Geotech Geol Eng (2015) 33:1221–1233

DOI 10.1007/s10706-015-9897-1

ORIGINAL PAPER

Correlations of SPT, CPT and DPL Data for Sandy Soil

in Tanzania
Mwajuma Ibrahim Lingwanda .
Stefan Larsson . Dalmas L. Nyaoro

Received: 1 April 2014 / Accepted: 18 May 2015 / Published online: 23 May 2015
Ó Springer International Publishing Switzerland 2015

Abstract In Tanzania, standard penetration test methods. Results indicate that DPL data correlate
(SPT) is the most commonly used in situ test for better with CPT than SPT data, with lower magnitudes
foundation design site investigations. In an effort to of transformation uncertainty. The local SPT–CPT
increase the amount of geotechnical information at correlations compare fairly well to those in the
low cost, the quicker and much cheaper dynamic literature. The established correlations extend the
probing of light (DPL) hammer is sometimes per- function of DPL data to analysis and design.
formed along with SPT to supplement the expensive
SPT. Nevertheless, the information gathered with DPL Keywords SPT  CPT  DPL  Correlation 
has been applicable only for site stratification. Re- In situ tests
cently, the static cone penetration test (CPT) has also
been introduced in the country with a view to
combining these methods in site investigations. In 1 Introduction
this study, side by side testing was performed with the
three in situ methods and correlations established The standard penetration test (SPT) is the most
through regression analysis and arithmetic mean commonly used in situ test method for geotechnical
investigations in Tanzania. One of the great advan-
tages of this test is that it offers retrieval of samples for
M. I. Lingwanda (&)
Department of Built Environment Engineering, Mbeya further processing in the laboratory. Nevertheless, the
University of Science and Technology, test is known to be costly and time-consuming, making
P. O. Box 131, Mbeya, Tanzania it unattractive to extensive investigation programs.
e-mail: [email protected]; Geotechnical engineers in Tanzania have been pairing
[email protected]
SPT with the light dynamic probing test (DPL) to
S. Larsson increase the amount of site information whereby DPL
Department of Civil and Architectural Engineering, KTH results are used for soil stratification only. To date only
Royal Institute of Technology, Brinellvägen 23, one study has been reported (Ibrahim and Nyaoro
100 44 Stockholm, Sweden
e-mail: [email protected] 2011) to determine local correlations between SPT
N values and DPL N10 values. Based on data collected
D. L. Nyaoro from projects conducted in different areas in Dar es
Department of Transportation and Geotechnical Salaam, the study indicated a fairly good correlation
Engineering, University of Dar es Salaam,
P. O. Box 35131, Dar es Salaam, Tanzania between N and N10 in different soil types and at
e-mail: [email protected] different penetration depths.

123
1222 Geotech Geol Eng (2015) 33:1221–1233

Around the world, the DPL test is less popular in geotechnical information, such correlations are still
comparison to SPT and the static cone penetration test useful especially in low to moderate risk projects
(CPT) as also mentioned by Spagnoli (2007). Conse- (Robertson 2012). Data were collected from an
quently, there is little research and information extensive side by side testing program performed
available concerning DPL and its application. How- specifically for this study where the three pen-
ever, the test has several advantages over SPT etrometers’ data were collected within a diameter of
including simplicity, continuity, speedy testing as 0.5 m to minimize the effects of soil variability.
well as low investment cost. The main drawback with
DPL is the fact that it cannot in most cases be
applicable to depths greater than 8 m below the 2 Test Site
ground (Steffanoff et al. 1988) making it reliable only
for shallow investigations. The test site is situated on the University of Dar es
A perfectly-performing soil testing method is only Salaam Mwalimu Nyerere campus, south-west of the
an impossible fantasy as there is always a tradeoff university center and about 100 m from the Msewe
between one attribute and another. A combination of river. The geological map of the study area (Geolo-
testing methods in a single investigation program gical Survey Division Dodoma 1963), identifies
provides better information than any of the individual presence of kaolinic sandstones, fine to medium
methods on its own (Rogers 2006). With a combina- grained, off-white in color overlaid by clay-bound
tion of methods, the transformation error from mea- sands. A significant portion of the western part of Dar
sured to evaluated property can be significantly es Salaam has such geological characteristics. The site
reduced (Ching et al. 2010, 2013; Müller et al. measures 30 m by 30 m and inclines towards the south
2013). In this regard, the CPT has recently been at an average slope of 7 %. The site is intended to
introduced in Tanzania to widen the choice of in situ serve for the collection of in situ and laboratory data
testing methods and more importantly to facilitate for the ongoing research aimed at improving the
combinations of testing methods and thereby max- quality and usefulness of geotechnical information
imize geotechnical investigation information at an while reducing the cost of such work. A plan of the
affordable cost. Among its many advantages, the CPT area is presented in Fig. 1, which also indicates the
is fast, economical and has good repeatability although relative location of the test points.
in normal procedures it cannot facilitate collection of
samples for laboratory testing.
Both SPT and CPT are well established methods West - East distance (m)
with a rich history of performance in different parts of -6 0 6 12 18 24 30 36
-6
the world. This has enabled correlation of soil
behavior indices such as relative density, shear A12 A11 A10 A1 A2 A3
0
strength and stiffness modulus with both SPT and
CPT test results. There are also many correlations
North - South distance (m)

B12 B11 B10 B1 B2 B3

6
between CPT and SPT data. Notably, Schmertmann
(1970), Robertson et al. (1983), Kulhawy and Mayne C12 C11 C10 C1 C2 C3
12
(1990), Jefferies and Davies (1993) are well cited and
used. However, this is not the case with DPL and in D12 D11 D10 D1 D2 D3
fact is the main reason for its under-utilization. Most 18

