Examining the swelling potential of cohesive soils with high plasticity according to their index properties using GIS

2.1 Study area

Aksaray is located in the Central Anatolian Region of Türkiye and is located at the geographical coordinates between 38° and 39° N Latitude and 33°–35° E Longitude (Figure 1a). This study covers an area of 70 km² in the selected region in Aksaray City (Figure 1b). Although many researchers in Türkiye have similar studies on the geotechnical properties of the soil using GIS, there is a lack of such study for the city center of Aksaray. Thus, this research will provide a starting point for creating a geographic database for the Aksaray city center, which has a rapidly increasing and constantly expanding area in terms of industrialization and population growth. The larger the database, the easier it will be to visualize the characteristics and behavior of the ground in the area. Since the study area is already a dense residential and industrial area, this study will also contribute to the existing settlements whether additional measures are needed and to make a safer settlement planning study.

Figure 1

Location maps of Aksaray (Türkiye) (a) and the study area (b).

2.2 Geology of study area

The location of the study area in the geology of Turkey is tectonically located within the Central Anatolian Crystalline Complex (CACC), a triangular area extending between Ankara Sivas and Nigde [48,49]. Metamorphic, ophiolitic, and intrusive rocks, which are the basement rocks of the CACC, are not present in the study area. In the study area, Upper Miocene-aged continental clastics (m3pl), Pleistocene-aged unsorted continental clastics (Q1), Quaternary-aged basalt (Qb), and alluvium (Qa) units are present (Figure 2). Previous studies have reported that in the northeast region of the study area, there exist alluvial deposits comprising predominantly of silty sand, marginally sandy gravel, and silty clay ranging from 1.5 to 0 m. It has been stated that the clayey layers in the upper part of the alluvium covering all units unconformably are dense and brown. The sandy layers have a loose structure, and the rounded pebbles in the low gravel sand are of volcanic origin [50,51,52].

Figure 2

Geological map of the study area [53].

2.3 Soil sampling and index properties for swelling potential analyses

Samples were taken from 15 points by digging pits at depths of 0–1.5 m. Experiments were carried out at Aksaray University Applied Geological Laboratory to determine the index properties of the samples. The experiments performed on soil samples are water content test [54], specific gravity test [55], Atterberg limits [56], and grain size analysis [57]. The obtained test results were evaluated according to the Unified Soil Classification System (USCS) [58], and the soils were classified.

The soil’s swelling potential was evaluated using data from literature and empirical formulas. The study classified soils using index properties and created spatial maps to determine soil swelling potential. The results of previous researchers were used as a basis [12,29,30,31,35,36,59,60,61]. The parameters studied were water content, #200 sieve pass, LL, PI, swelling percentage, and swelling pressure (Tables 1–4).

The PI is considered an important factor in determining the swelling potential in the literature. However, the interpretation of the swelling potential of the soil can be difficult due to the intertwining of the PI value ranges and other parameters in the tables and graphs. For this reason, the appropriate PI value ranges determined by previous researchers are given in Table 3.

In the classifications used, soil samples may have the same swelling potential, but it should be noted that the amount of swelling may differ [28,62]. For this purpose, swelling index values were calculated using equation (1) [63]. The relationship between LL and swelling index in swelling soils (Figure 3) was evaluated using the swelling pressure data obtained according to ASTM D 4546-08 standard [64].

where SI = Swelling index, W% = natural water content of the soil, LL = liquid limit value.

Figure 3

Relationship between LL and SI for expansive soil [34,63].

Additionally, there are several empirical formulas available to calculate the swelling percentage, which is a crucial parameter in determining the swelling properties [14,65,66,67,68,69]. The study employed the swelling percentage calculation from the study by Vijayvergiya and Ghazzaly [65], which uses the LL and initial water content. They defined the linear swelling percentage using Atterberg limits. Equation (2) was proposed by performing linear multiple regression analyses.

where S = linear free swelling percentage, WL = water limit, W = is the initial water content.

The graphs and equations provided above helped to determine the swelling index, swelling pressure, and potential swelling percentage values. Table 4 evaluated these values, and a comment was made regarding the swelling potential of the soils in the study area.

A geographic database was designed to determine the location, specific gravity, sieve analysis, Atterberg limits, water content, soil class, and swelling potential for each collected sample examined above. The data were converted to electronic format by creating Excel spreadsheets that can be used in the GIS program. The sample locations’ coordinates were recorded and then referenced to their exact locations using the ArcMap program with the georeferencing option. The Geodatabase file was created and entered into the GIS program as point features after editing the data (Figure 4). The version used for the GIS program for the current study is ArcMap10.2. The data for this study were prepared in two parts. The index properties of the soils were determined by experiments and used to determine the soil class in the first part. The data were analyzed spatially. The second part of the study identified regions with swelling soil by utilizing the index properties of the soils. The study area’s index properties and swelling potential were used to create zone maps with the assistance of GIS.

