fluids

Article
Effect of the Non-Stationarity of Rainfall Events on
the Design of Hydraulic Structures for Runoff
Management and Its Applications to a Case Study at
Gordo Creek Watershed in Cartagena de
Indias, Colombia
Alvaro Gonzalez-Alvarez 1, *, Oscar E. Coronado-Hernández 1 ID
, Vicente S. Fuertes-Miquel 2 ID

and Helena M. Ramos 3 ID

1 Facultad de Ingeniería, Departamento de Ingenieria Civil y Ambiental, Universidad Tecnológica de Bolívar,
Cartagena 131001, Colombia; [email protected]
2 Departamento de Ingeniería Hidráulica y Medio Ambiente, Universitat Politècnica de València,
46022 Valencia, Spain; [email protected]
3 Department of Civil Engineering, Architecture and Georesources, CERIS, Instituto Superior Técnico,
University of Lisbon, 1049-001 Lisbon, Portugal; [email protected]
* Correspondence: [email protected]; Tel.: +57-5-653-5200 (ext. 257)




Received: 7 January 2018; Accepted: 17 April 2018; Published: 20 April 2018


Abstract: The 24-h maximum rainfall (P24h-max ) observations recorded at the synoptic weather station
of Rafael Núñez airport (Cartagena de Indias, Colombia) were analyzed, and a linear increasing
trend over time was identified. It was also noticed that the occurrence of the rainfall value (over the
years of record) for a return period of 10 years under stationary conditions (148.1 mm) increased,
which evidences a change in rainfall patterns. In these cases, the typical stationary frequency
analysis is unable to capture such a change. So, in order to further evaluate rainfall observations,
frequency analyses of P24h-max for stationary and non-stationary conditions were carried out (by
using the generalized extreme value distribution). The goodness-of-fit test of Akaike Information
Criterion (AIC), with values of 753.3721 and 747.5103 for stationary and non-stationary conditions
respectively, showed that the latter best depicts the increasing rainfall pattern. Values of rainfall were
later estimated for different return periods (2, 5, 10, 25, 50, and 100 years) to quantify the increase
(non-stationary versus stationary condition), which ranged 6% to 12% for return periods from 5 years
to 100 years, and 44% for a 2-year return period. The effect of these findings were tested in the
Gordo creek watershed by first calculating the resulting direct surface runoff (DSR) for various return
periods, and then modeling the hydraulic behavior of the downstream area (composed of a 178.5-m
creek’s reach and an existing box-culvert located at the watershed outlet) that undergoes flooding
events every year. The resulting DSR increase oscillated between 8% and 19% for return periods from
5 to 100 years, and 77% for a 2-year return period when the non-stationary and stationary scenarios
were compared. The results of this study shed light upon to the precautions that designers should take
when selecting a design, based upon rainfall observed, as it may result in an underestimation of both
the direct surface runoff and the size of the hydraulic structures for runoff and flood management
throughout the city.

Keywords: rainfall frequency analysis; non-stationary; climate change; runoff management

Fluids 2018, 3, 27; doi:10.3390/fluids3020027 www.mdpi.com/journal/fluids

Fluids 2018, 3, 27 2 of 19

1. Introduction
Climate change is gradually coming to affect all living species by giving rise to new sets of
meteorological conditions: record breaking high temperatures, melting glaciers, increasing sea water
levels, increasing severity of droughts, floods, tornadoes, and hurricanes, to mention just a few.
Hydraulic structures have not been the exception; most of these structures were designed was based
on rainfall patterns that, in some cases, no longer conform to the realities of today’s weather conditions.
Generally speaking, sizing a hydraulic structure for an ungauged watershed starts with a hydrological
analysis that includes: (a) a rainfall frequency analysis; (b) the watershed delineation and estimation of
its morphometric parameters; and (c) the direct surface runoff (DSR) calculation using a rainfall-runoff
model, assuming it has the same return period of the generating rainfall.
Usually, a rainfall frequency analysis for an ungauged watershed consists of trying to find the
probability distribution that best represents the behavior of the extreme events (of different durations)
being analyzed, in order to subsequently obtain the value of the precipitation event associated with a
given return period to be later used for the DSR estimation. The traditional methods for frequency
analysis of extreme values for hydraulic structures design (streamflow, rainfall or rainfall intensity)
assume that: (a) the values of the variable are random and independent; (b) the probability distribution
functions are stationary; and (c) the return period (Tr) is the inverse of the probability of exceedance (p).
These assumptions imply that the probability of an event producing excessive precipitations does not
change over time (the sample is homogeneous). This approach cannot be applied when the variable
being analyzed shows noticeable change of pattern (either increase or decline), which may indicate
non-stationary conditions [1–4]. In these cases, the typical definition of the return period (Tr = 1/p) has
to be redefined to account for non-stationarity.
Cartagena de Indias, a city on the Caribbean coast of Colombia, has been lately undergoing
recurring floods that can be mainly attributed to different factors that go from uncontrolled
development projects—due to the lack of an updated territorial management plan that fits the city’s
realities—to poorly designed hydraulic structures for runoff management that do not take future
conditions (especially rainfall pattern changes and increase of impervious areas) into account. Most of
the runoff of Cartagena de Indias discharges into La Virgen swamp (ciénaga de La Virgen), a waterbody
of approximately 502 km2 , whose southern shoreline has been populated over the years by illegal
low-income settlements that suffer the consequences of both a rising sea level (the swamp is connected
to the Caribbean Sea) and an obsolete stromwater system. Furthermore, the cities of Cartagena de
Indias (downstream) and Turbaco (upstream) share several watersheds that also drain into the La
Virgen swamp. The upstream areas of these watersheds (like the one selected in this study) are mostly
rural, which have been gradually converted into impervious areas by local developers that offer more
competitive prices than those in Cartagena de Indias. This dynamic will most likely (and rapidly)
change the landscape, which brings with it more challenging hydrologic, and hydraulic, conditions.
In this context, Colombian legislation has mandated to analyze the possible effects of climate
change on the future rainfall pattern if a stormwater system (channels, pipes, culverts, etc.) is to be
designed [5]. A regional increasing trend of both rainfall and streamflow has been found in some
areas of the northern portion of the Colombian Caribbean region [6,7]. The studies that carry out
assessments for stationary (SC) versus non-stationary (NSC) conditions of streamflow and/or rainfall,
typically encompass only performing a statistical analysis (of a trend) without quantifying how the
two scenarios may affect real life conditions, both hydrologically and hydraulically speaking. In this
study, multiannual 24-h maximum rainfall (P24h-max ) data (from the synoptic weather station located
at the Rafael Núñez airport) was used to: (a) assess the trend of the P24h-max observations over time,
specifically for Cartagena de Indias, (b) quantify the rainfall values obtained for several return periods
(2, 5, 10, 25, 50, and 100 years) via stationary and non-stationary frequency analyses, (c) carry out a
hydrological and hydraulic analysis on the Gordo creek watershed under SC and NSC for different
return periods in order to understand the recurring floods reported that affect a commercial and
Fluids 2018, 3, 27 3 of 19

industrial area downstream, and (d) point out the importance of accounting for the effect of climate
Fluids 2018, 3, x FOR PEER REVIEW 3 of 19
change in the decision making process in runoff management, especially in flood-prone areas.
2.
2. Study
Study Area
Area and
and Data
Data

2.1.
2.1. Study
Study Area
Area
The
The Gordo
Gordo creek
creek watershed
watershed is is located
located within
within the
the jurisdiction
jurisdiction ofof the
the city
city of
of Turbaco
Turbaco (Colombia);
(Colombia);
the
the outlet
outlet is
is at
at the east-southern borderline
the east-southern borderline between
between thethe cities
cities of
of Cartagena
Cartagena de de Indias
Indias and
and Turbaco
Turbaco
(Figure
(Figure 1). It has a total area of 258 hectares. The creek runs south to north, with a total length of of
1). It has a total area of 258 hectares. The creek runs south to north, with a total length 2962
2962 m
m and an average slope of 0.00454. The watershed’s centroid and outlet
and an average slope of 0.00454. The watershed’s centroid and outlet are located at the geographicalare located at the
geographical
coordinates 10coordinates 10°20′04″
◦ 200 0400 N; 75 ◦ 270 2900 N;
W 75°27′29″
and 10◦ 22W and0010°22′34.83″
0 34.83 N; 0075°27′37.51″
N; 75◦ 270 37.51 W respectively.
W respectively.