SPT–CPT correlations are general, they utilized data E12 E11 E10 E1 E2 E3
from different soil conditions and with a relatively 24

large distance between the correlated information. F12 F11 F10 F1 F2 F3

This study aims to correlate DPL test results with both 30

SPT and CPT to facilitate transformation of DPL data

and increase its applicability. SPT to CPT correlations 36
CPT SPT DPL UNSUCCESSFUL CPT
are also developed and compared to those in the
literature. Like other empirical interpretations of Fig. 1 Relative location of tests on the site

123
Geotech Geol Eng (2015) 33:1221–1233 1223

The soil-testing program began with the establish- number of blows per 300 mm penetration denoted as
ment of a regular grid pattern over the area while N. Simultaneously, undisturbed and disturbed samples
making horizontal distances of 6 m between adjacent were collected from the boreholes for further labora-
marked points. CPT tests were then performed at each tory testing. DPL tests were performed while record-
of the 36 points reaching 10 m below the ground. As ing measurements in number of blows per 100 mm
indicated in Fig. 1, reliable CPT results could not be penetration denoted as N10.
obtained at some locations mainly due to failure of the
anchoring system of the CPT equipment. Eight 3.2 Determination of Correlations
locations were then selected randomly out of the 36
points for further DPL and SPT testing. At each of the It is common practice to convert measured SPT N
selected points, a DPL test was then performed close values to equivalent measurements at 60 % of the
to the previously performed CPT, followed by drilling theoretical driving energy to maintain consistency in
and SPT while keeping all the three test profiles within data collected with different equipment of different
a 0.5 m diameter. Along with SPT data collection, energy levels. In the present study, the hammer system
samples were retrieved for laboratory testing, which is known to be capable of delivering 70 % of
included determination of the fines percentage (F), maximum theoretical energy on average. The stan-
mean grain size (D50), natural water content (wn), dardized SPT N value is therefore given by Eq. 1 as
Liquid limit (LL), bulk density and oedometer con- follows:
strained modulus of the soil.
N  70 %
N60 ¼ ð1Þ
60 %
3 Methods where N is the raw number of blows per 300 mm
penetration and N60 the standardized N value.
3.1 Equipment and Testing With CPT data, the common practice is to utilize
pore pressure corrected cone resistance and disregard
Test equipment for SPT consisted of the standard sleeve friction measurements as they are considered to
sampler, a 63.5 kg automatic trip hammer with a drop be less accurate (Lunne et al. 1997). Accordingly,
height of 760 mm capable of delivering on average most CPT–SPT correlations with the exception of
70 % of the theoretical energy as measured during those based on a soil behavior type index such as that
equipment calibration by the manufacturer. Drilling by Jefferies and Davies (1993) were established with
was done using a hand auger that creates 200 mm qc (corrected or not). Examples include Robertson
diameter boreholes. The CPT equipment included a et al. (1983), Kulhawy and Mayne (1990), Akca
stand-alone pusher capable of providing a maximum (2003).
force of 150 kN, a measuring system comprising a CPT measurements are obtained continuously and
10 cm2 tip area and 150 cm2 friction sleeve, also recorded at 10 mm intervals, DPL at 100 mm intervals
capable of measuring pore water pressures. The DPL while SPT data represent discrete measurements at
equipment consisted of a 10 kg hammer with a falling 1.5 m apart. It was necessary to condition data so that
height of 500 mm and rods of 22 mm diameter fitted reference is made at common locations while creating
to a 25.2 mm diameter cone of 90° apex angle. This data pairs for correlations. Also of importance is to
equipment complies with the international (ISSMGE) consider the influence zones created in each test since
standard described by Steffanoff et al. (1988). they have different sampler sizes and a different
The CPT test was performed at the standard rate of driving mechanism. Rogers (2006) described the zone
20 mm/s while recording measurements at 10 mm of influence with the SPT sampler to be in the order of
intervals. Measurements of cone resistance (qc), sleeve 4–7 times the sampler diameter plus the 300 mm that
friction (fc), pore water pressure, cone inclination and are presumably tested. With the standard sampler of
the total driving force were recorded at each location. 51 mm diameter and taking an average of 5 sampler
For the eight test locations, SPT was performed at diameters, the SPT influence zone in this study was
1.5 m discrete intervals to a depth of 10 m below the considered to be 560 mm. The influence zone for CPT
ground level and results were recorded as standard also varies and according to Robertson (2009)

