Professional Documents
Culture Documents
In Situ Evaluation of Soil Contaminated by Total Petroleum Hydrocarbons Using Membrane Interface Probe A Case Study From Nanjing, China
In Situ Evaluation of Soil Contaminated by Total Petroleum Hydrocarbons Using Membrane Interface Probe A Case Study From Nanjing, China
In Situ Evaluation of Soil Contaminated by Total Petroleum Hydrocarbons Using Membrane Interface Probe A Case Study From Nanjing, China
https://doi.org/10.1007/s10064-022-02639-6
ORIGINAL PAPER
Received: 24 August 2021 / Accepted: 1 March 2022 / Published online: 10 March 2022
© Springer-Verlag GmbH Germany, part of Springer Nature 2022
Abstract
Delineation of contaminated soils is important for their remediation; however, reliable and efficient methods remain lack-
ing. In this study, a contaminated site around an abandoned refuelling station in Southeastern China was investigated using
resistivity piezocone penetration (RCPTU) and membrane interface probe (MIP) tests to evaluate the spatial distribution of
organic contaminants. Various influences on the MIP results were analyzed, such as contaminant concentrations, the soil–
groundwater interface and soil properties. Based on RCPTU and sampling results, MIP data were quantitatively evaluated in
terms of relative percentage differences. Results from a photoionization detector (PID) and flame ionization detector (FID)
were found to overestimate and underestimate the total petroleum hydrocarbon concentrations, respectively. In general, the
PID was slightly more accurate than FID. Maps of regional contaminant concentrations were derived from geostatistical
analysis of MIP data and were compared with contaminant concentrations measured in soil samples in the laboratory. This
paper provides a method for interpreting MIP data with consideration of the influence of soil properties. It demonstrates the
effectiveness of using MIP to determine contaminant distributions and increase the efficiency of site remediation programs.
Keywords Site investigation · Membrane interface probe · Geostatistics · Resistivity cone penetration test · Volatile organic
contaminants
13
Vol.:(0123456789)
140 Page 2 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
Regions contaminated by volatile organic compounds hydraulic conductivity (kh) were obtained using RCPTU
(VOCs) can be effectively detected using in situ MIP. tests. The influences of Ic and kh on MIP data were evalu-
Although MIP have proven to be a useful testing technique, ated. Geostatistics was used to obtain contaminant distribu-
experience and research indicate that its applications and tion maps based on MIP data with consideration of these
results are influenced by multiple factors, such as detection influences. The analysis of the influencing factors is valuable
limits (Bronders et al. 2009), the temperature and pressure for conducting field identification tests at contaminated sites
conditions during the test (Costanza et al. 2002), carry- while avoiding data misinterpretation.
over effects (Bumberger et al. 2012), and soil permeability
(Adamson et al. 2014). Therefore, the signal intensity of
MIP tests is not always consistent with soil sampling results. Site condition
Misinterpretation of MIP results may lead to the inaccurate
delineation of contaminated zones, resulting in suboptimal Nanjing is the capital of Jiangsu Province, located in south-
remediation outcomes. east China. The study area was an abandoned gasoline sta-
Previous studies have used MIP for the qualitative inves- tion located at a machinery plant that had been used for
tigation of contaminated sites (Griffin and Watson 2002; refuelling and for storing gasoline for over 60 years. Due
McAndrews et al. 2003; Kurup et al. 2006; Bumberger to the ageing of gasoline tanks caused by long-term stor-
et al. 2012; McCall et al. 2014). All reached the conclu- age and improper operation, gasoline leakage occurred and
sion that MIP can delineate the subsurface distribution of contaminated the surrounding soil. The site was planned to
organic contaminants in a semi-quantitative manner. This is be remediated by demolishing the existing buildings then
because various factors, such as soil water content, soil type, used as residential and commercial land.
and operating procedures, influence the accuracy of MIP The site was roughly square in shape with a length of
data (Christy 1998; Coztanza and Davis 2000; Myers et al. about 100 m in the east–west direction and 85 m in the
2002; Bronders et al. 2009). Although field survey experience north–south direction. The maximum elevation difference
has exposed the potential limitations of MIP in engineer- around the site was approximately 8 m. The geomorphol-
ing applications, the influence of soil mass heterogeneity ogy of the site was characterized as a Yangtze River terrace
has been neglected, since it is costly to determine such soil and gully area. According to field drilling and laboratory
properties. Therefore, there is a need to better understand the geotechnical tests (ASTM D1140-17 2017), the stratum
influences on MIP tests and to better interpret their results. being investigated in this study was deposited during the
The resistivity piezocone (RCPTU) is an economical and Quaternary period and is classified as silty clay. The stable
rapid in situ geotechnical investigation method (Lunne et al. groundwater level was located at a depth of 1.7–4.5 m from
2009; Robertson 2009; Cai et al. 2012; Duan et al. 2021a). the ground surface with a seasonal variation of about 1.0 m.
Various soil properties, such as soil type and horizontal per- Detailed soil coring, groundwater sampling, and labora-
meability, can be directly interpreted based on the acquired tory testing of contaminants were carried out at the site,
data, such as cone tip resistance (qt), sleeve friction (fs), and followed by MIP tests and, finally, RCPTU tests. Sampling
pore water pressure (u2; Cai et al. 2012; Liu et al. 2013). and in situ testing were completed within 20 days, so the
Therefore, a geological model of a site based on RCPTU influence of environmental change was not considered.
data can provide a basis for analyzing the influences on MIP Figure 1 presents an aerial map of the contaminated site and
tests. the MIP (M) and RCPTU (R) testing locations. According
Visualization of the distribution of contaminants at a to the soil sampling results, test locations with excessive
given site can provide useful information to guide site reme- contaminants were distinguished and are marked as M4, M5,
diation. Therefore, the development of contaminant distribu- M9, M10, M11, M12, M18, and M34 in Fig. 1. Due to the
tion maps is desirable for geo-environmental engineering. limitation in the extent of the area that could be investigated,
Geostatistics is an effective method for obtaining a three- the RCPTU testing locations were limited to the gasoline
dimensional map of the contaminant distribution at a site tank storage zone.
