 Al'tltiiMtlc Scilla
Exceedance Probablty

Figure 7-1. Schematic for Computation of Expected Annual Damage.

7-2
 EM 1110-2-1415
5 Mar 93

(2) Combine the stage-damage and the stage-discharge relations into an intermediate
discharge-damage function. Make certain that the stage datum for the stage-damage
and stage-discharge functions is consistent for the index location.

(3) Combine the discharge-exceedance frequency (in events per year) and
discharge-damage function into a damage-exceedance frequency relationship.

(4) Compute the area beneath the damage-exceedance frequency relation (expected
annual damage) for each index location and sum to obtain the total expected annual
flood damage.

(5) Repeat step (I) for each alternative flood plain management plan under
investigation, i.e., revise the three basic evaluation functions as necessary.
(6) Repeat steps (2)-(4).

(7) Subtract results of step (4) (with project) for each plan from results of step (4)
for without-project measures. The differences will be expected annual damage
reduction (raw damage reduction benefits) for each plan.

7-3. Equivalent Annual Damage.

a. To determine the expected annual benefit it is necessary to account for the

changes in expected annual damage that might occur over the life of the project. This
adjustment can be of substantial significance. Watershed runoff characteristics may be
changing with time due to changes in land use, there may be long-term adjustments in
alluvial channel flow regimes that would cause the rating curve to change with time, and
the damage potential of structures and facilities will certainly change with time resulting
in changed stage-damage functions.
b. To develop a single measure of the damage potential, the expected annual damage
must be evaluated over time, at say 10 year intervals with revised evaluation functions at
each interval. The revised expected annual damage is discounted to the base period and
then the raw damage value is amortized over the life of the project to obtain equivalent
annual damage. The computer program "Expected Annual Flood Damage Computation"
(54) has the capability to make these computations, and describes in detail the basic
concepts presented in this chapter.

7-3
 EM 1110-2-1415
5 Mar 93

CHAPTER 8

STATISTICAL RELIABILITY CRITERIA

8-1. Objective. One principal advantage of analytical frequency analysis is that there are
means for evaluating the reliability of the parameter estimates. This permits a more
complete understanding of the frequency estimates and provides criteria for
decision-making. For instance, a common statistical index of reliability is the standard
error of estimate, which is defined as the root-mean-square error. In general, it is
considered that the standard error is exceeded on the positive side one time out of six
estimates, and equally frequently on the negative side, for a total of one time in three
estimates. An error twice as large as the standard error of estimate is considered to be
exceeded one time in 40 in either direction, for a total of one time in 20. These
statements are based on an assumed normal distribution of the errors; thus, they are only
approximate for other distributions of errors. Exact statements as to error probability
must be based on examination of the frequency curve of errors or the distribution of the
errors. Both the standard error of estimate and the confidence limits are discussed in this
chapter.

8-2. Reliability of Frequency Statistics. The standard errors of estimate of the mean,
standard deviation, and skew coefficient, which are the principal statistics used in
frequency analysis, are given by the following equations:

S;c = S/(N)y, (8-1)

Ss • S/(2N)"' (8-2)

SG = {6N(N-1)/[(N-2)(N+l)(N+3)]}Vz (8-3)

where:

the standard error of estimate for the mean

the standard error of estimate for the standard deviation

the standard error for estimate for the skew coefficient, and Sand N are
defined in Section 3-2.

These have been used to considerable advantage, as discussed in Chapter 9, in drawing

maps of mean, standard deviation and skew coefficient for regional frequency studies.

8-3. Reliability of Frequency Curves. The reliability of analytical frequency

determinations can best be illustrated by establishing confidence limits. The error of the
estimated value at a given frequency based on a sample from a normal distribution is a
function of the errors in estimating the mean and standard deviation. (Note that in
practical application there are errors introduced by not knowing the true theoretical
distribution of the data, often termed model error.) Criteria for construction of
confidence limits are based on the non-central t distribution. Selected values are given in
Table F-9. Using that appendix, the confidence limit curves shown on Figure 8-1

8-1
EM 1110-2-1415
5 Mar 93

were calculated. While the expected frequency is that shown by the middle curve, there is
one chance in 20 that the true value for any given frequency is greater than that indicated
by the .05 curve and one chance in 20 that it is smaller than the value indicated by the .95
curve. There are, therefore, nine chances in 10 that the true value lies between the .05
and .95 curves. Appendix E and Example 1 in Appendix 12 of Bulletin 17B (40) provide
additional information and example computations.

a ObH"ed Annual Peaka

-- - - - CompuUd Frequency Cu"e
With Expected Probability Adju1tment
..,~

/
...

~
Confidence Limit Cu"•

•
~
u 18 4
/

J .·· ~-··
...
D

~
IL. .···
.Jt
••
~
ll ... ···
... .

•
:I
c 18 3
f ......
... ·
....,.. .

c ~-
G:
l.........·

99.99 99.9 99 98 58 1 .1 .81

Pe~ent C~e Excaedance

Figure 8-1. Frequency Curve with Confidence Limit Curves.

8-2
 EM 1110-2-1415
5 Mar 93

CHAPTER 9
REGRESSION ANALYSIS AND APPLICATION TO REGIONAL STUDIES

9-1. Nature and Application.

a. General. Regression analysis is the term applied to the analytical procedure for
deriving prediction equations for a variable (dependent) based on given values of one or
more other variables (independent). The dependent variable is the value sought and is to
be related to various explanatory variables which will be known in advance, and which
will be physically related to the dependent variable. For example, the volume of
spring-season runoff from a river basin (dependent variable) might be correlated with the
depth of snow cover in the watershed (explanatory variable). Recorded values of such
variables over a period of years might be graphed and the apparent relation sketched in by
eye. However, regression analysis will generally permit a more reliable determination of
the relation and has the additional advantage of providing a means for evaluating the
reliability of the relation or of estimates based on the relation.
b. Definitions. The function relating the variables is termed the "regression
equation," and the proportion of the variance of the dependent variable that is explained
by the regression equation is termed the "coefficient of determination," which is the
square of the "correlation coefficient." Correlation is a measure of the association between
two or more variables. Regression equations can be linear or curvilinear, but linear
regression suffices for most applications, and curvilinear regression is therefore not
discussed herein. Often a curvilinear relation can be linearized by using a logarithmic or
other transform of one or more of the variables.
9-2. Calculation of Regression Equations.
a. Simple Regression. In a simple regression (one in which there is only one
independent, or explanatory, variable), the linear regression equation is written:

Y = a+ bX (9-1)

in which Y is the dependent variable, X is the independent variable, "a" is the regression
constant, and "b" is the regression coefficient. The coefficient "b" is evaluated from the
tabulated data by use of the following equations:

(9-2a)

or

(9-2b)

in which y is the deviation of a single value y; from the mean ( Y) of its series, x is
similarly defined, Sy and Sx are the respective standard deviations and R is computed by

9-1
 EM III0-2-14J5
5 Mar 93

Equation 9- J J. The regression constant is obtained from the tabulated data by use of the
following equation:

a "" Y- bX (9-3)

All summations required for a simple linear regression can be obtained using Equations 9-
8 and 9-9a.

b. Multiple Regression. In a multiple regression (one in which there is more than

one explanatory variable) the linear regression equation is written:

(9-4)

In the case of two explanatory variables, the regression coefficients are evaluated from the
tabulated data by solution of the following simultaneous equations:

L{x 1) 2 b 1 + I<x 1x2 )b 2 • L(yx 1) (9-5)

L{x 1x2)b 1 + L(x 2) 2b 1 - L{yx 2)

In the case of three explanatory variables, the b coefficients can be evaluated from the
tabulated data by solution of the following simultaneous equations:

L{x 1) 2b 1 + I<x 1x 2)b 2 + L(x 1x3 )b3 • (yx 1) (9-6)

L{x 1x2 )b 1 + L(x 2) 2b 2 + L(x 2x3 )b3 • (yx 2)
I<x 1x3 )b 1 + L(x 2x3 )b 2 + L{x3 ) 2b 3 • (yx3 )

For cases of more than three explanatory variables, the appropriate set of simultaneous
equations can be easily constructed after studying the patterns of the above two sets of
equations. In such cases, solution of the equations becomes tedious, and considerable time
can be saved by use of the Crout method outlined in reference (51) or (52). Also,
programs are available for solution of simple or multiple linear regression problems on
practically any type of electronic computer. For multiple regression equations, the
regression constant is determined as follows:

(9-7)

In Equations 9-2, 9-5 and 9-6, the quantities I<x) 2 , L(yx) and I<x 1x2 ) can be determined
by use of the following equations:

9-2
 EM 1110-2-1415
5 Mar 93

l:<x)2 = L(X)2 - (2)()2/N (9-8)

L(yx) = l:<XY) - L X L y /N (9-9a)

I<x,xz) • l:<X 1X 2) - I x, I X 2/N (9-9b)

9-3. The Correlation Coefficient and Standard Error.

a. General. The correlation coefficient is the square root of the coefficient of

determination, which is the proportion of the variance of the dependent variable that is
explained by the regression equation. A correlation coefficient of 1.0 would correspond to
a coefficient of determination of 1.0, which is the highest theoretically possible and
indicates that whenever the values of the explanatory variables are known exactly, the
corresponding value of the dependent variable can be calculated exactly. A correlation
coefficient of 0.5 would correspond to a coefficient of determination of 0.25, which
would indicate that 25 percent of the variance is accounted for and 75 percent
unaccounted for by the regression equation. The remaining variance (error variance)
would be 75 percent of the original variance and the remaining standard error would be
the square root of 0.75 (or 87 percent) multiplied by the original standard deviation of the
dependent variable. Thus, with a correlation coefficient of 0.5, the average error of
estimate would be 87 percent of the average errors of estimate based simply on the mean
observed value of the dependent variable without a regression analysis.

b. Determination Coefficient. The sample coefficient of multiple determination (R 2 )

can be computed by use of the following equation:

b, 2hx,) + bzlhxz) ... + bN L{yxn)

(9-10)
L(y)2

In the case of simple correlation, Equation 9-10 resolves to:

(9-11)

An unbiased estimate of the coefficient of determination is recommended for most

applications, and is computed by the following equation:

iP • l-(l-R 2)(N-1)/df (9-12)

The number of degrees of freedom (df), is obtained by subtracting the number of

variables (dependent and explanatory) from the number of events tabulated for each
variable.

c. Standard Error. The adjusted standard error (Se) of a set of estimates is the
root-mean-square error of those estimates corrected for the degrees of freedom. On the

9-3
EM I I 10-2-1415
5 Mar 93

average, about one out of three estimates will have errors greater than the standard error
and about one out of 20 will have errors greater than twice the standard error. The
adjusted error variance is the square of the adjusted standard error. The adjusted
standard error or error variance of estimates based on a regression equation is calculated
from the data used to derive the equation by use of one of the following equations:

L(Y) 2
- - b 1 L(yx 1)- b2

df
lJYX 2) ••• - b" L(Yxn)
(9-13a)

(9-13b)

= (l-R 2 )S~ (9-13c)

Inasmuch as there is some degree of error involved in estimating the regression

coefficients, the actual standard error of an estimate based on one or more extreme values
of the explanatory variables is somewhat larger than is indicated by the above equations,
but this fact is usually neglected.
d. Reliability. In addition to considering the amount of variance that is explained by
the regression equation, as indicated by the determination coefficient or the standard
error, it is important to consider the reliability of these indications. There is some chance
that any correlation is accidental, but the higher the correlation and the larger the sample
upon which it is based, the less is the chance that it would occur by accident. Also, the
reliability of a regression equation decreases as the number of independent variables
increases. Ezekiel (8) gives a set of charts illustrating the reliability of correlation
coefficients. It shows, for example, that an unadjusted correlation coefficient (R) of 0.8
based on a simple linear correlation with 12 degrees of freedom could come from a
relationship that has a true value as low as 0.53 in one case out of 20. On the other hand,
the same unadjusted correlation coefficient based on a multiple linear correlation with the
same number of degrees of freedom but with seven independent variables, could come
from a relationship that has a true value as low as zero in one case out of 20. With only 4
degrees of freedom, an unadjusted correlation coefficient of 0.97 would one time in 20
correspond to a true value of 0.8 or lower, in the case of simple correlation, and as low as
zero in a seven-variable multiple correlation. Accordingly, extreme care must be
exercised in the use of multiple correlation in cases based on small samples.
9-4. Simple Linear Regression Example.

a. General. An example of a simple linear regression analysis is illustrated on

Figures 9-1 and 9-2. The data for this example are the concurrent flows at two stations
in Georgia for which a two-station comparison is desired (see Section 3-7). The long
record station is the Chattooga, so the flows for this station are selected as X; therefore
the flows for the short record station (Tallulah) are assigned to Y.

b. Physical Relationship. The values in the table are the annual peak flows for the
water years 1965-1985 (21 values). These two stations are less than 20 miles apart and are
likely to be subject to the same storm events; therefore, the first requirement of a

9-4
 EM 1110-2-1415
5 Mar 93

Chattooga Riv~r Tallulah

----
Riv~r
t---· ------
Flow log Flow log
Y~ar X' X y' y

1965 27200 4.434568 7440 3.871572 :ex = 82.55075 :CY • 73.07071

1966 13400 4.127104 5140 3.710963 :cx2= 325.93995 :cr2= 255.61486
1967
1968
1969
15400
5620
14700
4-187520
3.749736
4.167317
2800
3100
2470
3.447158
3.491361
3.392696
x= 3.93099 '. 3.47956

1970 3480 3.541579 2010 3.303196 c :c xn 2 288.~7484

1971 3290 3.517195 976 2.989449 ( :C X)2

1972 7440 3.871572 2160 3.334453 -N-- 324.50602
1973 19600 4.292256 15500 3.929418
1974 6400 3.806179 4660 3.668385 c :c n 2
-jj- 254.25311
1975 6340 3.802089 2410 3.382017
1976 18500 4.267171 6530 3.814913 :ex tY
1977 13000 4-113943 3580 3.553883 --N-- 287.24008
1978 7850 3.894869 4090 3.611723
1979
1980
14800
10900
4.170261
4.037426
6240
2880
3.795184
3.459392
.z 1.43393 (by equation 9·8)
1981 4120 3.614897 1600 3.204119 liY = 1-13351 (" 9·9a)
1982 5000 3.698970 1960 3.292256 y2 = 1.26115 (" 9·8)
1983 7910 3.898176 3260 3.513217
1984 4810 3.682145 2000 3.301029
1985 4740 3.675778 1010 3.004321

b .. 1.13351/1.433922 (by equation 9·2al

= 0.79049

• = 3.47956 - (0.79049)(3.93099) (by equation 9·3>

• 0.37213

R2 • (1.13351)2/(1.43393)(1.36115) (by equation 9·11)

• 0.658290

i2 • 1 • (1·0.65892)(21·1)/(21·2) (by equation 9·12)

• 0.64031

Computations for standard ~rror:

s e 2 • (1·0.640312)(1.36115)/(21·1) (by equation 9·13b>

• 0.02448

se • 0.15646

Regression equation: Y = 0.37213 • 0.79049X (by equation 9·1)

Y'• 2.356X• 0 • 79 (without logarithms)

Figure 9-1. Computation of Simple Linear Regression Coefficients.

9-5
EM 1110-2-1415
5 Mar 93

regression analysis (logical physical relationship) is satisfied. Because runoff is a

multiplicative factor of precipitation and drainage area. the logarithmic transformation is
likely to be appropriate when comparing two stations with different drainage areas. A
linear correlation analysis was made. as illustrated on Figure 9-1. using equations given in
Section 9-2. The annual peaks for the each station are plotted against each other on
Figure 9-2.

c. Regression Equation. The regression equation is plotted as Curve A on Figure

9-2. This curve represents the best estimate of what the annual peak Tallulah River
would be given the observed annual peak on the Chattooga River. Although not
computed in Figure 9-1. Curve B represents the regression line for estimating the annual
peak flow for the Chattooga River given an observed annual peak on the Tallulah River.

10s~----------~T~W~O_-~ST~A~T~I~O~N~C~O~MP~A~R~I~S~I~O~N~E~XA~M~P~L~E~-----------.

en
u.
u

Curve 8

::.:::
c
IJJ
Q.
Curve A
...J
c
:::.
z
z
c
a:
UJ
....>a:
~
c(
...J
:::.
...J
...J
c(
~

CHATTOOGA RIVER ANNUAL PEAK (X'), CFS

Cu~e A • legreaaion line with Y as dependent variable

curve 8 • legreaaion line with X as dependent variable

Figure 9-2. Illustration of Simple Regression.

9-6
 EM 1110-2-1415
5 Mar 93

d. Reliability. In addition to the curve of best fit, an approximate confidence

interval can be established at a distance of plus and minus 2 standard errors from Curve
A. Because logarithms are used in the regression analysis, the effect of adding (or
subtracting) twice the standard error to the estimate is equivalent to multiplying (or
dividing) the annual peak values by the antilogarithm of twice the standard error. In this
case, the standard error is 0.156, and the antilogarithm of twice this quantity is 2.05.
Hence, values of annual peak flow represented by the confidence interval curves are those
of Curve A multiplied and divided respectively by 2.05. There is a 95 percent chance that
the true value of the dependent variable (Y) for a single observed independent value (X)
will lie between these limits. The confidence interval ~ D..Q1 correct for repeated
predictions using the same sample (6).
9-5. Factors Responsible for Nondetermination.
a. General. Factors responsible for correlations being less than 1.0 (perfect
correlation) consist of pertinent factors not considered in the analysis and of errors in the
measurement of those factors considered. If the effect of measurement errors is
appreciable, it is possible in some cases to evaluate the standard error of measurement of
each variable (see Paragraph 9-3c) and to adjust the correlation results from such effects.
b. Measurement Errors. If an appreciable portion of the variance of Y (dependent
variable) is attributable to measurement errors and these errors are random, then the
regression equation would be more reliable than is indicated by the standard error of
estimate computed from Equation 9-13. This is because the departure of some of the
points from the regression line on Figure 9-2 is artificially increased by measurement
errors and therefore exaggerates the unreliability of the regression function. In such a
case, the curve is generally closer to the true values than to the erroneous observed values.
Where there is large measurement error of the dependent variable, the standard error of
estimate should be obtained by taking the square root of the difference of the error
variance obtained from Equation 9-13 and the measurement error variance. If well over
half of the variance of the points from the best-fit line is attributable to measurement
error in the dependent variable, then the regression line would actually yield a better
estimate of a value than the original measurement. If appreciable errors exist in the
values of an explanatory variable, the regression coefficient and constant will be affected,
and erroneous estimates will result. Hence, it is important that values of the explanatory
variables be accurately determined, if possible.
c. Other Factors. In the example used in Section 9-4 there may well be factors
responsible for brief periods of high intensities that do not contribute appreciably to
annual precipitation. Consequently, some locations with extremely high mean annual
precipitation may have maximum short-time intensities that are not correspondingly high,
and vice versa. Therefore, the station having the highest mean annual precipitation would
not automatically have the highest short-time intensity, but would in general have
something Jess than this. On the other hand, if mean annual precipitation were made the
dependent variable, the station having the highest short-time intensity would be expected
to have something less than the highest value of mean annual precipitation. Thus, by
interchanging the variables, a change in the regression line is effected. Curve B of Figure
9-2 is the regression curve obtained by interchanging the variables Y and X. As there is a
considerable difference in the two regression curves, it is important to use the variable
whose value is to be calculated from the regression equation as the dependent variable in
those cases where important factors have not been considered in the analysis.

9-7
 EM 1110-2-1415
5 Mar 93

d. A verase Slooe. If it is obvious that all of the pertinent variables are included in
the analysis, then the variance of the points about the regression line is due entirely to
measurement errors, and the resulting difference in slope of the regression lines is entirely
artificial. In cases where all pertinent variables are considered and most of the
measurement error is in one variable, that variable should be used as the dependent
variable. Its errors will then not affect the slope of the regression line. In other cases
where all pertinent variables are considered, an average slope should be used. An average
slope can be obtained by use of the following equation:

(9-14)

9-6. Multiole Linear Regression Example.

a. General. An example of a multiple linear regression analysis is illustrated on

Figure 9-3. In this case, the volume of spring runoff is correlated with the water
equivalent of the snow cover measured on April I, the winter low-water flow (index of
ground water) and the precipitation falling on the area during April. Here again, it was
determined that logarithms of the values would be used in the regression equation.
Although loss of 4 degrees of freedom of 12 available, as in this case, is not ordinarily
desirable, the adjusted correlation coefficient attained (0.94) is particularly high, and the
equation is consequently fairly reliable. The computations in Figure 9-3 were made with
the HEC computer program MLRP (reference 50).

b. Logarithmic Transformation. In determining whether logarithms should be used

for the dependent variable as above, questions such as the following should be considered:
"Would an increase in snow cover contribute a greater increment to runoff under
conditions of high ground water (wet ground conditions) than under conditions of low
ground water?" If the answer is yes, then a logarithmic dependent variable (by which the
effects are multiplied together) would be superior to an arithmetic dependent variable (by
which the effects are added together). Logarithms should be used for the explanatory
variables when they would increase the linearity of the relationship. Usually logarithms
should be taken of values that have a natural lower limit of zero and a natural upper limit
that is large compared to the values used in the study.

c. Function of Multiple Regression. It should be recognized that multiple regression

performs a function that is difficult to perform graphically. Reliability of the results,
however, is highly dependent on the availability of a large sampling of all important
factors that influence the dependent variable. In this case, the standard error of an
estimate as shown on Figure 9-3 is approximately 0.038, which, when added to a
logarithm of a value, is equivalent to multiplying that value by 1.09. Thus, the standard
error is about 9 percent, and the 1-in-20 error is roughly 18 percent. As discussed in
Paragraph 9-3d, however, the calculated correlation coefficient may be accidentally high.

9-7. Partial Correlation. The value gained by using any single variable (such as April
precipitation) in a regression equation can be measured by making a second correlation
study using all of the variables of the regression equation except that one. The loss in
correlation by omitting that variable is expressed in terms of the partial correlation
coefficient. The square of the partial correlation coefficient is obtained as follows:

9-8
 EM 1110-2-1415
5 Mar 93

INPUT DATA

OBS NO OBS ID LOG Q LOG SNO LOG Gil LOG PRCP

1 1936 .939 .399 .325 .710
2 1937 .945 .343 .385 .634
3 1938 1.052 .369 .408 .886
4 1939 .744 .246 .428 .581
5 1940 .666 • 181 .316 1.027
6 1941 1.081 .297 .460 1.315
7 1942 1.060 .299 .511 1.097
8 1943 .892 .354 .379 .707
9 1944 1.021 .295 .395 1.240
10 1945 .920 .321 .376 1.091
11 1946 .755 . 168 .413 1.038
12 1947 .960 .280 .410 .979

STATISTICS OF DATA
STANDARD
VARIABLE AVERAGE VARIANCE DEVIATION

LOG SNO .2960 .0050 .0704

LOG GW .4005 .0028 .0531
LOG PRCP .9421 .o5n .2392
LOG Q .9196 .0181 • 1346 DEPENDENT
VARIABLE
UNBIASED CORRELATION COEFFICIENTS (R)

VARIABLE LOG SNO LOG GW LOG PRCP LOG Q

LOG SNO 1.0000 .0000 ·.0459 .6308

LOG GW .DODO 1.0000 .1275 .4170
LOG PRCP ·.0459 .1275 1.0000 .2011
LOG Q .6308 .4170 .2011 1.0000

REGRESSION R~~YLT~
PARTIAL
INDEPENDENT REGRESSION DETERMINATION
VARIABLE COEFFICIENT COEFFICIENT

LOG SNO 1.621806 .9106

LOG GW 1.012912 .6814
LOG PRCP .273390 .7451

UNBIASED STANDARD
REGRESSION R R ERROR OF
CONSTANT SQUARE SQUARE ESTIMATE
•• 223698 .9437 .9226 .0375

Figure 9-3. Example Multiple Linear Regression Analysis.

9-9
 EM 1110-2-1415
5 Mar 93

(9-15)

in which the subscript to the left of the decimal indicates the variable whose partial
correlation coefficient is being computed, and the subscripts on the right of the decimal
indicate the independent variables. An approximation of the partial correlation can
sometimes be made by use of beta coefficients. After the regression equation has been
calculated, beta coefficients are very easy to obtain by use of the following equation:

(9-16)

The beta coefficients of the variables are proportional to the influence of each variable on
the result. While the partial correlation coefficient measures the increase in correlation
that is obtained by addition of one more explanatory variable to the correlation study, the
beta coefficient is a measure of the proportional influence of a given explanatory variable
on the dependent variable. These two coefficients are related closely only when there is
no interdependence among the various explanatory variables. However, some explanatory
variables naturally correlate with each other, and when one is removed from the equation,
the other will take over some of its weight in the equation. For this reason, it must be
kept in mind that beta coefficients indicate partial correlation only approximately.
9-8. Verification of Regression Results. Acquisition of basic data after a regression
analysis has been completed will provide an opportunity for making a check of the results.
This is done simply by comparing the values of the dependent variable observed, with
corresponding values calculated from the regression equation. The differences are the
errors of estimate, and their root-mean-square is an estimate of the standard error of the
regression-equation estimates (Paragraph 9-3). This standard error can be compared to
that already established in Equation 9-13. If the difference is not significant, there is no
reason to suspect the regression equation of being invalid, but if the difference is large,
the regression equation and standard error should be recalculated using the additional data
acquired.

9-9. Regression by Graphical Techniaues. Where the relationships among variables used
in a regression analysis are expected to be curvilinear and a simple transformation cannot
be employed to make these relationships linear, graphical regression methods may prove
useful. A satisfactory graphical analysis, however, requires a relatively large number of
observations and tedious computations. The general theory employed is similar to that
discussed above for linear regression. Methods used will not be discussed herein, but can
be found in references 8 and 27.
9-10. Practical Guidelines. The most important thing to remember in making
correlation studies is that accidental correlations occur frequently, particularly when the
number of observations is small. For this reason, variables should be correlated only when
there is reason to believe that there is a physical relationship. It is helpful to make
preliminary examination of relationships between two or more variables by graphical
plotting. This is particularly helpful for determining whether a relationship is linear and
in selecting a transformation for converting curvilinear relationships to linear
relationships. It should also be remembered that the chance of accidentally high

9-10
 EM 1110-2-1415
5 Mar 93

correlation increases with the number of correlations tried. If a variable being studied is
tested against a dozen other variables at random, there is a chance that one of these will
produce a good correlation, even though there may be no physical relation between the
two. In general, the results of correlation analyses should be examined to assure that the
derived relationship is reasonable. For example, if streamflow is correlated with
precipitation and drainage area size, and the regression equation relates streamflow to
some power of the drainage area greater than one, a maximum exponent value of one
should be used, because the flow per square mile usually does not increase with drainage
area when other factors remain constant.