123
1224 Geotech Geol Eng (2015) 33:1221–1233

increases in magnitude with soil strength and stiffness se

COVtrans;X!Y ¼  ð5Þ
and may reach up to 15 times cone area. The study site Y
has stiff soils, implying that the standard CPT cone can where X ! Y means a transformation from parameter
make up to 550 mm of influence zone. With smaller X to Y and Y is the mean of parameter Y.
dimensions, the DPL is expected to create influence The second method involves determination of the
zones less than those for CPT and SPT. The 560 mm arithmetic average of the ratios of the two variables to
zone was taken as reference for both tests and a be correlated. These ratios were calculated separately
moving window average of 560 mm was taken for each profile and the grand mean ratio was then
through DPL and CPT data to establish pairs corre- determined.
sponding to SPT data.
Two approaches were used to determine correla- 3.3 Outliers Detection Using Tukey Mean–
tions; the statistical ordinary least squares (OLS) Difference Plots
method of regression and the simple arithmetic mean
method. In the statistical method, scatter plots of data Tukey mean–difference plots (Tukey 1977; Altman
pairs were plotted and different trends were fitted to and Bland 1983) are useful when assessing the
the data. To determine the best fit, Eq. 2 (Eisenhauer agreement between two methods. The plots can also
2003) was used to calculate the coefficient of deter- be successfully applied to detect outliers in correlation
mination (R2) which was applied to compare different analysis. Agreement and correlation between two
alternatives. For linear regressions, the coefficient of methods are distinguished by the nature of the linear
correlation (r) was also determined using Eq. 3 and relationship between data pairs. While correlations are
was checked for significance by comparing to critical considered strong if the data points lie along any
values as recommended in the Pearson product– straight line, agreements are strong if the data lie along
moment correlation table. Equation 4 was then ap- the equality line. In the analysis of agreement, a Tukey
plied to determine the standard error of estimate (se). mean–difference plot is established by plotting the
P 2 means against the differences of data pairs from two
2 Y^i  Y
R ¼P 2
ð2Þ methods. The plots are meaningful if the two methods
ðYi  YÞ are similar in terms of physical measurements and
units. In this study, the agreement is established
R2 ¼ r 2 ð3Þ between data obtained from one in situ method, say
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi CPT data, and data derived from a correlation with
P 2
Yi  Y^i another method, say SPT data.
se ¼ ð4Þ
n1 The following is a sequence of the procedure used
to determine the best agreement and, accordingly,
where Yi is the ith value of a data set, Y the mean of the
outliers in the established correlations:
data set and Y^i the ith fitted value, n the number of data
points, and Se is the standard error of estimate. 1. First the linear correlation between two data sets is
Some degree of uncertainty is induced when a established, say qc versus N60
measured parameter is transformed to a desired 2. The CPT data is expressed in terms of SPT data in
property (Phoon and Kulhawy 1999; Akbas and the form of Eq. 6.
Kulhawy 2010) using a model such as the empirical
models derived by regression analysis. The value se is qc jN60 ¼ aN60 þ b ð6Þ
an indicator of the magnitude of transformation
uncertainty. The coefficient of variation of the trans- 3. The mean and difference between measured qc
formation uncertainty (COVtrans ) from parameter X to and those derived from Eq. 6 are determined. In
Y is determined by the ratio between se and the mean this case, the assumption is that the ‘true’ qc value
as indicated by Eq. 5. The coefficient of variation is a lies between the measured value and that derived
better parameter of comparison between two quanti- from N60 value. The standard deviation (SD) of
ties of different units than the standard deviation. the differences is also calculated.

123
Geotech Geol Eng (2015) 33:1221–1233 1225

4. The Tukey mean–difference plot is prepared and Measured Soil Resistance

boundary lines of 2SD are indicated on the plot. 0 10 20 30 40 50 60 70 80 90
0
5. Data beyond the boundaries are considered to be
outliers and filtered from the data set. The residue 1
is used to establish new correlations.
2
6. The best agreement between, say, qc and qc jN60 is
achieved when the difference between the values 3
is zero. In Eq. 6, the coefficients a and b are
adjusted to arrive at a zero difference and hence 4

Depth (m)
the best agreement over the present interval.
5

The procedure is repeated for the relationship between 6

CPT–DPL data and for SPT–DPL data.
7

8
4 Results
9
4.1 Field and Laboratory Observations
10

Results from the eight locations, namely A2, A12, B11,

11
C2, D3, D10, E12 and F2, as indicated in Fig. 1, where
CPT qc (MPa) CPT 10fs (MPa)
the three in situ tests were conducted side by side, are DPL N10 SPT N
considered for analysis. Typical in situ measurements
are shown in Fig. 2 as obtained at location B11 where fs Fig. 2 In situ measurements at location B11
values are magnified 10 times to enable them to be
presented in the same figure. A summary of soil et al. 1983). However, both profiles indicated a general
properties from laboratory measurements is provided increase in both qc and fs with depths up to around 7 m,
in Fig. 3. Visual observation from borehole profiles followed by a decrease then another slight increase at
and lab results indicate a variable site consisting of about 8 m depth. Very high values of qc and fs were
clayey to silty sand with lenses of clean sand around obtained at this site, where the average maximum qc
4.5–6 m deep. The sand lenses are generally less than was 26 MPa and for fs was 1.44 MPa. The overall
1 m thick and were observed in 5 out of the 8 mean qc was 14 MPa and fs was 0.61 MPa. It is
boreholes. important to note that the high magnitudes of cone
Six N values were obtained from each of the eight resistance and sleeve friction caused the equipment to
10 m deep borehole, making a total of 48 SPT data occasionally operate near its full capacity and at
points. The average N60 value for the soil was found to location D10 the operation had to stop before the
be 44 from a range of 19–89, which according to intended 10 m penetration could be reached, due to
Kulhawy and Mayne (1990) can be interpreted as equipment overload. Pore pressure measurements
medium-dense to dense soil. The DPL reached refusal were insignificant, indicating absence of ground water
at different depths in the eight locations ranging from in the penetrated zone. Consequently, the equipment is
5.5 m at point C2 to 7.9 m at point F2 with an average considered as CPT instead of CPTu. In this case, there
penetration of 6.9 m. The average range of measured was no need to perform end area correction of
DPL values was from 5 to 66 blows per 100 mm measured values as such corrections after all are
penetration with an overall mean of 29. The lowest N10 unimportant in sands (Robertson 2012).
values were generally recorded near the ground
surface and highest at the points of termination. 4.2 Correlations and Agreements
An overall observation in CPT data profiles indi-
cates a rapid variation of qc with depth, which is A total of 47N60 to qc data pairs were available and
typical characteristic of sandy deposits (Robertson plotted as indicated in Fig. 4. Using the method of