Study area
Figure 1. Study area location.

The
The watershed’s
watershed’s soil
soil cover
cover in
in the
the upstream
upstream area
area is
is mostly
mostly rural
rural (bush
(bush and
and forest)
forest) while
while the
the
downstream
downstream is in a busy commercial, industrial, and residential area (Table 1). The soil texture
is in a busy commercial, industrial, and residential area (Table 1). The soil texture is
is
predominantly clay-loam with a poor infiltration capacity when saturated.
predominantly clay-loam with a poor infiltration capacity when saturated.

Table
Table 1.
1. Gordo
Gordo creek
creek watershed’s soil cover
watershed’s soil cover distribution.
distribution.

Developed Areas Rural Areas

Item Developed Areas Rural Areas Total
Item Impervious Urban Residential Bush Forest Total
Area (ha)Impervious
2.9 Urban
12.7 Residential
23.4 Bush Forest
109.5 109.5 258
Area (ha) 2.9 12.7 23.4 109.5 109.5 258
The watershed’s downstream area has a main road with a box-culvert that gives access to several
industries and businesses.
The watershed’s The mostarea
downstream critical
has aportion of the
main road watershed’s
with downstream
a box-culvert that givesarea (circled
access area
to several
in Figure 2) is composed of the last 178.5-m reach (of the creek) and the box-culvert that controls
industries and businesses. The most critical portion of the watershed’s downstream area (circled area the
watershed
in Figure 2)outlet. Flooding
is composed of events
the lastoccur every
178.5-m year(ofduring
reach the rainy
the creek) season
and the (September
box-culvert that through
controls
November), which impedes regular transit for several hours. Also, people and vehicles
the watershed outlet. Flooding events occur every year during the rainy season (September through have been
stranded,
November),andwhich
have required
impedes rescuing on thefor
regular transit flooded road
several (Figure
hours. 3a–d).
Also, people and vehicles have been
stranded, and have required rescuing on the flooded road (Figure 3a–d).
Fluids 2018, 3, 27 4 of 19
Fluids 2018,
Fluids 2018, 3,
3, xx FOR
FOR PEER
PEER REVIEW
REVIEW 44 of
of 19
19

Figure 2.
Figure
Figure 2. Study
2. Study area location.
Study area
area location.
location.

Figure
Figure 3.3. (a)(a)
3. Access
Access road
road to industrial
to industrial
industrial area area (2007);
(2007); (b) Bus
Bus(b)stranded
Bus stranded in flooded
in flooded
flooded area(c)
area (2007);
(2007); (2007);
Box-
Figure (a) Access road to area (2007); (b) stranded in area (c) Box-
(c) Box-culvert
culvert (watershed (watershed
outlet outlet
at at cross-section
cross-section K0+00) K0+00)
(2015); (2015);
(d) Area(d)downstream
Area downstream
of the of the watershed
watershed outlet
culvert (watershed outlet at cross-section K0+00) (2015); (d) Area downstream of the watershed outlet
outlet
(2010).(2010). The locations
The locations
locations wherewhere the pictures
the pictures
pictures were are
were taken
taken taken are shown
shown in Figure
in Figure
Figure 2. 2.
(2010). The where the were are shown in 2.
Fluids 2018, 3, 27 5 of 19

2.2. Rainfall Data

Although the Institute of Hydrology, Meteorology, and Environmental Studies of Colombia
(IDEAM, in Spanish) operates another two weather stations in Cartagena de Indias (Escuela
Naval-CIOH and Santa Ana), the one located at the airport Rafael Núñez has the longest and most
complete data of records for several meteorological variables (Table 2); for this reason, it is the most
used in hydrological and hydraulic analysis and designs.

Table 2. Description of Rafael Núñez airport weather station.

IDEAM Height Geographic Coord. Operating Period

Station Type
ID N◦ (masl) North West Start Finish
Rafael Núñez Airport Synoptic (*) 14015020 2 10◦ 260 3100 75◦ 310 1300 15 March 1941 Still active
(*) Meteorological variables measured hourly (synoptic time) at a synoptic weather station include: precipitation,
cloud conditions, wind direction and velocity, atmospheric pressure, air temperature, visibility, and humidity.
Precipitation at Rafael Núñez station is measured by means of a Hellmann-Fuess pluviograph (standard model;
rain recorder 95).

For this study, continuous multiannual observations of 24-h maximum rainfall (P24h-max ) from
1941 to 2015 (Table 3; gray cell is the maximum value registered) were used. Years 2016 and 2017 have
not yet been officially reported by IDEAM.

Table 3. Multiannual P24h-max observations of Rafael Núñez airport (1941–2015).

Year 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951
P24h-max (mm) 60.0 35.0 30.0 89.0 71.0 60.0 60.0 107.0 54.0 85.0 93.0
Year 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962
P24h-max (mm) 41.0 51.0 90.0 110.0 95.0 40.0 109.0 68.0 109.0 65.0 75.0
Year 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973
P24h-max (mm) 59.0 69.0 89.0 76.0 67.0 89.0 129.0 157.0 104.7 120.0 74.1
Year 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984
P24h-max (mm) 126.4 101.6 54.4 60.5 68.6 120.7 135.9 124.4 98.0 63.4 102.7
Year 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995
P24h-max (mm) 164.5 64.9 171.3 115.0 201.8 77.8 32.5 161.5 133.4 54.8 76.3
Year 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
P24h-max (mm) 99.4 99.6 85.6 108.5 116.2 76.2 73.5 161.8 149.0 76.4 122.3
Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 - -
P24h-max (mm) 183.1 95.3 61.3 150.7 146.1 94.2 88.0 116.0 51.8 - -

3. Methodology
The methodology used in this research is presented in the flowchart of Figure 4. Each step is
described in detail in the following sub-sections.
Fluids 2018, 3, 27 6 of 19
Fluids 2018, 3, x FOR PEER REVIEW 6 of 19

P24h-max data

Rainfall Frequency Analysis

Stationary Freq. Analysis Non-stationary Freq. Analysis

• GEV-Type 1 (Gumbel) GEV with time-dependent
• GEV-Type 3 (Weibull) location parameter

Design rainfall
(stationary & non-stationary)
Tr = 2, 5, 10, 25, 50, and 100 years

Design DSR (Q)

(stationary & non-stationary)
Tr = 2, 5, 10, 25, 50, and 100 years

Hydraulic modeling
(stationary & non-stationary)
Tr = 2, 5, 10, 25, 50, and 100 years

Figure 4. Proposed methodology flowchart.

Figure 4. Proposed methodology flowchart.