Figure 4

The boundaries of the study area.

2.4 Geostatistical analysis

The Kriging method is used as a spatial analysis interpolation tool to analyze the obtained data and produce digital maps. Kriging is a spatial interpolation method named after Danie Krige, a South African mining engineer who worked in mining geology. The Kriging interpolation method estimates the optimum values of the data at other points by using data from known close points [72]. Additionally, Kriging generates estimates of the uncertainty surrounding each interpolated value. In general, kriging weights are calculated so that points closer to the point of interest are given more weight than distant points. Point clusters are also considered, so point clusters are less weighted. This helps reduce bias in estimates [73]. Maps were created by applying the ordinary kriging method using semivariogram models created for each soil parameter. The ordinary Kriging interpolation method is obtained by the following equation (3) [42]:

where Z(x) is the predicted value of the unknown location x. p represents the number of known points. Z _i represents the observed value of known point i, and λ _i represents the weight of known point i.

The method relies on variograms to describe spatial changes. Variogram ensures spatial continuity of data and explains the effect of spatial change in swelling potential according to the soil index properties on distance and direction as a function. In other words, it is defined as a curve that shows how the swelling potential in the study area changes with distance. Below is the semivariogram equation (4) [41].

where γ(h) is the semivariogram value, h is the distance between the samples (lag distance), and Z(x _i ) – Z(x _i + h) is the value of a target variable at some sampled location and the value of the neighbor at distance x _i + h.

The difference between the values of regional variables and the distance between these variables is called a semivariogram [74,75]. This function is defined as the variance of the difference between two variables at a distance h from each other. The experimental semivariogram allows the acquisition of crucial data regarding the spatial variability of the environmental variable. The semivariogram must be known at all distances to estimate the values of unsampled points. This can be achieved by integrating a function into the “theoretical semivariogram” values, resulting in the modeling of the semivariogram [76]. The theoretical semivariogram with the mathematical model is derived by fitting it to the experimental semivariogram.

The most appropriate semivariogram models for the soil index parameters examined in the study were selected from the circular, spherical, Gaussian, and exponential models by evaluating the estimation error margins, particularly the root mean square standardized error (RMSSE) and the mean standardized error (MSE). Equations (5) and (6) were used to calculate RMSSE and MSE values, respectively [77].

where Z i is the predicted value, Z i ⁎ is the actual value, n is the number of observation, and σ ( Z i ) is the standard deviation value.

To create the most accurate interpolation map, it is essential to minimize the RMSSE and MSE errors. The closeness of the RMSSE value to one and the MSE value to zero indicates the accuracy of the interpolation [78,79,80,81]. The “Geostatistical Analyst” tool in ArcGIS 10.2 software was utilized for geostatistical analysis and the generation of maps.

3.1 Laboratory analysis results of samples

Soil samples were tested according to ASTM standards. The water content, specific gravity, LL, plastic limit (PL), PI, gravel, sand, fine grain content (silt and clay) ratios, and soil classification for the depth of 0–1.5 m in the study area are presented in Table 5.

The specific gravity test yielded results ranging from 1.77 to 2.64. According to the USCS, 5 out of the total 15 samples obtained from the study area are in the high plasticity silty (MH) class, while 1 sample belongs to high plasticity clay (CH), another to low plasticity clay (CL), and 3 to low plasticity silty (ML). Furthermore, the study discovered that two samples belong to the low plasticity clay-low plasticity silty class (CL-ML), and the remaining three samples to the silty sand (SM) soil class.

Before the application of the interpolation method, it was first necessary to ascertain whether the soil parameters exhibited a normal distribution. For this reason, descriptive analyses were conducted on the parameters, and logarithmic transformations were applied to the data that did not exhibit a normal distribution. Analyzing the distribution of data through statistics such as arithmetic mean, mode, median, skewness, and Kurtosis coefficients are referred to as descriptive methods. For the normal distribution test, the fact that the mean and median values are close to each other indicates that the dataset has a normal distribution [82,83]. As illustrated in Table 5, the mean and median values are nearly identical, suggesting that our dataset is approximately normally distributed. In addition, Skewness and Kurtosis values are two values that should be checked for normal distribution tests. To demonstrate a normal distribution, the values of skewness and kurtosis should be as close to +3 or –3 as possible [84,85]. Among the soil parameters, only water content and PI were subjected to logarithmic variation due to their skewness values and the differences between the mean and median [86]. Nevertheless, the prediction error values of the map with logarithmic transformation in the PI were found to be higher, thus necessitating the creation of the map for these data in its original form.