123
1226 Geotech Geol Eng (2015) 33:1221–1233

Soil D50 mm F% Wn% LL%

Strata 0.1 0.6 1.1 0 20 40 2 8 14 20 40 60
0

2 S
C
3 L

4
Depth (m)

5 S
P
6

7 S
M
8 I

10

Fig. 3 Laboratory test results; stratification, mean grain size distribution, percentage of fines, natural water content, liquid limit

least squares, the expression qc ¼ 0:14N60 þ 8:61 was 30 all data

obtained for all data with a moderate correlation of qc= 0.14N60 + 8.61
r = 0.675 and se = 3.45. The transformation uncer- R² = 0.456
r = 0.615
tainty from N60 to qc (COVtrans N60!qc ) is therefore 25 n = 47
determined to be 24.6 %. The 95 % confidence sε = 3.45
interval for the expression with all data is also
indicated in the respective figure. A Tukey mean– 20
difference plot for qc and qc jN60 which enabled the
CPT qc (MPa)

best agreement to be determined as qc jN60 ¼

0:204N60 þ 5:35; is presented in Fig. 5. After screen- 15
ing outliers from the data set, the new expression
obtained is qc ¼ 0:15N60 þ 7:49 with r = 0.766 and
se = 2.97, which makes COVtrans N60!qc drop to 10

21.2 %. filtered
The magnitudes of sleeve friction are substantial qc = 0.15N60 + 7.49
5 R² = 0.587
and could not merely be ignored. Adding fs values r = 0.766
to corresponding qc values improves correlation with n = 44
sε = 2.97
N60, as can be seen from Fig. 6. The new relation-
0
ship obtained is ðqc þ fs Þ ¼ 0:15N60 þ 8:97 with a 10 30 50 70 90
moderate to high correlation of r = 0.692 and a SPT N60
slightly smaller se = 3.42 as compared to qc–N60
outliers
correlation. A similar procedure is applied as for the Linear (filtered)
qc–N60 relationship to establish Tukey mean–differ- Linear (boundaries, all data)
ence plot and detect outliers. The best agreement is
obtained as ðqc þ fs ÞjN60 ¼ 0:209N60 þ 5:75. The Fig. 4 Statistical correlation of qc and N60

123
Geotech Geol Eng (2015) 33:1221–1233 1227

10 30 all data
(qc+fs)= 0.15N60 + 8.97
R² = 0.479
r = 0.692
25
n = 47
s = 3.42
5
difference of qc and qc|N60

20

CPT (qc + fs) MPa

0 15

10

-5
filtered
(qc +fs) = 0.16N60 + 7.87
5 R² = 0.604
r = 0.777
n = 44
-10 s =2.96
0
5 15 25 10 30 50 70 90
average of qc and qc |N60 SPT N60

filtered data, best agreement criteria outliers

boundaries, mean + or - 2SD Linear (filterd)
all data, correlation criteria
Linear (boundaries, all data)
Linear (filtered data, best agreement criteria)
Linear (all data, correlation criteria)
Fig. 6 Statistical correlation of (qc ? fs) and N60
Fig. 5 Tukey mean–difference plot for qc and qc jN60

0:434N10 þ 0:16: After removing outliers, the corre-

new relationship after removing outliers is ðqc þ fs Þ ¼ lation line is qc ¼ 0:38N10 þ 1:57 with r = 0.880,
0:16N60 þ 7:87 with r = 0.777 and se = 2.96 leading se = 1.81 and transformation uncertainty = 12.9 %.
to COVtrans N60!qcþfs of 20 %. When sleeve friction is taken into account, there is a
In the arithmetic average method, the ratio qc =N60 small improvement in the correlation. As shown in
was determined for each data pair in each individual Fig. 9, the linear expression for all data is ðqc þ fs Þ ¼
borehole and the grand average determined to be 0.37 0:38N10 þ 2:45 which has a correlation of r = 0.847,
with a standard deviation of 0.12 for the filtered data. se = 2.16 and COVtrans N10!qcþfs is 14.8 %. The best
To evaluate variation of the ratio, a scatter is plotted of agreement was determined to be ðqc þ fs ÞjN10 ¼
qc =N60 against depth as shown in Fig. 7, where a 0:445N10 þ 0:33 and the new correlation for the
general decreasing trend in the ratio with depth was filtered data is ðqc þ fs Þ ¼ 0:40N10 þ 1:66 with
observed. Outliers detected by Tukey mean–differ- r = 0.891 and se = 1.78. The transformation uncer-
ence method are also plotted in the scatter diagram. tainty falls to 12.2 %. The arithmetic average for the
In a manner similar to CPT–SPT correlations, 40 ratio ðqc þ fs Þ=N10 was found to be 0.46 with a
data pairs of N10 to qc relationships were obtained and standard deviation of 0.06 for the filtered data.
the corresponding correlation determined as before. For the relationship between N10 and N60, 31 data
Figure 8 is the scatter diagram for the case where pairs were available and the resulting correlations are
sleeve friction was not considered. Considering all shown in Fig. 10. A linear regression line N60 ¼
data, a linear relationship of qc ¼ 0:36N10 þ 2:34 was 1:38N10  9:67 with a strong correlation of r = 0.781
obtained with a strong correlation of r = 0.834 and and se = 9.13 was obtained implying that
se = 2.18, implying a COVtrans N10!qc of 15.6 %. The COVtrans N10!N60 is 20.7 %. A Tukey mean–difference
best agreement between measured qc and those plot identified only one data point as an outlier. The
derived from N10 was determined to be qc jN10 ¼ best agreement was obtained as N60 jN10 ¼ 1:4N10 