3.1. Rainfall Stationary Frequency Analysis

3.1. Rainfall Stationary Frequency Analysis
The stationary frequency analysis is based on the following assumptions: (a) the cumulative
The stationary frequency analysis is based on the following assumptions: (a) the cumulative
distribution functions (CDFs) do not change over time [2] and (b) the variable analyzed must be
distribution functions (CDFs) do not change over time [2] and (b) the variable analyzed must be
random and independent [1]. In hydrology, there are several CDFs for the analysis of extreme values,
random and independent [1]. In hydrology, there are several CDFs for the analysis of extreme values,
namely: gamma, extreme value (EV) or Fisher-Tippett (types 1, 2, and 3), generalized extreme value
namely: gamma, extreme value (EV) or Fisher-Tippett (types 1, 2, and 3), generalized extreme value
(GEV), log-normal, Pearson 3, Log-Pearson 3, among others. The GEV was used for the frequency
(GEV), log-normal, Pearson 3, Log-Pearson 3, among others. The GEV was used for the frequency
analysis carried out in this section, as it has shown to work best in Colombia.
analysis carried out in this section, as it has shown to work best in Colombia.
The expression for the GEV distribution is given by Equation (1), where F(Z) is the CDF, z is the
The expression for the GEV distribution is given by Equation (1), where F(Z) is the CDF, z is
random variable, k is the shape (or shift) parameter, β is the mode (location parameter), and α is the
the random variable, k is the shape (or shift) parameter, β is the mode (location parameter), and α is
dispersion (scale parameter). The GEV distribution may adopt one of the three types of the EV
the dispersion (scale parameter). The GEV distribution may adopt one of the three types of the EV
distribution depending upon the value of k [8–10]: (a) when k equals zero, EV is type 1 (Gumbel) [11];
distribution depending upon the value of k [8–10]: (a) when k equals zero, EV is type 1 (Gumbel) [11];
(b) when k is less than zero, EV is type 2 (Frechet) [12]; and (c) when k is greater than zero, EV is type
(b) when k is less than zero, EV is type 2 (Frechet) [12]; and (c) when k is greater than zero, EV is type 3
3 (Weibull) [13].
(Weibull) [13].
/
−
( ) = exp − 1 − z − β 1/k (1)
" #
F ( Z ) = exp − 1 − k (1)
α
The estimated GEV’s parameters were: α = 31.35 mm, β = 78.097 mm, and k = 0.06. Table 4 shows
The estimated
the values GEV’sfor
of the P24h-max parameters
each of thewere: α = 31.35 used.
two methods mm, βGray
= 78.097
cellsmm,
showand k = 0.06. values.
maximum Table 4 shows
the values
Lastly,ofthe
theCDF
P24h-max for each of the
goodness-of-fit wastwo methods
carried used.
out via Gray cells[13]
Chi-square show
andmaximum values.
Kolmogorov-Smirnov
Lastly, the CDF goodness-of-fit was carried out via Chi-square [13] and Kolmogorov-Smirnov
[14–17] tests. Values of 0.06001 and 2.2876 were obtained for the Kolmogorov-Smirnov and [14–17]
Chi-
tests.
square Values of 0.06001 and
tests respectively. 2.2876
These were indicate
results obtainedthat
for the Kolmogorov-Smirnov and Chi-square
GEV distribution accurately representstests
the
behavior of the rainfall data analyzed.
Fluids 2018, 3, 27 7 of 19

respectively. These results indicate that the GEV distribution accurately represents the behavior of the
rainfall data analyzed.

Table 4. P24h-max values for stationary freq. analysis.

P24h-max
Tr (year)
Gumbel Weibull
2 88.6 89.2
5 124.4 125.0
10 148.1 148.1
25 178.1 176.4
50 200.3 196.9
100 222.4 216.8
Bold values indicate the largest value of the two CDF used.

3.2. Rainfall Non-Stationary Frequency Analysis

The non-stationary analysis considers the following assumptions [2–4]: (a) the probability of
exceeding the random variable over the years (pj ) of the CDFs may increase or decrease, and (b)
the extreme values of the analyzed variable are independent in time. Parameters of the CDFs are
considered time-varying in order to compute the probability of exceeding occurring, which implies
that extreme values are not identically distributed over time.
To define the return period under non-stationary environments (Tr,NSC ), it is necessary to
understand the concept of the waiting time. The waiting time is a random variable (X) defined
as the occurrence of a value exceeding the design value for the first time. In stationary conditions,
the probability of exceeding remains constant over time, which implies that a geometric distribution
can be used to compute the expected value (E(x) = Tr, SC = 1/p). Commonly, this is called the return
period [8]. In contrast, under non-stationary conditions, the return period (Tr) is computed with a
non-homogeneous geometric distribution (Equation (2)) [18]:
∞ x
∑∏


Tr, NSC = 1 + 1 − pj (2)
x =1 j =1

where pj is the time-varying probability of exceeding, and the subscript j represents the projecting year.
The GEV distribution with a temporal trend in the location parameter (βt ) can be expressed as
(Equation (3)) [2,18]: "  −1/k #
z − βt

F ( Z, β t ) = exp − 1 + k (3)
α

The P24h-max data at the Rafael Núñez airport weather station (Table 4) exhibits an increasing
linear trend over time (dotted black line in Figure 5), which confirms the non-stationary occurrence of
the analyzed variable.
To estimate the P24h-max values, the non-stationary frequency analysis followed these steps [4,18]:
(a) selecting the CDF under non-stationary condition (GEV or Gumbel); (b) defining a model (constant,
linear, or establishing other models) of each parameter (location, shape, and scale) with its trend
(increasing or decreasing); (c) estimating parameters by likelihood method; (d) calculating the
goodness-of-fit test of the Akaike Information Criterion (AIC) [19] and then selecting the CFD with the
minimum value of it; and (e) estimating the P24h-max for various return periods.
For this study, the CDF under non-stationary condition (GEV distribution) was selected
considering the distribution used in the stationary scenario of Section 3.1. The GEV’s parameters were
estimated by means of a code [2,18] programmed in R software (version 3.3.1, R Development Core
Team, Auckland, New Zealand) with the library nsextremes. The resulting linear trend lines for the
location (increasing trend) and scale (remains constant) parameters are shown in Figure 5 (blue and
Fluids 2018, 3, 27 8 of 19

green lines for the location and scale parameters, respectively). The values of the aforementioned
parameters were: βt = 79.543 mm + [(0.4949 mm/year) (tmean – 1978)], α = 31.1 mm, and k = −0.106.
The tmean term in the previous expression represents an analyzed year centered over its own mean
between 1941 and 2015. A sensitivity analysis showed that variations of both the scale and the shape
parameters do not bring an adequate trend for the P24h-max when the results were compared with the
findings of IDEAM. The AIC test was used in order to evaluate the GEV distribution goodness-of-fit
by comparing both scenarios (stationary and non-stationary). The obtained values were 747.5103
(non-stationary) and 753.3721 (stationary). The results of the P24h-max for several return periods are
presented
Fluids 2018, 3,in Table
x FOR 5. REVIEW
PEER 8 of 19