The spatial distribution maps of specific gravity and USCS in the soil samples are presented in Figure 5a and b. According to the USCS, soils exhibiting high plasticity (CH-MH) have a specific gravity ranging from 1.77 to 2.29. These soils are situated in the northern, central, and western regions (Sample Nos: M-1-2-3-7-8-9) of the study area (Figure 5b).

Figure 5

The spatial distribution maps of (a) specific gravity and (b) USCS.

3.2 Determination of swelling soils in the study area

To assess the index properties of samples collected from the study area, we used tables and empirical correlations provided by previous researchers. These are based on swelling potential data given in Section 2. The index properties of the soil samples in the study area are categorized as having low, medium, high, and very high swelling potential using the Kriging interpolation technique in ArcGIS software. Spatial distribution maps are also presented. In certain clay soils, typically found in arid climates and unsaturated with water, there is significant swelling when the water content increases, leading to a substantial surge in volume. The process of such an increase is called swelling. Expansive soils refer to clay soils that undergo swelling when wetted and shrink when dried [87]. It is common knowledge that for soil to demonstrate swelling properties, it must typically fall under the CH or CL classification in the USCS [25]. However, soils classified as ML, MH, and SC may also experience swelling [71].

The study by Chen [29] shows that the swelling rate of soils is affected by the initial water content. Clays with natural water content below 15% were found to be problematic in terms of swelling. It has also been found that clays with water content above 30% will swell at lower values. Soil samples within the study region underwent swelling potential evaluations based on water content, as per the study by Chen [29]. The resulting spatial data are presented in Figure 6a. Expansive soils comprise sand and silt alongside a substantial quantity of clay minerals. Due to the differing swellability of various clay minerals, the natural swelling is not simply dependent on the clay content. Based on the classification established by Chen [29], soils with more than 60% passing through the #200 sieve present a high potential for swelling. The sieve analysis from the study area has revealed regions with potentially high swelling propensity, which have been spatially displayed in Figure 6b.

Figure 6

Maps showing swelling potential in terms of (a) water content and (b) percentage of particles passing the #200 sieve in the study area.

Based on the natural water content swelling potential data in the study by Chen [29], some samples (M-1-3-6-7-14) show medium swelling potential, while others (M-2-4-5-8-10-11-12-13-15) have high swelling potential. Another classification system based on soil passing through a 200 mesh sieve identified some samples (M-6-10-11-12-13) as having medium swelling potential. In contrast, M-2 and M-8 show extremely high swelling potential, while M-1-3-4-7-9-14-15 have high swelling potential.

At the point where soil transitions from a liquid to a plastic consistency, the quantity of water present as a percentage of the weight of the oven-dried soil is known as its LL [88]. The PI is the numerical difference between the LL and the PL of a soil sample, and it indicates the soil’s range of plasticity. Previous studies [28,59] devised the Swelling Potential Classification Table using the LL and PI to determine the likelihood of swelling. When the soil samples in the study area were evaluated according to the LL and PI diagram created by Holtz and Gibbs [59], a significant proportion of them were found to have moderate to very high swelling potential (Figure 7).

Figure 7

Swelling potential classification card [59].

The LL values of soil samples collected from the study area were compared with the ranges of LL values of soils with swelling potential determined by prior studies. The spatial representation of the obtained results is illustrated on the map (Figure 8). Based on the swelling potential table established by previous literature [59,61] using LL value, samples M-6 of intermediate level and M-1-7-9 display a high propensity for swelling. M-2-8 samples, conversely, exhibit a very high swelling potential based on the LL value (Figure 8a). According to the LL value-swelling potential table of previous studies [29,36], sample M-4 has medium swelling potential, while samples M-6-7 have high swelling potential. Samples M-1-2-3-8-9 have a very high swelling potential (Figure 8b). The table for swelling potential evaluates the LL outcomes of soil specimens using the LL value specified by previous studies [31] and [60]. Sample M-7 has a moderately high swelling potential, whereas samples M-1-2-3-8-9 boast a notably high swelling potential (Figure 8c). Among the soil samples assessed for swelling potential in the table corresponding to the LL value developed by Coduto [35], the M-6 sample was classified as medium, and the samples M-1 and M-7 were classified as high. Samples M-2-3-8-9 showed a very high swelling potential, as depicted in Figure 8d.

Figure 8

The swelling potential maps in terms of LL.