(Zou et al. 2020) and, thus, was used in the present study.
The purpose of this work was to evaluate the applicability
of MIP tests to the identification of organic contaminated Testing program
zones with consideration of the influence of soil variation.
An experiment was carried out at an abandoned gas station Sampling and testing program
located in Nanjing, southeastern China. An extensive site
investigation program was conducted using different in situ A total of 39 soil cores were obtained from the same loca-
and laboratory tests, including RCPTU and MIP tests. Data tions as MIP tests were conducted in Fig. 1 using a direct-
on field stratigraphy, soil behavior type (Ic), and horizontal push drill rig with a sealed plastic tube inside a 2.25-in
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 3 of 20 140
rod. The average maximum coring depth was 7.5 m. The and M12 (corresponding to R4, R5, R7, R9, and R11, respec-
direct-push drill rig was also used to sample groundwater tively) were used for wash-sieving tests (ASTM D1140-17
at points M4, M9, M11, and M12. After the hollow drill 2017) to determine the soil type and corresponding fines con-
rod was pushed to the groundwater surface, a pumping tube tent (FC). The depths of the soil cores used for wash-sieving
was inserted into the groundwater through the hollow rod tests at the five points were 1 m, 2 m, 3 m, 4 m, 5 m, 6 m,
and samples were extracted by a water pump (Fig. 2). The and 7 m. The mass percentage of soil particles < 0.075 mm
groundwater sampling locations were selected based on soil in diameter and its corresponding depth were recorded for
cores exhibiting obvious contamination. The collected plas- further analysis with RCPTU data.
tic tube (soil cores) and groundwater were packed in airtight Soil cores from depths of 1 m, 6 m, and 9 m at eight sam-
polyethylene bags and refrigerated in a laboratory at 3–4 °C. pling points (M7, M15, M20, M25, M30, M31, M36, and
The geotechnical testing program included wash sieving M39) were used for flexible-wall permeation tests to evaluate
tests and flexible-wall permeation tests (ASTM D5084 2016). the field permeability. Soil specimens were fully saturated
Parts of the soil cores from locations M6, M13, M5, M10, for 24 h under 80 kPa of vacuum pressure before conducting
Fig. 2 Groundwater sampling
Hollow rod
Pumping tube
Collecting groundwater
13
140 Page 4 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
permeation tests. During permeation, tap water was applied volatilized and diffused across the semipermeable membrane
from the lower base to the upper side of the specimens under of the probe. An inert carrier gas inside the probe system
seepage pressure to avoid air entrapment. The cell pressure transmits the vapors in real-time at a constant rate for analy-
and constant flow pressure were set to 200 kPa, which is sis in a ground-surface detection system.
lower than the yield stress of the specimens. A total of 39 MIP tests were carried out in this study. The
Soil and groundwater samples below 1 m depth were used average test depth was 10.35 m and the layout of test loca-
to analyze the types and concentrations of contaminants. The tions is shown in Fig. 1. The MIP device used in this study
portable X-ray fluorescence (XRF) technique was applied is shown in Fig. 3. Detailed information on the device and
to detect metal contaminants and gas chromatography/mass its operating procedures is presented in Section S1. Typical
spectrometry (GC–MS) was used to assess organic contami- MIP test results are presented in Fig. S1 in the Supplemen-
nants. In view of the complexity of the operational proce- tary Material.
dures, their details can be found in the literature (Turner For contaminated site investigation, MIP has been used as
et al. 2018; Chen et al. 2020; Leupin et al. 2020). an economic and efficient tool for obtaining high-resolution
characterization data (Griffin and Watson 2002; McAndrews
Field MIP tests et al. 2003; Gilmore 2010). Three different detectors (FID,
PID, and XSD) were used in the MIP tests, which have dif-
The detection of VOCs in a subsurface soil mass can be con- ferent responses to different organic contaminants. A PID
ducted using MIP, which are a direct-push technology devel- is generally used to detect aromatic hydrocarbons, while an
oped by Geoprobe Systems (Salina, Kansas, USA; Dietrich XSD is used to detect halogenated VOCs and is insensitive
and Leven 2006). An MIP system typically consists of a to non-halogenated hydrocarbons. An FID will respond to
set of detectors, including a photoionization detector (PID), any high-concentration VOC; however, its sensitivity is low
flame-ionization detector (FID), and halogen-specific detec- according to the manufacturer’s specifications (Geoprobe
tor (XSD). Each of these provides a gross response to the 2009). Based on the sampling results, total petroleum hydro-
exposed total contaminant load (ASTM D7352-18 2018). carbons (TPH) were the dominant contaminant present in
The MIP system is used to collect depth-continuous data the soil. Since the XSD detector is more suitable for detect-
in soils where the probe can be advanced using static pen- ing potential halogenated hydrocarbons, only FID and PID
etration (ASTM D7352-18 2018). As the probe advances, signals were processed to analyze the relationship between
organic contaminants in adjacent soils and groundwater are TPH concentration and MIP signal intensity.