123
1228 Geotech Geol Eng (2015) 33:1221–1233

qc/N60 25
all data
0.1 0.3 0.5 0.7 0.9 qc= 0.36N10 + 2.34
0
R² = 0.696
r = 0.834
1 20 n = 40
s = 2.18
2

3
15

CPT qc MPa
4
Depth (m)

5
10

7 filtered
5 qc = 0.38N10 + 1.57
8 R² = 0.775
r = 0.880
n = 38
9
s = 1.81
0
10 0 20 40 60
DPL N10
11
outliers
outliers Linear (filtered)
Linear (clayey sand-silty sand boundary) Linear (boundaries, all data)

Fig. 7 Variation of the ratio qc/N60 with depth Fig. 8 Statistical correlation of qc and N10

considered reasonable for representing the site under

11: The new relationship for the filtered data is analysis. It can be seen that the presence of clean sand
N60 ¼ 1:01N10 þ 0:44, r = 0.724 and se = 7.14 lenses caused a wide variation in the soil mean grain
where the transformation uncertainty is 16.2 %. The size (D50) and fines percentage (F) around the region.
arithmetic average method led to a ratio of N60 =N10 ¼ The sand may also be responsible for the very low
1:03 and a standard deviation of 0.22 for the filtered measured natural water content (wn) between 4.5 and
data. 6 m depth.
Between 0 and 7 m depth DPL and CPT results are
comparable, where both trends indicate an increase in
5 Discussion soil resistance with depth. Similarly, SPT results
indicate increase in soil resistance with depth compa-
5.1 Lab and Field Results rable to CPT up to around 7 m depth. However, an
obvious difference can be seen at depths greater than
Visual inspection of boreholes suggested a very 7 m, where N values increase faster than qc values. A
variable soil stratification. This is corroborated by comparison with DPL values for depth larger than 7 m
the results of laboratory tests presented in Fig. 3. The might have helped explain this wider difference
presence of lenses of clean sand at different depths in between N and qc but is not possible because DPL
the boreholes made it difficult to demarcate the terminated at a shallower depth. Rogers (2006)
boundary between clayey sand and silty sand layers. pointed out that cone resistance values of CPT are
In fact, the bottom part is not entirely made up of silty affected by soil water content such that dry soils tend
sand but only dominated by it as the top layer is to offer higher resistances. This would mean that there
dominated by clayey sand. What is presented in Fig. 3 is an actual increase in resistance to penetration with
is a simplification of the stratification, which is depth which is captured by SPT but misinterpreted by

123
Geotech Geol Eng (2015) 33:1221–1233 1229

25 90
all data all data
(qc + fs) = 0.38N10 + 2.45 N60 = 1.38N10 + 9.67
R² = 0.718 80 R² = 0.610
r = 0.847 r = 0.780
20 n = 40 n = 31
70 s = 9.13
s = 2.16

60
CPT (qc+fs) MPa

15

SPT N60
50

40
10
30

filtered filtered
(qc + fs) = 0.40N10 + 1.66 20
5 N60= 1.01N10 + 0.44
R² = 0.794 R² = 0.524
r = 0.891 10 r = 0.724
n = 38 n = 30
s = 1.78 s = 7.14
0 0
0 20 40 60 0 10 20 30 40 50 60 70
DPL N10 DPL N10

outliers outlier
Linear (filtered) Linear (filtered)
Linear (Arthmetic average) Linear (boundaries, all data)
Linear (boundaries, all data)
Fig. 10 Statistical correlation of N60 and N10
Fig. 9 Statistical and arithmetical correlation of (qc ? fs) and
N10
these phenomena may have contributed to incompa-
rability of CPT and SPT in the lower part of the site
CPT, due to the effect of simultaneously increased stratum and are a subject for further research.
water content with depth. Both CPT and SPT measurements are known to be
A certain degree of incomparability is to be dependent on overburden stress levels and as a result,
expected between the results of two different methods normalization of data to minimize such effects has
although they are both conducted in situ. Mayne been suggested in both tests. For CPT data, stress
(2006) pointed out the difference in strain rates arising normalization enables proper profiling, as documented
in the process of each test as a factor that considerably by Eslami and Fellenius (2004). However, Fellenius
affects compatibility between the methods. According (2009) suggested such normalization be implemented
to Robertson (2012), the response due to shearing of for testing depths greater than 30 m to avoid misin-
soils during static loading is either dilative or terpretation of soil profiles at shallow depths. Over-
contractive in both drained and undrained conditions. burden stress normalization for SPT data have been
This also may explain the widened gap between CPT studied by, among others, Liao and Whitman (1986),
and SPT measurements as the depth increased. On one and Skempton (1986). It is also recommended by
hand, the CPT dependency on relative density and many researchers including, Kulhawy and Mayne
confining stress results to lower values of qc in dense (1990) and Rogers (2006). Application of overburden
silty soils at high confining pressures because dela- stress normalization generally scales down measure-
tancy causes soils in such a state to act as loose soils ments taken at high stress levels while increasing those
and therefore contract (Boulanger 2003). On the other measured at low stresses. Mahler and Szendefy (2009)
hand, performing SPT on silty to clayey sands if reported improvement in data scatter due to stress
compared to pile driving under similar conditions, normalization resulting in better correlations between
high rebounds are to be anticipated (Hussein et al. heavy dynamic probing and qc. Moss et al. (2006)
2006) resulting to false high N values. One or both of recommends similar treatment for normalization of