Figure 5. P24h-max
Figure and
5. P24h-max GEV
and parameters
GEV parameters(location andand
(location scale) behavior
scale) over
behavior time
over (Rafael
time Núñez
(Rafael airport
Núñez station).
airport
station).
Table 5. P24h-max values for non-stationary freq. analysis.
To estimate the P24h-max values, the non-stationary frequency analysis followed these steps [4,18]:
Tr (year) condition
(a) selecting the CDF under non-stationary P24h-max
(GEV(mm)
or Gumbel); (b) defining a model
2 128.1
(constant, linear, or establishing other models) of each parameter (location, shape, and scale) with its
5
trend (increasing or decreasing); (c) estimating parameters140.6
by likelihood method; (d) calculating the
10 161.5
goodness-of-fit test of the Akaike Information
25 Criterion (AIC)
189.1[19] and then selecting the CFD with
the minimum value of it; and (e) estimating 50 the P24h-max for various
213.1 return periods.
For this study, the CDF under 100 non-stationary condition
244.8 (GEV distribution) was selected
considering the distribution used in the stationary scenario of Section 3.1. The GEV’s parameters were
estimated
3.3. Curve by means
Number of aEstimation
(CN) code [2,18] programmed in R software (version 3.3.1, R Development Core
Team, Auckland, New Zealand) with the library nsextremes. The resulting linear trend lines for the
Given that the Gordo creek watershed is composed of several soil cover types, an area-based
location (increasing trend) and scale (remains constant) parameters are shown in Figure 5 (blue and
composite CN was estimated following the methodology established in the National Engineering
green lines for the location and scale parameters, respectively). The values of the aforementioned
Handbook, Section 4 (NEH-4) [20], where a series of lookup tables are given for different combinations
parameters were: βt = 79.543 mm + [(0.4949 mm/year) (tmean – 1978)], α = 31.1 mm, and k = −0.106. The
of hydrological soil groups (HSGs) and soil cover. The HSG of the watershed is C, as per the description
tmean term in the previous expression represents an analyzed year centered over its own mean between
of the soil texture provided in Section 2.1 (clay-loam soil texture). The values shown in those tables
1941 and 2015. A sensitivity analysis showed that variations of both the scale and the shape
represent what is called the average antecedent runoff condition (ARC-II). The ARC is a concept
parameters do not bring an adequate trend for the P24h-max when the results were compared with the
introduced into the CN methodology to try to account for the potential for runoff generation in a
findings of IDEAM. The AIC test was used in order to evaluate the GEV distribution goodness-of-fit
watershed due to the soil water content after several rainfall events [21]. For that, the total amount of
by comparing both scenarios (stationary and non-stationary). The obtained values were 747.5103
cumulative rainfall of the 5 previous days (5-day antecedent precipitation, P5 ) is estimated to establish
(non-stationary) and 753.3721 (stationary). The results of the P24h-max for several return periods are
whether the soil is in dry, average/normal or wet (Table 6) [21]; a set of equations are then proposed to
presented in Table 5.
adjust the CN accordingly (Equations (4) and (5)) [8,22].
Table 5. P24h-max values for non-stationary
CN freq. analysis.
II
CNI = (4)
2.281 P−24h-max
Tr (year) 0.01281CN
(mm) II
2 128.1
5 140.6
10 161.5
25 189.1
50 213.1
Fluids 2018, 3, 27 9 of 19

CNII
CNIII = (5)
0.427 − 0.00573CNII

Table 6. ARC types based on P5 .

ARC Type Dormant Season Growing Season

I (dry) <12.7 mm <36 mm
II (normal) 12.7–27.9 mm 36–53 mm
III (wet) >27.9 mm >53 mm

In 2010–2011, Colombia underwent the longest and most devastating rainy season ever recorded.
Floods were reported throughout the country, especially in the Caribbean coastal region. In this study,
a CN adjustment was carried out for ARCIII to mimic a saturated soil as the most critical scenario in
the Gordo creek watershed in that season, during which more than 53 mm of rain fell in 5 consecutive
days. Table 7 shows the values of CN composite obtained for average and wet conditions (gray cells).

Table 7. Composite CN for Gordo creek watershed.

CN Developed Areas CN Rural Areas

Item
Impervious Urban Residential Bush Forest
CNII 98 88 85 86 60
CNIII 99.1 94.4 92.9 93.4 77.5
Area (ha) 2.9 12.7 23.4 109.5 109.5
CNII sub-areas 86.9 73
CNII composite 75.1
CNIII sub-areas 93.8 85.5
CNIII composite 86.7

3.4. Time of Concentration Estimation

The time of concentration may be defined as the time it takes a drop of water to travel from the
hydraulically most remote point to the watershed outlet (or any other point of interest) [23]. It is one
of the most critical parameters for any hydrological analysis and the subsequent design of a hydraulic
structure aimed at managing stormwater. Many methodologies (empirical and semi-empirical) have
been proposed for its estimation [24,25]. For this study, the Kirpich method was used (Equation (6)) [26],
as the Gordo creek watershed fits some of the morphometric characteristics needed for using this
formula [25].

Tc = 0.0195L0.77 S−0.385 (6)

In Equation (6), Tc is the time of concentration (min), L is the length of the main
watercourse/channel (m) and S is the average slope of the main watercourse/channel (m/m). After
plugging the L and S values provided in Section 2.1, the Tc equals 73.3 min.

3.5. Direct Surface Runoff (DSR) Estimation

The DSRs produced by the rainfall under both stationary and non-stationary conditions for
different return periods (2, 5, 10, 25, 50, and 100 years) were estimated (Table 8) via Hydrologic
Engineering Center Hydrologic Modeling System (HEC-HMS), which is a CN-based hydrological
model developed by the U.S. Corps Army of Engineers [27]. HEC-HMS compiles several
methodologies for hydrological analysis integrated into one tool capable of quickly modeling different
scenarios at the same time.
Apart from the adjusted CN value estimated in Section 3.3, the HEC-HMS simulations for the
various return periods were made under the following conditions: (a) 3-h rainfall as it is the most
 different return periods (2, 5, 10, 25, 50, and 100 years) were estimated (Table 8) via Hydrologic
Engineering Center Hydrologic Modeling System (HEC-HMS), which is a CN-based hydrological
model developed by the U.S. Corps Army of Engineers [27]. HEC-HMS compiles several
methodologies for hydrological analysis integrated into one tool capable of quickly modeling
different
Fluids 2018, scenarios
3, 27 at the same time. 10 of 19
Apart from the adjusted CN value estimated in Section 3.3, the HEC-HMS simulations for the
various return periods were made under the following conditions: (a) 3-h rainfall as it is the most
typical duration historically
typical duration historicallyobserved
observedininCartagena
Cartagena dede Indias;
Indias; (b) (b)
the the
3-h 3-h rainfall
rainfall corresponds
corresponds to 79% to
79% of the P
of the P24h-max24h-max , and its distribution was estimated via multiannual probabilistic
, and its distribution was estimated via multiannual probabilistic analysis of 30 analysis of
30
pluviographs [8] [8]
pluviographs (Figure
(Figure 6) 6) (P90%
(P90% wasthe
was theone
oneused;
used;ititindicates
indicatesthethe probability
probability that
that the
the rainfall
rainfall
pattern observed falls to the left of the P90% curve); (c) no rainfall area reduction
pattern observed falls to the left of the P90% curve); (c) no rainfall area reduction was needed due was needed dueto
to the size of the watershed; (d) hydrograph generation via SCS-unit hydrograph
the size of the watershed; (d) hydrograph generation via SCS-unit hydrograph method; and (e) a lag method; and (e)
atime
lag of
time0.6ofTc0.6 Tc (Tlag
(Tlag = 0.6Tc).
= 0.6Tc). Table Table 8 summarizes
8 summarizes the peakthe peak flow values
flow values for stationary
for stationary and
and non-
non-stationary conditions.
stationary conditions.

Figure 6. 3-h rainfall distribution pattern at Rafael Núñez airport weather station.
Figure 6. 3-h rainfall distribution pattern at Rafael Núñez airport weather station.

Table 8. Peak flow values.