Fig. 3 Membrane interface
probe: a Probe used in this
study; b Schematic diagram of
the diffusion process
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 5 of 20 140
Field resistivity piezocone (RCPTU) tests influence them, especially the FC, as illustrated by Gilmore
(2010) and Adamson et al. (2014). However, it is difficult
Field resistivity piezocone testing is widely accepted as an to obtain the FC distribution in the field. For RCPTU tests,
efficient method for continuous high-resolution data acqui- the Ic parameter is considered the important indicator of FC
sition in geotechnical investigations (Lunne et al. 2009; changes (Davies 1999). Although Davies (1999) proposed
Robertson 2009; Cai et al. 2012; Liu et al. 2013; Duan an Ic–FC relationship based on the mining waste in Canada
et al. 2021b). The structure of the RCPTU probe used in (Eqs. 4, 5, 6), it is suggested that a site-specific relationship
this study is shown in Fig. S2, which is in accordance with be established through field sampling.
the international standard for piezocones (ASTM D5778-
20 2020). Detailed information on the RCPTU device and
FC = 0% For Ic < 1.31 (4)
its operating procedures is presented in Section S2 in the
Supplementary Material. FC = 42.4Ic − 54.9(%) For 1.31 < Ic < 3.6 (5)
Ic = (3.47 − lg(Qtn ))2 + (lg(Fr ) + 1.22)2 (1) where v = 0.02 m/s is the standard penetration rate, 𝛾w is
the unit weight of water (= 10 kN/m3), r0 = 1.785 cm is the
Qtn = [(qt − 𝜎v0 )∕Pa ](Pa ∕𝜎 � v0 )
c
(2) radius of a standard cone, σ'v0 is the effective overburden
stress (kPa), and Δu2 − u0 is the excess pore water pressure
(kPa), where u2 is the measured pore water pressure behind
Fr = fs ∕(qt − 𝜎v0 ) (3) the cone (kPa) and u0 is the initial static pore water pressure
where Qtn is the normalized cone tip resistance (dimension- (kPa).
less) as expressed by Eq. (2), σvo is the vertical total stress
(MPa), Pa is the atmosphere pressure (0.1 MPa), σ'vo is the Interpretation of MIP data
vertical effective stress (MPa), c is a stress exponent based
on the soil behavior type, and Fr is the normalized friction In this section, a quantitative method was established to
ratio (dimensionless), as expressed by Eq. (3). evaluate the accuracy of the MIP test results by statistical
The MIP testing results are not directly determined comparison of MIP data and influencing factors. Two meth-
by the contaminants as inherent soil properties can also ods were employed for statistical processing: (1) the relative
percentage difference (RPD) between MIP signals and influ-
encing factors at corresponding depths and (2) linear regres-
Table 1 Soil behavior type classification using Ic (Robertson 2009) sion analysis to determine the influence of soil properties on
Range of Ic Soil classification MIP signals, with the coefficient of determination (R2) used
as the criterion for the goodness of fit. The detailed stepwise
Ic < 1.31 Gravelly sand to clayey sand
procedure of the RPD calculation is as follows:
1.31 < Ic < 2.05 Sand: clean, silty sand
2.05 < Ic < 2.60 Sand mixtures: silty sand to sandy silt
1. At each depth, the concentrations of individual petro-
2.60 < Ic < 2.95 Silt mixtures: clayey silt to silty clay
leum hydrocarbons were summed to obtain TPH con-
2.95 < Ic < 3.6 Clay, silty clay
centrations from the sampling results, which were used
3.6 < Ic Organic soils, peats
as a benchmark for comparison.
13
140 Page 6 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
2. The TPH concentration data were log-transformed to be across the entire test interval. RPD values less than zero
more consistent with its expected distribution. indicate that the MIP response underestimated the TPH
3. At each point, the FID and PID signals recorded the geo- concentration, while values greater than zero indicate
metric mean values of signals received within 15 cm of overestimation.
each depth (i.e., the value at 1 m depth is the average of
signals received from depths of 0.85 m to 1.15 m). This
made the dataset more representative and minimized the Geostatistical analysis of MIP data
influences of single anomalous signals.
4. Each of the FID and PID signals was log-transformed. The FID and PID data were processed using geostatistical
Then, the baseline signal (i.e., the minimum signal for methods to obtain a three-dimensional map of the contami-
detectors over each test) of each log was determined and nant distribution at the site. Such maps can provide useful
subtracted from log-transformed MIP signals. information to guide site remediation. The main procedures
5. The MIP signals were normalized using the maximum include trend removal, normality checking, Box-Cox data
and minimum responses at that location. This estab- transformation, semi-variogram analysis, Kriging interpola-
lished a set range between zero (no response) to one tion, and data back-transformation. A flowchart of the geo-
(maximum response) across the entire test interval. statistical method is shown in Fig. 4. Geostatistics is widely
Similar normalization of discreet TPH values obtained used in contaminated site investigation and has proven effec-
from soil data was also conducted and the minimum tive in evaluating data uncertainty and determining contami-
value was set to zero. nant plumes. The geostatistical procedures are introduced
briefly and the results are presented in “Mapping the con-
The RPD between the normalized MIP data and nor- taminant distribution” For detailed information, interested
malized TPH concentration was calculated. This served readers may refer to Zou et al. (2017, 2020) and Bheemasetti
as an overall indicator of the accuracy of the MIP detector et al. (2017) or contact the corresponding author.
Trend removal
Trend superposition
Normality of
residuals ?