9-11. Regional Frequency Analysis.

a. General. In order to improve flood frequency estimates and to obtain estimates

for locations where runoff records are not available, regional frequency studies may be
utilized. Procedures described herein consist of correlating the mean and standard
deviation of annual maximum flow values with pertinent drainage basin characteristics by
use of multiple linear regression procedures. The same principles can be followed using
graphical frequency and correlation techniques where these are more appropriate.

b. Frequency Statistics. A regional frequency correlation study is based on the two

principal frequency statistics: the mean and standard deviation of annual maximum flow
logarithms. Prior to relating these frequency statistics to drainage basin characteristics, it
is essential that the best possible estimate of each frequency statistic be made. This is
done by adjusting short-record values by the use of longer records at nearby locations.
When many stations are involved, it is best to select long-record base stations for each
portion of the region. It might be desirable to adjust the base station statistics by use of
the one or two longest-record stations in the region, and then adjust the short-record
station values by use of the nearest or most appropriate base station. Methods of adjusting
statistics are discussed in Section 3-7.
c. Drainage-Basin Characteristics. A regional analysis involves the determination of
the main factors responsible for differences in precipitation or runoff regimes between
different locations. This would be done by correlating important factors with the
long-record mean and with the long-record standard deviation of the frequency curve for
each station (the long-record values are those based on extension of the records as
discussed in Section 3-7). Statistics based on precipitation measurements in mountainous
terrain might be correlated with the following factors:

- Elevation of station
- General slope of surrounding terrain
- Orientation of that slope
- Elevation of windward barrier
- Exposure of gage
- Distance of leeward controlling ridge

9-11
 EM 1110-2-1415
5 Mar 93

Statistics based on runoff measurements might be correlated with the following factors:

- Drainage area (contributing)

- Stream length
- Slope of drainage area or of main channel
- Surface storage (lakes and swamps)
- Mean annual rainfall
- Number of rainy days per year
- Infiltration characteristics
- Urbanized Area

d. Linear RelationshiPs. In order to obtain satisfactory results using multiple linear

regression techniques, all variables must be expressed so that the relation between the
independent and any dependent variable can be expected to be linear, and so that the
interaction between two independent variables is reasonable. An illustration of the first
condition is the relation between rainfall and runoff. If the runoff coefficient is sensibly
constant, as in the case of urban or airport drainage, then runoff can be expected to bear a
linear relation to rainfall. However, in many cases initial losses and infiltration losses
cause a marked curvature in the relationship. Ordinarily, it will be found that the
logarithm of runoff is very nearly a linear function of rainfall, regardless of loss rates,
and in such cases, linear correlation of logarithms would be most suitable. An illustration
of the second condition is the relation between rainfall, D, drainage area, A, and runoff,
Q. If the relation used for correlation is as follows:

Q • aD+ bA + c (9-17)

then it can be seen that one inch change in precipitation would add the same amount of
flow, regardless of the size of drainage area. This is not reasonable, but again a
transformation to logarithms would yield a reasonable relation:

log Q = d log D + e log A + log f (9-18)

or transformed:

(9-19)

Thus, if logarithms of certain variables are used, doubling one independent quantity will
multiply the dependent variable by a fixed ratio, regardless of what fixed values the other
independent variables have. This particular relationship is reasonable and can be easily
visualized after a little study. There is no simple rule for deciding when to use

9-12
 EM 1110-2-1415
5 Mar 93
logarithmic transformation. It is usually appropriate, however, when the variable has a
fixed lower limit of zero. The transformation should provide for near-uniform variance
throughout the range of data.
e. Example of Regional Correlation. An illustrative example of a regional
correlation analysis for the mean log of annual flood peaks (Y) with several basin
characteristics is shown on Figure 9-4. In this example, the dependent variable is
primarily related to the drainage area size, but precipitation and slope added a small
amount to the adjusted determination coefficient. The regression equation selected for the
regional analysis included only drainage area as an independent variable.
f. Selection of Useful Variables. In the regression equations shown on Figure 9-4,
the adjusted determination coefficient increases as variables are deleted according to their
lack of ability to contribute to the determination. This increase is because there is a
significant increase in the degrees of freedom as each variable is deleted for this small
sample of 20 observations. Both the adjusted determination coefficient and standard error
of estimate should be reviewed to determine how many variables are included in the
adopted regression equation. Even in the case of a slight increase in correlation obtained
by adding a variable, consideration of the increased unreliability of R as discussed in
Section 9-3 might indicate that the factor should be eliminated in cases of small samples.
The simplest equation that provides an adequate predictive capability should be selected.
In this example,· there is some loss in determination in only using drainage area, but this
simple equation is adopted to illustrate regional analysis. The adopted equation is:

log Y = 1.586 + 0.962 log (AREA) (9-20)

or
Y • 38.5 AREA · 962 (9-21)

The R 2 for this equation is 0.839.

g. Use of Map. Many hydrologic variables cannot be expressed numerically.
Examples are soil characteristics, vegetal cover, and geology. For this reason, numerical
regional analysis will explain only a portion of the regional variation of runoff
frequencies. The remaining unexplained variance is contained in the regression errors,
which varies from station to station. These regression errors are computed by subtracting
the predicted values from the observed values for each station. These errors can then be
plotted on a regional map at the centroid of each station's area, and lines of equal values
drawn (perhaps using soils, vegetation, or topographic maps as a guide). Combining this
regional error with the regression equation should be much better than using the single
constant for the entire region. In smoothing lines on such a map, consideration should be
given to the reliability of computed statistics. Equations 8-1 and 8-2 can be used to
compute the standard errors of estimating means and standard deviations. In Figure 9-5
for example, Station 5340 (observation 11) had 66 years of record and the standard error
for the mean was 0.028. There is about one chance in three that the mean is in error by
more than 0.029 or about one chance in twenty that the mean is in error by more than
0.056 (twice the standard error). Figure 9-6 shows a map of the errors and Figure 9-5
shows the regional map values for each station and evaluates the worth of the map. The
map has a mean square error of 0.0112 compared to that of 0.0356 for the regression
equation alone.

9-13
EM 1110-2-1415
5 Mar 93

JII~T RATA

015110 OIS ID AREA SLOPE LEIIGTN LAICES ELEY FC.EST PIECIP SOILS MEAN
1 5090 292.0 6.3 31.5 1.5 1.230 35.0 40.0 3.5 3.783
2 5140 185.0 14.3 30.3 1.0 1.024 52.0 38.2 3.2 3.783
3 5180 282.0 44.0 29.8 1.0 1.740 64.0 35.0 3.3 4.030
4 5200 298.0 zo. 1 37.3 1.0 1.600 30.0 36.5 3.3 4.044
5 5205 171.0 24.4 44.8 1.0 1.447 33.0 36.0 3.2 4.333
6 5260 114.0 35.8 17.5 1.0 1.3&3 30.0 34.0 3.0 3.751
7 5270 52.2 29.9 17.4 1.0 1.489 21.0 33.0 3.0 2.637
I 5280 66.8 12.6 20.1 1.0 1.305 41.0 33.6 3.0 3.186
9 5305 17.5 45.2 18.9 1.0 1.123 27.0 34.6 3.0 3.348
10 5320 215.0 17.7 27.5 1.1 .966 57.0 35.5 3.0 3.99S
11 5340 383.0 21.3 36.7 3.5 1.370 54.0 42.0 3.3 4.122
12 5375 15.7 291.0 5.6 1.0 1.350 11.0 40.0 3.5 2.722
13 5380 43.8 52.2 22.1 1.0 1.300 15.0 43.0 3.5 3.078
14 5390 274.0 39.6 32.3 1.6 1.200 70.0 43.0 2.8 3.930
15 5445 136.0 37.4 22.7 1.0 1.100 94.0 43.0 4.9 3.590
16 5415 604.0 22.1 44.5 1.0 1.900 13.0 37.0 3.2 4.092
17 5495 37.7 54.8 15.2 1.0 1.350 67.0 37.8 3.2 3.284
11 5500 173.0 45.7 24.2 1.0 1.700 65.0 39.0 3.9 3.116
19 5520 443.0 24.4 56.0 1.3 1.600 89.0 44.0 3.2 4.275
20 5525 23.1 115.7 10.0 1.0 1.100 80.o 47.0 4.3 3.249

STATJ~TI'~ QF ~ATA
STAIIDAIIO
VARIABLE AVEIIAGE YARIAIICE DEVIATIOII

AREA 2.1411 .2249 .4743

SLOPE 1.5105 .1255 .3542
LENGTH 1.3&32 .0550 .2345
STORAGE .0540 .0171 .1308
ELEY 1.4339 .0707 .2659
FC.EST 1.7293 .0353 .117'9
PRECIP 1.5845 .0020 .0444
SOILS .5230 .0034 .0511
IlEAl! 3.6524 .2455 .4955 DEPENDENT VAlUABLE

Y!!IA~ ~R,~!TIQ! ~FF!'l'NTI

VARIABLE AREA SLOPE LEIIGTN I.MES ELEV FOREST PIECJP SOILS lEAN
AREA 1.0000 ·.6327 .9304 .2749 .0000 .0000 .0000 .0000 .9159
SLOPE ·.6327 1.0000 ·.7144 ·.1318 .1147 .3635 .1867 .2053 ·.4521
LENGTH .9304 ·.7144 1.0000 .2345 .0000 .0000 .0000 •• 1096 .8263
STORAGE .2749 ·.1311 .2345 1.0000 .0000 .0000 .2112 .0000 .2596
ELEV .DODO .1117 .0000 .0000 1.0000 .27'91 .0000 .5297 .0000
FatEST .0000 .3635 .0000 .0000 . .27'91 1.0000 .6117 .4304 .0000
PIECIP • 0000 .1867 .0000 .2112 .0000 .6117 1.0000 .5412 .0000
SOILS .0000 .2053 ·.1096 .0000 .5297 .4304 .5412 1.0000 .0000
NEAll .9159 ·.4521 .8263 .2596 .0000 .0000 .0000 .0000 1.0000

SU!!M!RY OF UG!ESSIQ! ANALJSES FOR !!UN LQG OF A!!I!JAL r,AJCS

AD .lUSTED STANDARD lEAN
REGRESS lOll • • • • • • • • • • ltEGIIESSIOII c::aEFFICIENT • • • DETEIMINATIOII ERROR OF SQUARE
IXIISTAIIT AREA SLOPE LENGTH LMES ELEV FOREST PIECIP SOILS c::aE FFI CI ENT ESTIMATE ERROR
'~!!Gl '~!!Gl m11n U!!Gl U!S!!!'l '~!!Gl '~!!Gl '~!!Gl
·1 .522 1.261 0.112 ·0.321 ·0.272 ·0.114 0.055 1.739 0.140 0.1097 0.2162 0.0257
·1.633 1.269 0.169 ·0.364 ·0.289 ·0.169 0.054 1.174 0.1254 0.2070 0.0257
·1.1011 1.267 0.17'9 ·0.350 -0.301 -0.165 2.023 o.aw o. 1990 0.0251
-1.661 1.130 0.243 -0.251 -o. 167 1.753 0.1469 0.1939 0.0263
-1.130 1.104 0.250 ·0. 129 1.399 0.1537 0.1196 0.0269
-1.034 1.069 0.191 1.319 0.1514 0.1165 0.0278
-1.134 0.975 1.699 0.1553 0.1115 0.0302
1.586 0.962 0.1390 0.1911 0.0356

Figure 9-4. Regression Analysis for Regional Frequency Computations.

9-14
 EM 1110-2-1415
5 Mar 93
R!XHOW. ANALYSIS Win! RmU:SSictl ct1 0RAIN1GE M1.F.A ctn.Y

K\1' YEARS CE STANDARD

CIIS l«:l STAl'IOO OBSEIMil CX'f1fU'I"ff) ERRCR VALUE OIFF onr2 REtXIUl OEVIAI'IOO sy
1 5090 3. 783 3.957 -0.174 -0.18 0.006 0.000036 43 0.190 0.029
2 5140 3.783 3.766 0.017 0.01 0.007 0.000049 52 0.195 0.027
3 5180 4.030 3.942 0.088 0.09 -0.002 0.000004 39 0.289 0.046
4 5200 4.044 3.965 0.079 0.07 0.009 0.000081 27 0.256 0.049
5 5205 4.333 4.362 -0.029 0.08 -0.109 0. 011881 50 0.251 0.035
6 5260 3.751 3.564 0.187 0.11 0.077 0.005929 35 0.293 0.050
7 5270 2.637 3.238 -0.601 -0.22 -0.381 0.145161 31 0.206 0.037
e 5280 3.186 3.341 -0.155 -0.17 0.015 0.000225 45 0.186 0.028
9 5305 3.348 3.403 -0.055 -0.04 -0.015 0.000225 44 0.128 0.019
10 5320 3.995 3.829 0.166 0.16 0.006 0.000036 66 0.288 0.035
11 5340 4.122 4.070 0.052 0.02 0.032 0.001024 66 0.227 0.028
12 5375 2.722 2.736 -0.014 -0.05 0.036 0.001296 40 0.323 0.051
13 5380 3.078 3.164 -0.086 -0.08 -0.006 0.000036 60 0.226 0.029
14 5390 3.930 3.930 0.000 0.00 0.000 0.000000 41 0.261 0.041
15 5445 3.590 3.638 -0.048 -0.15 0.102 0.010404 39 0.278 0.045
16 5485 4.092 4.260 -0.168 -0.08 -0.088 0.001744 61 0.242 0.031
17 5495 3.284 3.102 0.182 0.04 0.142 0.020164 39 0.242 0.039
18 5500 3.816 3.738 0.078 0.08 -0.002 0.000004 66 0.237 0.029
19 5520 4.275 4.131 0.144 0.17 -0.026 0.000676 54 0.277 0.038
20 5525 3.249 2.910 0.339 0.20 0.139 0.019321 39 0.291 0.047

Sun 0.06 -0.058 0.224296

Av.rqe 0.003 -0.003 0.0112

Figure 9-5. Regional Analysis Computations for Mapping Errors.

Figure 9-6. Regional Map of Regression Errors.

9-15
 EM 1110-2-1415
5 Mar 93

h. Summary of Procedure. A regional analysis of precipitation or flood-flow

frequencies is generally accomplished by performing the following steps:
(I) Select long-record base stations within the region as required for extension of
records at each of the short-record stations.
(2) Tabulate the maximum events of each station.
(3) Transform the data to logarithms and calculate X. Sand, if appropriate, G
(Equations 3-1, 3-2 and 3-3) for each base station.
(4) Calculate X and S for each other station and for the corresponding values of the
base station, and calculate the correlation coefficient (Equation 3-16).
(5) Adjust all values of X and S by use of the base station, (Equations 3-17 and 3-
19). (If any base station is first adjusted by use of a longer-record base station, the
longer-record statistics should be used for all subsequent adjustments.)
(6) Select meteorological and drainage basin parameters that are expected to correlate
with X and S, and tabulate the values for each drainage basin or representative area.
(7) Calculate the regression equations relating X and S to the basin characteristics,
using procedures explained in Section 9-2, and compute the corresponding
determination coefficients.

(8) Eliminate variables in turn that contribute the least to the determination
coefficient, recomputing the determination coefficient each time, and select the
regression equation having the highest adjusted determination coefficient, or one
with fewer variables if the adjusted determination coefficient is nearly the same.
(9) Compute the regression errors for each station, plot on a suitable map, and draw
isopleths of the regression errors for the regression equations of X (see Figures 9-5
and 9-6 for an example) and S considering the standard error for each computed, or
adjusted, X and S. Note that an alternate procedure is to add the regression constant
to each error value and develop a map of this combined value. This procedure
eliminates the need to keep the regression constant in the regression equation as the
mapped value now includes the regression constant.
(10) A frequency curve can be computed for any ungaged basin in the area covered
within the mapped region by using the adopted regression equations and appropriate
map values to obtain X and S, and then using the procedures discussed in Section 3-2
to compute several points to define the frequency curve. (It may also be necessary to
develop regional (generalized) values of the skew coefficient if the Pearson type III
distribution is considered appropriate. The next section describes the necessary steps
to compute a generalized skew coefficient.)
i. Generalized Skew Determinations. Skew coefficients for use in hydrologic studies
should be based on regional studies. Values based on individual records are highly
unreliable. Figure 9-7 is a plot of skew coefficients sequentially recomputed after adding
the annual peak for the given year. Note that, after 1950, the skew coefficient was at a
at a minimum of about 0.5 in 1954 and maximum of about 1.9 in 1955, only one year

9-16
 EM 1110-2-1415
5 Mar 93

apart. The procedures for developing generalized skew values are generally set forth in
Bulletin 17B (pages I 0-15).
In summary, it is recommended that:

(I) the stations used in the study have 25 or more years of data,

(2) at least 40 stations be used in the analysis, or at least all stations surrounding the
area within 100 miles should be included,

(3) the skew values should be plotted at the centroid of the basins to determine if
any geographic or topographic trends are present,

(4) a prediction equation should be developed to relate the computed skew

coefficients to watershed and climate variables,

(5) the arithmetic mean of at least 20 stations, if possible, in an area of reasonably

homogeneous hydrology should be computed, and

(6) then select the method that provides the most accurate estimation of the skew
coefficient (smallest mean-square error).

In addition to the above guidelines, care should be taken to select stations without
significant man-made changes such as reservoirs, urbanization, etc.

9-17
EM 1110-2-1415
5 Mar 93

a.h Kl at Shoenw~Wra. PA

Year

IUh Kl at ShoerniiU'a. PA
2.1
2.0
1.1
r'
~~
1.1
1.7
l
., I. I
II 1\
&-
...s 1.11 ~
0 1.4 ~
1:
.2 l.ll
1\ 1\
\ ~
~ 1.2
8 1.1 \ .A
0 v
lit
CD 1.0
II\ \
ll5 0.1 1
0.1
lA
0.7 \...
0.1
v ...t"""\
0.1
0.4
1106 1111 ,... ""
Year
..
, 11171 111116

Figure 9-7. Annual Peaks and Sequential Computed Skew by Year.

9-18
 EM 1110-2-1415
5 Mar 93

CHAPTER 10
ANALYSIS OF MIXED POPULATIONS

I 0-1. Definition. The term mixed population, in a hydrologic context, is applied to data
that results from two or more different, but independent, causative conditions. For
example, floods originating in a mountainous area or the northern part of the United
States at a given site could be caused by melting snow or by rain storms. Along the Gulf
and Atlantic coasts, floods can be caused by general cyclonic storms or by intense tropical
storms. A frequency curve representing the events caused by one of the climatic
conditions may have a significantly different slope (standard deviation) than for the other
condition. A frequency plot of the annual events, irrespective of cause, may show a
rather sudden change in slope and the computed skew coefficient may be comparatively
high. In these situations, a frequency curve derived by combining the frequency curves
of each population can result in a computed frequency relation more representative of the
observed events.

10-2. Procedure.

a. The largest annual event is selected for each causative condition. As Bulletin 17B
(46) cautions, "If the flood events that are believed to comprise two or more populations
cannot be identified and separated by an objective and hydrologically meaningful
criterion, the record shall be treated as coming from one population." Also, Bulletin 17B
states, "Separation by calendar periods in lieu of separation by events is not considered
hydrologically reasonable unless the events in the separated periods are clearly caused by
different hydrometeorologic conditions."

b. The frequency relations for each separate population can be derived by the
graphical or analytical techniques described in Chapter 2·and then combined to yield the
mixed population frequency curve. The individual annual frequency curves are combined
by "probability of union." For two curves, the equation is:

(10-1)

where:

Pe = Annual exceedance probability of combined populations for a selected

magnitude.

P1 = Annual exceedance probability of same selected magnitude for population

series 1.

Annual exceedance probability of same magnitude selected above for

population series 2.

c. Figure I 0-1 illustrates a combined annual-event frequency curve derived by

combining a hurricane event frequency curve with a nonhurricane event curve for the
Susquehanna River at Harrisburg, PA. For more than two population series, n curves may
be more easily combined by the following form:

10-1
EM 1110-2-1415
5 Mar 93

(10-2)

NOTE: The exceedance probability (percent chance exceedance divided by 100) must be
used in the above equations.

d. If partial duration curves are to be added, the equation is simply P = P 1 + P2 .

This assumes that the events in both series are hydrologically independent. cWhen the
combined curve is used in an economic analysis, the events in both series must also be
economically independent.

10-3. Cautions.

a. If annual flood peaks have been separated by causative factors, a generalized skew
must be derived for each separate series to apply the log-Pearson Type III distribution as
recommended by Bulletin 17B. Plate 1 of Bulletin 17B or any other generalized skew map
based on the maximum annual event, irrespective of cause, will not be applicable to any
of the separated series. Derivation of generalized skew relations for each series can
involve much effort.

b. Some series may not have an event each year. For example, tropical storms do not
occur every year over most drainage areas in the United States, and quite often there are
only a few flood events for the series. Extensive regionalization may be necessary to
reduce the probable error in the frequency relations which results from small sample sizes.

c. Sometimes frequency relations of particular seasons are of interest, i.e., quarterly

or monthly, and the curves are combined to verify the annual series curve. The combined
curve will very likely fit the annual curve only in the middle parts of the curve. The
lower end of the curve will have a partial duration shape as many small events have been
included in the analysis. Also, it is possible that the slope of the frequency relation will
be higher at the upper end of the curve as the one season or month with the maximum
event included in its series will likely have a higher slope than that of the annual series.

d. A basic assumption of this procedure is that each series is independent of the

other. Coincidental frequency analysis techniques must be used where dependance is a
factor. For instance, the frequency curves of two or more tributary stations cannot be
combined by the above equation to derive the frequency curve of a downstream site. This
is because the downstream flow is a function of the summation of the coincident flows on
each of the tributaries.

10-2
 EM 1110-2-1415
5 Mar 93

.
D Annual Nonhurricane Event1, 19%9-85

-. Computed Frequency Curve, Nonhurricane Seriee

Annual Hurricane Event., 1929-85
- - Graphical Frequency Curve, Hurricane Seriea
~•u 11 5
. . "'
...,·
D . I

~~
""•• ~
..•
II.

:I ~
__,p-
lt. "'
c 114
c
« ~ I
lc
I
I
u:J
88.811 88.8 88 1 .1 .11

w..t~c.....- ...,. .,., PA

D Annual Event., 19%9-85

- Combined Frequency Curve

•
~
0 11 15
/

/
iD
....
""••
II.
... ~
•
:I
11 4
~
«
E ~a
Ia

89.811 98.9 81 151 11 1 .1 .11

Figure 10-1. Nonhurricane, Hurricane, and Combined Flood Frequency Curves.

10-3
 EM III0-2-I4I5
5 Mar 93

CHAPTER II
FREQUENCY OF COINCIDENT FLOWS

II-I. Introduction. In many cases of hydrologic design, it is necessary to consider

only those events which occur coincidentally with other events. For example, a pump
station is usually required to pump water only when interior runoff occurs at a time that
the main river stage is above interior ponding levels. In constructing a frequency curve of
interior runoff that occurs only at such times, data selected for direct use should be
limited to that recorded during high river stages. In some cases, such data might not be
adequate, but it is possible in these cases where the two types of events are not highly
correlated to make indirect use of noncoincident data in order to establish a more reliable
frequency curve of coincident events.

II-2. A Procedure for Coincident Frequency Analysis.

a. Objective. Determine an exceedance-frequency relationship for a variable C.

Variable C is a function of two variables, A and B.

b. Selection of pominant Varjable. The variable that has the largest influence on
variable C is designated as variable A; the less influential variable is designated as variable
B. The significance of "influential" will be indicated by means of an example. Figure
11-1 shows water surface profiles along a tributary near the junction with a main river.
Stage on the tributary (variable C) is a function of main river stage and tributary
discharge. In Region I, main river stage, will tend to have the dominant influence on
tributary stage, whereas in Region II, tributary discharge will tend to dominate. The
boundary between Regions I and II cannot be precisely defined and will vary with
exceedance frequency. Stage-frequency determinations will be least accurate in the
vicinity of the boundary where both variables have a substantial impact on the combined
result.

c. Procedure.

(1) Construct a duration curve for variable B. Discretize the duration curve with a
set of "index" values of B. Index values should represent approximately equal ranges
of magnitude of variable B. The area under the resulting discretized duration curve
should equal the area under the original duration curve. The number of index values
of B required for discretization depends on the range of variation of B and the
sensitivity of variable C to B. Therefore, the number of points selected should
adequately define the relationships.

(2) For each of the index values of variable B, develop a relationship between
variable A and the combined result C. In the illustration (Figure 11-1) the
relationship linking variables A, B and C would be obtained with a set of water
surface profile calculations for various combinations of main river stage and
tributary discharge.

(3) If variables A and Bare independent of each other, construct an

exceedance-frequency curve of variable A. If the variables are not independent,

11-1
EM II 10-2-1415
5 Mar 93

construct a conditional exceedance-frequency curve of variable A for each index

value of variable B.
(4) Using the relationship developed in step (2) and frequency curve(s) developed in
step (3), construct a conditional exceedance-frequency curve of variable C for each
index value of variable B.
(5) For a selected magnitude of variable C, multiply the exceedance-frequencies
from each curve developed in step (4) by the corresponding proportions of time
represented, and sum these products to obtain the exceedance-frequency of variable
C. Repeat this step for other selected magnitudes of C until a complete
exceedance-frequency curve for variable C is defined. This step is an application of
the total probability theorem.
d. Seasonal Effects. The duration of frequency curves from steps (I) and (3) are
assumed to represent stationary processes. That is, it is assumed that probabilities and
exceedance frequencies obtained from the curves do not vary with time. In order for this
assumption to be reasonably valid, it is generally necessary to follow the above procedure
on a seasonal basis. Once seasonal exceedance-frequency curves have been obtained (step
e), they may be combined to obtain an all-season exceedance-frequency curve.
e. Assumption of lndeoendence. Although step (3) enables application of the
procedure to situations where variables A and Bare not independent, data is generally not
available to establish the conditional exceedance frequency curves required by that step.
Consequently, application of the procedure presented here is generally limited to
situations where it is reasonable to assume that variables A and B are independent.

IUIOI I

IOIIIAL IIAJI
IJYEI STAGE

,IOfJL£ 1 100 Tl. IIA I I I I VEl STAGE IIAII IJVEI

AID 100 Tl. Q 01 Till.

,IOfiL£ 2 IOIIIAL IIAJI II VEl STAG£

AID 100 Tl. Q 01 Till.

'IOfJL£ ] 100 Tl. IIA II liVE I STAU

AID IOIIIAL Q 01 Till.

Figure 1I -1. Illustration of Water Surface Profiles in Coincident Frequency Analysis.

11-2
 EM 1110-2-1415
5 Mar 93

CHAPTER 12
STOCHASTIC HYDROLOGY

12-1. Introduction.
a. A stochastic process is one in which there is a chance component in each
successive event and ordinarily some degree of correlation between successive events.
Modeling of a stochastic process involves the use of the "Monte Carlo" method of adding a
random (chance) component to a correlated component in order to construct each new
event. The correlated component can be related, not only to preceding events of the same
series, but also to concurrent and preceding events of series of related phenomena.
b. Work in stochastic hydrology has related primarily to annual and monthly
streamflows, but the results often apply to other hydrologic quantities such as
precipitation and temperatures. Some work on daily streamflow simulation has been done.
12-2. Apolications.
a. Hydrologic records are usually shorter than 100 years in length, and most of them
are shorter than 25 years. Even in the case of the longest records, the most extreme
drought or flood event can be far different from the next most extreme event. There is
often serious question as to whether the extreme event is representative of the period of
record. The severity of a long drought can be changed drastically by adding or
subtracting I year of its duration. In order that some estimate of the likelihood of more
severe sequences can be made, the stochastic process can be simulated, and long sequences
of events can be generated. If the generation is done correctly, the hypothetical sequence
would have as equal likelihood of occurrence in the future as did the observed record.
b. The design of water resource projects is commonly based on assumed recurrence
of past hydrologic events. By generating a number of hydrologic sequences, each of a
specified desired length, it is possible to create a much broader base for hydrologic design.
While it is not possible to create information that is not already in the record, it is possible
to use the information more systematically and more effectively. In selecting the number
and length of hydrologic sequences to be generated, it is usually considered that 10 to 20
sequences would be adequate and that their length should correspond to the period of
project amortization.
c. It must be recognized that the more hydrologic events that are generated, the
more chance there is that an extreme event or combination of events will be exceeded.
Consequently, it is not logical that a design be based on the most extreme generated event,
but rather on some consideration of the total consequences that would prevail for a given
design if all generated events should occur. The more events that are generated, the less
proportional weight each event is given. If a design is tested on 10 sequences of
hydrologic events, for example, the benefits and costs associated with each sequence
would be divided by 10 and added in order to obtain the "expected" net benefits.
12-3. Basic Procedure. Successful simulation of stochastic processes in hydrology has
been based generally on the concept of multiple linear regression, where the regression

12-1
EM 1110-2-1415
5 Mar 93

equation determines the correlated component, and the standard error of estimate
determines the random component. Figure 12-1 illustrates the general nature of the
process. In this case, a low degree of correlation is illustrated, in order to emphasize
important aspects of the process. It can be seen that, if every estimate of the dependent
variable is determined by the regression line (Figure 12-1 a), the estimated points would be
perfectly correlated with the independent variable and would have a much smaller range
of magnitude than the actual observed values of the dependent variable. In order to avoid
such unreasonable results, it is necessary to add a random component to each estimate
(Figure 12-1 b), and this random component should conform to the scatter of the observed
data about the regression line.
6.0

n
__,...
4.0

0 /p
~--""
0
J,/~ 0

~~
0

1.0
v
0.0
0.0 1.0 2.0 3.0 4.0 e.o
lndlpelldent variable

Figure 12-la. Data Estimation from Regression Line.