123
1230 Geotech Geol Eng (2015) 33:1221–1233

SPT and CPT data in case of side by side measure- OLS line. The difference between values determined
ments from both tests. In the present study, measure- by the simple average method and those from OLS are
ments were generally shallow and no results were higher for stiffer soils. Besides simplification, the
stress-normalized. arithmetic mean method was intended to serve for
estimation of values below the tested range (below
5.2 CPT–SPT Correlations N60 = 20). Another alternative is to use the expression
0:53
qc ¼ 1:97N60 which appears to be more consistent
In Fig. 11, a comparison is presented of different with the OLS line even for stiffer soils.
suggestions for predicting CPT data from SPT data or Schmertmann (1970) made recommendations re-
vice versa. In this study, the method of ordinary least garding CPT to SPT data ratios for different soil types
squares (OLS) produces the expression qc ¼ and cautioned that they are generally conservative.
0:15N60 þ 7:49; which is valid between the tested The observation that qc ¼ 0:37N60 this study is similar
range of N60 = 20–90 for clayey to silty sand soil. The to ðqc þ fs Þ ¼ 0:35N60 recommended by Schmert-
arithmetic mean method yields a much simpler mann (1970) for fine to medium and slightly silty
expression, qc ¼ 0:37N60 . It can be seen from the sands. It is also similar to the findings by Kara and
figure that for low N60 values of up to around 40, the Gündüz (2010) using the arithmetic average method
arithmetic mean method yields conservative CPT that qc ¼ 0:42N60 for sand. Schmertmann (1970) also
values, where as for N60 values greater than 40 it suggested the dependency of CPT to SPT data ratios to
results in values higher than those derived from the test depth, where he observed a decrease in the ratio
with depth. This condition is also observed in the
35 current study and presented in Fig. 7. In the current
study, however, the variation may be contributed by
more than just depth. The soil at the study site appears
30
to be very variable but the top 5 m is dominated by
clayey sand while the lower part is dominated by silty
sand. This distinction results in qc ¼ 0:45N60 for the
25
upper clayey sand and qc ¼ 0:29N60 for the lower silty
sand and is consistent with the general observation that
CPT–SPT data ratios are lower in silty soils (Sch-
CPT qc MPa

20
mertmann 1970; Akca 2003).
Robertson et al. (1983) made a significant contri-
15 bution to the development of CPT to SPT correlations
by considering the effect of actual SPT driving energy
as well as soil type in terms of mean grain size (D50).
10 The soil at the study site was found to have a mean D50
of 0.38 mm and a standard deviation of 0.22 mm.
Assuming normal distribution of soil properties at the
5 study site, 95 % of the soil is expected to have a D50
value of between 0.16 and 0.60 mm. This range of D50
and the obtained ratio qc =N60 ¼ 0:37 from the arith-
0 metic mean method plot very close to the chart
0 20 40 60 80 100
SPT N60
recommended by Robertson et al. (1983).
Kulhawy and Mayne (1990) presented relationships
Linear (This study, filtered) between qc and N values in relation to D50 and the soil
Power (This study, filtered)
Linear (Acka (2003) ) fine content. In fact the D50 to qc/N relationship was an
Linear (Schmertmann (1970)) extension and modification of the recommendations
Linear (this study, filtered, arithmetic average)
made by Robertson et al. (1983) by including new data
Fig. 11 Comparison of CPT–SPT data correlations obtained at that time. With relation to D50, Kulhawy