No
Back transformation
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 7 of 20 140
1
Sand Min Max
Mean
3
Depth (m)
Sand mixtures
4
Silt mixtures 6
13
140 Page 8 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
4 4
Depth (m)
Depth (m)
6
Sand mixture
6
8
8
10
10 Silt mixture
12
1.31 2.05 2.60 2.95 3.60 0 3 0 200 0 3 0 200 0 3 0 200 0 3 0 200 0 3 0 200
A Aÿ
ÿ
Ic R6 R2 R5 R8 R9
Relationship between FC and Ic based on field To derive the FC–I c relationship and consider the
RCPTU data potential influences of contaminants, the FC values were
determined directly by wash-sieving tests from five drill-
Based on the Ic values calculated using Eq. (1) and the soil ing samples, including uncontaminated soil samples from
classification criterion in Table 1, the soil stratum mainly locations R4 and R5 and contaminated soil samples from
contained sand and sand mixtures with thicknesses of about R7, R9, and R11. Figure 7 shows the site-specific relation-
2.5 m and 2 m, respectively, underlain by a silt mixture. The ship between the FC and Ic values based on the RCPTU
calculated Ic values and stratification of section A-A’ are test results and Eq. (1). The fitted results indicate that the
shown in Fig. 6. relationships are almost the same for contaminated and
FC=42.2Ic-51.7
R2=0.88
45
R2=0.86
30
For uncontaminated soils
FC=42.7Ic-53.9
R2=0.83
15
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 9 of 20 140
0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8
0 0 0 0 0 0
1 1 1 1 1
1
2
2 2 2 2
2
3
3 3 3 3
3 4
4 4 4 4
Depth (m)
5
4
5 5 5 6 5
5
6 6 6 7 6
6 8
7 7 7 7
9
7
8 8 8 8 Normalized log (FID)
10
8 Normalized log (TPH)
9 9 9 11 9
M4 M5 M9 M10 M11 M12 Mismatch area
10 9 10 10 12 10 Carry-over area
non-contaminated soils. Therefore, the effect of contami- Influence of TPH concentration on MIP response
nants can be ignored. Therefore, the site-specific relation-
ship was fitted using both contaminated and uncontami- Gas chromatography was used to detect VOCs in the carrier
nated soil data, with the coefficient of determination R2 gas stream at the site. Figures 8 and 9 show the FID and PID
being 0.86. This strong correlation confirms the feasibility response data and soil analysis results at six locations: M4,
of using Ic instead of FC to analyze the impacts on MIP M5, M9, M10, M11, and M12. Due to the limited coring
testing. depth and quantity of soil samples, the analysis results of
0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8
0 0 0 0 0 0
1 1 1 1 1
1
2
2 2 2 2 2
3
3 3 3 3
3 4
4 4 4 4
Depth (m)
5
4
5 5 5 6 5
5
6 6 6 7 6
6 8
7 7 7 7
9
7
8 8 8 8 Normalized log (PID)
10
8 Normalized log (TPH)
9 9 9 11 9
M4 M5 M9 M10 M11 M12 Mismatch area
Carry-over area
10 9 10 10 12 10
Normalized log (PID and TPH concentration)
13
140 Page 10 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
soil sample cannot completely correspond to MIP data. In Table 3 Evaluation of MIP response accuracy
general, the responses of the MIP tests matched the sampling Calculation type RPD of normalized FID RPD of normalized
results. In the field tests, the FID and PID responses did not PID
exceed the upper detection limits. Their peak values were
Median Mean Median Mean
mostly found within depths of 1–5 m, indicating that the
soil contaminants mostly occurred in the shallow layers. The Directional −8.00 −5.63 9.39 8.89
peaks in the FID profiles appeared at 2–3.4 m and 4.3–5.2 m Non-directional 17.68 19.41 15.21 15.87
in log M4, 2.5–2.8 m in log M9, 2.8–3.2 m in log M10, and
3–4.5 m in log M12. This suggests a relatively high contami-
nation in these ranges; however, this was not supported by M11 and the distribution is relatively shallow (i.e., the M10
the sampling data. Similar results were recorded in the PID peak occurred at 2–3 m and the M11 peak was at 3.5–4 m).
profiles, with peaks at 2–3.5 m in log M4 and 2.5–2.8 m in Although points M10 and M11 were close and in a similar
log M9. This discrepancy may have been caused by the loss soil stratum, due to the lower TPH concentration at M10,
or dilution of the contaminant mass before analysis of core there was no carry-over effect similar to that at M11, so the
sub-samples, or by spatial variability (Adamson et al. 2014). MIP results are more consistent with the sampling results.
Another possible explanation is provided by Zhao et al. The above methods for qualitative evaluation of MIP
(2019), who observed faster desorption of contaminants in data focus on checking the visual profile of every data log
the presence of moisture and attributed it to the steam distil- to assess its overall trend and anomalies. Statistical com-
lation of contaminants. Another common problem with MIP parisons of MIP data and total TPH concentrations in soil
tests is the carry-over effect (Bumberger et al. 2012), which require the establishment of a quantitative method for
was also observed in this study. These effects often occur assessing the reliability of the MIP contaminant distribu-
after passing high-concentration zones that contain mobile tions. RPD and linear regression analysis were employed
or residual phases. Disproportionately high MIP signals as quantitative methods. The RPD of the normalized MIP
were obtained in both the FID and PID profiles in 1.5–2 m response is presented statistically in Fig. 10. Directional and
in log M5 and < 4 m in log M11. According to the sampling non-directional mean and median values of normalized MIP
results, the TPH concentration at M10 was lower than that at response RPDs were calculated (Table 3). Non-directional
75
RPD of Normalized FID response
RPD of Normalized PID response
50
RPD of normalized FID/PID (%)
0
0 2 4 6 8 Depth (m)
-25
-50
-75
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 11 of 20 140
RPDs were obtained from the absolute values of each RPD. by TPH. In reality, TPH contains different proportions of
The directional RPD, obtained from the raw values, is an organic contaminants that cause various responses in detec-
indicator of the sensitivity of MIP detectors to TPH. The tors. (2) Organic contaminants other than TPH were not
non-directional RPD was calculated to measure the vari- considered due to their low contents. However, they may
ability associated with the dataset. still have affected the MIP response signals. (3) The MIP
Table 3 shows that the RPD of the mean and median val- penetrates to a specific depth and the surrounding contami-
ues of the directional normalized FID are negative and the nants undergo desorption due to the heated probe. Organic
normalized PID values are positive. This indicates that the contaminants diffuse as vapor through the thin film and are
detection results of the FID underestimate the real TPH val- transmitted by the carrier gas and are then analyzed by the
ues, whereas the PID results overestimate them. The same detectors at the surface. The thermal desorption of contami-
trend can also be seen in Fig. 10, where the 95% CI of the nants is often controlled by many factors, such as soil type
normalized PID RPD is higher than that of the normalized and water content, which often have obvious spatial differ-
FID RPD. This may be because the PID detector is more ences at a site. (4) The portable gas chromatograph used
sensitive than the FID detector. The median non-directional in the MIP field system is less accurate than an equivalent
MIP response RPD indicates that the TPH detection error laboratory instrument.