6.0

4.0
FIMdarnEmlr
() . ~
__,...

i ~·.
~
.. 0
3.0
-----

I .. I 0
• vv i--""'
• 0

~
~ 0

1.0
~ •
0.0
0.0 1.0 2.0 3.0 4.0 6.0

kldapelldent variable

Figure 12-1 b. Data Estimation with Addition of Random Errors.

12-2
 EM 1110-2-1415
5 Mar 93
12-4. Monthly Streamflow Model.

a. In accordance with the above basic procedure, a simulation model for generating
values of a variable which can be defined only partially by a deterministic relation is:

( 12-1)

where:

Y = dependent variable
a = regression constant
b 1,b 2 = regression coefficients

x,.x2 = independent variables

Z = random number from normal standard population with zero mean and
unit variance
Sy = standard deviation of dependent variable
R = multiple correlation coefficient

b. This type of simulation model can be used to generate related monthly streamflow
values at one or more stations. Multiple linear regression theory is based on the assumed
distribution of all variables in accordance with the Gaussian normal distribution.
Therefore, mathematical integrity requires that each variable be transformed to a normal
distribution, if it is not already normal. It has been found that the logarithms of
streamflows are approximately normally distributed in most cases. For computational
efficiency it is convenient to work with deviations from the mean which have been
normalized by dividing by the standard deviation. This deviate is sometimes called the
Pearson Type Ill deviate and can be computed as follows:

(12-2)

where:

t = Pearson Type Ill deviate

= month number
J =
year number
X = logarithm of flow
X = mean of flow logarithms
S = standard deviation of flow logarithms

12-3
EM 1110-2-1415
5 Mar 93

c. If these deviates exhibit a skewness, they can be further transformed, if

necessary, to a distribution very close to normal by use of the following approximate
Pearson Type III transform equation:

(12-3)

where:

K = normal standard deviate

• month number

G = skew coefficient
= Pearson Type III deviate as defined in Equation 12-2

An equation for generating monthly streamflow is:

(12-4)

where:

K' • monthly flow logarithm, expressed as a normal standard deviate

f3 • beta coefficient, defined as b 1 ,..,S~,.jS 1,k where m is a station not equal to
k and b is the regression coefficient.

= month number for value being generated

k = station number for value being generated

n • number of interrelated stations

R = multiple correlation coefficient

Z = random number from normal standard population

12-4
 EM 1110-2-1415
5 Mar 93

For the case of a single station, this resolves to:

(12-5)

d. Note that Equation 12-5 is very similar to Equation 12-1. The differences result
from using normal standard deviates. When this is done, the regression constant, a, equals
zero, the regression coefficients, b, become beta coefficients, /3, and the standard
deviation, S, does not appear in the random component since it equals I. Note also that
one of the independent variables is the flow for the preceding month in order to preserve
the inherent serial correlation. The flow value in the original units is computed by
reversing the transformation process, i.e., from normal standard deviate to Pearson Type
III deviate, to logarithm of flow and finally flow value.
e. A step-by-step procedure for generating monthly streamflows for a number of
interrelated locations having simultaneous records is as follows:
(1) Compute the logarithm of each streamflow quantity. If a value of zero
streamflow is possible, it is necessary to add a small increment, such as 0.1
percent of the mean annual flow, to each monthly quantity before taking the
logarithm.
(2) Compute the mean, standard deviation and skew coefficient of the values
for each location and each month, using equations given in Chapter 2.
(3} For each month and location, subtract the mean from each event and divide
by the standard deviation (Equation 12-2).
(4) Transform these "standardized" quantities to a normal distribution by use of
Equation 12-3.
(5) Arrange the locations in any sequence, and compute a regression equation
for each location in turn for each month. In each case, the independent
variables will consist of concurrent monthly values at preceding stations and
preceding monthly values at the current and subsequent stations.
(6) Generate standardized variates for each location in turn for each month,
starting with the earliest month of generated data. This is accomplished by
computing a regression value and adding a random component. The random
component, according to Equation 12-5, is a random selection from a normal
distribution with zero mean and unit standard deviation, multiplied by the
alienation coefficient which is (I - R 2 )'1r.

(7) Transform each generated value by reversing the transform of Equation 12-
3 with the appropriate skew coefficient, multiplying by the standard deviation
and adding to mean in order to obtain the logarithm of streamflow.
(8) Find the antilogarithm of the value determined in step (7) and subtract the
small increment added in step ( 1). If a negative value results, set it to zero.

12-5
EM 1110-2-1415
5 Mar 93

f. It is obviously not feasible to accomplish the above computations without the use
of an electronic computer. A computer program, HEC-4 Monthly Streamflow Simulation
(51) can be used for this purpose.
12-5. Data Fill ln. Ordinarily, periods of recorded data at different locations do not
cover the same time span, and therefore, it is necessary to estimate missing values in order
to obtain a complete set of data for analysis as described above. In estimating the missing
values, it is important to preserve all statistical characteristics of the data, including
frequency and correlation characteristics. To preserve these characteristics, it is necessary
to estimate each individual value on the basis of multiple correlation with the preceding
value at that location and with the concurrent or preceding values in all other locations. A
random component is also required, as indicated in Equation 12-1.
12-6. Application In Areas of Limited Data. The streamflow generation models discussed
so far have assumed that sufficient records were available to derive the appropriate
statistics. For instance, the monthly streamflow model requires four frequency and
correlation coefficients for each of the 12 months, or 48 values for one station simulation.
A model has been developed (51) that combines the coefficients into a few generalized
coefficients for the purpose of generating monthly streamflow at ungaged locations.
(Procedures for determining generalized statistics for use in generating daily flows have
not yet been developed.) The generalized model considers the following:
- season of maximum runoff
- lag to season of minimum runoff
- average runoff
- variation between maximum and minimum runoff
- standard deviation of flows
- interstation and serial correlations of flows

12-7. Daily Streamflow Model.

a. Generation of daily streamflows can be accomplished in a manner very similar to
the generation of monthly streamflow quantities. Although a computer program has been
prepared for this purpose, it is capable only of generating flows at a single location and
does not provide a totally satisfactory hydrograph. Since it is desired in many reservoir
operation studies to use a monthly interval most of the time, and to perform daily
operation computations for only a few critical periods, the program has been designed to
generate daily flows after the monthly total runoff has been generated by another
program. Flows for any particular day are correlated with flows for the preceding day
and for the second antecedent day.
b. A procedure that will give a reasonable shaped hydrograph, as well as coordinated
hydrographs at many locations in a basin, would consist of (I) stochastic generation of

12-6
 EM 1110-2-1415
5 Mar 93

precipitation over the basin, and (2) using a precipitation- runoff model to derive the
resulting streamflow.

12-8. Reliability. While the simulation of stochastic processes can add reliability in
hydrologic design, the techniques have not yet developed to the stage that they are
completely dependable. All mathematical models are simplified representations of the
physical phenomena. In most applications, simplifying assumptions do not cause serious
discrepancies. It is important at this "state of the art," however, to examine carefully the
results of hydrologic simulation to assure that they are reasonable in each case.

12-7
 EM 1110-2-1415
5 Mar 93

APPENDIX A

SELECTED BIBLIOGRAPHY

A. REFERENCES AND TEXTBOOKS

1. Beard, L. R., "Statistical Methods in Hydrology," U.S. Army Corps of Engineers,

Sacramento, CA, January 1962.

2. Beard, L. R., "Flood Flow Frequency Techniques," Technical Report CRWR-119,

Center for Research in Water Resources, Austin, TX, October 1974.

3. Benjamin, J. R. and C. A. Cornell, Probability. Statistics. and Decision for Civil

Engineers, McGraw-Hill Book Co., New York, NY, 1970.
4. Crippen, J. R., "Composite Log-Type III Frequency-Magnitude Curve of Annual
Floods," U.S. Geological Survey Open-File Report 78-352, Reston, VA, 1978.

5. Dalrymple, T., "Flood-Frequency Analysis," U.S. Geological Survey Water-Supply

Paper 1543-A, GPO, Washington, D.C., 1960.

6. Dixon, W. J. and F. J. Massey, Jr., Introduction to Statistical Analysis, 3rd ed.,

McGraw-Hill Book Co., 1969.

7. Draper, N. R. and H. Smith, Applied Regression Analysis, John Wiley & Sons, Inc.,
1966.
8. Ezekiel, M. and Karl A. Fox, Methods of Correlation and Regression Analysis, 3rd
ed., John Wiley and Sons, 1959.

9. Frederick, R. H., et. al., "Five-To Sixty-Minute Precipitation Frequency for the
Eastern and Central U.S.," NOAA Technical Memorandum NWS Hydro- 35, Silver
Spring, MD, June 1977.

10. Frederick, R. H., "Interduration Precipitation Relations for Storms-Southeast States,"

NOAA Technical Report NWS 21, Silver Springs, MD, March 1979.

11. Haan, C. T., Statistical Methods jn Hydrology, The Iowa State University Press,
Ames, IA, 1977.

12. Hardison, C. H., "Storage Analysis for Water Supply," Chapter B2 of Hydrologic
Analysis and Interpretation, Techniques of Water-Resources Investigations, U.S.
Geological Survey, GPO, Washington, D.C., 1973.

13. Hazen, A., "Storage to be Provided in Impounding Reservoirs for Municipal Water
Supply", Trans. ASCE 77, p 1539-1559, 1914.

14. Hershfield, D. M., "Rainfall Frequency Atlas of the U.S.," Technical Paper No. 40,
Weather Bureau, U.S. Department of Commerce, Washington, D.C., May 1961.

A-I
 EM 1110-2-1415
5 Mar 93

15. Inter-Agency Committee on Water Resources," Methods of Flow Frequency

Analysis," Bulletin 13, Subcommittee on Hydrology, Washington, D.C., April 1966.
16. James, G. and R. C. James, Mathematics Dictionary, 3rd ed., D. Van Nostrand Co.,
Inc., Princeton, NJ, 1968.
17. Kite, G. W., "Flood Frequency and Risk," Inland Waters Directorate, Ottawa, Canada,
1974.
18. Langbein, W. B., and K. T. lseri," General Introduction and Hydrologic Definitions,"
U.S. Geological Survey Water-Supply Paper 1541-A, GPO, Washington, D.C., 1960.
19. Massey, B. C. and E. E. Schroeder," Application of a Rainfall-Runoff Model in
Estimating Flood Peaks for Selected Small Natural Drainage Basins in Texas," U.S.
Geological Survey Open- File Report 77-792, Reston, VA, December 1977.
20. McCuen, R. H., et al., "Flood Flow Frequency for Ungaged Watersheds - A
Literature Evaluation," ARS, U.S. Department of Agriculture, Beltsville, MD,
November 1977.
21. Miller, J. F., "Two- to Ten-Day Precipitation for Return Periods of 2 to 100 Years in
the Contiguous U.S.," Technical Paper No. 49, Weather Bureau, U.S. Department of
Commerce, Washington, D.C., 1964.
22. Miller, J. F., "Two- to Ten-Day Rainfall for Return Periods of 2 to 100 Years in the
Hawaiian Islands," Technical Paper No. 51, Weather Bureau, U.S. Department of
Commerce, Washington, D.C., 1965.
23. Miller, J. F., "Two- to Ten-Day Precipitation for Return Periods of 2 to 100 Years in
Alaska," Technical Paper No. 52, Weather Bureau, U.S. Department of Commerce,
Washington, D.C., 1965.
24. Miller, J. F., et al., "Precipitation-Frequency Atlas of the Western U.S.," NOAA Atlas
2, Volume I. - Montana: Volume II.- Wyoming: Volume III. - Colorado: Volume IV.
- New Mexico; Volume V. - Idaho: Volume VI. -Utah: Volume VII. - Nevada:
Volume VIII. - Arizona; Volume IX. - Washington: Volume X. - Oregon: Volume XI.
- California: Silver Spring, MD, 1973.
25. Owen, D. B., "Factors for One-Sided Tolerance Limits and for Variables Sampling
Plans," Monogram SCR-607, Sandia Corporation, available from NTIS, Springfield,
VA 22151, March 1963.
26. Riggs, H. C., "Frequency Curves," Chapture A2 of Hydrologic Analysis and
Interpretation, Techniques of Water-Resources Investigations, U.S. Geological
Survey, GPO, Washington, D.C., 1968.
27. Riggs, H. C., ."Some Statistical Tools in Hydrology," Chapter AI of Hydrologic
Analysis and Interpretations, Techniques of Water-Resources Investigations, GPO,
Washington, D.C., 1968.

A-2
 EM 1110-2-1415
5 Mar 93

28. Riggs, H. C., "Low-Flow Investigations," Chapture Bl of Hydrologic Analysis and

Interpretation, Techniques of Water-Resources Investigations, U.S. Geological
Survey, GPO, Washington, D.C., 1972.
29. Riggs, H. C., "Regional Analysis of Streamflow Characteristics," Chapter B3 of
Hydrologic Analysis and Interpretation, Techniques of Water Resources
Investigations, U.S. Geological Survey, GPO, Washington, D.C., 1973.

30. Sauer, V. B., W. 0. Thomas, Jr., V. A. Stricker and K. V. Wilson, "Flood

Characteristics of Urban Watersheds in the United States," U.S. Geological Survey
Water Supply Paper 2207, GPO, Washington, D.C., 1983.

31. Searcy, J. K., and C. H. Hardison, "Double-Mass Curves," U.S. Geological Survey
Water-Supply Paper 1541-B, GPO, Washington, D.C., 1960.

32. Searcy J. K., "Flow-Duration Curves," U.S. Geological Survey, Water-Supply Paper
1541-A, GPO, Washington, D.C., 1960.

33. Sokolov, A. A., et al., Flood Flow Computation. Methods Compiled from World
Experience, The UNESCO Press, Paris, France, 1976.
34. U.S. Army Corps of Engineers, "Stream Flow Volume-Duration-Frequency Studies,"
CW-152 Technical Report No. I, Washington District, Washington, D.C., June 1955.
35. U.S. Army Corps of Engineers," Frequency of New England Floods," CW-151
Research Note No. I, Sacramento District, Sacramento, CA, July 1958.

36. U.S. Army Corps of Engineers, "Probability Estimates Based on Small

Normal-Distribution Samples," CW- 151 Research Note No. 2, Sacramento District,
Sacramento, CA, March 1959.
37. U.S. Army Corps of Engineers, "Frequency Curves from Incomplete Samples,"
CW-151 Research Note No. 3, Sacramento District, Sacramento, CA, September 1959.

38. U.S. Army Corps of Engineers, "Characteristics of Low Flow

Volume-Duration-Frequency Statistics," CW-152 Technical Bulletin No. I,
Washington District, Washington, D.C., June 1960.
39. U.S. Army Corps of Engineers, "Hydrologic Data Management, "Volume 2 of
Hydrologic Methods for Water Resources Development, the Hydrologic Engineering
Center, Davis, CA, April 1972.

40. U.S. Army Corps of Engineers, "Regional Frequency Study, Upper Delaware and
Hudson River Basins, New York District," Hydrologic Engineering Center, Davis,
CA, November 1974.

41. U.S Army Corps of Engineers, "Hydrologic Frequency Analysis," Volume 3 of

Hydrologic Methods for Water Resources Development, the Hydrologic Engineering
Center, Davis, CA, April 1975.

42. U.S. Army Corps of Engineers, "Reservoir Yield," Volume 8 of Hydrologic

Engineering Methods for Water Resources Development, the Hydrologic Engineering
Center, Davis, CA, June 1977.

A-3
 EM 1110-2-1415
5 Mar 93

43. U.S. Army Corps of Engineers, "Generalized Skew for State of New Jersey," Special
Project Memo 480, Hydrologic Engineering Center, Davis, CA, December 1977.
44. U.S. Army Corps of Engineers, "Hydrologic Analysis of Ungaged Watersheds with
HEC-1 ," Hydrologic Engineering Center, Davis, CA, April 1982.
45. U.S. Army Corps of Engineers, Engineering Manual EM 1110-2-1413, "Hydrologic
Analysis of Interior Areas," January 1987.
46. Water Resources Council, "Guidelines for Determining Flood Flow Frequency,"
Bulletin 17B, Hydrology Committee, Washington, D.C .• March 1982.
47. Water Resources Council, "A Uniform Technique for Determining Flood Flow
Frequencies," Bulletin 15, Hydrology Committee, Washington, D.C .• December 1967.
48. World Meteorological Organization, "International Glossary of Hydrology," First
Edition, WMO No. 385, Geneva, Switzerland, 1974.
49. World Meteorological Organization, "Guide to Hydrometeorological Practices," WMO
No. 168, Third Edition, Geneva, Switzerland, 1974.

B. COMPUTER PROGRAMS
50. U.S. Army Corps of Engineers, "Multiple Linear Regression," Computer Program
704-G l-L2020, Hydrologic Engineering Center, Davis, CA, September 1970.
51. U.S. Army Corps of Engineers, "HEC-4, Monthly Streamflow Simulation," Computer
Program 723-X6-L2340, Hydrologic Engineering Center, Davis, CA, February 1971.
52. U.S. Army Corps of Engineers, "Regional Frequency Computation," Computer
Program 723-X6-L2350, Hydrologic Engineering Center, Davis, CA, July 1972.
53. U.S. Army Corps of Engineers, "Flood Flow Frequency Analysis, "Computer Program
723-X6-L 7550, Hydrologic Engineering Center, Davis, CA, February, 1982.
54. U.S. Army Corps of Engineers, "Expected Annual Flood Damage Computation,"
Computer Program 761-X6-L7580, Hydrologic Engineering Center, Davis, CA, June
1977.
55. U.S. Army Corps of Engineers," HEC-5, Simulation of Flood Control and·
Conservation Systems." Computer program 723-X6-L2500, Hydrologic Engineering
Center, Davis, CA, April 1982
56. U.S. Army Corps of Engineers, "HEC-1, Flood Hydrograph Package," Computer
Program 723-X6-L2010, Hydrologic Engineering Center, Davis, CA, September
1981.

A-4
 EM 1110-2-1415
5 Mar 93

57. U.S. Army Corps of Engineers, "Storage, Treatment, Overflow, Runoff Model
'STORM'," Computer Program 723-58-L 7520, Hydrologic Engineering Center,
Davis, CA, August 1977.
58. U.S. Army Corps of Engineers, "Statistical Analysis of Time Series Data 'ST ATS' ,"
under development, Hydrologic Engineering Center, Davis, CA, May 1987.

A-5
 EM 1110-2-1415
5 Mar 93

APPENDIX B

GLOSSARY

These definitions have been collected from four major sources: (I) Guidelines for
Determining Flood Flow Frequency ( 17B), reference 46; (2) International Glossary of
Hydrology (WMO), reference 48; (3) General Introduction and Hydrologic Definitions
(USGS), reference 18; and (4) Mathematics Dictionary (MD), reference 16.

TERM DEFINITION

Analytical Frequency A predefined method of estimating the parameters that define a

Analysis selected theoretical frequency distribution.

Annual Event The most extreme event, either maximum or minimum, in the
year. (17B)
Annual Series A general term for a set of any kind of data in which each item is
the maximum or minimum in a year. (17B)
Array A list of data in order of magnitude; in flood- frequency analysis
it is customary to list the largest value first, in a low-frequency
analysis, the smallest first. ( 17B)
Autocorrelation See "Serial Correlation."
Base Discharge Usually refers to the discharge above which independent
instantaneous peak flows are collected for a partial duration
frequency analysis.
Biased The expected value of a statistic obtained from random sampling
is not equal to the parameter or quantity being estimated. (MD)
Broken Record A systematic record which is divided into separate continuous
segments because of deliberate discontinuation of recording for
significant periods of time. (17B)
Chi-Square The distribution of sample variances drawn from a normal
Distribution distribution. Used to compute confidence intervals for the
population variance estimated from a sample.
Class Interval A convenient sized interval into which data may be grouped. The
upper and lower bounds of the class interval are called "class
limits." (MD)

B-1
EM 1110-2-1415
5 Mar 93

Climatic Year A continuous 12-month period during which a complete annual

cycle occurs, arbitrarily selected for the presentation of data
relative to hydrologic or meteorologic phenomena. The climatic
year is usually designated by the calendar year during which most
of the 12 months occur. See "Water Year." (USGS)
Coefficient of A numerical measure or index of the lack of symmetry in a
Skewness frequency distribution. Function of the third moment of
magnitudes about their mean, a measure of asymetry. Also called
"coefficient of skew" or "skew coefficient." (17B)

Coefficient of Statistical parameter describing the change of a stochastic variable

Variation in time or space, expressed as the ratio of the standard deviation
to the mean. (WMO)

Confidence Limits Computed values on both sides of an estimate of a parameter that

show for a specified probability the range in which the true value
of the parameter lies. ( 17B)
Correlation The interdependence between two sets of numbers. (MD)
Covariance The first product moment of two variates normalized by their
respective mean values. (WMO)
Cumulative Frequency Relation of event magnitude to percentage of events exceeding (or
Curve not exceeding) that magnitude.
Deviation The difference between the magnitude of an event and the mean
of all the events in the sample.
Depth-Duration- Curve showing the frequency relationship of precipitation depth
Frequency for a given storm duration.
Distribution Function describing the relative frequency with which events of
various magnitudes occur. (17B)

Double Mass Curve Plot of successive accumulated values of one variable against the
contemporaneous accumulated values of another variable. (WMO)

Drought A period of abnormally dry weather sufficiently prolonged for

the lack of precipitation to cause a serious hydrological imbalance.
(WMO)

Duration Curve A cumulative frequency curve that shows the percent of time that
specific values are equalled or exceeded. (USGS)

Error Variance Square of the standard error.

Exceedance Frequency See "Percent chance exceedance."

Exceedance Interval See "Recurrence interval."

B-2
 EM 1110-2-1415
5 Mar 93

Exceedance Probability that a random event will exceed a specified magnitude

Probability in a given time period, usually one year unless otherwise
indicated. ( 17B)
Expected Probability The average of the true probabilities of all magnitude estimates
for any specified flood frequency that might be made from
successive samples of a specified size. (17B)

F Distribution The random sampling distribution of the ratio of two independent

estimates of the variance of a normal distribution. (MD)

Flow-Duration Curve See "Duration Curve."

Frequency The number of events in a sample (or the population) that meet
specified criteria.
Frequency Analysis Procedure involved in interpreting a record of events in terms of
future probabilities of occurrence. (WMO)
Frequency Curve A graphical representation of a frequency distribution. Usually a
cumulative frequency curve with the abscissa a probability grid
and the ordinate the event magnitude.
Generalized Skew A skew coefficient derived by a procedure which integrates
values obtained at many locations. ( 17B)

Geometric Mean The Nth root of the product of N values or the antilogarithm of
the mean logarithm of a set of values.
Graphical Frequency The development of a frequency curve by drawing a smooth
Analysis curve through plotted points while considering known constraints.
Plotting positions are computed based on the order number and
the total number of values represented, and then plotted on the
appropriate probability paper.

Histogram Univariate frequency diagram with rectangles proportional in area

to the class frequency, erected on a horizontal axis with width
equal to the class interval. (WMO)

Historic Data Information about significant events before or after the period of
"systematic" data collection. (derived from 17B)

Homogeneity Records (samples) from the same population. ( 17B)

Incomplete Record A streamflow record in which some peak flows are missing
because they were too low or high to record or the gage was out
of operation for a short period because of flooding. ( 17B)
Level of Significance The probability of rejecting a hypothesis when it is in fact true,
At a "10-percent" level of significance the probability is 1/10.
(17B)

B-3
EM 1110-2-1415
5 Mar 93

Log-Pearson Type Ill Application of the Pearson Type III distribution to the logarithms
Distribution of the data.
Mass Curve Curve of an accumulative quantity versus time. (WMO)
Mean The expected value of a random variable, the first moment. The
arithmic mean (or average) of a sample is an estimate of the
population mean.
Mean Daily The mean of daily values in a specified period, or the mean of
values within one day. (derived from WMO)
Median The value at which one-half of ordered observations lie on either
side. If there is no middle value, the median is the average of the
two middle values.
Method of Moments A standard statistical computation for estimating the moment of a
distribution from the data of a sample. (17B)
Mixed Populations A sample whose events have come from two or more different
populations, i.e., data not homogenous.
Mode The most frequent value of a set of numbers. (MD)
Non-Central t A distribution that combines the probable error in the mean and
Distribution the standard deviation for samples from a normal distribution.
Used in the development of confidence limit curves about a
frequency curve computed from sample statistics.
Nonstationary Not stationary with respect to time. See "Stationary Process."
Normal Distribution A probability distribution that is symmetrical about the mean,
median, and mode (bell shaped). It is the most studied
distribution in statistics, even though most data are not exactly
normally distributed, because of its value in theoretical work and
because many other distributions can be transformed into the
normal. It is also known as Gaussian, the Laplacean, the
Gauss-Laplace, or the Laplace-Gauss distribution, or the Second
Law of Laplace. (17B)
Outlier Outliers (extreme events) are data points which depart from the
trend of the rest of the data. (17B)
Parameter A characteristic descriptor of the population, such as mean or
standard deviation. Parameter estimates are called statistics.
Parent Population See "Population".
Partial-Duration Series A list of flood peaks that exceed a chosen base stage or discharge,
regardless of the number of peaks occurring in a year. (USGS)

B-4
 EM 1110-2-1415
5 Mar 93
Pearson Type III Family of asymmetrical, theoretical frequency distributions of
Distribution which the normal distribution is a special case.
Percent Chance The probability, expressed as a percentage, with which values
Exceedance exceed a specified magnitude.
Percent Chance The probability, expressed as a percentage, with which values
Non-Exceedance will not exceed a specified magnitude.

Plotting Position Percent chance of exceedance (or non-exceedance) of an observed

value estimated from its position in the array.
Population The entire (usually infinite in hydrologic application) number of
data from which a sample is taken or collected. For example,
total number of past, present, and future floods at a location on a
river is the population of floods for that location even if the
floods are not measured or recorded. (17B)
Probability The ratio of the number of random events with some particular
size or other attribute to the total number of equally likely events.
Random Any event in the population has an equal chance of being
selected.
Recurrence Interval The average time interval between actual occurrences of a
hydrological event of a given or greater magnitude. In an annual
flood series, the average interval in which a flood of a given size
is exceeded as an annual maximum. In a partial duration series,
the average interval between floods of a given size, regardless of
their relationship to the year or any other period of time. The
distinction holds even though for large floods, recurrence
intervals are nearly the same for both series. (17B)
Regional Analysis Extension of the results of the frequency analysis of point data to
an area. (WMO)
Regression An analytical procedure that derives estimation or prediction
equations for a variable (dependent) based on given values of one
or more other variables (independent). Commonly, the principle
of minimum squared error (least squares) is used in the derivation.
Return Period See "Recurrence Interval."
Risk The probability of a potential outcome (success) being realized
within a specified number of events (trials). ln a hydrologic
context, the probability that one or more events will exceed a
given annual event, that has an assumed "true" percent chance
exceedance, during a specified number of years. Risk is
computed by the binomial distribution. (For contrast, see
uncertainty.)