Table 8. Peak flow values.
Peak Flow (m3/s)
Tr (year) (m3 /s)
Peak FlowNSC
Tr (year) SC
2 SC
14.3 25.3NSC
2 5 24.3
14.3 29.025.3
5 10 24.3
31.2 35.229.0
10 25 31.2
40.2 43.635.2
25 40.2 43.6
50 50 47.1
47.1 51.051.0
100 100 53.9
53.9 60.960.9

3.6. Channel
3.6. Channel Hydraulic Modeling
Hydraulic Modeling
A one-dimension
A one-dimension (1-D)
(1-D) hydraulic
hydraulic modeling
modeling waswas carried
carried out
out using
using the
the Hydrologic
Hydrologic Engineering
Engineering
Center River
Center River Analysis
Analysis System
System (HEC-RAS)
(HEC-RAS) software,
software, developed
developed by by the
the U.S.
U.S. Corps
Corps Army
Army of of Engineers.
Engineers.
The simulation was performed under gradually varied flow and subcritical conditions (Froude (Froude
The simulation was performed under gradually varied flow and subcritical conditions number
number less than 1). Also, the following inputs were utilized in the simulation: (a) Gordo
less than 1). Also, the following inputs were utilized in the simulation: (a) Gordo creek’s bathymetry, creek’s
bathymetry, that includes ten cross-sections (Figure 7) of the last 178.5 m of the stream (the most
that includes ten cross-sections (Figure 7) of the last 178.5 m of the stream (the most critical portion of critical
portion
the of the watershed)
watershed) which consistwhich
of aconsist
naturalofcross-section
a natural cross-section with three
with three culverts; (b)culverts; (b) design
design DSR DSR
for several
return periods estimated in Section 3.5; (c) hydraulic control section governed by a subcritical flow at the
downstream box culvert (section K0+000), which produces a backwater profile in the entire channel.
The natural channel exhibits different cross-sections, as presented in Figure 8a,b.
Figure 9a,b show the plant view and the longitudinal profile of Gordo creek watershed outlet
(gray-shaded areas show the location of culverts). A description of the culverts is presented in Table 9.
The entire natural channel presents a Manning’s coefficient of 0.050 (natural minor streams with some
Fluids 2018, 3, 27 11 of 19

Fluids 2018, 3, x FOR PEER REVIEW 11 of 19

Fluids 2018, 3, x FOR PEER REVIEW 11 of 19
weeds and stone), while culverts have a value of 0.014 (concrete in culverts with bends, connections,
for several
andseveral return periods estimated in Section 3.5; (c) hydraulic control section governed by a
some debris)
for return[8].
periods estimated in Section 3.5; (c) hydraulic control section governed by a
subcritical flow at the downstream box culvert (section K0+000), which produces a backwater profile
subcritical flow at the downstream box culvert (section K0+000), which produces a backwater profile
Table 9. Description
in the entire channel. The natural channel of culverts.
exhibits different cross-sections, as presented in Figure
in the entire channel. The natural channel exhibits different cross-sections, as presented in Figure
8a,b.
8a,b. Channel Section
Figure 9a,b show the plant view and the longitudinal profile of Gordo creek watershed outlet
Slope (%)
Figure 9a,b show the plant view and the longitudinal profile of Gordo creek watershed outlet
(gray-shaded
From areas show To the locationType
of culverts). A description
Number of Cells of the culverts is presented in Table
Dimensions
(gray-shaded areas show the location of culverts). A description of the culverts is presented in Table
9. TheK0+000
entire naturalK0+008.44
channel presents a Manning’s
Box culvert coefficient
2 of 0.050 (natural
2.5 (natural
m minor streams
× 2.0 mminor with
1.25 with
9. The entire natural channel presents a Manning’s coefficient of 0.050 streams
K0+012.44 K0+062.44 Pipe culvert 2 Diameter of 1.6
some weeds and stone), while culverts have a value of 0.014 (concrete in culverts with bends,m 0.37
someK0+067.11
weeds and stone), while culverts have a value 2of 0.014 (concrete
K0+097.11 2.0 m × in
2.0 culverts with
0.033bends,
connections, and some debris) [8]. Box culvert m
connections, and some debris) [8].

Figure 7.
Figure7. Plant view
Plantview
7.Plant of
viewof the
ofthe critical
thecritical point
criticalpoint of
pointof flooding.
offlooding.
flooding.
Figure
.05 .05 .05
11.0 .05 .05 .05 Legend
11.0 Legend
WS T -NSC = 2 years
WS
WST -NSC
T -SC =
=25 years
years
10.5 WS T -SC = 5 years
Ground
10.5 Ground
Ineff
IneffSta
Bank
10.0 Bank Sta
10.0
(m)
(m)
Elevation

9.5
Elevation

9.5

9.0
9.0

8.5 (a)
8.5 (a)
8.0
8.0 0 5 10 15 20 25
0 5 10 15 20 25
Station (m)
Station (m)

Figure 8. Cont.
Fluids 2018, 3, 27 12 of 19
Fluids2018,
Fluids 2018,3,3,x xFOR
FORPEER
PEERREVIEW
REVIEW 1212ofof1919

.05 .05 .05

14 .05 .05 .05
Legend
14 Legend
WS T -NSC = 2 years
WS T -NSC = 2 years
WS T -SC = 5 years
WS T -SC = 5 years
13
13 Ground
Ground
Bank Sta
Bank Sta

12
12
Elevation (m)
Elevation (m)

11
11

10
10

9
9

8 (b)
(b)
8 0 5 10 15 20 25 30 35
0 5 10 15 20 25 30 35
Station (m)
Station (m)

Figure8.8.Natural
Figure Naturalcross-sections
cross-sectionsofofGordo
Gordocreek.
creek.(a)
(a)Cross-section
Cross-sectionatatK0+018.5;
K0+018.5;(b)
(b)cross-section
cross-sectionatat
Figure 8. Natural
K0+178.5. WS cross-sections
indicates the ofsurface
water Gordolevel.
creek. (a) Cross-section at K0+018.5; (b) cross-section at
K0+178.5. WS indicates the water surface level.
K0+178.5. WS indicates the water surface level.

(a)
(a)

(b)
(b)

Figure 9. Hydraulic modeling with HEC–RAS: (a) Plan view; (b) longitudinal profile. WS indicates the
Figure9.9.Hydraulic
Figure Hydraulicmodeling
modelingwith
withHEC–RAS:
HEC–RAS:(a)
(a)Plan
Planview;
view;(b)
(b)longitudinal
longitudinalprofile.
profile.WS
WSindicates
indicates
water surface level.
the water surface level.
the water surface level.
 Channel Section
Slope (%)
From To Type Number of Cells Dimensions

K0+000 K0+008.44 Box culvert 2 2.5 m × 2.0 m 1.25

K0+012.44
Fluids 2018, 3, 27 K0+062.44 Pipe culvert 2 Diameter of 1.6 m 0.37 13 of 19
K0+067.11 K0+097.11 Box culvert 2 2.0 m × 2.0 m 0.033

4. Results and Discussion

The analysis of of the
the multiannual
multiannual PP24h-max observations recorded
24h-max observations recorded at at the
the Rafael Núñez airport
weather station not only evidenced an increasing increasing trendtrend (Figure
(Figure 5),5), but
but also,
also, in
in Table 3, it could be
noticed that, before 1985 rainfall events above 148.1 mm (estimated rainfall value for a 10-year return
period under stationary conditions in Table Table 4)4) just occurred
occurred once (in 1970). After 1985, rainfall events
of that magnitude (or higher) have occurred nine times (2011 rainfall was included as it is close
enough). Colombian legislation recommends a 50-year return period for the design of open channels
for watersheds of an area less than 1000 ha [5]. In In Tables
Tables 4 and 5, the rainfall for this return period, period,
even for stationary
stationary conditions,
conditions,was wasestimated
estimatedtotobe benearly
nearly200200mm.
mm.However,
However, in Table 3, a rainfall
in Table 3, a rainfall of
of 201.8
201.8 mm mm hashas been
been alreadyrecorded
already recordedinin1989. 1989.FromFromthis,
this,ititmay
maybe be inferred
inferred thatthat any hydraulic
structure to be designed for such return period should be also evaluated for a higher value to test its
hydraulic performance.
The estimated P24h-max (Sections3.1
24h-max (Sections 3.1and
and3.2)3.2)and
andthethe peak
peak flow
flow (Section
(Section 3.5)
3.5) for several return
periods under stationary and non-stationary conditions depicted
periods under stationary and non-stationary conditions depicted in Figure 10 in Figure 10 also showed an increase,
also showed an
from 6% to
increase, 44%6%
from for P
to24h-max
44% for(Table 10) and
P24h-max from10)
(Table 8%andto 77%
fromfor8%peakto flow
77%(Table
for peak11), depending
flow (Tableupon 11),
the return period.
depending upon the Thereturn
increase seen in
period. TheP24h-max
increase forseen
returnin periods of 5return
P24h-max for to 100periods
years areofquite similar
5 to 100 to
years
each other, with an average value of 1.09. This value is in line with those observed
are quite similar to each other, with an average value of 1.09. This value is in line with those observed in some areas of
the Caribbean
in some areas ofregion of Colombia
the Caribbean [6,7].ofDespite
region Colombia the [6,7].
fact that a 44%the
Despite increase
fact thatin Pa 24h-max for the in
44% increase 2-year
P24h-
return period looks, at a glance, to be too high when compared to the remaining
max for the 2-year return period looks, at a glance, to be too high when compared to the remaining values, it may be
construed as an
values, it may beindication
construedthat a more
as an conservative
indication that a moreapproach might be
conservative conducted
approach when
might designing
be conducted
with
whenthis low return
designing withperiod as the
this low probability
return period asofthe exceeding is high.
probability of exceeding is high.