by the FID is slightly higher than that of the PID. The mean
values reflect that the errors of the two detectors are almost Soil‑groundwater interface
the same. However, Fig. 10 shows that the 95% CI of the FID
is much wider than that of the PID, indicating that the PID An MIP can detect organic contaminants below groundwater
has better performance. level due to the presence of a semi-permeable membrane
The correlations between TPH concentration and the FID that keeps water from entering the probe. The MIP probe
and PID response signals were also investigated by plotting used in this field test was equipped with a temperature sensor
them logarithmically, as shown in Fig. 11a and b, respec- to record real-time changes in probe temperature. When the
tively. The TPH concentration is positively correlated with probe penetrates to the depth of the soil-groundwater inter-
the MIP signal but the data points are rather scattered around face (i.e., in partially saturated soils), intensive temperature
the fitted trendline. The coefficients of determination for the fluctuations are recorded. McAndrews et al. (2003) attrib-
FID and PID models are 0.23 and 0.24, respectively. This uted this to the noticeably greater specific heat of saturated
indicates that many other factors affected the accuracy of soils compared to unsaturated soils. After passing through
the MIP tests in the determination of TPH concentration in the unsaturated zones and entering saturated soil, the tem-
the subsoil. perature fluctuations steadied. As the probe penetrated to the
The poor fit of the regression equation may be because (1) desired depth, it was reheated to the standard temperature
MIP data are the values of the total electrical signal caused and then driven to the next test depth.
6.2 7.0
Log FID=0.085 Log TPH +5.25 Log PID = 0.355 Log TPH + 5.04
6.0 R2=0.23 R2 = 0.24
6.5
Log PID response (μV)
Log FID response (μV)
5.8
6.0
5.6
5.5
5.4
5.2 5.0
1.5 2.0 2.5 3.0 3.5 4.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Log TPH (mg/kg) Log TPH (mg/kg)
(a) (b)
Fig. 11 Plot between Log TPH and Log MIP response: a for FID; b for PID
13
140 Page 12 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
1 1 1 1 1
1
2
2 2 2 2
2
3
3 3 3 3
3 4
4 4 4 5 4
4
Depth (m)
5 5 5 6 5
5
6 6 6 7 6
6 8
7 7 7 7
9
7
8 8 8 8
10
9 8 9 9 9
11
M4 M5 M9 M10 M11 M12
10 9 10 10 12 10
Real-time temperature Soil-water junction zone
The real-time temperatures of MIP tests conducted at 3.6–3.8 m, 3.8–4.4 m, and 2.8–3.2 m, respectively. In Figs. 8
high-concentration sites M4, M5, M9, M10, M11, and M12 and 9, the positions of mismatch between the MIP response
are presented in Fig. 12. From the temperature anomaly loop and soil sampling results are approximately the same as
in Fig. 12 (i.e., repeated fluctuation loop), the depth of the those of the soil-groundwater interface or are below the
water-soil interface at each test site location can be roughly groundwater level. This indicates that the soil-groundwater
estimated. After passing through the soil-groundwater inter- interface and groundwater levels are key influences on the
face, the temperature loop became more regular. The approx- MIP response.
imate depths of the soil-groundwater interfaces at locations Table 4 shows the groundwater sampling and analysis
M4, M5, M9, M10, M1,1 and M12 were primarily deter- results, specifically, the maximum contaminant concentra-
mined as having ranges of 1.2–1.5 m, 2.2–2.4 m, 2.5–2.7 m, tions at different locations. According to the contaminant
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 13 of 20 140
concentrations, both light non-aqueous phase liquids (LNA- XSD data that the halogenated VOCs were concentrated in
PLS) and DNAPLS were found. The LNAPLS floated on groundwater at depths 2.2–3.2 m, 4–4.5 m, and 4.9–5.7 m.
the surface of groundwater due to being less dense. There-
fore, when using an MIP to detect NAPLS, LNAPLS mostly Influence of inherent soil properties on MIP
exist at the soil-groundwater interface but DNAPLS diffuse response
downward due to gravity. Therefore, compared to LNA-
PLS, DNAPLS may have a lesser influence on MIP tests The principle of the MIP testing mechanism is to extract
at the soil-groundwater interface. Both TPH and LNAPLS contaminants from soil particle surfaces or pores by vola-
were detected in shallow groundwater at test locations M4, tilizing them into gases using heat. The detection of con-
M9, M11, and M12. This may be the cause of the mis- taminants by MIP is influenced by many inherent soil prop-
match between Figs. 8 and 9, where a higher MIP signal erties. During a demonstration of the site characterization
was obtained at the soil-groundwater interface than in the and analysis penetrometer system (SCAPS), the presence
sample analysis. of fine-grained soils was found to reduce the correlation
Unlike the results of sample analysis, BTEX and halo- between soil trichloroethylene (TCE) data and MIP data.
genated VOCs were detected in groundwater samples from Furthermore, while the MIP performed best in sand, data
locations M4 and M9. Their concentrations were higher collected in clayey soils were generally better than those
at M4 than M9. Although its sensitivity to BTEX is low, from silt or sand mixed with silt/clay. No further explanation
the XSD detector is highly sensitive to halogenated VOCs. of the influence of soil type on test results is provided in the
The XSD test results at locations M4 and M9 are shown in literature. Therefore, there is a need to better understand
Fig. 13. The baseline XSD responses at M4 and M9 were the basic physical and chemical processes that occur during
identified as 2.1 × 105 μV and 2.2 × 105 μV, respectively. the desorption of organic contaminants from soil particles
Although high contents of BTEX and TPH were detected in during heating.