Sample An element, part, or fragment of a "population." Every

hydrologic record is a sample of a much longer record. ( 17B)

B-5
EM 1110-2-1415
5 Mar 93

Sampling Error The difference between a random sampling statistic and the
parameter of the population from which the random sample was
drawn. (MD)
Serial Correlation A measure of the interdependence between an observation at a
given time period and that of a preceeding time period. Also
called autocorrelation.
Skew Coefficient See "Coefficient of Skewness."
Stage The height of a water surface above an established datum plane.
(USGS)
Standard Deviation A measure of the dispersion or precision of a series of statistical
values such as precipitation or stream flow. It is the square root
of the sum of squares of the deviations from the arithmetic mean
divided by the number of values or events in the series. It is now
standard practice in statistics to divide by the number of values
minus one in order to get an unbiased estimate of the variance
from the sample data. (17B)
Standard Error Standard deviation of the sampling distribution of a statistical
parameter. (WMO)
Stationary Process All of the generating moments of the frequency distribution
remain fixed with respect to time.
Statistic An estimate of a population parameter obtained from a sample of
the population.
Stochastic Process Process in which both the probability and the sequence of
occurrence of the variables are taken into account. (WMO)
Student's t- A distribution used in evaluation of variables which involve
Distribution sample standard deviation rather than population standard
deviation. ( 17B)
Systematic Record Information collected by a systematic data collection program.
(derived from 17B)
t-Distribution See "Student's t-Distribution."
Test of Significance A test mode to learn the probability that a result is accidental or
that a result differs from another result. For all the many types
of tests, there are standard formulae and tables. ( 17B)
Transformation The change of numerical values of data to make later
computations easier, to linearize a plot or to normalize a skewed
distribution by making it more nearly a normal distribution. The
most common transformations are those changing ordinary
numerical values into their logarithms, square roots or cube roots;
many others are possible. (17B)

B-6
 EM I I 10-2-1415
5 Mar 93

Trend A statistical term referring to the direction or rate of increase or

decrease in magnitude of the individual members of a time series
of data when random fluctuations of individual members are
disregarded. (USGS)
Unbiased The expected value of a statistic obtained from random sampling
is equal to the parameter or quantity being estimated. (MD)

Uncertainty The inherent error in an analysis caused by not knowing either

the true model or the model parameters. In frequency analysis,
model uncertainty comes from assuming a theoretical frequency
distribution and parameter uncertainty comes from estimating the
parameters for the selected distribution by sample statistics.

Variance A measure of the amount of spread or dispersion of a set of

values around their mean, obtained by calculating the mean value
of the squares of the deviations from the mean, and hence equal
to the square of the standard deviation. ( 17B)

Watershed The divide separating one drainage basin from another and in the
past has been generally used to convey this meaning. However,
over the years, use of the term to signify drainage basin or
catchment area has come to predominate, although drainage basin
is preferred. (USGS)

Water Year Continuous twelve-month period selected in such a way that all
solid and liquid precipitation runs off during this period. Thus,
carryover is reduced to a minimum. (WMO) In U.S. Geological
Survey reports, it is the twelve-month period. October 1 through
September 30. The water year is designated by the calendar year
within which most of the twelve-months occur. (USGS)

B-7
 EM 1110-2-1415
5 Mar 93

APPENDIX C

COMPUTATION PROCEDURE
FOR
EXTREME VALUE (GUMBEL) DISTRIBUTION

(Reproduced from reference 48.)

5.2.6.1.2 Compwatiorw.l methods

It can be shown that most frequency functions applicable to hydrological
analysis can take the form

(5.1)

where X is the mean value, and s,. is the standard deviation of the variable
being studied. Value XTr denotes the magnitude of the event reached or exceeded
on an average once in Tr years. K is the frequency factor. I£ X is not normally
distributed, K depends on frequency and skewness coefficient. A commonly used
distribution of extreme values (annual series) is the double exponential distribu-
tion, which bas been widely applied by Gumbel (see Bibliography), and often
bears his name. In this method

YTr _
K== __ -YD
_
(5.2)
Sn

C-1
EM 1110-2-1415
5 Mar 93

II YllROI.OI. tc:.\ I. A!\' .\I. Y!'iiS ;, . :! I

where Y,, the rl'cluccd mf'an, and So, the rt•Juccd standard !leviation, are functions
only of samplt· sizf'; and YT,, the rcdut·l'd variate, is rl'latcd to r·cturn period by

Y1, = - - (0.83405 + 2.3025~1 lo:z log T, T'

1
) (5. 3\

TaLl!' 5.3 a-i''''s \'&lues of K computed by means of Eq. (5.:!) using Gumbel's
values for )·n, sn, and YT.r
Therc arc two basic methods for fitting data to the extrcmc value distribution.
One consists in computation of XTr by means of Eq. (5 .1), after a previous
computation of the values of Xand Sz (Table 5. 4 ). The other consists in plotting
data on suitable graph paper, known as extreme probability paper, and drawing
a line by inspection.

TABLE 5.3
Valu,.s of K based on Eq. (5. 2)

Return period (years)

n 2 5 10 25 50 100
10 -0.1355 1.0580 1.8483 2.8467 3.5874 4.3227
11 -0.1376 1.0338 1.8094 2.7894 3.5163 4.2379
12 -0.1393 1.0134 1.7766 2.7409 3.4563 4.1664
13 -0.1408 .9958 1..7484 2.6993 3.4048 4.1050
14 -0.1422 .9806 1.7240 2.6632 3.3600 4.0517

15 -0.1434 .9672 1.7025 2.6316 3.3208 4.0049

16 -0.1444 .9553 1.6835 2.6035 3.2860 3.9635
17 -0.1454 .9447 1.6665 2.5784 3.2549 3.9265
18 -0.1463 .9352 1.6512 2.5559 3.2270 3.8932
19 -0.1470 .9265 1.6373 2.5354 3.2017 3.8631

20 -0.1478 .9187 1.6247 2.5169 3.1787 3.8356

21 -0.1484 .9115 1.6132 2.4999 3.1576 3.8106
22 -0.1490 .9049 1.6026 2.4843 3.1383 3.7875
23 -0.1496 .8988 1.5929 2.4699 3.1205 3.7663
24 -0.1501 .8931 1.5838 2.4565 3.1040 3.7466

25 -0.1506 .8879 1.5754 2.4442 3.0886 3.7283

26 -0.1510 .8830 1.5676 2.4326 3.0743 3.7113
27 -0.1515 .8784 1.5603 2.4219 3.0610 3.6954
28 -0.1518 .8742 1.5535 2.4118 3.0485 3.6805
29 -0.1522 .8701 1.5470 2.4023 3.0368 3.6665

(continued)

C-2
 EM l l 10-2-1415
5 Mar 93

5.22 HYDROLOGICAL ANALYSIS

TABLE 5. 3 (continued)
Return period (years)
n 2 5 10 25 50 100

30 -0.1526 .86G~ 1.5~10 2.3934 3.02:ii 3.ti:i34

31 -0.1529 .8628 1.5353 2.3850 3.01S3 3.11410
32 -0.1532 .8594 1.5299 2.3770 30054 3.ti292
33 -0.1535 .8562 1.5248 2.3695 2.9901 3.6181
34 -0.1538 .8532 1.5199 2.3623 2.9873 3.607()

35 -0.1540 .8504 1.5153 2.3556 2.9789 3.5976

36 -0.1543 .8476 1.5110 2.3491 2.9709 3.5881
37 -0.1545 .8450 1.5068 2.3430 2.9633 3.5790
38 -0.1548 .8425 1.5028 2.3371 2.9561 3.5704
39 -0.1550 .8402 1.4990 2.3315 2.9491 3.5622

40 -0.1552 .8379 1.4954 2.3262 2.9~25 3.5543

41 -0.1554 .8357 1.4920 2.3211 2.9362 3.5467
42 -0.1556 .8337 1.4886 2.3162 2.9301 3.5395
43 -0.1557 .8317 1.4.854 2.3115 2.9243 3.5325
44 -0.1559 .8298 1.4824 2.3069 2.9187 3.5259

45 -0.1561 .8279 1.4794 2.3026 2.9133 3.5194

46 -0.1562 .8262 1.4766 2.2984 2.9081 3.5133
47 -0.1564 .8245 1.4739 2.2944 2.9031 3.5073
48 -0.1566 .8228 1.4712 2.2905 2.8983 3.5016
49 -0.1567 .8212 1.4687 2.2868 2.8937 3.4961

50 -0.1568 .8197 1.4663 2.2832 2.8892 3.4908

51 -0.1570 .8182 1.4639 2.2797 2.8849 3.4856
52 -0.1571 .8168 1.4616 2.2763 2.8807 3.4807
53 -0.1572 .8154 1.4594 2.2731 2.8767 3.4759
54 -0.1573 .8141 1.4573 2.2699 2.8728 3.4712

55 -0.1575 .8128 1.4552 2.2669 2.8690 3.4667

56 -0.1576 .8116 1.4532 2.2639 2.8653 3.4623
57 -0.1577 .8103 1.4512 2.2610 2.8618 3.4581
58 -0.1578 .8092 1.4494 2.2583 2.8583 3.4540
59 -0.1579 .8080 1.4475 2.2556 2.8550 3.4500

60 -0.1580 .8069 1.4458 2.2529 2.8518 3.4461

61 -0.1581 .8058 1.4440 2.2504 2.8486 3.4424
62 -0.1582 .8048 1.4424 2.2479 2.8455 3.4387
63 -0.1583 .8038 1.4407 2.2455 2.8426 3.4352
6~ --{}.1583 .8028 1.4391 2.2432 2.8397 3.4317

C-3
EM 1110-2-1415
5 Mar 93

HYDROLOGICAL ANALYSIS 5.23

T.uu 5.3 (continued)

Return period (years)
n !! 5 10 25 50 100

65 -0.1584 .8018 1.4376 2.2409 2.8368 3.4284

66 -0.1585 .8009 1.4361 2.2387 2.8341 3.4251
67 -0.1586 .8000 1.4346 2.2365 2.8314 3.4219
68 -0.158i .7991 1.4332 2.2344 2.8288 3.4188
69 -0.1587 .7982 1.4318 2.2324 2.8263 3.4158

70 -0.1588 .7974 1.4305 2.2304 2.8238 3.4128

it -0.1589 .7965 1.4291 2.2284 2.8214 3.4099
n -0.1590 .7957 1.4278 2.2265 2.8190 3.4071
73 -0.1590 .7950 14266 2.2246 2.8167 3.4044
74 -0.1591 .7942 1.4254 :.!.222tS :.!.8141: 3.4017

75 -0.1592 .7934 1.4242 2.2211 2.8122 3.3991

76 -0.1592 .7927 1.4230 2.2193 2.8101 3.3965
77 -0.1593 .7920 1.4218 2.2176 2.8080 3.3940
78 -0.1593 .7913 1.4207 2.2160 2.8059 3.3916
79 -0.1594 .7906 1.4196 2.2143 2.8039 3.3892

80 -0.1595 .7899 1.4185 2.2128 2.8020 3.3868

8! -0.1595 .7893 1.4175 2.2112 2.8000 3.3845
82 -0.1596 .7886 1.4165 2.2097 2.7982 3.3823
83 -0.1596 .7880 1.4154 2.2082 2.7963 3.3801
84 -0.1597 .7874 1.4145 2.2067 2.7945 3.3779

85 -0.1597 .7868 1.4135 2.2053 2.7927 3.3758

86 -0.1598 .7862 1.4125 2.2039 2.7910 3.3738
87 -0.1598 .7856 1.4116 2.2026 2.7893 3.3717
88 -0.1599 .7851 1.4107 2.2012 2.7877 3.3698
89 -0.1599 .7845 1.4098 2.1999 2.7860 3.3678

90 -0.1600 .7840 1.4089 2.1986 2.7844 3.3659

91 -0.1600 .7834 1.4081 2.1973 2.7828 3.3640
92 -0.1601 .7829 1.4072 2.1961 2.7813 3.3622
93 -0.1601 .7824 1.4064 2.1949 2.7798 3.3604
94 -0.1602 .7819 1.4056 2.1937 2.7783 3.3586

95 -0.1602 .7814 1.4048 2.1925 2.7769 3.3569

96 -0.1602 .7809 1.4040 2.1913 2.7754 3.3552
97 -0.1603 .7804 1.4033 2.1902 2.7740 3.3535
98 -0.1603 .7800 1.4025 2.1891 2.7726 3.3519
99 -0.1604 .7795 1.4018 2.1880 2.7713 3.3503

100 -0.1604 .7791 1.4010 2.1869 2.7700 3.3487

C-4
 EM 1110-2-1415
5 Mar 93

J.:.!4 HYDROLOGICAL ANALYSIS

Extreme probability paper has a linear ordinate for the variable being
5tudied. and the abscissa is a linear scale of the reduced variate [Eq. (5.3)]. For
convenience in plotting, the warped scale of Tr is also shown along tlw top of
Fig. 5. 9. Plotting positions are commonly determined by the formulae [16):

Tr = n +1 (5 .4!
m
or
n 0.4
Tr=--~
+ (5.51
m-0.3

where n is the number of years of record (the number of items in the annual
series) and m is the rank of the item on the aeries, m being 1 for the largest.
To illustrate the steps in numerical computation of the rainfall value for a
given return period, hypothetical values of ll serie:; of annual rainfall maxima are
given in the upper part of Table 5. 4. Computations are illustrated in the lower
part of the table for Tr of 10. Rainfall depths for return periods other than
10 years can be computed in a similar manner.

EXTREME PROBABILITY PAPER

•ETU•• P!lttOO (Y[A.Sl
I 01 s '0 n so too 200
100
: 1111 I I III'I;JI II II II II II II I I IIIII

/+
10

~
I
.b:(j,
."69

I
A-66

_!;;_"I
+'BD
I I
20 f--

0
I I I I I I I I I I I
t.O ell 10 M M ft "·'
~IIOIAIIUTY (' 00~+11]
1111111111!1!!1 1 11' d II' 1 It!!' ''11'1111'1' !IIIII' dt "''II' rl l!!!ll"'i.Lw.u.u.1.l.w.J.J
-l.tl II l.tl l.tJ ~0 4.0 ~0 4.11
IIEIHICI. I'AtfiArC

Figure 5.9- Example of extreme probability plot using data of Table 5.4.

C-5
EM 1110-2-1415
5 Mar 93

IIYOROLIH;IC.\L ANALYSIS .::> ••::>

')"

TABLE ; •. 4
Computation of eztreme values

II+ 1 f'2
"'l"t>ar I' Ill 1'-P (P-I'f
m
1960 37 i 1.72 -9 81 1,369
1961 20 11 1.09 -26 676 400
1962 32 8 1.5 -14 196 1,024
1963 60 3 4.0 + 14 196 3,600
1964 25 9 1.33 -21 441 625
1965 52 4 3.0 +6 36 2,704
1966 46 6 2.0 0 0 2,116
1967 70 2 6.0 + 24 576 4,900
1968 92 1 12.0 + 46 2,116 8,464
196~ 48 5 2.4 +2 4 2,304
1970 24 10 1.2 -22 484 576
Total 506 4,806 28,082
- n 50() ,
P

Sx
= l:P/n = IT =

. .
by square or devaataons:
4u.O

vi (P)Z-
n-1
Pi: p -- ~ -- 21.92
10

I:D (P)1 - _PDI: P ~

4806
Sx by short cut: \ / n_
1
= W = 21.92
For Tr = 2 and n = 11, K = -0.1376 (from Table 5.3).
Substituting into Eq. (5 .1):
P! = 46.0-0.1376 X 21.92 = 43.0
Similarly, for Tr = 10, K = 1.8094, and
PJO = 46.0 + 1.8094 X 21.92 = 85.7
In Fig. 5.9 the two+ 's show the above values for P:: and P10 and define
the line shown.
To illustrate the graphical method of fitting data to the extreme-value
distribution, reference is again made to Table 5. 4 and Fig. 5. 9. In the table,
values of the plotting position are given, and in Fig. 5. 9 the plotted points
are given, with rainfall values for each plotting position. The curve shown could
have been drawn by 6tting the plotted points by inspection.
In this example, for convenience, a record of only 11 years is used. Such a
record gives a fairly stable value for return periods of as much as five years, but
for longer return periods the short record bas a large sampling error and the
computations should not be taken as precise estimates.

C-6
 EM 1110-2-1415
5 Mar 93

5.2G IIYDROI.fii;IC.-'L AI\ AI. \~I,;

For some types of data, instead of using the extreme-value distribution, a

better fit of the data, or a closer approach to linearity, may be obtained from
one of several other types of distribution, such as thl' normal or lognormal
Pearson Type III distribution. A commonly used distribution is one in which
the magnitude scale is logarithmic and the probability or return-period scale
is the normal distribution. This distribution and the plottin~ papl'r used with
it are widely known as log-normal. For discussion of additional distributions
and of additional methods for fitting distributions, reference may be made to
textbooks, and periodical statistical literature [17, 18, 63-66].
An advantage of fitting data to a distribution is achievement of objectivity.
This advantage has the corollary of standard treatment of data, so that a deci-
sion is based on differences in data rather than differences in subjective inter-
pretation of data. A third advantage of linearity in plotting points, and of close
fit to a particular distribution, is the facility for extrapolating beyond the range
of the data. However, it should be remembered that extrapolation involves con-
siderable sampling error.
For evaluation of the accuracy of the computed values XTr it would be
desirable to compute the confidence interval with limits:

within which, with given confidence levels, one may expect to find the true
precipitation value xT.Values of t(a) for selected confidence levels are as
follows:

a=95% t(a) = 1.960

a=90% t(a) = 1.645
a=80% t(a) = 1.282
a=68% t(a) = 1.000

In most cues, values of s., the standard error of estimate, can be computed
by means of the formula
s,.
Se = P-rr·--
'1/n
(5.6)

In particular, for the Gumbel distribution the following relation [19] exists:

fh. = 1/1 + 1.14K + 1.10K 2 (5. 7)

where K is the numerical value defined by Eq. (5. 2) and readily obtainable from
Table 5.3. However, for convenience in the use of Eq. (5.6) values of lh-./v'~
can be obtained directly from Table 5. 5. Thus, in determining the 80 per cent
~:onfidence interval for P 10 = 85.7 in Table 5.4, for example,

t(a)s. = 1.282 X 0. 7783 X 21.92 = 21.9

C-7
EM 1110-2-1415
5 Mar 93

HYDROLOGICAL ANALYSIS 5.27

The lower and upper limits of the confidence interval are therefore 85.7- 21.9
and 85.7 + 21.9, respectively, which means that there is an 80 per cent probability
that the true value or PJO lies between 63.8 and 107.6. Similarly, the lower and
upper limits of the same confidence interval for P 1 == 43.0 are 35.1 and 50.9,
respectively, the value of fh.!V~ being 0.2803 (Table 5. 5).

TABLE 5.5
Valuu of fh-./Vn for tue in Eq. (5.6)
Return period (years)
n 2 5 10 25 50 100
10 .2942 .5863 .8285 1.1472 1.3873 1.6273
11 .2803 .5522 .7733 1.0761 1.3007 1.5252
12 .2681 .5232 .7358 1.0161 1.2275 1.4389
13 .2574 .4982 .6992 .9645 1.1646 1.3648
14 .2479 .4763 .6673 .9196 1.1100 1.3005
15 .2393 .4569 .6392 .8801 1.0620 1.2439
16 .2316 .4397 .6142 .8450 1.0193 1.1937
17 .2246 .4242 .5918 .8136 .9811 1.1488
18 .2182 .4102 .5716 .7853 .9467 1.1083
19 .2123 .3974 .5532 .7596 .9155 1.0716
20 .2068 .3857 .5365 .7361 .8871 1.0382
21 .2018 .3750 .5211 .7146 .8610 1.0075
22 .1971 .3651 .5069 .6948 .8370 .9793
23 .1927 .3559 .4937 .6765 .8148 .9532
24 .1886 .3473 .4815 .6595 .7942 .9290
25 .1847 .3394 .4702 .6437 .7750 .9064
26 .1811 .3319 .4595 .6289 .7571 .8854
27 .1777 .3249 .4496 .6150 .7403 .8657
28 .1745 .3183 .4402 .6020 .7245 .8472
29 .1714 .3121 .4314 .5898 .7097 .8298
30 .1685 .3062 .4230 .5782 .6957 .8134
31 .1657 .3007 .4152 .5673 .6825 .7978
32 .1631 .2954 .4077 .5569 .6699 .7831
33 .1606 .2904 .4006 .5471 .6581 .7692
34 .1582 .2856 .3938 .5377 .6468 .7559
35 .1559 .2811 .3874 .5289 .6360 .7433
36 .1537 .2767 .3813 .5204 .6257 .7313
37 .1516 .2726 .3754 .5123 .6159 .7198
38 .1496 .2686 .3698 .5045 .6066 .7088
39 .1476 .2648 .3645 .4971 .5976 .6983

( conlinU«i)

C-8
 EM 1110-2-1415
5 Mar 93

5.28 HYDROLOGICAL A:"ALYSJ~

TABLE 5.5 (continued)

Return period (years)

II 2 5 10 25 50 100

40 .1457 .:!ott .3:J!I:l .4900 .58!:10 .fi883

41 .143!1 .257fi .:154'. .4832 .5808 .liiRf.
42 .1422 .:1:,4:1 .34!11i .4 7611 Si2H . fi(i!J.'3
43 .1405 .2510 .3451 .4703 .5o5:1 .()()04
44 .138!1 .2479 .3407 .4o43 .5580 .6518

45 .1373 .2449 .3365 .4584 .5509 .643()

4(; .1358 .2419 .3324 .4528 .::i441 .635()
47 .1344 .23!:11 .3284 .4474 .5375 .ti279
48 .1330 .2364 .3241i .4431 .5312 .6205
4!1 .1316 .2338 .3209 .4370 .5251 .6133

50 .1303 .2312 .3174 .4321 .5192 .6064

51 .1290 .2288 .3139 .4274 .5134 .5996
52 .1277 .2264 .3106 .4228 .5079 .5931
53 .1265 .2241 .3073 .4183 .5025 .5868
54 .1253 .2218 .3042 .4140 .4973 .5807

55 .1242 .2196 .3012 .4098 .4922 .5748

56 .1230 .2175 .2982 .4057 .4873 .5690
57 .1219 .2155 .2953 .4018 .4825 .5635
58 .1209 .2135 .2925 .3979 .4779 .5580
59 .1198 .2115 .2898 .3942 .4734 .5527

60 .1188 .2096 .2872 .3905 .4690 .5476

61 .1179 .2078 .2846 .3870 .4647 .5426
62 .1169 .2060 .2821 .3836 .4606 .5377
63 .1160 .2042 .2796 .3802 .4565 .5330
64 .1150 .2025 .2772 .3769 .4525 .5284

65 .114:1 .2008 .2749 .3737 .4487 .5239

66 .1133 .1992 .2726 .3706 .4449 .5195
67 .1124 .1976 .2704 .3676 .4413 .5152
68 .1116 .1960 .2683 .3646 .4377 .5110
69 .1108 .1945 .2GG1 .3617 .4342 .5069

70 .1100 .1930 .2G41 .3589 .4208 .5029

71 .1092 .1911i .2621 .3561 .4274 .4990
,_
~.,

.1084 .1902 .2001 .3534 .4242 .4952

73 .1077 .1888 .2582 .3507 .4210 .4914
74 .1070 .1874 .2563 .3481 .4179 .4878

C-9
EM 1110-2-1415
5 Mar 93

II YllROI.III;IC.\1. .\:-;AL YSIS 5.:!!l

TABLE 5. 5 (continued)
Return period (years)
•I
5 i•J 25 50 100
"
/;} .lOll:.! .18tit .:.!544 .J45G .4148 .484:.!
j{i .to:>:) .1848 .2526 .3431 .4118 .480i
ii .1041'\ .1835 .2508 .3407 .408!.1 .4ii3
78 .104:.! .1823 .2491 .3383 .4060 .473B
79 .103:> .1810 .2474 .3300 .4032 .4 7011

80 .1028 .1798 .2457 .3337 .4005 .41li4

81 .1022 .178{; .2441 .3315 .3978 .4G43
82 .1016 .li75 .2425 .3293 .3951 .4612
83 .1010 .1764 .2409 .3271 .3925 .4581
84 .10C4 c~·)
.2394 .3250 .3900 4e::::')
. ·~- • V!J-

85 .0998 .1742 .2379 .3229 .3875 .4522

86 .0992 .1731 .2364 .3209 .3850 .4494
87 .0986 .1i20 .2349 .3189 .3826 .4466
88 .0980 .1710 .2335 .3169 .3803 .4438
89 .0975 .1700 .2321 .3150 .3780 .4411

90 .0969 .1690 .2307 .3131 .3757 .4384

91 .0964 .1680 .2293 .3113 .3734 .4358
92 .0959 .1670 .2280 .3094 .3712 .4332
93 .0954 .1661 .2267 .3076 .3691 .4307
94 .0948 .1652 .2254 .3059 .3670 .4282

95 .0943 .1642 .2241 .3041 .3649 .4258

96 .0939 .1633 .2229 .3024 .3628 .4234
97 .0934 .1624 .2217 .3007 .3608 .4210
98 .0929 .1616 .2204 .2991 .3588 .4187
99 .0924 .1607 .2193 .2975 .3569 .4164

100 .0919 .1599 .21~1 .2959 .3549 .4142

An example of the magnitude of error in extrapolation beyond the range

of the data may be found in the record of maximum annual 24-hour rainfall at
Hartford, Connecticut, U.S.A. Based on the 50 years of record through 1954,
the 100-year value was found to be 155 mm. The maximum event during this
period was 170 mm. In 1955 a hurricane produced 307 mm in 24 hours. The
computation of 100-year 24-hour rainfall based on the 51 years of record through
1955 resulted in a new estimate of 218 mm, a 40 per cent increase. Even the
10-year value was increased substantially by this one event.

C-10
 EM 1110-2-1415
5 Mar 93

5.30 HYDROLOGICAL ANALYSIS

It may happen, however, that during a definite period of Tr years, pre-

cipitation of the m2gnitude P ~ PTr does not occur at all, or that it occurs several
times. The probability that, during a given period of t years, a respective
phenomenon will occur n times, is equal to

Pratt= (n! (tt~ n)!) pn (1- p)'--n (5.8)

where p = 1/Tr. Assuming, for example, that t = T, = 100 years, then the
probabilities for various values of n are:
n 0 1 2 3 4 5
Pranoo 0.366 0.370 0.185 0.061 0.015 0.003

The overall probability of P-r, or greater event occurring in t years is

discussed in Sec. A . 5. 7. 3.

C-11
 EM 1110-2-1415
5 Mar 93

APPENDIX D

HISTORIC DATA 1
Flood information outside that in the systematic record can often be
used to extend the record of the largest events to a historic period much
longer than that of the systematic record. In such a situation, the follow-
ing analytical techniques are used to compute a historically adjusted log-
Pearson Type III frequency curve.
1. Historic knowledge is used to d~fine the historically longer period
of "H" years. The number "Z" of events that are known to be the largest in
the historically longer period "H" are given a weight of 1.0. The remaining
"N" events from the systematic record are given a weight of (H-Z)/(N+L) on the
assumption that their distribution is representative of the (H-Z) remaining
years of the historically longer period.
2. The computations can be done directly by applying the weights to
each individual year's data using equations 6-1, 6-2a, 6-3a, and 6-4a.
Figure 6-1 is an example of this procedure in which there are 44 years of
systematic record and the 1897, 1919 and 1927 floods are known to be the
three largest floods in the 77 year period 1897 to 1973. If statistics have
been previously computed for the current continuous record, they can be
adjusted to give the equivalent historically adjusted values using equations
6-1, 6-2b, 6-3b, and 6-4b, as illustrated in Figure 6-2.
3. The historically adjusted frequency curve is sketched on logarithmic-
probabiiity paper through points established by use of equation 6-5. The
individual flood events should also be plotted for comparison. The histor-
ically adjusted plotting positions for the individual flood events are
computed by use of equation 6-8, in which the historically adjusted order
number of each event 11 rii" is computed from equations 6-6 and 6-7. The com-
putations are illustrated in Figures 6-1 and 6-2, and the completed plotting
1s shown in Figure 6-3.
4. The following example illustrates the steps in application of the
historic peak adjustment only. It does not inc 1ude the fi na 1 step of
weighting with the generalized skew. The historically adjusted skew developed
by this procedure is appropriate to use in developing a generalized skew.
1
Reproduction of Appendix 6 of Bulletin 17B.