123
Geotech Geol Eng (2015) 33:1221–1233 1231

and Mayne proposed the use of Eq. 7 while Eq. 8 is and DPL data is better when the friction component is
proposed to relate the ratio with F; taken into account for CPT and the expression
ðqc þ fs Þ ¼ 0:40N10 þ 1:66 is therefore proposed for
qc =Pa =N ¼ 5:44D0:26
50 ð7Þ
transforming DPL data to equivalent CPT. As indi-
F cated in Fig. 9, the arithmetic method which resulted
qc =Pa =N ¼ 4:25  ð8Þ in the ratio ðqc þ fs Þ=N10 ¼ 0:46 yields results very
41:3
close to those using statistical methods and therefore
where Pa is the atmospheric pressure and the other may be used to simplify estimation but with caution
symbols are as defined previously. because it results in un-conservative estimates of CPT
Substituting the range 0.16–0.60 mm of D50 in values for DPL values greater than 30. Because the
Eq. 7, the obtained ratio qc/N is between 0.34 and 0.48 DPL could barely reach 7 m deep penetration the
which indicates that the ratio obtained in this study falls correlation of CPT and DPL is dominated by data from
within the range predicted by the Kulhawy and Mayne the top clayey sand and should therefore be considered
(1990) method. Similarly, substituting the value of F more suitable for such soil types. The strong correla-
obtained from lab results of 2.95–21.67 % in Eq. 8, the tion between CPT and DPL data calls for application
ratio qc/N so obtained ranges from 0.42 to 0.38. The of DPL for more than site stratification purposes. For
ratio qc =N60 ¼ 0:37 obtained in this study falls slightly example, Nilsson (2013) reports the possibility of
outside the predicted range. Other studies, for example using same soil classification methods meant for CPT
Chin et al. (1988), conclude that the ratio qc/N is better with DPL data as long as the tip and friction
related to F than to D50, whereas Mahler and Szendefy resistances for the DPL are measured separately.
(2009) found the relationship between qc and number Data filtering using Tukey mean–difference plots
of blows of a heavy dynamic probe based on F less has enabled outliers in the data sets to be detected and
reliable in comparison to the D50 based relation. helped to improve the coefficients of correlation,
Akca (2003) presented a study about correlations of except in the case of SPT–DPL data, where the
SPT and CPT data for sandy soils, where he used the coefficient decreased. However, there is a decrease in
OLS method to determine the expression qc ¼ the standard error of estimate, leading to a decrease in
0:32N60 þ 8:23 to be applied for sand. A comparison transformation uncertainty from N10 to N60. The
in Fig. 11 shows that this expression may result in very relationship between SPT and DPL data especially
high values of qc if used for data from the present within a range of N10 between 10 and 50 may be
study. The ratios qc =N60 ¼ 0:77 and qc =N60 ¼ 0:70 expressed by N60 ¼ 1:01N10 þ 0:44 or the simpler
were also established for sand and silty sand soils ratio N60 =N10 ¼ 1:03. These expressions are generally
respectively by means of the arithmetic average valid for clayey sand because of its dominance in the
method. These two are also very high as compared shallow depth penetrated by DPL where measure-
to the findings of the current study. The study by Akca ments were taken. It is also observed that the data
(2003) considered SPT energy normalization from scatter increases with higher N values and correlations
N to N60 which is similar to the current study, but it is for stiff soils are expected to be poor. This observation
also noticeable that the range of data from his study is consistent with Figuiredo et al. (2013), who pointed
was very wide with N60 ranging from near zero to as out the need for further research to consider energy
high as 150. Another important factor is the separation dissipations in DPL measurements to improve the poor
distance between correlated tests. While his study correlation with SPT. The relationship found by
considered tests that are up to 80 m apart, the current Ibrahim and Nyaoro (2011), which is also basically
study minimized this separation to 0.5 m to reduce the for clayey sand, is less conservative compared to this
effect of soil variability in the derived correlations. study as can be seen from Fig. 12. Their difference is
more pronounced as the stiffness increases. The
5.3 CPT–DPL and SPT–DPL Correlations expression N60 ¼ N10 may therefore be applied in-
stead to mitigate the differences as it also is also close
This study has revealed that a strong correlation exists to the results using the arithmetic average method.
between CPT and DPL data. A similar observation to If the three test methods are compared in terms of
that from CPT–SPT, the relationship between CPT physical make of equipment, delivery of driving

123
1232 Geotech Geol Eng (2015) 33:1221–1233

90 uncertainty. It implies that, if these three tests are

performed together in a site investigation scheme, DPL
80 data should better be expressed in terms of CPT data to
facilitate transformation to desired design values. Con-
70
sideration of sleeve friction measurements resulted in a
60
small improvement in both SPT–CPT and DPL–CPT
correlations and may be considered insignificant. The
derived correlations also serve as a future framework for
SPT N60

50
development of more refined relationships (Marchetti
40 2011).
There are many factors that influence correlations
30 developed at different conditions such as site
geology, normalization and general treatment of
20
data, degree of control of soil variability effects, the
range of soil strength/stiffness and methods of
10
determining correlations. Consequently it is difficult
0 to meet all those criteria when comparing correla-
0 10 20 30 40 50 60 70 80 90 tion expressions suggested by different authors. It
DPL N10 implies that each suggestion is rather unique to a
Linear (This study, filtered) specific case and it remains as once pointed out by
Linear (This study, filtered, arthmetic average)
Linear (Equality line)
Schmertmann (1978) that local correlations are
Linear (Ibrahim and Nyaoro (2011)) better and should be preferred to generalized ones.
The correlations developed in this study are consid-
Fig. 12 Comparison of SPT–DPL data correlations ered most suitable for the local silty to clayey sand,
energy and measurement intervals, SPT resembles stiff to very stiff soil deposits found to dominate a
DPL in terms of dynamic driving energy but they significant part of Dar es Salaam. Moreover, SPT–
differ in regard to size of sampler and make of tip as CPT correlations developed in this study compare
well as DPL being continuous while SPT is a discrete fairly well with previously suggested correlations by
test. DPL and CPT are similar in physical shape as Schmertmann (1970), Robertson et al. (1983) as
well as both being continuous, making them more well as Kulhawy and Mayne (1990) and these may
sensitive to soil variations. Because this study has therefore still be applied for general cases. It is also
proved a better correlation between CPT and DPL as important to note that the incomparability between
compared to CPT–SPT and SPT–DPL, this may mean different suggested correlations was found to in-
that the two factors of continuity and make of a crease with soil stiffness.
sampler have a stronger influence on measurements
Acknowledgments The authors wish to acknowledge the
than the mode of driving energy. support of Sida (Swedish International Development Cooperation
Agency) through the project Continuous Penetration Test as
Alternative to Intensive Borehole Investigation under the Rural and
6 Conclusions Urban Infrastructure Development Project (University of Dar es
Salaam, Tanzania).