the laboratory analysis of groundwater samples from loca- The RCPTU tests performed at the study site enabled the
tion M9, the halogenated VOC content was low. This is also classification of the continuous soil profile using Eqs. (1, 2, 3),
reflected in the XSD response curve derived from the field and the influence of soil type on the accuracy of MIP tests can
MIP tests (Fig. 13), which is fairly smooth and has no peak. be reflected by the relationship between Ic and RPD. Follow-
Compared with location M9, the groundwater samples from ing the calculation method discussed in the previous section,
M4 contained more halogenated VOCs. The XSD response the Ic-values at this site were found to range between 1.35
curve of M4 (Fig. 13) has markedly high signal intervals. and 2.90. According to the soil behavior types determined
Considering that the groundwater depth began at 1.2–1.5 m from the Ic-values defined by Robertson (2009; Table 1), the
at location M4, it can be preliminarily determined from the site’s soil contains sand, sand mixtures, and silt mixtures. The
mean and median RPDs of the FID and PID were calculated
(Table 5). For the FID detector, the average value was positive
MIP XSD Response (× 105 μV) for 2.05 < Ic < 2.60 and negative for other intervals. This indi-
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4
0 cates that the FID-measured values are greater than the true
values for sandy layers and less than the true values for mixed
1 sand–silt layers, whereas the PID-measured values are greater
than the true values for all Ic intervals. By using the absolute
2
mean, it can be found that the deviations in FID and PID
3 readings increase with increases in Ic-values. No significant
correlations between Ic and MIP RPDs were found; however,
4
it is obvious that when Ic increases (i.e., increasing fines con-
Depth (m)
5 tent in the soil mixture), there is high deviation in the MIP test
results. This trend is similar to the effect of fine-grained soils
6
on MIP TCE data that was demonstrated in SCAPS.
7 Soil permeability affects the gasification of contaminants
from soil media. Soil with high FC or a compact structure
8 has low permeability, which makes it more difficult to detect
contaminants. Soil layers with a low permeability coefficient
9 M4 XSD Response
M9 XSD Response have a high water retention capacity (Otalvaro et al. 2016).
10 Therefore, a considerable amount of the organic contaminants
retained in the soil was covered by groundwater and could not
Fig. 13 MIP XSD responses at M4 and M9 locations be directly volatilized and absorbed by the carrier gas. The kh
13
140 Page 14 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
Table 5 Statistical data of RPD Soil behavior index Detector Mean Absolute Median Absolute
of normalized Mip signals in mean median
different soil types
1.31 < Ic < 2.05 RPD of normalized FID −6.59 18.17 −3.44 13.82
RPD of normalized PID 12.11 12.32 7.13 13.23
2.05 < Ic < 2.60 RPD of normalized FID 8.83 16.02 −6.12 19.71
RPD of normalized PID 14.33 16.63 9.88 13.88
2.60 < Ic < 2.95 RPD of normalized FID −16.02 20.42 −13.88 11.71
RPD of normalized PID 16.90 18.83 14.11 25.11
of the soil, as determined by flexible-wall permeation tests, and 30%. Figure 14 shows that the FID RPD is positive
varied from 1.04 × 10−3 to 1.68 × 10−7 cm/s. The kh of the site when k h < 1 0 −5 m/s, as is the PID RPD when kh < 1 0 −4.
varies greatly and is relatively high in shallow sand layers. To With increases in kh, the RPD value gradually decreased
compare and analyze MIP data and kh data at the same depth and became negative. This indicates that soil with low
position, kh were calculated using Eq. (7) to investigate its permeability leads to MIP results that are greater than the
influence on the MIP test results. true values, and vice-versa. This may be due to larger soil
It should be noted that the parameters applied for cal- particles having a lower contact area with the MIP heating
culating k h in Eq. (7) are affected by testing practices. block. Another explanation may be that soil with a large
During RCPTU operation, the discontinuity caused by the particle size can quickly dissipate excess pore water pres-
process of connecting rods to increase penetration depth sure, which absorbs part of the heat and affects the ther-
causes an uneven penetration rate and transient dissipa- mal desorption of organic contaminants during testing.
tion of pore water pressure, resulting in errors in kh esti-
mates at those depths. The relationship between normal- Mapping the contaminant distribution
ized MIP signal RPD with log(kh) is presented in Fig. 14.
The correlation coefficient of the fitted FID signal is The MIP data was modified using the average coefficient
R2 = 0.41 and the PID signal is higher at 0.44. Based on from Table 5 and corresponding Ic values. Geostatistics was
the flexible-wall permeation tests, the horizontal perme- then used to determine the contaminant distribution. The
ability coefficient at the site was calculated to be between trend function was estimated using ordinary least-squares
approximately 1.2 × 10−3 and 1 × 10−7 m/s, so kh estimates regression of the FID and PID signals. Since FID and PID
that differed significantly from this range were excluded. signals directly reflect the presence of contaminants, theoret-
Besides, high RPD deviations may be caused by other ically, a signal from uncontaminated soil should be constant
influences that may greatly affect the k h-RPD relation- and there will be no need to fit separate models to data from
ship; hence, it was also excluded from the model. In this contaminated and uncontaminated soil. Zou et al. (2020)
section, the range of RPD values studied was between −30 reported that a quadratic function performs well in capturing
30 30
20 20
Trend: Y = -7.4X - 36.2
10 R2 = 0.41 10
0 0
-12 -10 -8 -6 -4 -2 0 -10 -8 -6 -4 -2 0
Log(kh) (m/s) Log(kh) (m/s)
-10 -10
-20 -20
Trend: Y = -5.9X - 24.4
-30 -30 R2 = 0.44
-40 -40
-50 -50
(a) (b)
Fig. 14 Correlation between MIP RPD and log(kh): a for FID; b for PID
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 15 of 20 140
Fig. 15 Semi-variogram func-
tions in the horizontal direction:
a for FID; b for PID
(a)
(b)
13
140 Page 16 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
exponential semi-variogram model of Eq. (10) was found ordinary Kriging interpolation and back-transformation, per
to be a good fit to the averaged semi-variogram cloud. The as Eq. (11). The 3D maps based on FID and PID data are
parameters used in the fitted exponential semi-variogram shown in Figs. 17 and 18.
model are shown in Table 6.