D-1
EM 1110-2-1415
5 Mar 93

DEFINITION OF SYMBOLS
E = event number when events are ranked in order from greatest magnitude
to smallest magnitude. The event numbers "E" will range from 1 to
(Z + N).

X = logarithmic magnitude of systematic peaks excluding zero flood

events, peaks below base, high or low outliers
= logarithmic magnitude of a historic peak including a high outlier
that has historic information
N = number of X's
M = mean of X's
,
M = historically adjusted mean
.,
m = historically adjusted order number of each event for use in formulas
to compute the plotting position on probability paper
s = standard deviation of the X's
.,
s = historically adjusted standard deviation

-G
G = skew coefficient of the X's
= historically adjusted skew coefficient
K = Pearson Type III coordinate expressed in number of standard devia-
tions from the mean for a specified recurrence interval or percent
chance
Q = computed flood flow for a selected recurrence interval or percent
chance
.,
pp = plotting position in percent
..,
p = probability that any peak will exceed the truncation level (used
in step 1, Appendix 5)
z = number of historic peaks including high outliers that have historic
i nfonna ti on
H = number of years in historic period
l = number of low values to be excluded, such as: number of zeros,
number of incomplete record years (below measurable base), and low
outliers which have been identified
a =constant that is characteristic of a given plotting position formula.
For Weibull formula, a= 0; for Beard formula. a = 0.3; and for
Hazen formula, a = 0.5
w = systematic record weight
D-2
 EM 1110-2-1415
5 Mar 93

EQUATIONS

w = t!....:..1.
N+ L ( 6-1)

- _ WlX +lXz
M- H-WL (6-2a)
- 2 ~ 2
-2 Wl (X - M) + l (X - M}
s -- {H-WL-1)
z (6-3a)

G = H-WL
(H-WL-1) (H-WL-2)
[Wl(X- M) +l(Xz- M)
3

s
3
J (6-4a)

~ WNM +l X
M - z (6-2b)
- H-WL
2 - 2 - 2
_2 W (N - 1 )S + WN (M - M) + l (Xz - M)
S = (H-WL-1) (6-3b)

G = H-WL [w(N- 1)N(N- 2)S3G + 3W (N- 1) (M- M)S2

(H-WL-1) (H-WL-2)5
3
- - + WN (fl- ii) +1 (Xz -M>'] (6-4b)

Log Q = M + KS (6-5)
rii = E; when: 1 ~ E ~ Z (6-6)
rii =WE- (W- 1) (Z + 0.5); when: (Z +1) ~ E ~ (Z + N+L) (6-7)

pp = rii - a
H + 1 - 2a 100 (6-8)

D-3
EM 1110-2-1415
5 Mar 93

Fii!W'! n-1 HISTORICALLY WEIGHTED LOG PEARSON TYPE !II ·ANNUAL PEAKS
Station: 3-6065, Bic s.u.dy Riwr at Bruceton, TN. D.A. 205 oq...,... mil.,.
Record: 1897, 1919. !927, 1930-1973 <H,......!
Hiftoncal penod: 1897-1973 (77 ,.....,
N • 44; Z • 3; H • 77

Q (c{s) LocY Deputure Weipt EftJit Weilbtecf Plotting

.y (rom NIIJI\ber ordar pae1tion

....
Yaar
-x -lotr -w •E Nlllllber (Weibu.ll!
pp
s•OC-ill
1897 25,000 4.39794 0.68212 1.00 I 1.00 1.28
1919 21,000 4.32222 0.60640 1.00 2 2.00 2.56
1927 18,600 4.26717 0.65136 1.00 3 3.00 3.85

1935 17,000 4.23045 0.51464 1.68182 4 4.34 5.66

1937
1946
13,800
12.000
4.13988
4.07918
0.42407
0.36337
- 5
6
6.02
7.71
7.72
9.88
1972 12.000 4.07918 0.36337 7 9.39 12.04
1956 11.800 4.07188 0.35607 8 11.07 14.19

1942 10,100 4.00432 0.28851 9 12.75 16.35

1950 9.880 3.99475 0.27895 10 14.43 18.50
!930 9,100 3.95904 0.7A323 11 16.12 20.67
1967 9.060 3.95713 0.24132 12 17.80 22.82
1932 7,820 3.89321 0.1'1'740 13 19.48 24.97

1973 7,640 3.88309 0.16728 14 21.16 27.13

1962 7,480 3.87390 0.15809 15 22.84 29.28
1965 7,180 3.86612 0.14031 16 24.63 31.45
N 26.21 33.60
1936 6,740 3.82M6 0.11285 17
6,130 3.78746 0.07165
~ 18 27.89 35.76
1948 CIC
cc
1939
1945
!934
195:0
5,940
5.630
5,580
5,480
3.'1'7379
3.75051
3.74&63
3.73878
0.05798
O.O:W70
0.03082
0.02297
..
.....

....
>::::
c;;
19
20
21
22
29.57
31.25
32.94
34.62
37.91
40.06
42.23
44.38
1944 5,340 3.72754 0.01173 23 36.30 46.54
r-'
1951 5,230 3.71850 0.00219 !:: 24 37.98 48.69
1957 5,150 3.71181 ~.00400 25 39.66 50.65
II
1971 5,080 3.70586 ~-00995 26 41.35 53.01
:z; 43.03 55.17
1953 5,000 3.69897 ~.011184 27
1949 4,740 3.675'78 ~.04003
N 28 44.71 57.32

!970 4,330 3.63649 ~.07ll32

; 29 46.39 69.47
1938 4,270 3.63043 ~.06536 II 80 48.07 61.62
!952 4,260 3.62941 ~.06640 31 49.76 63.79
1947 3,980 3.591188 ~.11613 == 32 51.44 65.95
1943 3,780 3.5'1'749 ~.13832 33 53.12 68.10

1961 3.770 3.5'7634 ~-1~'7 34 54.80 70.25

1958 3350 3.62504 ~.11107'7 35 56.49 72.42
1954 3,320 3.62114 ~.19467 36 58.1'7 74.58
1933 3,220 3.60786 ~.20716 37 69.65 76.73
1964 3,100 3.49136 ~.22445 38 61.53 78.88

1968 3,080 3.481166 ~.22726 39 63.21 81.04

1969 2,800 3.44716 ~.268&5 40 64.90 83.21
1963 2,740 3.""5 ~.27806 41 66.68 65.36
1959 2,400 3.38021 ~.33580 42 68.26 87.51
1931 2,060 3.31387 ~.40194 43 69.94 89.67

1966 1,920 3.28330 ~.43251 44 71.62 91.82

1940 1,680 3.22531 ~.49050 45 73.31 93.99
1960 1,460 3.16435 ~.65146 L-... 46 74.99 96.14
1941 1,200 3.07918 ~.63663 1.68182 47 76.67 98.29

D-4
 EM 1110-2-1415
5 Mar 93

figure 6-1. HI5TORICALLY W[IGHT[O LOG PEARSO~·TYPE II! - AN~UAL PEAKS (Continued)

5o1vlny (Eq.~ Solving (Eg. 6-3a)

lX • 162.40155 I.x 2 = 3.09755
lj I.~= 273.13018 wlx 2 = 5.20952
I.xz= 12.98733 r.zl = -1.13705
-
286.11751 6.34657
M = 286.11751/77 = 3.715~1 = 6.34657/(77- 1) = 0.08351
s3 = 0.02413
Solving (Eg. 6-4a)
lx 3 = -0.37648 (77) {0.07485)
3
lollx = -0.63317 G=-------
Ix;
= 0.70802
(76) (75) (0.02413)
0.07485
Solving (Eg. 61 Page 13)
N• 77
A = -0.33 + 0.08 {0.0418) • -0.32666
B• 0.94 - 0.26 (0.0418) • 0.92913

15 EG • 10 [-0.32666- 0.92913[0.88649)). 10[-1.150325) • 0.070 74

Solving !Eg. 9.51 Page 12)
O.JO~(C.0418) + 0.07074(-0.2)
G• • -0.00409
w
.302 + 0.07074

Solving (Eg. 6-5)

Q
:0 K (S) (K) H+ (~ (K) • Log Q
Gw • -0. 00409 ~ •. 28898 H • 3.71581 (ft
3;s)

99 -2.32934 -0.67313 3.04269 11103

95 -1.64599 -0.47566 3.24014 1 1738
90 -1.28196 -0.37046 3.34535 21215
80 -0.84141 -0.2431~ 3.47266 21969
50 0.00067 0.00019 3.71600 51200
20 0.84180 0.24326 3.95907 91100
10 1.28110 0.37021 4.08602 121190
4 1.74929 0.50551 4.22132 161646
2 2.05159 0.59289 4.30868 201355
1 2.32340 0.67142 4.38723 241391
.1 3.08455 0.89138 4.60719 401475
.01 3.71054 1.07227 4.78808 61.387

Solving ([g. 6-61 Solving !Eg. 6-7)

z• 3 (Z + 1) = 4
For E • 1 ; iii • E = 1 (Z + N) = 47
For E • 2; ~ • E = 2 For 4:!: E:!': 47:
For E • 3; m• E = 3 iii • ( 1.682) (E) - (0.682) (3. 5)
Solving (Eq. 6-8) iii= (1.682) (E) - 2.387
~or Weibull: a • 0. PP = (100) (m);(7b)

D-5
EM 1110-2-1415
5 Mar 93

Figure 6-2. HISTORICALLY WEIGHTED LOG-PEARSO~ TYPE III -ANNUAL PEAKS

Results of Standard Computation for the Current Continuous Record
Big Sandy River at Bruceton, TN. DA- 205 square miles
#3-6065 ( 44 years)
N = number of observations used = 44
t-1 = mean of logarithms = 3.69094
S = standard deviation of logarithms = 0.26721
s2 = o.o7140 s3 = o.Ol908
G = coefficient of skewness (logs) =-0.18746

Adjustment to Historically Weighted 77 Years

Historic Peaks (Z = 3 Years)

I

Year Y
2
3
(ft ;s) Log v = x lx -
z z I z
M l
I z
(X - M)2

1897 25,000 4.39794 ,0.68213 I 0.46531 0.31740

1919 21,000 4.32222 0.60641 0.36774 0.22300
1927 18,500 4.26717 I 0.55136 0.30400 0.16762
I
Summation 12.98733 1.83990 1.13705 0.70802

N=44 Z•3 H=77

Solving (Eg. 6-1): W= (77-3)/44 • 1.68182
Solving (Eg. 6-2b): M.
(1.68182) (44) (3.69094) + (12.98733) = 3 •71581
77
Solving (Eg. 6-3b):
(M- M) = -0.02487, (M- H) 2 = 0.000619; (M- M) 3 = -0.0000154
52 .. {1.68182)(43)(o.o714o> + <1.6~~82)(44)(o.ooo619) + <1.13705) .. 0 . 08351
s2 .. o.o8351 s = o.28898 s3 .. o.o2413
Solving (Eq. 6-4b):
G_ n [ (1.68182) (43l {42) co.ol908l (-o.l8746) +
- (76) (75) (0.02413) 44
(3}(1.68182)(43)(-0.02487)(0.07140) + (1.68182)(44)(-0.0000154) + (0.70802)]
G= o.0418
D-6
 IOD.ooo1 - - - ,1 ,--.--,-,---,---.----r---r----,r--r-,---,--,---.---,--..,--.--
- - - - - - - t - - ' - - ' - - - L - - L - - . L . - _ ___ t __ _.l_I_..J___
I_ I -f_--r--- -- -- -- -- -
_...]___!
----- r:~-~~
: I:.
~-
EXPLANATION • --1----· - - - -
0 Historical peaks (1897, 1919 and 1927) plotted as
largest in 77 year period (1897-1973).
L__-t-~--:~~=:=------t-r-----t---_-_-----+----·--i------ --·----· ; /
:--
r-
• Recorded peaks (44 years, 1930-1973) plotted I

such that each point represents 1.68 years

in the longer 77 year period.
Points plotted by Weibull Plotting Position formula.
---- 1-- ---
- Weighted Log Pearson Type Ill computed curve.
N=44. Z=3. H=77 G= 0.04
.,.y
N

---- M=3.71581 Gw=-0.004

~ ~--- -- ,_ .-.--y--- - -
""5=0.289
- . -- --~-- -t-- -~-- ·: v .I ;
0
~ ~
~ 10000!-----

-
r----- -·-·· r---·· - - - - - - - - -
-·~· --·- - :J;·~-.
,:':
. . I i
I

0 - ---1--- t-·-~ - - ---- - ; - - i -

l--------11----t--+----t--+---+--1-----+-·---r----- -+---- - -· --+·- !

t-------_ - - - - - ~_,-~-----_ -~--~.:!-~=:(~,. -- ~-~!G ~ .~,;~I

--_--+----_---_-,,'--
--4 __ >- ..-., 1..",;,., !. .
~ ~ t--- __ ANNUAl P[AI\S t--t------
HISfORICAllY WEIGHTED LOG
PEAR~PN TYPE Ill
/ia FIGURE 6-J
---- - --- r- -r----4---+---~E--f'L---t----+--+- -+--+-+---- -- ··-- .-

/v.-
..

- - - - + - - + - - --f----,t--:r..r+--f---+--+-+- - '------ --- ---

/.
I.OO~ql,.,g,.----..,-,99L,-9-,9~9~8---=q+q~,...--±~:--·-::9':-8--9:!-,-S----}90-:----)-::0-~I0--;60;;---;~';;--0-41;:;0--t,l0:-----ti0:---~1~0-~----7--L--0:-:\--~0'-:-1~0-I--,O:i:-(I\:---
[XC£E0ANC[ PI\OBABILITY. IN PERCENT
 EM 1110-2-1415
5 Mar 93

APPENDIX E

EXAMPLES OF RELIABILITY TESTS

FOR THE
MEAN AND STANDARD DEVIATION

Mean

Analysis of 19 annual flood peaks indicates that X = 3.8876 and S = 0.4681. It is

hypothesized on the basis of a regional analysis that the true mean is 3.7782. Sample
means. where the population variance is unknown, are distributed like the t-distribution.
If I" = 3. 7782, what is the probability of obtaining an X greater than 3.8876?

For N = 19 X = 3.8876 S • 0.4681 I" ... 3.7782

t =
X-~.£
(S 2/N)y,
.. 3.8876 - 3. 7782
((0.4681 )2/19)"'

Prob (X > 3.8876) • Prob (t > 1.0 19)

From a table for the t-distribution (Appendix F-4), there is a 16.2% chance that a sample
with 18 degrees of freedom will have an X of 3.8876 or more.

Standard Deviation

Find the 90% confidence interval for the population variance from a sample for which
S2 • 14.4818 and N ""20. Sample values of variance (standard deviation squared) are
distributed like the Chi-square (x2) distribution. From a table (Appendix F-5) for the
Chi-square distribution with 19 degrees of freedom:

Exceedance
Probability Chi-square
0.95 10.117
0.05 30.144
The 90% confidence interval is computed as:

(N - 1) s2 (N-1) 2 s
<O<
X~
2
Xs

19(14.4918) 19(14.14818)
<O<
30.144 10.117

9.128 <o< 27.197

E-1
 EM 1110-2-1415
5 Mar 93

APPENDIX F

STATISTICAL TABLES

F-1
 EM 1110-2-1415
5 Mar 93

Table F-1

MEDI6N PLQTTINQ PQSITIQNS

Plotting Positions in Percent Chance Exceedance

Order . Number o! Values in Array CN> Order

lml 2 3 4 ~ 6 7 8 ~ 10 11 11: 13 ~ml
l 50.00 29.29 20.63 15.91 12.94 10.91 9.43 8.30 7.41 6.70 6.11 5.61 5.19 l
70.71 50.00 38.64 31.47 26.55 22.95 20.21 18.06 16.32 14.89 13.68 12.66 2
52 98.68 79.37 61.36 50.00 42.18 36.48 32.13 28.71 25.94 23.66 21.75 20.13 3
51 96.77 98.65 84.09 68.53 57.82 50.00 44.04 39.35 35.57 32.44 29.82 27.60 4
so 94.86 96.70 98.62 87.06 73.45 63.52 55.96 50.00 45.19 41.22 37.89 35.06 5
49 92.95 94.76 96.64 98.60 89.09 77. OS 67.87 60.65 54.81 50.00 45.96 42.53 6
48 91.04 92.81 94.65 96.57 98.57 90.57 79.79 71.29 64.43 58.78 54.04 50.00 7
47 89.13 90.87 92.67 94.55 96.50 98.54 91.70 81.94 74.06 67.56 62.11 57.47 8
46 87.22 88.92 90.68 92.52 94.43 96.43 98.50 92.59 83.68 76.34 70.18 64.94 9
45 85.31 86.97 88.70 90.50 92.37 94.32 96.35 98.47 93.30 85.11 78.25 72.40 10
44 83.40 85.03 86.72 88.47 90.30 92.21 94.19 96.27 98.44 93.89 86.32 79.87 11
43 81.50 83.08 84.73 86.45 88.23 90.10 92.04 94.06 96.18 98.40 94.39 87.34 12
42 79.59 81.14 82.75 84.42 86.17 87.98 89.88 91.86 93.93 96.10 98.36 94.81 13
41 77.68 79.19 80.76 82.40 84.10 85.87 87.73 89.66 91.68 93.79 96.00 98.32
40 75.77 77.24 78.78 80.37 82.03 83.76 85.57 87.46 89.43 91.49 93.64 95.91 98.28 40
39 73.86 75.30 76.79 78.35 79.97 81.65 83.41 85.25 87.17 89.18 91.29 93.49 95.81 39
38 71.95 73.35 74.81 76.32 77.90 79.54 81.26 83.05 84.92 86.88 88.93 91.08 93.33 38
37 70.04 7l.U 72.82 74.30 75.83 77.43 79.10 80.85 82.67 84.57 86.57 88.66 90.85 37
36 68.13 69.46 70.84 72.27 73.77 75.32 76.95 78.64 80.41 82.27 84.21 86.24 88.38 36
35 66.23 67.51 68.85 70.25 71.70 73.21 74.79 76.44 78.16 79.96 81.85 83.83 85.90 35
34 64.32 65.57 66.87 68.22 69.63 71.10 72.64 74.24 75.91 77.66 79.49 81.41 83.43 34
33 62.41 63.62 64.88 66.20 67.57 68.99 70.48 72.03 73.66 75.35 77.13 78.99 80.95 33
32 60.50 61.68 62.90 64.17 65.50 66.88 68.32 69.83 71.40 73.05 74.77 76.58 78.47 32
31 58.59 59.73 60.92 62.15 63.43 64.77 66.17 67.63 69.15 70.74 12.41 74.16 76.00 31
30 56.68 57.78 58.93 60.12 61.37 62.66 64.01 65.42 66.90 68.44 70.05 71.75 73.52 30
29 54.77 55.84 56.95 58.10 59.30 60.55 61.86 63 . .22 64.64 66.13 67.69 69.33 71. OS 29
28 52.86 53.89 54.96 56.07 57.23 58.44 59.70 61.02 62.39 63.83 65.33 66.91 68.57 28
27 50.95 51.95 52.98 54.05 55.17 56.33 57.55 58.81 60.14 61.52 62.98 64.50 66.09 27
26 49.05 50.00 50.99 52.02 53.10 54 . .22 55.39 56.61 57.89 59.22 60.62 62.08 63.62 26
25 47.14 48.05 49.01 50.00 51.03 52.11 53.23 54.41 55.63 56.91 58.26 59.66 61.14 25
24 45.23 46.11 47.02 47.98 48.97 50.00 51.08 52.20 53.38 54.61 55.90 57.25 58.67 24
23 43.32 44.16 45.04 45.95 46.90 47.89 48.92 50.00 51.13 52.30 53.54 54.83 56.19 23
22 41.41 42.22 43.05 43.93 44.83 45.78 46.77 47.80 48.87 50.00 51.18 52.42 53.71 22
21 39.50 40.27 41.07 41.90 42.77 43.67 44.61 45.59 46.62 47.70 48.82 50.00 51.24 21
20 37.59 38.32 39.08 39.88 40.70 41.56 42.45 43.39 44.37 45.39 46.46 47.58 48.76 20
19 35.68 36.38 37.10 37.85 38.63 39.45 40.30 41.19 42.11 43.09 44.10 45.17 46.29 19
18 33.77 34.43 35.12 35.83 36.57 37.34 38.14 38.98 39.86 40.78 41.74 42.75 43.81 18
17 31.87 32.49 33.13 33.80 34.50 35.23 35.99 36.78 37.61 38.48 39.38 40.31t 41.33 17
16 29.96 30.54 31.15 31.78 32.43 33.12 33.83 34.58 35.36 36.17 37.02 37.92 38.86 16
15 28.05 28.59 29.16 29.75 30.37 31.01 31.68 32.37 33.10 33.87 34.67 35.50 36.38 15
lit 26.14 26.65 27.18 27.73 28.30 28.90 29.52 30.17 30.85 31.56 32.31 33.09 33.91 14
13 24.23 24.70 25.19 25.70 26.23 26.79 27.36 27.97 28.60 29.26 29.95 30.67 31.43 13
12 22.32 .22.76 23.21 23.68 24.17 24.68 25.21 25.76 26.34 26.95 27.59 28.25 28.95 12
11 20.41 20.81 21.22 21.65 22.10 .22.57 23.05 23.56 24.09 24.65 25.23 25.84 26.48 11
10 18.50 18.86 19.24 19.63 20.03 20.46 20.90 21.36 21.84 22.34 22.87 23.42 24.00 10
9 16.59 16.92 17.25 17.60 17.97 18.35 18.74 19.15 19.59 20.04 20.51 21.01 21.53 9
8 14.69 14.97 15.27 15.58 15.90 16.24 16.59 16.95 17.33 17.73 18.15 18.59 19.05 8
7 12.78 13.03 13.28 13.55 13.83 14.13 14.43 14.75 15.08 15.43 15.79 16.17 16.57 7
6 10.87 11.08 11.30 11.53 11.77 12.02 12.27 12.54 12.83 13.12 13.43 13.76 14.10 6
5 8.96 9.13 9.32 9.50 9.70 9.91 10.12 10.34 10.57 10.82 11.07 11.34 11.62 5
4 7.05 7.19 7.33 7.48 7.63 7.79 7.96 8.14 8.32 8.51 8. 71 8.92 9.15 4
3 5.14 5.24 5.35 5.45 5.57 5.68 5.81 5.94 6.07 6.21 6.36 6.51 6.67 3
2 3.23 3.30 3.36 3.43 3.50 3.57 3.65 3.73 3.82 3.90 4.00 4.09 4.19 2
1 1.32 l 35 1,38 l. 40 l. 43 l. 46 1,50 •. ~3 •. ~6 1. 60 l. 64 l. 68 1.72 1
(m) 52 51 50 49 48 47 46 45 44 43 42 41 40 (m)

Order , Number o! Values in Arra:z!: 'Nl Order

F-2
 EM 1110-2-1415
5 Mar 93

Table F-1 (Cont)

MEDIAN PLQTTINQ PQSITIONS

Plotting Positions in Percent Chance Exceedance

Order . Number o! Values in Array (N) Order

'ml !4 15 16 17 18 19 ~0 ~! 22 23 24 25 26 tml
1 4.83 4.52 4.24 4.00 3.78 3.58 3.41 3.25 3.10 2.97 2.85 2.73 2.63 1
2 11.78 11.01 10.34 9.75 9.22 8.74 8.31 7.92 7.57 7.24 6.95 6.67 6. 42 2
3 18.73 17.51 16.44 15.50 14.65 13.90 13.22 12.60 12.03 11.52 11.05 10.61 10.21 3
4 25.68 24.01 22.54 21.25 20.09 19.05 18.12 17.27 16.50 15.80 15.15 14.55 14.00 4
5 32.63 30.51 28.65 27.00 25.53 24.21 23.02 21.95 20.97 20.07 19.25 18.49 17.79 5
6 39.58 37.00 34.75 32.75 30.97 29.37 27.93 26.62 25.43 24.35 23.35 22.43 21.56 6
7 46.53 43.50 40.85 38.50 36.41 34.53 32.83 31.30 29.90 28.62 27.45 26.37 25.37 7
8 53.47 50.00 46.95 44.25 41.84 39.68 37.74 35.97 34.37 32.90 31.55 30.31 29.16 B
9 60.42 56.50 53.05 50.00 47.28 44.84 42.64 40.65 38.83 37.17 35.65 34.24 32.95 9
10 67.37 63.00 59.15 55.75 52.72 50.00 47.55 45.32 43.30 41.45 39.75 38.16 36.74 10
11 74.32 69.49 65.25 61.50 58.16 55.16 52.45 50.00 47.77 45.72 43.65 42.12 40.53 11
12 81.27 75.99 71.35 67.25 63.59 60.32 57.36 54.68 52.23 50.00 47.95 46.06 44.32 12
13 88.22 82.49 77.46 73.00 69.03 65.47 62.26 59.35 56.70 54.28 52.05 50.00 48.11 13
14 95.17 88.99 83.56 78.75 74.U 70.63 67.17 64.03 61.17 58.55 56.15 53.94 51.89 14
95.48 89.66 84.50 79.91 75.79 72.07 68.70 65.63 62.83 60.25 57.88 55.68 15
39 98.24 95.76 90.25 85.35 80.95 76.98 73.38 70.10 67.10 64.35 61.82 59.47 16
38 95.70 98.19 96.00 90.78 86.10 81.88 78.05 74.57 71.38 68.45 65.76 63.26 17
37 93.16 95.59 98.14 96.22 91.26 86.78 82.73 79.03 75.65 72.55 69.69 67.05 18
36 90.62 92.98 95.47 98.09 96.42 91.69 87.40 83.50 79.93 76.65 73.63 70.84 19
35 88.08 90.38 92.79 95.34 98.04 96.59 92.08 87.97 84.20 80.75 77.57 74.63 20
34 85.54 87.77 90.12 92.60 95.21 97.98 96.75 92.43 88.48 84.85 81.51 78.42 21
33 83.01 85.17 87.45 89.85 92.39 95.07 97.92 96.90 92.76 88.95 85.45 82.21 22
32 80.47 82.56 84.77 87.10 89.56 92.17 94.93 97.86 97.03 93.05 89.39 86.00 23
31 77.93 79.96 82.10 84.35 86.74 89.26 91.93 94.77 97.79 97.15 93.33 89.79 24
30 75.39 77.35 79.42 81.60 83.91 86.35 88.94 91.68 94.60 97.72 97.27 93.58 25
29 72.85 74.75 76.75 78.86 81.08 83.44 85.94 88.59 91.42 94.43 97.64 97.37 26
28 70.31 72.14 74.07 76.11 78.26 80 . .53 82.9.5 8.5 . .51 88.23 91.13 94.24 97.55
27 67.77 69.54 71.40 73.36 75.43 77.63 79.95 82.42 8.5.05 87.84 90.83 94.03 97.47 27
26 6.5.23 66.93 68.72 70.61 72.61 74.72 76.96 79.33 81.86 84.55 87.43 90.51 93.81 26
2.5 62.69 64.33 66.0.5 67.86 69.78 71.81 73.96 76.24 78.67 81.26 84.03 86.99 90.16 25
24 60.16 61.72 63.37 65.11 66.95 68.90 70.97 73.16 75.49 77.97 80.62 83.46 86.51 24
23 .57.62 .59.12 60.70 62.37 64.13 65.99 67.97 70.07 72.30 74.68 77.22 79.94 82.86 23
22 .55.08 56 . .51 58.02 59.62 61.30 63.09 64.98 66.98 69.12 71.39 73.82 76.42 79.21 22
21 .52.54 53.91 .5.5.35 .56.87 58.48 60.18 61.98 63.89 65.93 68.10 70.42 72.90 75.56 21
20 50.00 51.30 52.67 54.12 55.65 57.27 58.99 60.81 62.74 64.81 67.01 69.37 71.91 20
19 47.46 48.70 50.00 51.37 52.83 54.36 55.99 57.72 59.56 61.52 63.61 65.85 68.26 19
18 44.92 46.09 47.33 48.63 50.00 51.45 53.00 54.63 56.37 58.23 60.21 62.33 64.60 18
17 42.38 43.49 44.65 45.88 47.17 48.55 50.00 51.54 53.19 54.94 56.81 58.81 60.95 17
16 39.84 40.88 U.98 43.13 44.35 45.64 47.00 48.46 50.00 51.65 53.40 55.28 57.30 16
15 37.31 38.28 39.30 40.38 41.52 42.73 44.01 45.37 46.81 48.35 50.00 51.76 53.65 15
14 34.77 35.67 36.63 37.63 38.70 39.82 41.01 42.28 43.63 45.06 46.60 48.24 50.00 14
13 32.23 33.07 33.95 34.89 35.87 36.91 38.02 39.19 40.44 41.77 43.19 44.72 46.35 13
12 29.69 30.46 31.28 32.14 33.05 34.01 35.02 36.11 37.26 38.48 39.79 41.19 42.70 12
11 27.15 27.86 28.60 29.39 30.22 31.10 32.03 33.02 34.07 35.19 36.39 37.67 39.05 11
10 24.61 25.25 25.93 26.64 27.39 28.19 29.03 29.93 30.88 31.90 32.99 34.15 35.40 10
9 22.07 22.65 23.25 23.89 24.57 25.28 26.04 26.84 27.70 28.61 29.58 30.63 31.74 9
8 19.53 20.04 20.58 21.14 21.74 22.37 23.04 23.76 24.51 25.32 26.18 27.10 28.09 8
7 16.99 17.44 17.90 18.40 18.92 19.47 20.05 20.67 21.33 22.03 22.78 23.58 24.44 7
6 14.46 14.83 15.23 15.65 16.09 16.56 17.05 17.58 18.14 18.74 19.38 20.06 20.79 6
5 11.92 12.23 12.55 12.90 13.26 13.65 14.06 14.49 14.95 15.45 15.97 16.54 17.14 5
4 9.38 9.62 9.88 10.15 10.44 10.74 11.06 11.41 11.77 12.16 12.57 13.01 13.49 4
3 6.84 7.02 7.21 7.40 7.61 7.83 8.07 8.32 8.58 8.87 9.17 9.49 9.84 3
2 4.30 4.41 4.53 4.66 4.79 4.93 5.07 5.23 5.40 5.57 5.76 5.97 6.19 2
l l. 76 1.81 1.86 1,91 1,96 2.02 2,08 2.14 2.21 2.28 2.36 2.45 2.53 1
(m) 39 38 37 36 35 34 33 32 31 30 29 28 27 (m)