Figure10.
10.PP24h-max and peak flow increase for various return periods.
Figure 24h-max and peak flow increase for various return periods.

Table 10. P24h-max values comparison.

P24h-max (mm)
Tr (year) Ratio (NSC/SC) Avg. Ratio
SC NSC P24h-max Diff. (NSC-SC)
2 89.2 128.1 38.9 1.44
5 125.0 140.6 15.6 1.12
10 148.1 161.5 13.4 1.09
25 178.1 189.1 11.0 1.06 1.09
50 200.3 213.1 12.8 1.06
100 222.4 244.8 22.4 1.10

For the peak flow, the values of the overall increase ratio have the same behavior of that shown by
P24h-max , with the maximum value occurring at a 2-year return period. A 44-percent increase in rainfall
(Table 10) caused a 77-percent peak flow rise (Table 11). Once again, this indicates the precautions
 25 178.1 189.1 11.0 1.06 1.09
50 200.3 213.1 12.8 1.06
100 222.4 244.8 22.4 1.10

For the peak flow, the values of the overall increase ratio have the same behavior of that shown
Fluids 2018, 3, 27 14 of 19
by P24h-max, with the maximum value occurring at a 2-year return period. A 44-percent increase in
rainfall (Table 10) caused a 77-percent peak flow rise (Table 11). Once again, this indicates the
precautions
that that the
the designers designers
must mustusing
take when take when using this
this return return
period. Forperiod. Forreturn
the other the other returnitperiods,
periods, may be
it may be that
observed observed that theinincrease
the increase P24h-maxinresulted
P24h-max resulted
in a peakinflow
a peak flow of
increase increase of arange.
a similar similar range.

Table 11. Peak flow values comparison.

Peak Flow (m3/s)

Peak Flow (m3 /s)
Tr (year) Ratio (NSC/SC) Avg. Avg.
RatioRatio
Tr (year) SC NSC Flow Diff. (NSC-SC) Ratio (NSC/SC)
SC NSC Flow Diff. (NSC-SC)
2 14.3 25.3 11.0 1.77
2 14.3 25.3 11.0 1.77
5 24.3 29.0 4.7 1.19
5 24.3 29.0 4.7 1.19
10 10 31.2
31.2 35.2
35.2 4.0 4.0 1.13 1.13
25 25 40.2
40.2 43.6
43.6 3.4 3.4 1.08 1.08 1.12 1.12
50 50 47.1
47.1 51.0
51.0 3.9 3.9 1.08 1.08
100 100 53.9
53.9 60.9
60.9 7.0 7.0 1.13 1.13

non-stationary and stationary difference and ratios

Figures 11 and 12 show respectively the non-stationary
between P 24h-maxand
P24h-max andpeak
peakflow forfor
flow different
differentreturn
returnperiods (DPmax,
periods (DPmax,DQp, Pratio
DQp, andand
Pratio Qratio). As
Qratio).
expected,
As bothboth
expected, show howhow
show the the
behavior of Pof
behavior , and the
P24h-max
24h-max watershed’s
, and response
the watershed’s to it (the
response peak
to it (theflow),
peak
have the
flow), same
have thetrend.
same trend.

Figure 11.
11. PP24h-max and peak flow variation for various return periods. DPmax and DQp are the P24h-max
Figure 24h-max and peak flow variation for various return periods. DPmax and DQp are the
and peak flow difference between between
P24h-max and peak flow difference non-stationary and stationary
non-stationary conditions.
and stationary conditions.
Fluids 2018, 3, x FOR PEER REVIEW 15 of 19

Figure 12. P24h-max and peak flow ratio variation for various return periods. Qratio and Pratio are the
Figure 12. P24h-max and peak flow ratio variation for various return periods. Qratio and Pratio are the
ratio of non-stationary to stationary values of peak flow and P24h-max respectively.
ratio of non-stationary to stationary values of peak flow and P24h-max respectively.

The water level reached after the simulations carried out under stationary and non-stationary
conditions indicates that modeling under non-stationary scenario better represents the current
hydrological conditions of the Gordo creek watershed, where rainfall events of low magnitude are
generating floods every year during the rainy season (during which the soil is saturated most of the
time) at the watershed’s outlet, when, in theory, they ought not to. This can be observed, for instance,
Fluids 2018, 3, 27 15 of 19

The water level reached after the simulations carried out under stationary and non-stationary
conditions indicates that modeling under non-stationary scenario better represents the current
hydrological conditions of the Gordo creek watershed, where rainfall events of low magnitude are
generating floods every year during the rainy season (during which the soil is saturated most of the
time) at the watershed’s outlet, when, in theory, they ought not to. This can be observed, for instance,
in the results of the hydraulic simulation at K0+018.5 shown in Table 12 (P24h-max , peak flow, and water
elevation) and Figure 13, where the bankfull level (9.94 m) is reached at a lower return period (floods
occur more frequently) under non-stationary conditions (Tr < 2-year). Under stationary conditions, a
2-year return period flow, defined as the mean annual flow (or annual maximum daily flow) [28,29],
did not result in a bankfull section, which is not what is currently occurring in the study area (floods
every year). The change in the rainfall pattern evidenced in previous sections may be the reason
for a shift towards more recurrent and higher-than-usual events that cause floods more frequently
than in the past. Likewise, statistically speaking, the ACI results obtained in Section 3.2 (747.5103 for
non-stationary and 753.3721 for stationary) demonstrated that the non-stationary condition is more
adequate. The AIC test serves to evaluate how good a model is for predicting future values [30].
The lower the value of the AIC the better the model.

Table 12. Maximum water level (elevations) at cross-section K0+018.5.

Max. Water Level Elevation Reached at Peak Flow

Bankfull Elev. Is at 9.94 m Tr
2 5 10 25 50 100
P24h-max (mm) 89.2 125.0 148.1 178.1 200.3 222.4
SC Q (m3 /s) 14.3 24.3 31.2 40.2 47.1 53.9
Water elev. (m) 9.56 10.18 10.70 10.70 11.02 11.22
P24h-max (mm) 128.1 140.6 161.5 189.1 213.1 244.8
NSC Q (m3 /s) 25.3 29.0 35.2 43.6 51.0 60.9
Water elev. (m) 10.24 10.56 10.75 10.92 11.13 11.29
Water elev. difference (m) 0.68 0.38 0.05 0.22 0.11 0.07
Values in bold indicate which year (return period) first exceeded the bankfull level under SC and NSC.
Fluids 2018, 3, x FOR PEER REVIEW 16 of 19

Figure 13.
13. Water
Waterlevel
levelatatcross-section K0+018.5
cross-section forfor
K0+018.5 different return
different periods
return under
periods stationary
under and non-
stationary and
stationary conditions.
non-stationary conditions.