In this study, a high degree of soil variability control

through closely spaced side by side in situ testing as well
as control of testing procedures was exercised through- References
out the data collection process, aiming for correlations
between the test data. A comparison of the correlations Akbas SO, Kulhawy FH (2010) Characterization and estimation
of geotechnical variability in ankara clay: a case history.
produced in this study indicates that CPT–DPL corre- Geotech Geol Eng 28:619–631
lations are better than SPT–DPL in terms of higher Akca N (2003) Correlation of SPT–CPT data from the United
coefficient of correlation and lower transformation Arab Emirates. Eng Geol 67:219–231

123
Geotech Geol Eng (2015) 33:1221–1233 1233

Altman DG, Bland JM (1983) Measurement in medicine: the Mahler A, Szendefy J (2009) Estimation of CPT resistance
analysis of method comparison studies. Stat 32:307–317 based on DPH results. Periodica Polytech 53(2):101–106
Boulanger RW (2003) Relating Ka to Relative State Parameter Marchetti S (2011) Discussion of CPT–DMT correlations by P.
Index. ASCE J Geotech Geoenviron Eng 129(8):770–773 K. Robertson. ASCE J Geotech Geoenviron Eng 137(4):
Chin CT, Duann SW, Kao TC (1988) SPT–CPT correlations for 441–442
granular soils. In: 1st international symposium on Mayne PW (2006) In-situ test calibrations for evaluating soil
penetration testing. Orlando, USA, 1:335–339 parameters. Overview paper, characterization and engi-
Ching J, Phoon K-K, Chen YC (2010) Reducing shear strength neering properties of natural soils II (Proc. Singapore
uncertainties in clays by multivariate correlations. Can Workshop)
Geotech J 47:16–33 Moss RES, Seed RB, Olsen RS (2006) Normalizing the CPT for
Ching J, Phoon K-K, Yu J-W (2013) Linking site investigation overburden stress. ASCE J Geotech Geoenviron Eng
efforts to final design savings with simplified reliability 132(3):378–387
based design methods. ASCE J Geotech Geoenviron Eng Müller R, Larsson S, Spross J (2013) Extended multivariate
140(3):04013032 approach for uncertainty reduction in the assessment of
Eisenhauer JG (2003) Regression through the origin. J Teach undrained shear strength in clays. Can Geotech J
Stat 23:76–80 51(3):231–245
Eslami A, Fellenius BH (2004) CPT and CPTu data for soil Nilsson TU (2013) Shear resistance by the DPL Nilsson Test.
profile interpretation: review of methods and a proposed ISĆ4 geotechnical and geophysical site characterization,
new approach. Iran J Sci Technol 28:70–86 Porto de Galinhas: 1249–1257
Fellenius BH (2009) Basics of foundation design (electronic Phoon K-K, Kulhawy FH (1999) Evaluation of geotechnical
edition). www.fellenius.net property variability. Can Geotech J 36(4):625–639
Figuiredo LC, Cunha RP, Conciani W (2013) An overview on Robertson PK (2009) Interpretation of cone penetration tests—a
existing dynamic cone penetration test research related to unified approach. Can Geotech J 46(11):1337–1355
the Central Area of Brazil. Geotech Geophys Site Charact Robertson PK (2012) Interpretation of in situ tests—some in-
4:1669–1675 sights. Mitchell Lecture-ISC 4, Brazil
Geological Survey Division Dodoma (1963) Report: the geo- Robertson PK, Campanella RG, Wightman A (1983) SPT–CPT
logical survey of Tanganyika Quarter Degree Sheet 186, correlations. ASCE J Geotech Eng Div 109(11):449–1459
First Edition for Dar es Salaam Rogers JD (2006) Subsurface exploration using the standard
Hussein MH, Woerner WA, Sharp M, Hwang C (2006) Pile penetration test and the cone penetrometer test. J Environ
drivability and bearing capacity in high-rebound soils. Eng Geosci 12(2):161–179
GeoCongress 2006:1–4 Schmertmann JK (1970) Static cone to compute static settlement
Ibrahim M, Nyaoro DL (2011) Correlations of DCPT and SPT for over sand. ASCE J Soil Mech Found Div 96(3):1011–1043
analysis and design of foundations. In: Proceedings of the 15th Schmertmann JH (1978) Guidelines for cone test, performance
African regional conference on soil mechanics and geotech- and design. US Federal Highway Administration, Report
nical engineering, Maputo, Mozambique, pp. 632–637 FHWA—TS-78209
Jefferies MG, Davies MP (1993) Use of CPTu to estimate Skempton AW (1986) Standard penetration test procedures and
equivalent SPT N60. Geotech Test J 16(4):458–468 the effects in sands of overburden pressure, relative density,
Kara O, Gündüz Z (2010) Correlation between CPT and SPT in particle size, ageing and overconsolidation. Géotechnique
Adapazari, Turkey. In: Proceedings of the 2nd interna- 36(3):425–444
tional symposium on cone penetration Testing, Huntington Spagnoli G (2007) An empirical correlation between different
Beach, California, pp. 2–18 dynamic penetrometers. Electron J Geotech Eng 12(Bun-
Kulhawy F, Mayne P (1990) Manual on estimating soil prop- dle C). http://www.ejge.com/2007/Ppr0729/Abs0729.htm.
erties for foundation design. EPRI, Hillview Steffanoff G, Sanglerat G, Bergdahl U, Melzer KJ (1988) Dy-
Liao SSC, Whitman RV (1986) Overburden correction factors namic probing: international reference test procedure. In:
for SPT in sand. ASCE J Geotech Eng 112(3):373–377 Proceedings of the 1st international symposium on
Lunne T, Robertson PK, Powell JJM (1997) Cone penetration penetration testing. Orlando, USA, 1:53–70
testing in geotechnical practice. Blackie Academic and Tukey J (1977) Exploratory data analysis. Addison-Wesley,
Professional, New York Reading

123