Order . Number o! Values in Arra;x ~Nl Order

F-3
 EM 1110-2-1415
5 Mar 93

Table F-2

DEVIATES FOR PEARSON TYPE-Ill DISTRIBUTION

Positive Skew

Sit. .
Coe!fici~t . . . . . . . . . . . . . . . . . . . . . . ~~t Chance Exeeedance . . . .
<G> 99.0 95.0 90.0 80.0 SO.O 20.0 10.0 5.0 2.0 1.0 0.5 0.2 0.1
0.0 -2.32535 -1.64485 -1.281.55 -o.84162 0.00000 0.114162 1.281.55 1.64485 2.05375 2.32.635 2.57583 2.87816 3.09023
0.1 -2.25258 -1.61594 -1.27037 -o.84611 -o.Ol662 0.83639 1.29178 1.67279 2.10697 2.39961 2.66965 2.99978 3.23322
0.2 -2.17840 -1.58607 -1.25824 -0.84986 -o.03325 0.83044 1.3010.5 1.69971 2.15935 2.47226 2.76321 3.12169 3.37703
0.3 -2.10394 -1.55527 -1.24516 -o.85285 -o.04993 0.82377 1.30936 1.72562 2.21081 2.54421 2.85636 3.24371 3.52139
0.4 -2.02933 -1.5Z357 -1.23114 -0.85S08 -0.06651 0.81638 1.31671 1.7S048 2.26133 2.61539 2.94900 3.36566 3.66608
0.5 -1.95472 -1.49101 -1.21618 -0.85653 -0.08302 0.80829 1.32309 1.77428 2.31084 2.68572 3.04102 3.48737 3.81090
0.6 -1.88029 -1.45762 -1.20028 -0.85718 -o.09945 0.799SO 1.328SO 1.79701 2.35931 2.75514 3.13232 3.60872 3.95567
0.7 -1.80621 -1.42345 -1.18347 -o.85703 -o.11578 0.79002 1.33294 1.81864 2.40670 2.82359 3.22281 3.72957 4.10022
0.8 -1.73271 -1.38855 -1.16574 -o.85607 -o.13199 0.77986 1.33640 1.83916 2.45298 2.89101 3.31243 3.114981 4.24439
0.9 -1.66001 -1.35299 -1.14712 -o.85426 -o.14807 0.7e902 1.33889 1.858S6 2.49811 2.95735 3.40109 3.96932 4.38807

1.0 -1 . .58838 -1.31684 -1.12762 -o.8S161 -o.l6397 0.757~ 1.311039 1.87683 2.54206 3.02256 3.48874 4.08802 4.53112
1.1 -1.51808 -1.28019 -1.10726 -o.84809 -o.17968 0.74537 1.34092 1.89395 2.58480 3.08660 3.57530 4.20582 4.67344
1.2 -1.44942 -1.24313 -1.08608 -0.84369 -o.l9S17 0.73257 1.34047 1.90992 2.62631 3.14944 3.66073 4.322.63 4.81492
1.3 -1.3112.67 -1.20578 -1.06413 -o.83841 -o.21040 0.7191.5 1.33904 1.92472 2.66657 3.21103 3.74497 4.43839 4.95549
1.4 -1.3181.5 -1.16827 -1.04144 -o.83223 -o.22535 0. 70.512 1.33665 1.93836 2. 70.556 3.27134 3.82798 4.55304 5.09S05
1.5 -1.25611 -1.13075 -1.01810 -o.82516 -o.23996 0.690SO 1.33330 1.9S083 2.7432.5 3.33035 3.90973 4.66651 5.23353
1.6 -1.19680 -1.09338 -0.99418 -o.81720 -o.25422 0.67.532 1.32900 1.96213 2.77964 3.38804 3.99016 4.77875 5.37087
1.7 -1.14042 -1.05631 -o.96977 -o.80837 -o.26808 0.659.59 1.32376 1.97227 2.81472 3.44438 4.06926 4.88971 5.50701
1.8 -1.08711 -1.01973 -o.94496 -o.79868 -o.281SO 0.64335 1.31760 1.98124 2.84848 3.49935 4.14700 4.99937 5.64190
1.9 -1.03695 -o.98381 -o.91988 -o.78816 -o.29443 0.62662 1.31054 1.98906 2.88091 3.SS295 4.22336 5.10768 5.77549

2.0 -o.98995 -o.94871 -o.89464 -o.77686 -o.3068S 0.60944 1.30259 1.99573 2.91202 3.60517 4.29832 5.21461 5.90776
2.1 -o.94607 -o.91458 -o.86938 -0.76482 -o.31872 0.59183 1.29377 2.00128 2.94181 3.65600 4.37186 5.32014 6.03865
2.2 -o.90521 -o.88156 -o.84422 -o.7S211 -o.32999 0.57383 1.28412 2.00570 2.97028 3.70543 4.44398 5.42426 6.16816
2.3 -0.86723 -o.84976 -o.81929 -o.73880 -o.34063 o.SSS49 1.27365 2.00903 2.99744 3.7.5347 4.51467 5.52694 6.29626
2.4 -o.83196 -o.81927 -o.79472 -o.7249S -o.3S062 0.53683 1.26240 2.01128 3.02330 3.80013 4.58393 5.62818 6.42292
2.5 -o.79B21 -o.79015 -o.77062 -o.71067 -o.359BZ 0.51789 1.25039 2.01247 3.04787 3.84540 4.65176 5.72796 6.54814
2.6 -o.76878 -o.76242 -o.74709 -o.69602 -o.~ 0.49872 1.23766 2.01263 3.07116 3.88930 4.71815 5.82629 6.67191
2.7 -o.74049 -o.73610 -o.72422 -o.681ll -o.37640 0.47934 1.22422 2.01177 3.09320 3.93183 4.78313 5.92316 6.79421
2.8 -o.71415 -o.71ll6 -o.70209 -o.66603 -o.38353 0.4S980 1.21013 2.00992 3.11399 3.97301 4.84669 6.01858 6.91505
2.9 -o.689S9 -o.68759 -o.6807.5 -o.65086 -o.38991 0.4401.5 1.19539 2.00710 3.13356 4.01286 4.90884 6.112S4 7.03443

3.0 -o.66663 -0.66532 -o.66023 -0.63S69 -o.39SS4 0.42040 1.18006 2.00335 3.15193 4.05138 4.96959 6.20506 7.15235

Source: Appendix 3, Bulletin 17B, reference (46).

F-4
 EM 1110-2-1415
5 Mar 93

Table F-2 (Cont)

DEVIATES FOR PEARSON TYPE-JII DISTRIBUTION

Negative Skew

Sk...
CceWcimt. . . . . . . . . . . . . . . . . . . . . . . Pvcmt Qwx:e ~. . . . . . . . . . . . . .
CGl 99.0 95.0 90.0 80.0 50.0 20.0 10.0 5.0 2.0 1.0 0.5 0.2 0.1
0.0 -2.32535 -1.154485 -1.28155 -0.81ol62 0.00000 0.84162 1.28155 1.154485 2.05375 2.32535 2.57583 2.87816 3.09023
-o.1 -2.391161 -1.67279 -1.29178 -0.83639 0.01662 0.84611 1.27037 1.61.594 1.99973 2.252511 2.48187 2.75706 2.94834
-o.2 -2.47226 -1.69971 -1.30105 -o.83044 o.o3325 0.8491!6 1.2.5824 1.58607 1.94499 2.17840 2.38795 2.63672 2.80786
-0.3 -2.54421 -1.72562 -1.30936 -0.82377 0.04993 0.85285 1.24516 1.55527 1.88959 2.10394 2.29423 2.51741 2.66915
-o.4 -2.61539 -1.75048 -1.31671 -0.81638 0.06651 0.85508 1.23114 1.52357 1.83361 2.02933 2.20092 2.39942 2.53261
-0.5 -2.68572 -1.77428 -1.32309 -0.80829 0.08302 0.85653 1.21618 1.49101 1.77716 1.95472 2.10825 2.28311 2.39867
-0.6 -2.75514 -1.79701 -1.32850 -0.79950 0.09945 0.85718 1.20028 1.45762 1.72033 1.88029 2.011544 2.16884 2.26780
-0.7 -2.82359 -1.818154 -1.33294 -0.79002 0.11578 0.85703 1.18347 1.42345 1.66325 1.80621 1.92580 2.05701 2.14053
-0.8 -2.89101 -1.83916 -1.331540 -o.77986 0.13199 0.85607 1.16574 1.38855 1.60604 1.73271 1.83660 1.94806 2.01739
-o.9 -2.95735 -1.~ -1.33889 -o.76802 0.14807 0.85426 1.14712 1.35299 1.54886 1.66001 1.74919 1.84244 1.89894

-1.0 -3.02256 -1.87683 -1.34039 -o.75752 0.16397 0.85161 1.12762 1.31684 1.49188 1.58838 1.66390 1.74062 1.78572
-1.1 -3.08660 -1.89395 -1.34092 -0.74537 0.17968 0.84809 1.10726 1.28019 1.43529 1.51808 1.58110 1.154305 1.67825
-1.2 -3.14944 -1.90992 -1.34047 -o.73257 0.19517 0.84369 1.08608 1.24313 1.37929 1.44942 1.50114 1.55016 1.57695
-1.3 -3.21103 -1.92472 -1.33904 -0.71915 0.21040 0.83841 1.06413 1.20578 1.32412 1.38267 1.42439 1.46232 1.48216
-1.4 -3.27134 -1.93836 -1.33665 -0.70512 0.22535 0.83223 1.04144 1.16827 1.26999 1.31815 1.35114 1.37981 1.39408
-1.5 -3.33035 -1.95083 -1.33330 -o.69050 0.23996 0.82516 1.01810 1.13075 1.21716 1.25611 1.28167 1.30279 1.31275
-1.6 -3.38804 -1.96Zl3 -1.32900 -0.67532 0.25422 0.81720 0.99418 1.09338 1.16584 1.19680 1.21618 1.23132 1.23805
-1.7 -3.44438 -1.97227 -1.32376 -o.8S959 0.26808 0.80837 0.96977 1.05631 1.11628 1.14042 1.15477 1.16534 1.16974
-1.8 -3.49935 -1.98124 -1.31760 -o.64335 0.28150 0.79868 0.94496 1.01973 1.06864 1.08711 1.09749 1.10465 1.10743
-1.9 -3.5529S -1.98906 -1.31054 -o.62662 0.29443 0.78816 0.91988 0.98381 1.02311 1.03695 1.04427 1.04898 1.05068

-2.0 -3.60517 -1.99573 -1.30259 -o.60944 0.30685 0.77686 0.89464 0.94871 0.97980 0.98995 0.99499 0.99800 0.99900
-2.1 -3.65600 -2.00128 -1.29377 -o.59183 0.31872 0.76482 0.86938 0.91458 0.93878 0.94607 0.94945 0.95131 0.95188
-2.2 -3.70543 -2.00570 -1.28412 -o.57383 0.329119 0.75211 0.84422 0.88156 0.90009 0.90521 0.90742 0.90854 0.90885
-2.3 -3.75347 -2.00903 -1.27365 -o.55549 0.34063 0.73880 0.81929 0.84976 0.86371 0.86723 0.86863 0.86929 0.86945
-2.4 -3.80013 -2.01128 -1.26240 -o.S3683 0.35062 0.72495 0.79472 0.81927 0.82959 0.83196 0.83283 0.83320 0.83328
-2.5 -3.84540 -2.01247 -1.25039 -o.51789 0.35992 0.71067 0.77062 0.79015 0.79765 0.79921 0.79973 0.79994 0.79998
-2.6 -3.88930 -2.01263 -1.23766 -o.49872 0.36852 0.69602 0.74709 0.76242 0.76779 0.76878 0.76909 0.76920 0.76922
-2.7 -3.93183 -2.01177 -1.22422 -0.47934 0.37640 0.68111 0.72422 0.73610 0.73987 0.74049 0.74067 0.74073 0.74074
-2.8 -3.97301 -2.00992 -1.21013 -o.4S980 0.38353 0.66603 0. 70209 o. 71116 0.71377 0. 71415 0. 71425 0. 71428 0. 71428
-2.9 -4.01286 -2.00710 -1.19539 -o.44015 0.38991 0.65086 0.68075 0.68759 0.68935 0.68959 0.68964 0.68965 0.68965

-3.0 -4.05138 -2.00335 -1.18006 -o.42040 0.39554 0.63569 0.66023 0.66S32 0.66649 0.66663 0.66666 0.66667 0.66667

Source: Appendix 3, Bulletin 17B, reference (46).

F-5
EM 1110-2-1415
5 Mar 93

Table F-3

NQRMAL QISTRIBU:TIQN
Percent Chance Exceedance for Given Normal Standard Deviate (K)

K .DO .01 .a, Incr-ents to K in Column 1

.03 .04 .0~ ,06 .07 .DB .09
.0 50.00 49.60 49.20 48.80 48,40 48.01 47.61 47.21 46.81 46.41
.1 46.02 45.62 45.22 44,83 44,43 44.04 43,64 43.25 42.86 42.47
.2 42.07 41.68 41.29 40.90 40.52 40.13 39.74 39.36 38.97 38.59
.3 38.21 37.83 37.45 37.07 36.69 36.32 35.94 35.57 35.20 34.83
.4 34.46 34.09 33.72 33.36 33.00 32.64 32.28 31.92 31.56 31.21
.5 30.85 30.50 30.15 29.81 29.46 29.12 28.77 28.43 28.10 27.76
.6 27.43 27.09 26.76 26.43 26.11 25.78 25.46 25.14 24.83 24.51
.7 24.20 23.89 23.58 23.27 22.96 22.66 22.36 22.06 21.77 21.48
.8 21.19 20.90 20.61 20.33 20.05 19.77 19.49 19.22 18.94 18.67
.9 18 .lol 18.14 17.88 17.62 17.36 17.11 16.85 16.60 16.35 16.11
1.0 15.87 15.62 15.39 15.15 14.92 llo. 69 14.46 14.23 14.01 13.79
1.1 13.57 13.35 13.14 12.92 12.71 12.51 12.30 12.10 11.90 11.70
1.2 11.51 11.31 11.12 10.93 10.75 10.56 10.38 10.20 10.03 9.85
1.3 9.68 9.51 9.34 9.18 9.01 8.85 8.69 8.53 8.38 8.23
1.4 8.08 7.93 7.78 7.64 7.49 7.35 7.21 7.08 6.94 6.81
1.5 6.68 6.55 6.43 6.30 6.18 6.06 5.94 5.82 5. 71 5.59
1.6 5.48 5.37 5.26 5.16 5.05 4.95 4.85 4.75 4.65 4.55
1.7 4.46 4.36 4.27 4.18 4.09 4.01 3.92 3.84 3.75 3.67
1.8 3.59 3.51 3. 44 3.36 3.29 3.22 3.14 3.07 3.01 2.94
1.9 2.87 2.81 2.74 2.68 2.62 2.56 2.50 2.44 2.39 2.33
2.0 2.28 2.22 2.17 2.12 2.07 2.02 1.97 1.92 1.88 1.83
2.1 1. 79 1.74 l. 70 1.66 1.62 1.58 1.54 1. 50 1.46 l. 43
2.2 1.39 1.36 1.32 1.29 1.25 1.22 1.19 1.16 1.13 1.10
2.3 1.072 1.044 1.017 .990 .96/o .939 . 911o .889 .866 .842
2.4 .820 .798 . 776 .755 .734 .714 .695 .676 .657 .639
2.5 .621 .604 .587 .570 .554 .539 .523 .508 .494 .480
2.6 .466 .453 ,440 .427 .415 .402 .391 .379 .368 .357
2.7 .347 .336 .326 .317 .307 .298 .289 .280 .272 .264
2.8 .256 .248 .240 .233 .226 .219 .212 .205 .199 .193
2.9 .187 .181 .175 .169 .164 .159 .154 .149 .11o4 .139
3.0 .135 .131 .126 .122 .118 .114 .111 .107 .104 .100
3.1 .0968 .0936 .0904 .0874 .0845 .0816 .0789 .0762 .0736 .0711
3.2 .0687 .0664 .0641 .0619 .0598 .0577 .0557 .0538 .0519 .0501
3.3 .0483 .0467 .0450 .0434 .0419 .0404 .0390 .0376 .0362 .0350
3.4 .0337 .0325 .0313 .0302 .0291 .0280 .0270 .0260 .0251 .0242
3.5 .0233 .0224 .0216 .0208 .0200 .0193 .0185 .0179 .0172 .0165
3.6 .0159 .0153 .0147 . 0142 .0136 .0131 .0126 .0121 .0117 .0112
3.7 .01078 .01037 .0099 .00958 .00920 .00884 .00850 .00817 .00784 .00753
3.8 .00724 .00695 .00667 .00641 .00615 .00591 .00567 .00544 .00522 .00501
3.9 .00481 .00462 .00443 .004Z5 .00408 .00391 .00375 .00360 .00345 .00331
4.0 .00317 .00304 .00291 .00279 .00267 .00256 .00245 .00235 .00225 .00216
5.0 .0 42867

6.0 .o 7987

Note- Values have been generated by use of computer routines for the normal
distribution.

F-6
 EM 1110-2-1415
5 Mar 93

Table F-4

PfR!:f:NTAQ:E fQJNTS QF TH:E QNE-TAILED t-DISTRIBUTIQN

..... . . . . . . . . . EKceedmce Pr!Dibility

OF 0.40 0.30 0.20 0,10 0,05 0,04 o,o~ 0,01 0,005 0.002 0.001 0.0005 0.0002 0.0001
1 .32.5 .727 1.376 3.078 6.314 7.916 15.895 31.821 63.657 159.153 318.309 636.619 1591..549 3183.099
2 .289 .617 1.061 1.886 2.920 3.320 4.849 6.965 9.925 15.764 22.327 31.599 49.985 70.700
3 .277 .584 .979 1.638 2.353 2.605 3.482 4 . .541 5.841 8.053 10.215 12.924 17.598 22.204
4 .271 .!i69 .941 1.533 2.132 2.333 2.999 3. 747 4.604 5.951 7.173 8.610 10.915 13.034
5 .267 .559 .920 1.476 2.015 2.191 2.757 3.365 4.032 5.030 5.893 6.869 8.363 9.678

6 .265 .553 .906 1.440 1.943 2.104 2.612 3.143 3.708 4.525 5.208 5.959 7.074 8.025
7 .263 . .549 .896 1.415 1.895 2.046 2.517 2.998 3.500 4.207 4.786 5.408 6.311 7.063
8 .262 . .546 .889 1.397 1.860 2.004 2.449 2.896 3.355 3.991 4.501 5.042 5.812 6.442
9 .261 . .543 .883 1.383 1.833 1.973 2.398 2.821 3.250 3.835 4.297 4.781 5.461 6.011
10 .260 . .542 .879 1.372 1.812 1.948 2.359 2.764 3.169 3.716 4.144 4.587 5.202 5.694

11 .260 . .540 .876 1.363 l. 796 1.928 2.328 2.718 3.106 3.624 4.025 4.437 5.004 5.453
12 .259 .539 .873 1.356 1. 782 1.912 2.303 2.681 3.055 3.550 3.930 4.318 4.847 5.263
13 .259 .538 .870 1.350 1.771 1.899 2.282 2.650 3.012 3.489 3.852 4.221 4.721 5.111
14 .258 .537 .868 1.345 1.761 1.887 2.264 2.624 2.977 3.438 3.787 4.140 4.616 4.985
15 .258 .536 .866 1.341 1.753 1.878 2.249 2.602 2.947 3.395 3.733 4.073 4.528 4.880

16 .258 .535 .865 1.337 1.746 1.869 2.235 2.583 2.921 3.358 3.686 4.015 4. 4.54 4.791
17 .257 .534 .863 1.333 1.740 1.862 2.224 2.567 2.898 3.326 3.646 3.965 4.390 4.714
18 .257 .534 .862 1.330 1.734 1.855 2.214 2.~ 2.878 3.298 3.610 3.922 4.334 4.648
19 .257 .533 .861 1.328 1.729 1.850 2.205 2.539 2.861 3.273 3.579 3.883 4.285 4.590
20 .257 .533 .860 1.325 1.725 1.844 2.197 2.528 2.845 3.251 3.~ 3.850 4.241 4.539

21 .257 .532 .859 1.323 1.721 1.840 2.189 2.518 2.831 3.231 3.527 3.819 4.203 4.493
22 .256 .532 .858 1.321 1.717 1.835 2.183 2.508 2.819 3.214 3.505 3.792 4.168 4.452
23 .256 .532 .858 1.319 1.714 1.832 2.177 2.500 2.807 3.198 3.485 3.768 4.137 4.415
24 .256 .531 .857 1.318 1.711 1.828 2.1n 2.492 2.797 3.183 3.467 3.745 4.109 4.382
25 .256 .531 .856 1.316 1.708 1.825 2.167 2.485 2.787 3.170 3.450 3.725 4.083 4.352

26 .256 .531 .856 1.315 1.706 1.822 2.162 2.479 2.779 3.158 3.435 3.707 4.060 4.324
27 .256 .531 .855 1.314 1.703 1.819 2.158 2.473 2.771 3.147 3.421 3.690 4.038 4.299
28 .256 .530 .855 1.313 1.701 1.817 2.154 2.467 2.763 3.136 3.408 3.674 4.018 4.275
29 .256 .530 .8.54 1.311 1.699 1.814 2.150 2.462 2.756 3.127 3.396 3.659 4.000 4 .2.54
30 .256 .530 .8.54 1.310 1.697 1.812 2.147 2.457 2.750 3.118 3.385 3.646 3.983 4.234

40 .255 .529 .851 1.303 1.684 1.796 2.123 2.423 2.704 3.055 3.307 3.551 3.864 4.094
50 .255 .528 .849 1.299 1.676 1.787 2.109 2.403 2.678 3.018 3.261 3.496 3.795 4.014
60 .2.54 .527 .848 1.296 1.671 1.781 2.099 2.390 2.660 2.994 3.232 3.460 3.750 3.962
70 .2.54 .527 .847 1.294 1.667 1.776 2.093 2.381 2.648 2.977 3.211 3.435 3.719 3.926
80 .2.54 .526 .846 1.292 1.664 1.m 2.088 2.374 2.639 2.964 3.195 3.416 3.696 3.899

90 .2.54 .526 .846 1.291 1.662 1.771 2.084 2.368 2.632 2.9.54 3.183 3.402 3.678 3.878
100 .2.54 .526 .845 1.290 1.660 1.769 2.081 2.364 2.626 2.946 3.174 3.390 3.664 3.862
110 .2.54 .526 .845 1.289 1.659 1.767 2.078 2.361 2.621 2.940 3.166 3.381 3.652 3.848
120 .2.54 .526 .845 1.289 1.658 1.766 2.076 2.358 2.617 2.935 3.160 3.373 3.642 3.837
IliF .253 .524 .842 1.282 1.645 1.751 2.0.54 2.326 2.576 2.878 3.090 3.291 3 . .540 3.719

Note- Values have been generated by use of computer routines for the inverse
t-distribution. A few values for exceedance probabilities of 0.005 and less
may differ plus or minus 0.001 from published tables.