T − A + (1 + pRt )1∕p p ≠ 0
{
Ro = (11)
T − A + exp(Rt ) p=0
�
(hx )2 + (hy ax ∕ay )2 ⎟⎪
⎧ ⎛ ⎞⎫
(10)
⎪ ⎜
𝛾(h) = c⎨1 − exp⎜−3
ax ⎟⎬ where R o represents the original residuals after back-
transformation, Rt represents the Box-Cox-transformed
⎪ ⎜ ⎟⎪
⎩ ⎝ ⎠⎭
residuals, and p is a transformation parameter estimated
where c is the sill value; hx and hy are the lag distances, and using the boccox function in Matlab software. The p-values
ax and ay are the autocorrelation distances in the horizontal for the FID and PID residuals are 0.33 and 0.49, respectively.
and vertical directions, respectively. The 3D maps of TPH concentrations based on FID and
After determining the semi-variogram, a three-dimensional PID data exhibited relatively high consistency, although
(3D) map of the MIP data of the site was obtained using with minor discrepancies. The FID results indicate high
Fig. 16 Semi-variogram func-
tions in the vertical direction: a
for FID; b for PID
(a)
(b)
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 17 of 20 140
Table 6 Parameters of fitted c ax ay of the PID at M34 was lower than that of the FID. Also, a
expotential semi-variogram high response signal was not observed in the area between
models FID data 320 200 2.5 the flowerbed and refuelling area, which is consistent with
PID data 500 180 2.5 the sampling results.
According to the map based on MIP, the high-concentration
TPH area can be determined. The A-A’ section passing
TPH concentrations at locations M4, M13, and M34. The through points M4, M10, and M11 was selected to conduct
results of the sample analysis show that the highest TPH contamination distribution analysis in the vertical direction
concentration was observed at M11, while that at M13 did (Figs. 17b, 18b). Two contaminant sources can be identi-
not exceed the screening value. The FID response signals at fied 23 m and 57 m from the south end of A-A’ section
point M12 were low; however, the sample analysis shows (i.e., points M4 and M11, respectively). The contaminants
that the TPH concentration at this point exceeded the screen- were observed to gradually diffuse from these two sources
ing level. Besides, the FID response signals show that TPH into the surrounding areas. The TPH was mainly retained
existed in the area between the flowerbed and refuelling area at shallow depths at point M4, especially < 2 m, whereas
and may have exceeded the screening level. However, the at M11, it diffused over a wider vertical range (up to 9 m).
sample analysis shows that the TPH concentration in this It seems that contaminant migration at point M4 was much
area was much lower than the screening level. The results more significant in the horizontal than the vertical direction,
based on the PID detector show that the TPH concentra- as observed from the wider contour curves. Besides, at point
tion was highest at location M11 and that the area around M11, vertical contaminant migration was more significant
M4 also had high TPH concentrations. Similar to the FID than at M4. However, horizontal migration was still domi-
results, the PID results underestimated the TPH concentra- nant, as reflected in the contour curve of the A-A’ section
tion at M12. Comparatively, the response signal intensity (i.e., Figs. 17b, 18b). This can be attributed to soils typically
Fig. 17 Distribution of TPH
concentration based on FID
detector: a planar area; b along
A-A’ section distribution
13
140 Page 18 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
exhibiting better horizontal hydraulic properties than verti- contaminant migration direction. Further sampling should
cal ones (Chapuis 2012; Dai et al. 2018). be carried out near M28 and M16 to determine whether the
Figures 17 and 18 show the contamination plumes in the contaminant concentrations exceed the screening levels. The
horizontal and vertical directions that need to be considered rest of this field can be considered uncontaminated and, thus,
for remediation purposes. For example, it can be concluded does not require remediation.
that the key area requiring remediation is near the gasoline In general, except for the misinterpretations at point
tank storage area near points M4 and M11. Although the M12, all points where the TPH concentration exceeded the
A-A’ section in Figs. 17 and 18 shows high-value signals screening level based on sample analysis were examined.
below 7 m at point M11, this may be caused by the carry- The vertical distribution of TPH concentration was identified
over effect, as previously discussed in “Influence of TPH to determine the presence of two contaminant sources. The
concentration on MIP response.” Besides, the TPH concen- TPH was mainly found at shallow depths and horizontal con-
tration approaches zero at about 7 m based on soil sample taminant migration was more significant than vertical migra-
analysis. Thus, remediation should be focused on the top tion, which can be attributed to soil having better hydraulic
7 m of soil. permeability in the horizontal direction. Field remediation
According to the soil stratum represented in Fig. 6, mixed measures could be planned based on the contaminant dis-
sandy soil occurs deeper in the northeast direction; further- tribution maps provided in this study. The MIP test results
more, Figs. 17 and 18 show that the contamination plume at were highly consistent with the sampling results, which fur-
M11 in the northeast is deeper than that at M4 in the south- ther verifies the accuracy and reliability of the MIP-assisted
west. This shows that the soil texture significantly affects the rapid field screening technique.