In
In terms
terms of
of the
the flood depth (elevation
flood depth (elevation above
above bankfull
bankfull level)
level) at
at the
the area
area of
of the
the watershed
watershed outlet,
outlet,
Table 13
Table 13 and
and Figure
Figure 1414 show
show the
the water
water depth
depth above
above the
the bankfull
bankfull level
level for
for different
different return
return periods.
periods.
This sensitivity
This sensitivity analysis
analysis reveals the necessity
reveals the necessity for
for sizing
sizing hydraulic
hydraulic structures
structures under
under non-stationary
non-stationary
conditions to avoid more recurring floods. Notwithstanding that the 1-D hydraulic simulation is
incapable of delineating the extension of the flooded area beyond the creek’s bank, the difference
between NSC and SC elevations in Table 13 is an indication of what locals and workers of the
industrial area have been noticing every year: floods in this area have been gradually worsening in
terms of both frequency of occurrence and area covered by water (as well as the water depth marks
Fluids 2018, 3, 27 16 of 19

conditions to avoid more recurring floods. Notwithstanding that the 1-D hydraulic simulation is
incapable of delineating the extension of the flooded area beyond the creek’s bank, the difference
between NSC and SC elevations in Table 13 is an indication of what locals and workers of the industrial
area have been noticing every year: floods in this area have been gradually worsening in terms of both
frequency of occurrence and area covered by water (as well as the water depth marks left at some
points that local people use as reference points). Further research is needed, though. Future work in
this area must include a more detailed survey (topographic study) covering more area, so as to assess
and more accurately quantify (numerically and spatially) the aforementioned anecdotal evidence.

Table 13. Flood depth at cross-section K0+018.5.

Flood Depth (m)

Tr (year) SC NSC NSC-SC
2 0.00 0.30 0.30
5 0.24 0.62 0.38
10 0.76 0.81 0.05
25 0.76 0.98 0.22
50 1.08 1.19 0.11
100 1.28 1.35 0.07
Values in bold indicate which return period first exceeded the bankfull level under SC and NSC.
Fluids 2018, 3, x FOR PEER REVIEW 17 of 19

Figure
Figure 14.
14. Flood
Flood depth
depth at
at cross-section
cross-section K0+018.5.
K0+018.5.

In addition to the already quantified effects of climate change on Cartagena’s P24h-max pattern and
An option for making decisions when designing some hydraulic structures would be to perform
the Gordo creek watershed’s hydrological response, the fact that almost 85% of the watershed is still
a benefit-cost analysis to compare a design under non-stationary conditions versus under stationary
rural indicates that any increase in the impervious area will both raise the peak flow and reduce the
condition, taking into account free board heights, allowing handling of the non-stationary peak flow
time of concentration. These two variables will worsen the situation at the outlet of Gordo creek
depending on the design return period.
watershed unless a series of measures are implemented. For instance, the combination of sustainable
In addition to the already quantified effects of climate change on Cartagena’s P24h-max pattern
urban drainage systems (SUDS) [31,32], stormwater storage vaults/tanks (both online and offline),
and the Gordo creek watershed’s hydrological response, the fact that almost 85% of the watershed is
aquifer storage and recovery (ASR) wells, infiltration ponds, among others, have proved to be
still rural indicates that any increase in the impervious area will both raise the peak flow and reduce
effective at managing stormwater by: (a) keeping the peak flow at the same magnitude (or lower)
the time of concentration. These two variables will worsen the situation at the outlet of Gordo creek
when pre-development and post-development scenarios are compared, (b) avoiding the design of
watershed unless a series of measures are implemented. For instance, the combination of sustainable
larger hydraulic structures, (c) minimizing the risk of floods, and (d) recharging aquifers, as a plus.
urban drainage systems (SUDS) [31,32], stormwater storage vaults/tanks (both online and offline),
aquifer
5. storage and recovery (ASR) wells, infiltration ponds, among others, have proved to be effective
Conclusions
at managing stormwater by: (a) keeping the peak flow at the same magnitude (or lower) when
P24h-max valuesand
pre-development associated to a given return
post-development periodare
scenarios arecompared,
a key variable when estimating
(b) avoiding design
the design DSR
of larger
for hydraulic
hydraulic structures
structures, (c) for stormwater
minimizing themanagement,
risk of floods,especially in ungauged
and (d) recharging watersheds.
aquifers, The values
as a plus.
obtained in the AIC test carried out in this study, indicated that a frequency analysis under non-
stationary conditions represents best the behavior of the rainfall patterns of the P24h-max observations
at the Rafael Núñez station. A stationary scenario is, in this case, further from the natural reality,
when compared to non-stationary conditions. This was also confirmed by both the increase in the
occurrence of P24h-max events greater than or equal to 148.1 mm (value for a 10-year return period
Fluids 2018, 3, 27 17 of 19

5. Conclusions
P24h-max values associated to a given return period are a key variable when estimating design
DSR for hydraulic structures for stormwater management, especially in ungauged watersheds.
The values obtained in the AIC test carried out in this study, indicated that a frequency analysis
under non-stationary conditions represents best the behavior of the rainfall patterns of the P24h-max
observations at the Rafael Núñez station. A stationary scenario is, in this case, further from the natural
reality, when compared to non-stationary conditions. This was also confirmed by both the increase
in the occurrence of P24h-max events greater than or equal to 148.1 mm (value for a 10-year return
period under stationary conditions), and the increasing linear trend over time of the overall P24h-max
observations. These findings may be an indication that the typical frequency analysis of rainfall under
stationary conditions may no longer be applicable when calculating the design rainfall associated with
a given return period for sizing hydraulic structures throughout Cartagena de Indias. Furthermore,
designing under stationary conditions may have direct implications in the decisions local authorities
will be taking in the coming years, given that millions of dollars will be invested in upgrading the
city’s stormwater system. The 1-D hydraulic simulation performed herein revealed that the Gordo
creek watershed outlet area can be flooded even with an event of 2-year return period (every year)—a
situation that had never been observed in the past according to what local people have affirmed.
The increase in the P24h-max , though, is not the sole factor to be taken into account when evaluating
the reasons behind the more frequent floods reported. Uncontrolled and unplanned urbanization of
vegetated areas may and will exacerbate the recurrent floods registered within the study area.
It is noteworthy to mention that sizing larger hydraulic structures for stormwater management as
the only solution for the larger rainfall events estimated under non-stationary conditions might be
unsuitable and unsustainable in cases where surface area limitations exist (especially in urban areas).
Therefore, implementing best management practices for controlling stormwater sources in old and
newly constructed areas shall be one of the city’s goals in maintaining the design peak flow values,
especially under increasing non-stationary conditions.

Acknowledgments: The authors want to thank Germán Castaño for contributing with relevant and helpful
technical information. Also, thanks to Orlando Viloria for helping to edit some of the blue-prints.
Author Contributions: Oscar E. Coronado-Hernández has carried out the hydrological and hydraulic simulations.
Rainfall frequency analysis has been performed by Alvaro Gonzalez-Alvarez and Oscar E. Coronado-Hernández.
Data and results analysis has been done by Alvaro Gonzalez-Alvarez. Helena M. Ramos and Vicente Fuertes-Miquel
have helped writing and revising the paper.
Conflicts of Interest: The authors declare no conflict of interest. The funding sponsors had no role in the design
of the study; in the collection, analyses and interpretation of data; in the writing of the manuscript; and in the
decision to publish the results.