F-7
 EM 1110-2-1415
5 Mar 93

Table F-5

VA LUES QF CI::JJ-SQJ.lAB.E DISTRIBUTIQN

Exceecllnce Prcability . ..
OF 0.99 0.95 0,90 0. 75 0 . .50 o,~ 0.10 0.05 0.01 0.005 0.001
1 .o3 ts7 .00393 .0158 .102 .~55 1.323 2.706 3.841 6.635 7.879 10.828
2 .020 .103 .211 .575 1.386 2.m 4.605 5.991 9.210 10.597 13.816
3 .llS .352 .584 1.213 2.366 4.108 6.251 7 .81.5 11.345 12.838 16.256
4 .2S7 . 711 1.064 1.923 3.357 5.385 7.779 9.488 13.277 14.860 18.467
5 .554 1.145 1.610 2.675 4.351 6.626 9.236 11.070 1.5.086 16.7.50 20.51.5

6 .872 1.635 2.204 3.455 5.348 7.841 10.845 12.592 16.812 18.548 22.458
7 1.239 2.167 2.833 4.255 6.346 9.037 12.017 14.067 18.475 20.278 24.322
8 1.846 2.733 3.490 5.071 7.344 10.219 13.362 1.5 . .507 20.090 21.955 26.125
9 2.088 3.325 4.168 5.898 8.343 11.389 14.684 16.919 21.666 23.589 27.877
10 2.558 3.940 4.865 6.737 9.342 12.549 1.5.987 18.307 23.209 25.188 29.588

11 3.053 4.575 5.578 7.584 10.341 13.701 17.275 19.675 24.725 26.757 31.264
12 3.571 5.226 6.304 8.438 11.340 14.845 18.549 21.026 26.217 28.299 32.909
13 4.107 5.892 7.042 9.299 12.340 1.5.984 19.812 22.362 27.688 29.819 34.528
14 4.660 6.571 7.790 10.165 13.339 17.117 21.064 23.685 29.141 31.319 36.123
1S 5.229 7.261 8.547 11.037 14.339 18.245 22.307 24.996 30.578 32.801 37.697

16 5.812 7.962 9.312 11.912 1.5.338 19.369 23.542 26.296 32.000 34.267 39.252
17 6.408 8.672 10.085 12.792 16.338 20.~89 24.769 27.587 33.409 35.719 40.790
18 7.01S 9.390 10.865 13.675 17.338 21.605 25.989 28.869 34.805 37.1S6 42.312
19 7.633 10.117 11.651 14.562 18.338 22.718 27.204 30.144 36.191 38.582 43.820
20 8.260 10.851 12.443 1.5.~52 19.337 23.828 28.412 31.410 37.566 39.997 45.31.5

21 8.897 11.591 16.240 13.344 20.337 24.935 2S.61S 32.671 38.932 41.401 46.797
22 9.542 12.338 14.041 17.240 21.337 26.039 30.813 33.924 40.289 42.796 48.268
23 10.196 13.091 14.848 18.137 22.337 27.141 32.007 35.172 41.638 44.181 49.728
24 10.856 13.848 1.5.659 19.037 23.337 28.241 33.196 36.41.5 42..980 45.559 51.179
25 11.524 14.611 16.473 19.939 24.337 29.339 34.382 37.652 44.314 46.928 52.620

26 12.198 1.5.379 17.282 20.843 25.336 30.435 35.563 38.885 45.842. 48.290 54.052
27 12.879 16.1.51 18.114 21.749 26.336 31.528 36.741 40.113 46.963 49.845 55.476
28 13.!165 16.928 18.939 22.657 27.336 32.621 37.916 41.337 48.278 .50.993 56.892
2S 14.256 17.708 19.758 23.567 28.336 33.711 39.087 42.557 49.588 52.336 58.302.
30 14.953 18.493 20.!189 24.478 29.336 34.800 40.256 43.773 .50.892 53.672 59.703

40 22.184 26 . .509 29.051 33.660 39.335 45.616 51.1105 55.759 63.691 66.766 73.402.
50 29.707 34.764 37.689 42.942 49.335 56.334 63.167 67.505 76.154 79.490 86.661
60 37.485 43.188 lt6.459 52.294 59.335 66.981 74.397 79.082 88.379 91.952 99.607
70 45.442. 51.739 55.32S 61.698 69.334 77.577 85.527 90.531 100.425 104.2.1.5 122.317
80 53.540 60.392 84.2.78 71.1~5 79.334 88.130 96.578 101.879 112.329 116.321 124.839

90 61.754 69.126 73.291 80.625 89.334 98.6.50 107.565 113.145 124.116 128.299 137.208
100 70.065 77.930 82.358 90.133 99.334 109.141 118.498 124.342 135.807 140.169 149.449

From Table 26.8, Percentage Points of the x 2 -Distribution, Handbook of Mathematical

Functions, National Bureau of Standards Applied Mathematical Series 55, U.S.
Government Printing Office, Washington, D.C., 1972, page 984 and 985.

F-8
 EM 1110-2-1415
5 Mar 93

Table F-6

VAL!.l&S QF TH&; F DISTRIBUTION

F Distribution Values for Upper 5% (a= .05)

. o.sr- o! Freedaa !or !Unuator .

IF 2 3 4 5 6 8 ~ 15 20 30 60 INF
1 161.4 199.5 215.7 224.6 230.2 234.0 2311.9 243.9 245.9 248.0 250.1 252.2 254.3
2 18.51 19.00 19.16 19.25 19.30 19.33 19.37 19.41 19.43 19.45 19.46 19.48 19.50
3 10.13 9.55 9.28 9.12 9.01 8.94 8.85 8.74 8.70 8.66 8.62 8.57 8.53
4 7.71 6.94 6.59 6.39 6.26 6.16 6.04 5.91 5.86 5.80 5.75 5.69 5.63
5 6.61 5.79 5.41 5.19 5.05 4.95 4.82 4.68 4.62 4.56 4.50 4.43 4.36
6 5.99 5.14 4.76 4.53 4.39 4.28 4.15 4.00 3.94 3.87 3.81 3.74 3.67
7 5.59 4.74 4.35 4.12 3.97 3.87 3.73 3.57 3.51 3.44 3.38 3.30 3.23
8 5.32 4.46 4.07 3.84 3.69 3.58 3.44 3.28 3.22 3.15 3.08 3.01 2.93
9 5.12 4.26 3.86 3.63 3.48 3.37 3.23 3.07 3.01 2.94 2.86 2. 79 2. 7l
10 4.96 4.10 3.71 3.48 3.33 3.22 3.07 2.91 2.85 2.77 2.70 2.62 2.54
ll 4.84 3.98 3.59 3.36 3.20 3.09 2.95 2.79 2.72 2.65 2.57 2.49 2.40
12 4.75 3.89 3.49 3.26 3.11 3.00 2.85 2.69 2.62 2.54 2.47 2.38 2.30
13 4.67 3.81 3.41 3.18 3.03 2.9Z 2.77 2.60 2.53 2.46 2.38 2.30 2.21
14 4.60 3. 74 3.34 3.11 2.96 2.85 2.70 2.53 2.46 2.39 2.31 2.22 2.13
15 4 . .54 3.68 3.2.9 3.06 2.90 2.79 2.64 2.48 2.40 2.33 2.25 2.16 2.07
16 4.49 3.63 3.24 3.01 2.85 2.74 2.59 2.42 2.35 2.28 2.19 2.11 2.01
17 4.45 3.59 3.20 2.96 2.81 2.70 2.55 2.38 2.31 2.23 2.15 2.06 1.96
18 4.41 3.55 3.16 2.93 2.77 2.86 2.51 2.34 2.27 2.19 2.11 2.02 l.9Z
19 4.38 3.52 3.13 2.90 2.74 2.63 2.48 2.31 2.23 2.16 2.07 1.98 1.88
20 4.35 3.49 3.10 2.87 2.71 2.60 2.45 2.28 2.20 2.12 2.04 1.95 1.84
21 4.32 3.47 3.07 2.84 2.68 2.57 2.42 2.25 2.18 2.10 2.01 1.9Z 1.18
22 4.30 3.44 3.05 2.82 2.66 2.55 2.40 2.23 2.15 2.07 1.98 1.89 l. 78
23 4.28 3.42 3.03 2.80 2.64 2.53 2.37 2.20 2.13 2.05 1.96 1.86 1. 76
24 4.26 3.40 3.01 2.78 2.62 2.51 2.36 2.18 2.11 2.03 1.94 1.84 1. 73
25 4.24 3.39 2.99 2.76 2.60 2.49 2.34 2.16 2.09 2.01 1.9Z 1.82 1.71
26 4.23 3.37 2.98 2.74 2.59 2.47 2.32 2.15 2.07 1.99 1.90 1.80 1.69
27 4.21 3.35 2.96 2.73 2.57 2.46 2.31 2.13 2.06 1.97 1.88 1.79 1.67
28 4.20 3.34 2.95 2.71 2.56 2.45 2.2.9 2.12 2.04 1.96 1.87 1.77 1.65
2.9 4.18 3.33 2.93 2.70 2.55 2.43 2.28 2.10 2.03 1.94 1.85 1.75 1.64
30 4.17 3.32 2.9Z 2.69 2.53 2.42 2.27 2.09 2.01 1.93 1.84 1.74 1.62
40 4.08 3.23 2.84 2.81 2.45 2.34 2.18 2.00 1.9Z 1.84 1.74 1.64 1.51
60 4.00 3.15 2.76 2.53 2.37 2.25 2.10 1.9Z 1.84 1.75 1.65 1.53 1.39
120 3.9Z 3.07 2.68 2.45 2.29 2.17 2.02 1.83 1.75 1.66 1.55 1.43 1.25
INF 3.84 3.00 2.60 2.37 2.21 2.10 1.94 1.75 1.67 1.57 1.46 1.32 1.00

From Table 26.9, Percentage Points of the F-Distribution, Handbook of Mathematical

Functions, National Bureau of Standards Applied Mathematical Series 55, U.S.
Government Printing Office, Washington, D.C., 1972, page 987.

F-9
EM 1110-2-1415
5 Mar 93

Table F-7
DEVIATES FQR THE &;XP&C:If:Q PRQBAIHLITY ADJUSTMENT

• • • • 0 . PEcct. Qwlce Exlceedlnce .. .... .

N 40 30. 20. 10. 5. 4, ~- 1, 0 5 o,;;: 0 l 0.05 0.02 0.01
2 .398 .890 1.686 3.769 7.733 9.695 19.467 38.972 77.963 1911.922 389.847 779.696 1949.242 3898.484
3 .333 .713 1.225 2.177 3.372 3.833 5.S99 8.042 11.460 18.203 25.781 36.487 57.718 81.637
lo .309 .653 1.094 1.831 2.631 2.913 3.893 5.077 6.530 9.003 11.420 14.449 19.675 24.825
5 .297 .623 1.031 1.680 2.335 2.556 3.285 4.105 5.044 6.519 7.858 9.432 11.957 14.278
6 .289 .604 .993 1.594 2.177 2.367 2.978 3.635 4.355 5.433 6.366 7.419 9.034 10.453
7 .283 .ssz .968 1.539 2.077 2.250 2.793 3.360 3.964 4.837 5.567 6.370 7.562 8.579
8 .279 .5112 .950 1.501 2.010 2.170 2.669 3.1BO 3.712 4.463 5.076 5.736 6.694 7.492
9 .276 .575 .937 1.472 1.960 2.113 2.581 3.053 3.537 4.207 4. 744 5.314 6.126 6.790
10 .274 .570 .927 1.451 1.923 2.069 2.516 2.959 3.408 4.022 4.507 5.014 5.728 6.304

11 .272 •.566 .918 1.433 1.893 2.035 2.464 2.887 3.310 3.881 4.328 4.791 5.434 5.947
12 .270 .562 .911 1.419 1.869 2.007 2.423 2.829 3.233 3.772 4.189 4.618 5.208 5.676
13 .269 .559 .906 1.407 1.850 1.985 2.390 2.782 3.170 3.684 4.078 4.481 5.030 5.462
14 .258 .556 .901 1.398 1.833 1.966 2.362 2.743 3.118 3.611 3.987 4.369 4.886 5.290
15 .267 .554 .897 1.389 1.819 1.949 2.338 2.711 3.074 3.551 3.912 4.276 4.768 5.149

16 .266 .552 .893 1.382 1.B07 1.936 2.318 2.683 3.037 3.499 3.848 4.198 4.668 5.030
17 .265 .551 .890 1.376 1.796 1.923 2.300 2.658 3.005 3.455 3.793 4.131 4.583 4.930
18 .264 .549 .887 1.370 1.787 1.913 2.285 2.637 2.978 3.417 3.746 4.074 4.510 4.844
19 .264 .548 .884 1.365 1.779 1.904 2.271 2.619 2.953 3.384 3.704 4.024 4.446 4.769
20 .263 .546 .882 1.361 Lm 1.895 2.259 2.602 2.932 3.354 3.668 3.979 4.391 4.703

21 .263 .545 .880 1.357 1.765 1.888 2.248 2.587 2.912 3.328 3.635 3.940 4.341 4.645
22 .262 .544 .878 1.353 1.759 1.881 2.239 2.574 2.895 3.304 3.606 3.905 4.297 4.594
23 .262 .544 .877 1.350 1.754 1.875 2.230 2.562 2.879 3.283 3.580 3.874 4.258 4.548
24 .262 .543 .875 1.347 1.749 1.869 2.222 2.551 2.865 3.264 3.557 3.845 4.222 4.506
25 .261 .542 .874 1.344 1.745 1.864 2.215 2.542 2.1152 3.246 3.535 3.820 4.190 4.469

26 .261 .541 .873 1.341 1.741 1.860 2.208 2.532 2.841 3.230 3.516 3.796 4.161 4.435
27 .261 .541 .871 1.339 1.737 1.855 2.202 2.524 2.830 3.216 3.498 3.775 4.134 4.403
28 .260 .540 .870 1.337 1.733 1.851 2.196 2.516 2.820 3.202 3.482 3.755 4.110 4.375
29 .260 .539 .869 1.335 1.730 1.848 2.181 2.509 2.811 3.190 3.466 3.737 4.087 4.349
30 .260 .539 .868 1.333 1.727 1.844 2.186 2.503 2.802 3.178 3.452 3.720 4.066 4.324

40 .258 .535 .862 1.320 1.706 1.820 2.151 2.456 2.742 3.097 3.354 3.602 3.921 4.1.56
50 .257 .533 .857 1.312 1.693 1.B06 2.131 2.429 2.707 3.051 3.298 3.535 3.838 4.061
60 .257 .532 .855 1.307 1.685 1.796 2.118 2.411 2.684 3.020 3.261 3.492 3.785 3.999
70 .256 .531 .853 1.303 1.679 1.nKI 2.108 2.399 2.688 2.999 3.235 3.462 3.748 3.957
ao .256 .530 .851 1.300 1.675 1.785 2.101 2.389 2.656 2.984 3.217 3.439 3.721 3.925

90 .255 .529 .850 1.298 1.671 1.781 2.096 2.382 2.647 2.971 3.202 3.422 3.700 3.901
100 .255 .529 .849 1.297 1.669 1.778 2.092 2.376 2.640 2.962 3.190 3.408 3.683 3.882
110 .255 .528 .849 1.295 1.666 1.775 2.088 2.372 2.634 2.954 3.181 3.397 3.670 3.867
120 .255 .528 .848 1.294 1.665 1.m 2.085 2.368 2.629 2.948 3.173 3.388 3.658 3.854
INF .253 .524 .842 1.282 1.645 1.751 2.054 2.326 2.576 2.878 3.090 3.291 3.540 3.719

Note- Values have been generated by use of computer routines for the inverse
t-distribution. This table is based on samples drawn from a normal distribution.
The above values (after adjustment for skew) may be used as approximate
adjustments to Pearson type-III distributions having small skew coefficients.

F-10
 EM 1110-2-1415
5 Mar 93

Table F-8
PER!;:ENT AQ&S FQR THE &XPE!;:TED PRQBABILITY ADJUSTMENT

.. . Pvcmt Ollnce Exceed~K!ce

N 40. 30. 20. 10. 5. 4. 2. 1. 0.5 02 0.1 0.05 0.02 0.01

2 43.5 37.1 30.8 24.3 20.37 19.43 17.12 15.43 14.13 12.81 12.011 11.342 10.602 10.125
3 42.3 34.7 27.1 19.1 14.~ 13.44 10.86 9.08 7.77 6.51 5.793 5.212 4.598 4.219
4 41.8 33.6 25.3 16.7 11.88 10.77 8.18 6.45 5.23 4.11 3.496 3.018 2.531 2.242
5 41.4 32.9 24.3 15.3 10.38 9.26 6.70 5.05 3.92 2.92 2.389 1.990 1.596 1.370
6 41.2 32.4 23.6 14.4 9.41 8.30 .5.78 4.19 3.14 2.23 1.768 1.427 1.101 .919
7 41.0 32.1 23.1 13.8 8.74 7.63 .5.16 3.62 2.63 1.80 1.384 1.086 .809 .658
8 40.9 31.8 22.7 13.3 8.24 7.14 4.70 3.22 2.28 1 . .50 1.127 .863 .623 .495
9 40.8 31.6 22.4 12.9 7.86 6.77 4.36 2.92 2.02 1.29 .947 .710 .498 .388
10 40.7 31.5 22.1 12.6 7.S6 6.47 4.09 2.69 1.82 1.13 .816 .599 .409 .313
11 40.7 31.3 22.0 12.4 7.32 6.23 3.88 2 . .50 1.67 1.01 .716 .516 .345 .259
12 40.6 31.2 21.8 12.2 7.12 6.04 3.71 2.36 1.54 .92 .638 .453 .296 .219
13 40.6 31.1 21.7 12.0 6.9.5 5.87 3.S6 2.23 1.44 .84 .577 .403 .258 .188
14 40.5 31.0 21 . .5 11.9 6.80 5.73 3.44 2.13 1.36 .78 . .525 .363 .228 .164
15 40.5 31.0 21.4 11.8 6.68 5.61 3.33 2.04 1.29 .73 .485 .330 .204 .145

16 40 . .5 30.9 21.3 11.6 6 . .57 5 . .50 3.24 1.97 1.23 .68 .4.50 .303 .184 .129
17 40.4 30.9 21.3 11.5 6.47 5.41 3.16 1.90 1.18 .6.5 .421 .280 .168 .116
18 40.4 30.8 21.2 11.5 6.39 .5.33 3.09 1.85 1.13 .61 .396 .261 .154 .106
19 40.4 30.8 21.1 11.4 6.31 5.26 3.03 1.80 1.09 .59 .374 .244 .143 .097
20 40.4 30.7 21.1 11.3 6.25 5.19 2.98 1. 7.5 1.06 .S6 .355 .230 .133 .089
21 40.4 30.7 21.0 11.2 6.19 5.13 2.93 1. 71 1.02 .54 .339 .217 .124 .083
22 40.3 30.7 21.0 11.2 6.13 5.08 2.88 1.67 1.00 -~ .324 .206 .117 .077
23 40.3 30.6 20.9 11.1 6.08 5.03 2.84 1.64 .97 . .50 .311 .197 .110 .072
24 40.3 30.6 20.9 11.1 6.03 4.99 2.80 1.61 .9.5 .49 .299 .188 .104 .068
25 40.3 30.6 20.9 11.0 .5.99 4.9.5 2.77 1.59 .93 .47 .289 .180 .099 .064

26 40.3 30.6 20.8 11.0 .5.9.5 4.91 2.74 1.S6 .91 .46 .279 .173 .094 .061
27 40.3 30 . .5 20.8 11.0 .5.92 4.87 2.71 1.54 .89 .4.5 .271 .167 .090 .058
28 40.3 30 . .5 20.8 10.9 5.88 4.84 2.68 1.~ .88 .44 .263 .161 .086 .0.55
29 40.3 30.5 20.7 10.9 5.85 4.81 2.66 1. .50 .86 .43 .25.5 .1S6 .083 .0~
30 40.2 30 . .5 20.7 10.9 5.82 4.78 2.63 1.48 .85 .42 .249 .151 .080 .0.50
40 40.2 30.4 20 . .5 10.7 .5.61 4.58 2.47 1.3.5 .75 .3.5 .204 .119 .060 .036
.50 40.1 30.3 20.4 10 •.5 .5.49 4.47 2.37 1.28 .70 .32 .179 .102 .049 .029
60 40.1 30.2 20.4 10.4 5.41 4.39 2.31 1.23 .66 .30 .164 .092 .043 .025
70 40.1 30.2 20.3 10.4 .5.35 4.33 2.26 1.19 .64 .28 .154 .085 .039 .022
80 40.1 30.2 20.3 10.3 .5.30 4.29 2.23 1.17 .62 .27 .146 .080 .036 .020
90 40.1 30.2 20.2 10.3 .5.27 4.26 2.20 1.15 .61 .26 .141 .076 .034 .019
100 40.1 30.1 20.2 10.3 .5.24 4.23 2.18 1.13 .59 .26 .136 .073 .032 .018
110 40.1 30.1 20.2 10.2 5.22 4.21 2.17 1.12 .59 .25 .132 .071 .031 .017
120 40.1 30.1 20.2 10.2 .5.20 4.19 2.15 1.11 .58 .25 .130 .069 .030 .016
INF 40.0 30.0 20.0 10.0 .5.00 4.00 2.00 1.00 . .50 .20 .100 .0.50 .020 .010

Note- Values have been generated by use of computer r.outines for the inverse normal
and inverse t-distributions. This table is based on samples drawn from a normal
distribution. The above values may be used as approximate adjustments to
Pearson type-III distributions having small skew coefficients.

F-11
EM 1110-2-1415
5 Mar 93

Table F-9

CONFIDENCE LIMIT DEVIATES FOR NORMAL DISTRIBUTION

99% and 98% Limits

Sqll.e
Size 0 Pvcmt. Clun:e Exceedmce o
N 9900 9.500 9000 eooo ~00 2000 10,0 5,0 ~.o 1,0 o5 02 01

Ccllfis!!!!s:• ~! - ,oos
10 -1o232 -0731 -0436 -0027 10028 2o490 3o367 4.121 4.991 So.581 60126 60790 702.58
15 -1.38.5 -0866 -0569 -017.5 0769 lo996 20718 3.336 4.048 40530 40975 5o517 50898
20 -10483 -.951 -08.51 -0259 .640 1.768 2o423 2.982 3.624 4o059 40460 40948 5o2.92
25 -1 •.554 -1.012 -0708 -0317 .SS9 1.632 2.249 2.774 3.378 3078.5 40161 4o618 40940
30 -10608 -1.0.58 -07.51 -0359 .S03 1.539 2.132 2.636 3.214 3o603 30962 40399 40707
40 -1.688 -1.125 -0813 -0419 o428 1.421 1.983 20460 3.00.5 3o372 30710 40122 40412
50 -10 744 -1.172 -08.57 -0460 .379 1.346 1o890 2o350 2.87.5 3.229 3.5.55 3o951 40229
60 -1.787 -1.207 -0890 -0492 .344 1.294 1.825 2.273 2078.5 3.129 30446 30832 40103
70 -1.822 -1.236 -0916 -0.516 .317 10254 1.776 2.216 2o719 30056 30366 3.744 40009
eo -10850 -1.259 -0937 -053.5 .295 1.224 1.738 20172 2o666 2.998 3o304 30675 30936
90 -1.874 -1.279 -09.55 -.552 .277 1o199 10708 2.136 2o624 2.952 30254 3o620 30878
100 -1.894 -1.295 -0970 -osss .263 1.177 1.682 2.106 2o.589 2.914 3o212 30.574 30829

Ccllficltllce lAval. • .~
10 -.5 . .581 -4.121 -3.367 -20490 -1.028 o027 .436 o731 1.038 1.232 1o405 1o61l 10 754
15 -4.530 -3.336 -2.718 -1.996 -0769 .17.5 .569 .81!6 1.182 1.38.5 1.567 1.785 1.936
20 -4.059 -2.982 -2.423 -1.768 -.640 .259 .8.51 .951 1.274 1.483 1.611 10897 2o054
25 -3.78.5 -2.774 -2.21o9 -1.632 -.SS9 o317 0708 1.012 1.340 1. .554 1. 747 1.979 2o140
30 -3.603 -2.636 -20132 -1.539 -.503 .359 o7.51 1.0.58 1.391 1.608 10804 2.040 2.205
loO -3.372 -2.460 -1.983 -10421 -o428 .lo19 o813 1.125 1.1o64 1.688 1o890 2.132 20302
50 -30229 -2.350 -1.890 -1.348 -0379 .460 .8.57 1.172 1 •.518 1.71o4 1.950 2o198 2.371
60 -3.129 -2.273 -1.825 -1.294 -o341o .492 o890 1.207 1.S.58 1.787 1.996 2.248 2o424
70 -3.056 -2.216 -1.776 -1.254 -.317 . .516 o916 1.236 1.590 1.822 2.033 2.288 20466
eo -2.998 -2.172 -1.738 -1.224 -0295 .53.5 o937 1.259 1.615 1.850 20063 2.320 2.SOO
90 -2.952 -2.136 -1.708 -1.199 -.277 .552 .955 1.279 1.637 1.874 2.089 2.348 2.529
100 -2.914 -2.106 -1.682 -1.177 -.263 .S6S .970 1.295 1.8.56 1.894 2.110 2.371 2.554
Ccllficltllce X..V.l • ,010
10 -1.314 -.804 -.508 -.107 .8112 2.243 3.048 3.738 lo.53S .5.074 .5 . .572 6.178 6.605
15 -1.1o.58 -.929 -.829 -.236 .678 1.841 2.521 3.102 3.770 4.222 4.639 .5.147 5 . .504
20 -1.550 -1.008 -.70S -.313 •.568 1.8.51 2.276 2.808 3.419 3.832 4.212 4.67.5 50001
25 -1.616 -1.064 -.757 -.364 .lo98 1.!136 2.129 2.633 3.211 3.601 3.960 4.398 4.706
30 -1.667 -1.107 -.797 -.lo03 .lo!50 1.457 2.030 2.515 3.071 3.1o47 3.792 4.212 4.508
40 -1.741 -1.169 -.854 -.457 .384 1.35.5 1.902 2.364 2.893 3.249 3.577 3.97.5 4.255
50 -1.793 -1.212 -.894 -.496 .340 1.291 1.821 2.2&9 2.781 3.125 3.442 3.826 4.097
60 -1.833 -1.24.5 -.924 -.524 .309 1.21o4 1.764 2.202 2.702 3.038 3.347 3.723 3.987
70 -1.865 -1.272 -.948 - . .545 .285 1.210 1.722 2.153 2.641o 2.974 3.278 3.647 3.906
80 -1.891 -1.293 -.968 -.563 .285 1.183 1.688 2.114 2.599 2.924 3.223 3 . .587 3.842
90 -1.913 -1.311 -.984 -.578 .250 1.160 1.661 2.082 2.561 2.883 3.179 3.538 3.791
100 -1.932 -1.325 -.998 -.391 .236 1.142 1.639 2.056 2.531 2.850 3.143 3.498 3.748

Q;m:icltllce Lew! • 990

10 -5.074 -3.738 -3.048 -2.243 -.892 .107 .508 .804 1.115 1.314 1.492 1.704 1.8.51
15 -4.222 -3.102 -2.521 -1.841 -.678 .236 .829 .929 1.251 l.lo.58 1.645 1.868 2.024
20 -3.832 -2.808 -2.276 -1.8.51 - •.568 .313 .705 1.008 1.336 1.550 1.743 1.974 2.135
25 -3.601 -2.633 -2.129 -1.!136 -.498 .364 .757 1.084 1.399 1.616 1.813 2.050 2.21.5
30 -3.1o47 -2.515 -2.030 -1.457 -.450 .403 .797 1.107 1.446 1.667 1.867 2.109 2.277
40 -3.249 -2.364 -1.902 -1.35.5 -.384 .457 .8.54 1.169 1.515 1.741 1.946 2.194 2.367
50 -3.125 -2.2&9 -1.821 -1.291 -.340 .496 .894 1.212 1.563 1.793 2.002 2.255 2.431
60 -3.038 -2.202 -1.764 -1.21o4 -.309 .524 .924 1.245 1.600 1.833 2.045 2.301 2o4eo
70 -2.974 -2.153 -1.722 -1.210 -.285 . .545 .948 1.272 1.630 1.865 2.079 2.338 2.519
eo -2.924 -2.114 -1.688 -1.183 -.265 .563 .968 1.293 1.653 1.891 2.107 2.368 2.551
90 -2.883 -2.082 -1.661 -1.160 -.250 .578 .984 1.311 1.674 1.913 2.131 2.394 2.578
100 -2.850 -2.056 -1.639 -1.142 -.236 .391 .998 1.326 1.691 1.932 2.151 2.416 2.601

F-12
 EM 1110-2-1415
5 Mar 93

Table F-9 (Cont)

CONFIDENCE LIMIT DEVIATES FOR NORMAL DISTRIBUTION

95% and 90% Limits

Scpl.e
Size . PIIRmt O.X:e Ellc..WX:e .
N 99,0 95.0 90.0 80.0 :!Q,O 6Q,Q !0,0 ~.a 2,0 1.0 .5 .2 .1