Fig. 18 Distribution of TPH
concentration based on PID
detector: a planar area; b along
A-A’ section
13
Bulletin of Engineering Geology and the Environment (2022) 81: 140 Page 19 of 20 140
13
140 Page 20 of 20 Bulletin of Engineering Geology and the Environment (2022) 81: 140
Chen R, Cai G, Congress S, Dong X, Duan W (2020) Dynamic prop- Leupin OX, Loon L, Gimmi T, Elam WT, Jakob A (2020) In-situ
erties and environmental impact of waste red mud-treated loess x-ray fluorescence to investigate iodide diffusion in opalinus clay:
under adverse conditions. B Eng Geol Environ (3) demonstration of a novel experimental approach. Chemosphere
Christy TM (1998) A permeable membrane sensor for the detection of 269:128674
volatile compounds in soil. In: Symposium on the application of Liu SY, Cai GJ, Du YJ, Zou HF, Fan RD, Puppala AJ (2013) Site
geophysics to engineering and environmental problems. Society investigation of a pesticide contaminated factory based on in-situ
of exploration geophysicists. pp. 65–72 resistivity piezocone tests in Southeastern China. In: Geotechni-
Cai G, Liu S, Puppala AJ (2012) Reliability assessment of CPTU- cal and geophysical site characterization proceedings of the 4th
based pile capacity predictions in soft clay deposits. Eng Geol international conference on site characterization ISC-4, Vol. 1, pp.
141:84–91 1757–1764. Taylor & Francis Books Ltd.
Chai JC, Agung PMA, Hino T, Igaya Y, Carter JP (2011) Estimating Lunne T, Robertson PK, Powell JJM (2009) Cone-penetration testing
hydraulic conductivity from piezocone soundings. Géotechnique in geotechnical practice. Soil Mech Found Eng 46(6):237–237
61(8):699–708 McAndrews B, Heinze K, Diguiseppi W (2003) Defining TCE plume
Chapuis RP (2012) Predicting the saturated hydraulic conductivity of source areas using the membrane interface probe (MIP). Soil Sedi-
soils: a review. B Eng Geol Environ 71(3) ment Contam 12:799–813
Costanza J, Davis W (2000) Rapid detection of volatile organic com- McCall W, Christy TM, Pipp D, Terkelsen M, Christensen A, Weber K,
pounds in the subsurface by membrane introduction into a direct Engelsen P (2014) Field application of the combined membrane-
sampling ion-trap mass spectrometer. Field Anal Chem Technol interface probe and hydraulic profiling tool (MiHpt). Groundwater
4:246–254 Monit R 34(2):85–95
Costanza J, Pennell KD, Rossabi J, Riha B (2002) Effect of temperature Myers KF, Davis WM, Costanza J (2002) Tri-service site characteriza-
and pressure on the MIP sample collection process. Remediation tion and analysis penetrometer system validation of the membrane
of Chlorinated and Recalcitrant Compounds – 2002. Paper 1F-08 interface probe. Eng Res Dev Center Vicksburg Ms Environ Lab.
Crumbling DM, Groenjes C, Lesnik B, Lynch K, Shockley J, VanEe J, Otalvaro IF, Neto MPC, Delage P, Caicedo B (2016) Relationship
McKenna G (2001) Applying the concept of effective data to con- between soil structure and water retention properties in a residual
taminated sites could reduce costs and improve cleanups. Environ compacted soil. Eng Geol 205:73–80
Sci Technol A 405 Rajabi H, Sharifipour M (2019) Geotechnical properties of hydrocarbon-
Dai S, Kim J, Xu Y, Waite WF, Kumar P (2018) Permeability anisot- contaminated soils: a comprehensive review. B Eng Geol Environ
ropy and relative permeability in sediments from the national gas 78(5):3685–3717
hydrate program expedition 02, offshore india. Mar Petrol Geol Ringwald P, Chapin C, Iceman C, Tighe ME, Wells EM (2021) Char-
108(1–2) acterization and within-site variation of environmental metal
Davies MP (1999) Piezocone technology for the geoenvironmental concentrations around a contaminated site using a community-
site characterization of mine tailings. PhD Thesis, Department of engaged approach. Chemosphere, 129915.
Civil Engineering, University of British Columbia, Vancouver, Robertson PK (2009) Interpretation of cone penetration tests—a unified
BC, Canada. approach. Can Geotech J 46(11):1337–1355
Dietrich P, Leven C (2006) Direct push-technologies: in groundwater Turner A, Chan CC, Brown MT (2018) Application of field-portable-
geophysics: a tool for hydrogeology, R Kirsch Springer, Berlin xrf for the determination of trace elements in deciduous leaves
Duan W, Congress SSC, Cai G, Puppala AJ, Dong X, Du Y (2021a) from a mine-impacted region. Chemosphere 209(OCT.), 928–934
Empirical correlations of soil parameters based on piezocone Zhao C, Dong Y, Feng Y, Li Y, Dong Y (2019) Thermal desorption
penetration tests (CPTU) for Hong Kong-Zhuhai-Macau Bridge for remediation of contaminated soil: a review. Chemosphere
(HZMB) Project Transp Geotech 100605 221:841–855
Duan W, Congress SSC, Cai G, Liu S, Dong X, Chen R, Liu X (2021b) Zou H, Liu S, Cai G, Bheemasetti TV, Puppala AJ (2017) Mapping
A hybrid GMDH neural network and logistic regression frame- probability of liquefaction using geostatistics and first order
work for state parameter-based liquefaction evaluation. Can Geo- reliability method based on CPTU measurements. Eng Geol
tech J (ja) 218:197–212
Geoprobe (2009) G eoprobe® membrane interface probe (MIP) standard Zou H, Liu S, Cai G, Puppala AJ (2020) Probabilistic identification of
operating procedure. Technical Bulletin No. MK3010 contaminated soils using resistivity piezocone penetration tests.
Gilmore AM (2010) High resolution investigation in a heterogeneous Acta Geotech 15(3):761–779
aquifer and evaluation of the membrane interface probe to assess Zhang LM, Dasaka SM (2010) Uncertainties in geologic profiles ver-
back diffusion potential. MS thesis, University of Guelph, Canada sus variability in pile founding depth. J Geotech Geoenviron Eng
Griffin TW, Watson K (2002) A comparison of field techniques for 136(11):1475–1488
confirming dense non aqueous phase liquids. Groundwater
22(2):48–59
Kurup PU, Issac B, Griffin EP (2006) Electronic nose–membrane inter-
face probe for geoenvironmental site characterization. J Geotech
Geoenviron Eng 132(9):1133–1142
13