References
1. Faber, B. Current methods for hydrologic frequency analysis. In Proceedings of the Workshop on
Nonstationarity, Hydrologic Frequency Analysis, and Water Management, Boulder, CO, USA, 13–15 January
2010; pp. 33–39. Available online: http://www.cwi.colostate.edu/nonstationarityworkshop/index.shtml
(accessed on 21 December 2017).
2. Obeysekera, J.; Salas, J.D. Quantifying the uncertainty of design floods under nonstationary conditions. J.
Hydrol. Eng. 2014, 19, 1438–1446. [CrossRef]
3. Wang, D.; Hagen, S.C.; Alizad, K. Climate change impact and uncertainty analysis of extreme rainfall events
in the Apalachicola River basin, Florida. J. Hydrol. 2012, 480, 125–135. [CrossRef]
4. Salas, J.D.; Obeysekera, J.; Vogel, R.M. Techniques for assessing water infrastructure for nonstationary
extreme events: A review. Hydrol. Sci. J. 2018. [CrossRef]
5. Ministry of Housing, City, and Development (MinVivienda); Republic of Colombia. Resolution 0330 of 8 June
2017, Technical Guidelines for the Sector of Potable Water and Basic Sanitation (RAS). 2017. Available online:
http://www.minvivienda.gov.co/ResolucionesAgua/0330%20-%202017.pdf (accessed on 28 October 2017).
Fluids 2018, 3, 27 18 of 19

6. Poveda, G.; Alvarez, D.M. The collapse of the stationarity hypothesis due to climate change and climate
variability: Implications for hydrologic engineering design. Rev. Ing. Univ. Andes 2012, 36, 65–76.
7. IDEAM, PNUD, MADS, DNP, CANCILLERÍA. New Scenarios of Climate Change for Colombia 2011–2100
Scientific Tools for Deparment-based Decision Making—National Emphasis: 3rd National Bulletin on
Climate Change. 2015. Available online: http://documentacion.ideam.gov.co/openbiblio/bvirtual/022964/
documento_nacional_departamental.pdf (accessed on 28 October 2017).
8. Chow, V.T.; Maidment, D.R.; Mays, L.W. Applied Hydrology, 1st ed.; McGraw-Hill: New York, NY, USA, 1988;
pp. 350–376.
9. Palutikof, J.P.; Brabson, B.B.; Lister, D.H.; Adcock, S.T. A review of methods to calculate extreme wind speeds.
Meteorol. Appl. 1999, 6, 119–132. [CrossRef]
10. Wilks, D.S. Statistical Methods in the Atmospheric Sciences, 3rd ed.; Academic Press: Oxford, UK; Waltham,
MA, USA, 2011; pp. 105–110.
11. Gumbel, E.J. The return period of flood flows. Ann. Math. Stat. 1941, 2, 163–190. [CrossRef]
12. Frechet, M. Sur la loi de probabilité de l’ecart máximum (On the probability law of maximum values).
Ann Soc. Pol. Math. 1927, 6, 93–116.
13. Weibull, W. A statistical theory of the strength of materials. In Proceedings of the Ingeniors Vetenskaps
Akademien (The Royal Swedish Institute for Engineering Research) No. 51, Stockholm, Sweden; 1939;
pp. 5–45. Available online: http://www.scirp.org/%28S%28czeh2tfqyw2orz553k1w0r45%29%29/reference/
ReferencesPapers.aspx?ReferenceID=1923153 (accessed on 19 April 2018).
14. Pearson, K. On the systemic fitting of curves to observations and measurements. Biometrika 1902, 1, 265–303.
[CrossRef]
15. Kolmogorov, A.N. Sulla determinazione empirica di una legge di distribuzione. G. Inst. Ital. Attuari 1933, 4,
83–91.
16. Smirnov, N.V. Estimate of deviation between empirical distribution functions in two independent samples.
Bull. Moscow Univ. 1939, 2, 3–16.
17. Smirnov, N.V. Table for estimating the goodness of fit of empirical distributions. Ann. Math. Stat. 1948, 19,
279–281. [CrossRef]
18. Obeysekera, J.; Salas, J.D. Frequency of recurrent extremes under nonstationarity. J. Hydrol. Eng. ASCE 2016,
21. [CrossRef]
19. Akaike, H. A new look at the statistical model identification. IEEE Trans. Autom. Control 1974, 19, 716–723.
[CrossRef]
20. USDA-Soil Conservation Service (USDA SCS). National Engineering Handbook, Section 4, Hydrology
(NEH-4); USDA-SCS: Washington, DC, USA, 1985. Available online: https://policy.nrcs.usda.gov/
OpenNonWebContent.aspx?content=18393.wba (accessed on 17 November 2017).
21. USDA Natural Resources Conservation Service (USDA NRSC). Conservation Engineering Division. Urban
Hydrology for Small Watersheds, Technical Release 55 (TR-55); USDA-NRCS: Washington, DC, USA, 1986.
Available online: https://www.nrcs.usda.gov/Internet/FSE_DOCUMENTS/stelprdb1044171.pdf (accessed
on 19 December 2017).
22. Mishra, S.K.; Singh, V. Soil Conservation Service Curve Number (SCS-CN) Methodology, 1st ed.; Springer:
Dordrecht, The Netherlands, 2003.
23. Ponce, V.M.; Hawkins, R.H. Runoff curve number: Has it reached maturity? J. Hydrol. Eng. ASCE 1996, 1,
11–19. [CrossRef]
24. Gericke, O.J.; Smithers, J.C. Review of methods used to estimate catchment response time for the purpose of
peak discharge estimation. Hydrol. Sci. J. 2014, 59, 1935–1971. [CrossRef]
25. Sharifi, S.; Hosseini, S.M. Methodology for identifying the best equations for estimating the time of
concentration of watersheds in a particular region. J. Irrig. Drain. Eng. ASCE 2011, 137. [CrossRef]
26. Kirpich, Z.P. Time of concentration of small agricultural watersheds. Civ. Eng. 1940, 10, 362–368.
27. Gonzalez, A.; Temimi, M.; Khanbilvardi, R. Adjustment to the curve number (NRCS-CN) to account for the
vegetation effect on hydrological processes. Hydrol. Sci. J. 2015, 60, 591–605. [CrossRef]
28. U.S. Geological Survey (USGS). Flood-frequency Analyses, Manual of Hydrology: Part 3, Flood-flow Techniques;
Government Printing Office: Washington, DC, USA, 1960. Available online: https://pubs.usgs.gov/wsp/
1543a/report.pdf (accessed on 17 December 2017).
Fluids 2018, 3, 27 19 of 19

29. U.S. Geological Survey (USGS). Theoretical Implications of under Fit Streams, Flood-flow Techniques; Government
Printing Office: Washington, DC, USA, 1965. Available online: https://pubs.usgs.gov/pp/0452c/report.pdf
(accessed on 17 December 2017).
30. Mohammed, E.A.; Far, B.H. Emerging Business Intelligence Framework for a Clinical Laboratory Through
Big Data Analytics 2015. Chapter 32. pp. 577–602. Available online: https://doi.org/10.1016/B978-0-12-
802508-6.00032-6 (accessed on 28 January 2018).
31. Ramos, H.M.; Pérez-Sánchez, M.; Franco, A.B.; López-Jiménez, P.A. Urban Floods Adaptation and
Sustainable Drainage Measures. Fluids 2017, 2, 61. [CrossRef]
32. New York City Department of Environmental Protection. Guidelines for the Design and Construction of
Stormwater Management Systems; New York City Department of Environmental Protection: New York, NY,
USA, 2012. Available online: http://www.nyc.gov/html/dep/pdf/green_infrastructure/stormwater_
guidelines_2012_final.pdf (accessed on 4 January 2018).

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).