Ccn!ida'lce lAM!l • .0~

10 -1.442 -.915 -.615 -.220 .715 1.931 2.647 3.259 3.964 4.440 4.879 5.414 5.791
15 -1.572 -1.026 -.721 -.328 .~ 1.638 2.264 2.797 3.409 3.822 4.203 4.667 4.993
20 -1.654 -1.095 -.786 -.392 .4611 1.494 2.079 2.576 3.145 3.529 3.883 4.313 4.616
25 -1.713 -1.145 -.831 -.436 .413 1.405 1.966 2.442 2.986 3.353 3.691 4.101 4.390
30 -1.757 -1.182 -.866 -.468 .373 1.344 1.889 2.351 2.878 3.233 3.560 3.957 4.237
40 -1.823 -1.236 -.916 -.515 .320 1.263 1.789 2.232 2.738 3.078 3.391 3.772 4.039
so -1.869 -1.274 -.950 -.547 .284 1.211 1.724 2.156 2.649 2.980 3 . .2.84 3.654 3.914
60 -1.903 -1.303 -.976 -.571 .258 1.174 1.679 2.103 2.586 2.911 3.209 3.571 3.825
70 -1.931 -1.326 -.997 -.590 .238 1.146 1.645 2.063 2.540 2.859 3.153 3.510 3.761
80 -1.954 -1.344 -1.014 -.605 .223 1.124 1.618 2.032 2.502 2.819 3.109 3.462 3.710
90 -1.973 -1.360 -1.028 -.618 .209 1.106 1.596 2.006 2.473 2.786 3.073 3.423 3.668
100 -1.989 -1.373 -1.040 -.629 .198 1.091 1.578 1.985 2.449 2.759 3.044 3.391 3.634

Ccn!ida'lce lAM!l • .975

10 -4.440 -3.259 -2.647 -1.931 -.715 .220 .615 .915 1.235 1.442 1.628 1.8SO 2.005
15 -3.822 -2.797 -2.264 -1.638 -.~ .328 .721 1.026 1.357 1.572 1.766 1.999 2.161
20 -3.529 -2.576 -2.079 -1.494 -.468 .392 .786 1.095 1.433 1.654 1.854 2.093 2.261
2S -3.353 -2.442 -1.966 -1.405 -.413 .436 .831 1.145 1.488 1.713 1.917 2.162 2.333
30 -3.233 -2.351 -1.889 -1.344 -.373 .4611 .866 1.182 1.529 1.757 1.964 2.214 2.388
40 -3.078 -2.232 -1.789 -1.263 -.320 .515 .916 1.236 1.590 1.823 2.034 2.289 2.467
so -2.91.!0 -2.156 -1.724 -1.2ll - .2.84 .547 .950 1.274 1.633 1.869 2.084 2.343 2.524
60 -2.911 -2.103 -1.679 -1.174 -.258 .571 .976 1.303 1.664 1.903 2.120 2.382 2.566
70 -2.859 -2.063 -1.645 -1.146 -.238 .590 .997 1.326 1.690 1.931 2.1SO 2.415 2.600
80 -2.819 -2.032 -1.618 -1.124 -.223 .605 1.014 1.344 1.711 1.954 2.175 2.441 2.628
90 -2.786 -2.006 -1.596 -1.106 -.209 .618 1.028 1.360 1.728 1.973 2.195 2.464 2.652
100 -2.759 -1.985 -1.578 -1.091 -.198 .629 1.040 1.373 1.744 1.989 2.213 2.483 2.672

Ccn!ida'lce Lev.1 • ,O:!Q

10 -1.563 -1.017 -.712 -.317 .580 1.702 2.355 2.911 3.549 3.981 4.379 4.862 5.203
15 -1.677 -1.114 -.802 -.406 .455 1.~ 2.068 2.566 3.136 3.520 3.874 4.304 4.607
20 -1.749 -1.175 -.BSB -.460 .387 1.370 1.926 2.396 2.934 3.295 3.628 4.033 4.318
2S -1.801 -1.217 -.898 -.497 .342 1.301 1.838 2.292 2.809 3.1SB 3.478 3.868 4.142
30 -1.840 -1.2SO -.928 -.525 .310 1.252 1.m 2.220 2.724 3.064 3.376 3.755 4.022
40 -1.896 -1 . .297 -.970 -.S65 .266 1.188 1.697 2.125 2.613 2.941 3.242 3.608 3.865
so -1.936 -1.329 -1.000 -.592 .237 1.146 1.646 2.065 2.542 2.862 3.157 3.515 3.766
60 -1.966 -1.354 -1.022 -.612 .216 1.116 1.609 2.022 2.492 2.807 3.096 3.448 3.695
70 -1.990 -1.374 -1.040 -.629 .199 1.093 1 . .581 1.990 2.454 2.765 3.051 3.399 3.643
80 -2.010 -1.390 -1.054 -.642 .186 1.076 1.559 1.964 2.425 2.733 3.016 3.360 3.601
90 -2.026 -1.403 -1.066 -.652 .175 1.061 1.542 1.944 2.400 2.706 2.987 3.328 3.567
100 -2.040 -1.414 -1.077 -.662 .166 1.049 1.527 1.927 2.380 2.684 2.963 3.301 3.539

Ccn!ida'lce Leoval • ,9SO

10 -3.981 -2.911 -2.355 -1.702 -.580 .317 .712 1.017 1.348 1.563 1.757 1.989 2.151
15 -3 . .520 -2.566 -2.068 -1.482 -.455 .406 .802 1.114 1.454 1.677 1.878 2.121 2.290
20 -3.295 -2.396 -1.926 -1.370 -.387 .460 .858 1.175 1.522 1.749 1.955 2.204 2.378
2S -3.1SB -2.2SZ -1.838 -1.301 -.342 .497 .898 1.217 1.569 1.801 2.011 2.264 2.441
30 -3.064 -2.220 -1.m -1.252 -.310 .525 .928 1.2SO 1.605 1.840 2.053 2.310 2.489
40 -2.941 -2.125 -1.697 -1.188 -.266 .S65 .970 1 •.297 1.657 1.896 2.113 2.375 2. 5.58
so -2.862 -2.065 -1.646 -1.146 -.237 .592 1.000 1.329 1.694 1.936 2.156 2.421 2.607
60 -2.807 -2.022 -1.609 -1.116 -.216 .612 1.022 1.354 1.722 1.966 2.188 2.456 2.644
70 -2.765 -1.990 -1 • .581 -1.093 -.199 .629 1.040 1.374 1.745 1.990 2.214 2.484 2.673
80 -2.733 -1.9611 -1.559 -1.076 -.186 .642 1.054 1.390 1.762 2.010 2.235 2.S07 2.697
90 -2.706 -1.944 -1.542 -1.061 -.175 .652 1.066 1.403 1.778 2.026 2.252 2.526 2.718
100 -2.684 -1.927 -1.527 -1.049 -.166 .662 1.077 1.414 1.791 2.040 2.267 2.542 2.735

F-13
EM 1110-2-1415
5 Mar 93

Table F-9 (Cont)

CQNFIDENCE LIMIT Of:V)A Tf:S FQR NQB.MAL DISTRIBUTIQN

80% and 50% Limits
s.pl.e
Size . ~ O.X:e !la:eedmce .
N 99,0 ~.0 90.0 !!0.0 500 6Q,O 1,0,0 ~.o ~.o l.O .5 .2 .1

Cal!i!!!!l!:• !.Awl • .100

10 -1.715 -1.144 -.828 -.429 .437 1.474 2.066 2.S68 3.144 3.532 3.889 4.324 4.629
15 -1.808 -1.222 -.901 -.499 .347 1.320 1.867 2.329 2.8S7 3.212 3 . .539 3.936 4.215
20 -1.867 -1.271 -.946 -.541 .297 1.240 1.765 2.208 2.712 3.052 3.364 3.743 4.009
25 -1.908 -1.306 -.978 - . .570 .264 1.190 1.702 2.132 2.623 2.952 3.255 3.623 3.882
30 -1.9/oO -1.332 -1.002 -.593 .239 1.154 1.657 2.080 2.561 2.884 3.181 3.541 3.794
40 -1.986 -1.369 -1.036 -.624 .206 1.106 1.~ 2.010 2.479 2.793 3.082 3.433 3.679
50 -2.018 -1.396 -1.059 -.645 .184 1.07.5 1.559 1.965 2.426 2.735 3.019 3.363 3.605
60 -2.042 -1.415 -1.on -.1562 .167 1.052 1.532 1.933 2.389 2.694 2.974 3.313 3.552
70 -2.061 -1.431 -1.091 -.674 .155 1.035 1 . .511 1.909 2.360 2.662 2.940 3.276 3.513
80 -2.on -1.444 -1.103 -.684 .144 1.021 1.495 1.8 2.338 2.638 2.913 3.247 3.482
90 -2.090 -1.454 -1.112 -.6113 .136 1.010 1.481 1.874 2.319 2.618 2.891 3.223 3.456
100 -2.101 -1.463 -1.120 -.701 .129 1.001 1.470 1.861 2.305 2.601 2.873 3.203 3.435

Cal!idlrlce ~~ • ,900
10 -3.532 -2.S68 -2.066 -1.474 -.437 .429 .828 1.144 1.489 1.715 1.919 2.16.5 2.337
15 -3.212 -2.329 -1.867 -1.320 -.347 .499 .901 1.222 1.576 1.808 2.019 2.273 2.451
20 -3.052 -2.208 -1.765 -1.240 -.297 .541 .91t6 1.271 1.630 1.867 2.082 2.342 2.523
25 -2.952 -2.132 -1.702 -1.190 -.264 .570 .978 1.306 1.689 1.908 2.126 2.390 2 . .574
30 -2.884 -2.080 -1.657 -1.154 -.239 .593 1.002 1.332 1.698 1.940 2.160 2.426 2.612
40 -2.793 -2.010 -1.598 -1.106 -.206 .624 1.036 1.369 1.740 1.986 2.209 2.479 2.668
so -2.73.5 -1.965 -1.559 -1.075 -.184 .645 1.059 1.3116 1.no 2.018 2.244 2.517 2.708
60 -2.694 -1.933 -1.532 -1.052 -.167 .1562 1.on 1.415 1.79Z 2.042 2.269 2.544 2.737
70 -2.1562 -1.909 -1 . .511 -1.03.5 -.155 .674 1.091 1.431 1.810 2.061 2.290 2.567 2.761
80 -2.638 -1.8 -1.495 -1.021 -.144 .684 1.103 1.444 1.824 2.on 2.307 2.585 2.780
90 -2.618 -1.874 -1.481 -1.010 -.136 .693 1.112 1.454 1.836 2.090 2.321 2.600 2.796
100 -2.601 -1.861 -1.470 -1.001 -.129 .701 1.120 1.463 1.847 2.101 2.333 2.613 2.810

Cal!idlrlce ~~ • .~
10 -2.008 -1.382 -1.043 -.625 .222 1.155 1.671 2.104 2.~ 2.927 3.231 3.599 3.858
15 -2.055 -1.422 -1.081 -.661 .179 1.083 1.sn 1.991 2.460 2.775 3.064 3.415 3.661
20 -2.085 -1.448 -1.104 -.683 .154 1.04.5 1 •.528 1.1132 2.390 2.697 2.978 3.320 3.560
25 -2.106 -1.466 -1.121 -.689 .137 1.020 1.497 1.895 2.346 2.648 2.925 3.261 3.497
30 -2.123 -1.479 -1.133 -.710 .125 1.002 1.47.5 1.8611 2.31.5 2.614 2.888 3.220 3. 4.53
40 -2.147 -1.499 -1.1.51 -.726 .108 .978 1.44.5 1.834 2.274 2.S68 2.838 3.165 3.395
50 -2.183 -1 •.513 -1.164 -.738 .096 .962 1.425 1.811 2.247 2 . .538 2.805 3.129 3.3.57
60 -2.176 -1 •.523 -1.173 -.747 .088 .9.50 1.411 1.795 2.227 2•.517 2.783 3.105 3.331
70 -2.186 -1.532 -1.181 -. 7.53 .081 .942 1.401 1.782 2.213 2.S01 2. 76.5 3.085 3.310
80 -2.194 -1 . .538 -1.187 -.759 .076 .93.5 1.392 1.772 2.202 2.489 2.752 3.070 3.294
90 -2.201 -1.544 -1.192 -.763 .071 .929 1.386 1.764 2.192 2.478 2.740 3.058 3.281
100 -2.207 -1.549 -1.196 -.767 .068 .925 1.380 1.758 2.184 2.470 2.731 3.048 3.270

Cal!idge ~1 • ,750
10 -2.927 -2.104 -1.671 -1.155 -.222 .625 1.043 1.382 1.759 2.008 2.235 2.508 2.700
1.5 -2.ns -1.991 -1 •.577 -1.083 -.179 .661 1.081 1.422 1.803 2.0.5.5 2.284 2.562 2.756
20 -2.697 -1.1132 -1 •.528 -1.04.5 -.154 .683 1.104 1.448 1.831 2.08.5 2.317 2.597 2.793
25 -2.648 -1.89.5 -1.497 -1.020 -.137 .699 1.121 1.466 1.6.51 2.106 2.339 2.621 2.819
30 -2.614 -1.8611 -1.475 -1.002 -.125 .710 1.133 1.479 1.867 2.123 2.357 2.641 2.839
40 -2 . .568 -1.834 -1.445 -.978 -.108 .726 1.151 1.499 1.888 2.147 2.383 2.668 2.868
so -2 . .538 -1.811 -1.425 -.962 -.096 .738 1.164 1•.513 1.903 2.183 2.400 2.688 2.889
60 -2 . .517 -1.795 -1.411 -.9.50 -.088 .747 1.173 1•.523 1.916 2.176 2.414 2.702 2.904
70 -2.S01 -1.782 -1.401 -.942 -.081 . 7.53 1.181 1.532 1.925 2.186 2.425 2.714 2.917
80 -2.489 -1.772 -1.392 -.935 -.076 .759 1.187 1..538 1.1132 2.194 2.434 2.724 2.927
90 -2.478 -1.764 -1.386 -.929 -.071 .763 1.192 1.544 1.938 2.201 2.441 2.731 2.935
100 -2.470 -1.758 -1.380 -.925 -.068 .767 1.196 1.549 1.944 2.207 2.447 2.739 2.943

F-14
 EM 1110-2-1415
5 Mar 93

Table F-9 (Cont)

~QNFID:ENC:E LIMIT Qf:VIAT:ES FQR NQRMAI.. DISTRIBUTIQN

50% Confidence Level

Slaple
Size . Perctmt Cllmce Ela:eedll'lce .
N 99,0 95.0 90,0 80.0 50,0 ~.o 10,0 ~.o ~.o 1.0 .5 .2 .1

Qm:idence IA\I'el • .sao

10 -2.410 -1.702 -1.324 -.868 .000 .868 1.324 1.702 2.126 2.410 2.670 2.984 3.205
1.5 -2.379 -1.681 -1.309 -.~ .000 -~ 1.309 1.681 2.100 2.379 2.635 2.945 3.163
20 -2.365 -1.671 -1.301 -.854 .000 .854 1.301 1.671 2.087 2.365 2.619 2.927 3.143
2.5 -2.357 -1.666 -1.297 -.851 .000 .851 1.297 1.666 2.080 2.357 2.610 2.917 3.132
30 -2.352 -1.662 -1.294 -.850 .000 .850 1.294 1.662 2.076 2.352 2.605 2.910 3.125
40 -2.345 -1.658 -1.291 -.847 .000 .847 1.291 1.658 2.070 2.345 2.597 2.902 3.116
50 -2.341 -1.655 -1.289 -.846 .000 .846 1.289 1.655 2.066 2.341 2.592 2.897 3.111
60 -2.339 -1.653 -1.288 -.846 .000 .846 1.288 1.653 2.061o 2.339 2.590 2.894 3.107
70 -2.337 -1.652 -1.287 -.845 .000 .845 1.287 1.652 2.063 2.337 2.588 2.892 3.105
eo -2.336 -1.651 -1.286 -.845 .000 .845 1.286 1.651 2.062 2.336 2.587 2.890 3.103
90 -2.335 -1.650 -1.286 -.845 .000 .845 1.286 1.650 2.061 2.335 2.585 2.888 3.101
100 -2.334 -1.650 -1.285 -.844 .000 .844 1.285 1.650 2.060 2.334 2.584 2.887 3.100

Source: Owen, D. B., reference (25).

F-15
EM 1110-2-1415
5 Mar 93

Tab1eF-IO

Mf;AN-SQL!AB,&; ERRQR QF STATIQN SKJ;;W CQEFFICIENT

St.at.ion Record Lenst.h, in Years CN or Bl

Sk!w 'Gl 10 20 30 40 so 60 70 80 90 100
0.0 0.468 0. 244 0.167 0.127 0.103 0.087 0.075 0.066 0.059 0. 054
0.1 0.476 0.253 0.175 0.134 0.109 0.093 0.080 0. 071 0.064 0.058
0.2 0.485 0.262 0.183 0.142 0.116 0.099 0.086 0.077 0.069 0.063
0.3 0.494 0.272 0.192 0.150 0.123 0.105 0.092 0.082 0.074 0.068
0.4 0.504 0.282 0.201 0.158 0.131 0.113 0.099 0.089 0.080 0.073
0.5 0.513 0.293 0.211 0.167 0.139 0.120 0.106 0.095 0.087 0.079
0.6 0.522 0.303 0.221 0.176 0.148 0.128 0.114 0.102 0.093 0.086
0.7 0.532 0.315 0.231 0.186 0.157 0.137 0.122 0.110 0.101 0.093
0.8 0.542 0.326 0.243 0.196 0.167 0.146 0.130 0.118 0.109 0.100
0.9 0.552 0.338 0.254 0.207 0.177 0.156 0.140 0.127 0.117 0.109
1.0 0.603 0.376 0.285 0.235 0.202 0.178 0.160 0.147 0.135 0.126
1.1 0.646 0.410 0.315 0.261 0.225 0.200 0.181 0.166 0.153 0.143
1.2 0.692 O.U8 0.347 0.290 0.252 0.225 0.204 0.187 0.174 0.163
1.3 0.741 0.488 0.383 0.322 0.281 0.252 0.230 0.212 0.197 0.185
1.4 0.794 0.533 0.422 0.357 0.314 0.283 0.259 0.240 0.224 0.211
1.5 0.851 0.581 0.465 0.397 0.351 0.318 0.292 0.271 0.254 0.240
1.6 0.912 0.623 0.498 0.425 0.376 0.340 0.313 0.291 0.272 0.257
1.7 0.977 0.667 0.534 0.456 0.403 0.365 0.335 0.311 0.292 0.275
1.8 1.047 0.715 0.572 0.489 0.432 0.391 0.359 0.334 0.313 0.295
1.9 1.122 0.766 0.613 0.523 0.463 0.419 0.385 0.358 0.335 0.316
2.0 1.202 0.821 0.657 0.561 0.496 0.449 0.412 0.383 0.359 0.339
2.1 1.288 0.880 0.704 0.601 0.532 0.481 0.442 0.410 0.385 0.363
2.2 1.380 0.943 0.754 0.644 0.570 0.515 0.473 0.440 0.412 0.389
2.3 1.479 1.010 0.808 0.690 0.610 0.552 0.507 0.471 0.442 0.417
2.4 1.585 1.083 0.866 0.739 0.654 0.592 0.543 0.505 0.473 0.447
2.5 1.698 1.160 0.928 0.792 0.701 0.634 0.582 0.541 0.507 0.479
2.6 1.820 1.243 0. 994 0.849 0.751 0.679 0.624 0.580 0.543 0.513
2.7 1.950 1.332 1.088 0.910 0.805 0.728 0.889 0.621 0.582 0.550
2.8 2.089 1.427 1.H2 0.975 0.862 0.780 0.716 0.666 0.624 0.589
2.9 2.239 1.529 1.223 1.044 0.924 0.836 0.768 0.713 0.669 0.631
3.0 2.399 1.638 1.311 1.119 0.990 0.895 0.823 0.764 0. 716 0.676

Note- Values computed from Equation 6 in Bulletin 17B, reference (46).

F-16
 EM 1110-2-1415
5 Mar 93
Table F-11

Q:UTLIER TEST K VALUES (IQ PERCENT SIQNIFICANCE LEVEL)

Sample Outlier Sample Outlier Sample Outlier Sample Outlier

Size K Value Size K Value Size K Value Size K Value

10 2.036 45 2. 727 80 2.940 115 3.064

11 2.088 46 2.736 81 2.945 116 3.067
12 2.134 47 2.744 82 2.949 117 3.070
13 2.175 48 2.753 83 2.953 118 3.073
14 2.213 49 2.760 84 2.957 119 3.075
15 2.247 so 2.768 85 2.961 120 3.078
16 2.279 51 2. 775 86 2.966 121 3.081
17 2.309 52 2.783 87 2.970 122 3.083
18 2.335 53 2.790 88 2.973 123 3.086
19 2.361 54 2.798 89 2.977 124 3.089
20 2.385 55 2.804 90 2.981 125 3.092
21 2.408 56 2.811 91 2.984 126 3.095
22 2.429 57 2.818 92 2.989 127 3.097
23 2.448 58 2.824 93 2.993 128 3.100
24 2.467 59 2.831 94 2.996 129 3.102
25 2.486 60 2.837 95 3.000 130 3.104
26 2.502 61 2.842 96 3.003 131 3.107
27 2.519 62 2.849 97 3.006 132 3.109
28 2.534 63 2.854 98 3.011 133 3.112
29 2.549 64 2.860 99 3.014 134 3.114
30 2.563 65 2.866 100 3.017 135 3.116
31 2.577 66 2.871 101 3.021 136 3.119
32 2.591 67 2.877 102 3.024 137 3.122
33 2.604 68 2.883 103 3.027 138 3.124
34 2.616 69 2.888 104 3.030 139 3.126
35 2.628 70 2.893 105 3.033 140 3.129
36 2.639 71 2.897 106 3.037 141 3.131
37 2.650 72 2.903 107 3.040 142 3.133
38 2.661 73 2.908 108 3.043 143 3.135
39 2.671 74 2.912 109 3.046 144 3.138
40 2.682 75 2.917 110 3.049 145 3.140
41 2.692 76 2.922 111 3.052 146 3.142
42 2.700 77 2.927 112 3.055 147 3.144
43 2. 710 78 2.931 113 3.058 148 3.146
44 2. 719 79 2.935 114 3.061 149 3.148

Note- Table contains one sided 10% significance level deviates for the normal
distribution. Source: Appendix 4, Bulletin 17B, reference (46).

F-17
EM 1110-2-1415
5 Mar 93

Table F-12
BINOMIAL RISK TABLES

Tables are probabilities of binomial risk for

various exceedance frequencies and record lengths.

10.0 Percent Chance Exceedance Event

Number
of . . . . . . . . . . . . . . . . . . Length of Period in Years ................ .
Exceedances 10 20 30 40 50 60 80 100

0 .3487 .1216 .0424 .0148 .0052 .0018 .0002 .0000

1 .3874 .2702 .1413 .0657 .0286 .0120 .0019 .0003
2 .1937 .2852 .2277 .1423 .0779 .0393 .0085 .0016
3 .0574 .1901 .2361 .2003 .1386 .0844 .0246 .0059
4 .0112 .0898 .1771 .2059 .1809 .1336 .0527 .0159
5 .0015 .0319 .1023 .1647 .1849 .1662 .0889 .0339
6 .0001 .0089 .0474 .1068 .1541 .1693 .1235 .0596
7 .0000 .0020 .0180 .0576 .1076 .1451 .1451 .0889
8 .0000 .0004 .0058 .0264 .0643 .1068 .1471 .1148
9 .0000 .0001 .0016 .0104 .0333 .0686 .1308 .1304
10 .0000 .0000 .0004 .0036 .0152 .0389 .1032 .1319

5,0 Percent Chance Exceedance Event

Number
of .................. Length of Period in Years • ••••••••• ., • 0 ••••

Exceedances 10 20 30 40 50 60 80 100

0 .5987 .3585 .2146 .1285 .0769 .0461 .0165 .0059

1 .3151 .3774 .3389 .2706 .2025 .1455 .0695 .0312
2 .0746 .1887 .2586 .2777 .2611 .2259 .1446 .0812
3 .0105 .0596 .1270 .1851 .2199 .2298 .1978 .1396
4 .0010 .0133 .0451 .0901 .1360 .1724 .2004 .1781
5 .0001 .0022 .0124 .0342 .0658 .1016 .1603 .1800
6 .0000 .0003 .0027 .0105 .0260 .0490 .1055 .1500
7 .0000 .0000 .0005 .0027 .0086 .0199 .0587 .1060
8 .0000 .0000 .0001 .0006 .0024 .0069 .0282 .0649
9 .0000 .0000 .0000 .0001 .0006 .0021 .0119 .0349
10 .0000 .0000 .0000 .0000 .0001 .0006 .0044 .0167

F-18
 EM 1110-2-141:
5 Mar 9:

Table F-12 (Cont)

BINOMIAL RISK TABLES

Tables are probabilities of binomial risk for

various exceedance frequencies and record lengths.

2.0 Percent Chance Exceedance Event

Number
of . . ... . .... . . .. . .. . Length of Period in Years ...............
Exceedances 10 20 30 40 so 60 80 10

0 . 8171 .6676 .5455 .4457 .3642 .2976 .1986 .13

1 .1667 .2725 .3340 .3638 . 3716 .3644 .3243 .27
2 .0153 .0528 .0988 .1448 .1858 .2194 .2614 .27
3 .0008 .0065 .0188 .0374 .0607 .0865 .1387 .18
4 .0000 .0006 .0026 .0071 .0145 .0252 .0545 .09
5 .0000 .0000 .0003 .0010 .0027 .0058 .0169 .03
6 .0000 .0000 .0000 .0001 .0004 .0011 .0043 .01
7 .0000 .0000 .0000 .0000 .0001 .0002 .0009 .00
8 .0000 .0000 .0000 .0000 .0000 .0000 .0002 .00
9 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .00
10 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .00

l,O ~~t~ent Ch!nce E~£eedance Event

Number
of .................. Length of Period in Years . ..............
Exceedances 10 20 30 40 so 60 80 10

0 .9044 .8179 .7397 .6690 .6050 .5472 .4475 .36

1 .0914 .1652 .2242 .2703 .3056 .3316 .3616 .36
2 .0042 .0159 .0328 .0532 .0756 .0988 .1443 .18
3 .0001 .0010 .0031 .0068 .0122 .0193 .0379 .06
4 .0000 .0000 .0002 .0006 .0015 .0028 .0074 .01
5 .0000 .0000 .0000 .0000 .0001 .0003 .0011 .00
6 .0000 .0000 .0000 .0000 .0000 .0000 .0001 .00
7 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .00
8 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .00
9 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .00
10 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .00

F-19
EM 1110-2-1415
5 Mar 93

Table F-12 (Coot)

BINOMIAL RISK TABLES

Tables are probabilities of binomial risk for

various exceedance frequencies and record lengths .

. 5 Percent Chance Exceedance Event

Number
of Length of Period in Years ................ .
Exceedances 10 20 30 40 50 60 80 100

0 .9511 .9046 .8604 .8183 . 7783 .7403 .6696 .6058

1 .0478 .0909 .1297 .1645 .1956 .2232 .2692 .3044
2 .0011 .0043 .0095 .0161 .0241 .0331 .0534 .0757
3 .0000 .0001 .0004 .0010 .0019 .0032 .0070 .0124
4 .0000 .0000 .0000 .0000 .0001 .0002 .0007 .0015
5 .0000 .0000 .0000 .0000 .0000 .0000 .0001 .0001
6 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
7 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
8 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
9 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
10 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000

.2 feicent Cban£e 'x£eedance Event

Number
of . . . . . . . .. . . . . . . . . . . Length of Period in Years ...................
Exceedances 10 20 30 40 50 60 80 100

0 .9802 .9608 .9417 .9230 .9047 .8868 .8520 .8186

1 .0196 .0385 .0566 .0740 .0907 .1066 .1366 .1640
2 .0002 .0007 .0016 .0029 .0045 .0063 .0108 .0163
3 .0000 .0000 .0000 .0001 .0001 .0002 .0006 .0011
4 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0001
5 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
6 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
7 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
8 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
9 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
10 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000

F-20