The PI serves as a reliable parameter for detecting expansive soils. Figure 9 illustrates the spatial distribution of swelling potential in the study area, using the PI value ranges for swelling soils identified by earlier researchers. Expansive soils generally remain plastic over a wider range of water content, meaning they have a higher PI. However, this does not mean that all high plasticity soils are swelling soils.

Figure 9

The swelling potential maps in terms of PI.

Based on the PI values determined by Krebs and Walker [30], the swelling potential status of the soil samples can be determined. Soil sample M-3 has a medium swelling potential, while samples M-2 and M-8 exhibit a high swelling potential (Figure 9a). In a separate investigation, the PI swelling potential table as found in previous studies [31,60] revealed that exclusively M-2 and M-8 soil samples exhibited substantial swelling potential (Figure 9b). There are no instances of soil exhibiting moderate swelling potential. When PI results are assessed using the limit values determined by a previous study [61], it can be deduced that samples identified as M-3 and M-9 are situated within the medium swelling potential area. Samples labeled M-2 and M-8 demonstrate a significant potential for swelling according to the findings presented in Figure 9c. Although there is an area displaying high-level swelling potential between the medium and very high-level swelling potential regions, no sample was located in this area. Based on the PI-swelling potential table recommended by Yildirim and Acar [15], it is evident that Sample M-3 exhibits a high swelling potential, while samples M-2-8 demonstrate a very high swelling potential (Figure 9d). Although it is a mid-level swelling potential area, none of the samples taken from the field were located in this area.

Soil sample swelling index values were calculated using equation (1). The swelling pressure values were determined by evaluating the results obtained from the graphic of the LL-swell index. Furthermore, equation (2) from the study by Vijayvergiya and Ghazzaly [65] was used to calculate the swelling percentage. Table 6 presents the results obtained. As illustrated in Table 6, the mean and median values are nearly identical, suggesting that our dataset is approximately normally distributed.

The analysis of swelling pressure and percentage values provided by Coduto [71] was used to assess the soil’s swelling potential. The corresponding spatial locations are showcased on the map in Figure 10. Based on the evaluation of swelling pressure presented in previous studies [70] and [71], it can be concluded that samples labeled as M-1-4-6-7-10-11-12-13 display a medium potential for swelling. Soil samples numbered M-2-3-8-15 have high swelling potential (Figure 10a). When you examine the swelling potential evaluation table about the swelling percentage value provided by the same researchers, it becomes clear that samples labeled M-1-4-7-10-12-13-15 possess a moderate swelling potential. However, samples M-2 and M-3 have a high swelling potential, and sample M-8 has a very high swelling potential (Figure 10b).

Figure 10

Maps showing the swelling potential based on (a) swelling pressure and (b) swelling percentage.

In the production of ordinary kriging maps, the most appropriate models were selected and used from semivariograms, which are an effective tool to best evaluate the spatial variations of soil parameters. The semivariogram provides a clear description of the spatial structure of variables and gives insight into possible processes affecting the data distribution [86,89]. Before the construction of the experimental semivariograms, the anisotropy condition was initially evaluated, and it was determined that there was no appreciable discrepancy between the calculated variograms. In this context, Gaussian for water content and LL, spherical for passing #200 sieve, and circular semivariogram models for PI, swelling pressure, and percentage were applied (Table 7). The ratio of Nugget value to Sill value is employed as a percentage expression to classify the spatial dependence of soil parameter variables. The ratio in question is classified as strongly spatially dependent if it is ≤25%, moderately dependent if it is between 25 and 75%, and weakly dependent if it is more than 75% [80,90,91,92]. The Nugget/Sill ratio (%) (Nugget effect), calculated for the modeled semivariograms, demonstrates a strong spatial dependence for the LL and a moderate spatial dependence for all other parameters. The strong spatial dependence between samples shows that the similarity between samples does not disappear at short distances and continues even at long distances [93]. The range values obtained from semivariogram models indicate the maximum distance at which the similarity between two sampling or prediction points persists [93]. The calculated range values were 14,475 m for water content and 5,199 m for sieve no. 200, 5,610 m for LL, 4,957 m for PI, 5,199 m for swelling pressure, and 5,870 m for swelling percentage. The aforementioned values indicate that in future studies, narrower intervals should be sampled for parameters with small range values and wider intervals for parameters with large range values [94]. Furthermore, the study demonstrated that a correlation can be established between sample pairs within a specific distance range, which corresponds to the calculated range value for soil parameters. Conversely, no correlation was observed between sample pairs beyond this distance [81]. In all soil parameters, the highest RMSSE values of 1.16 and MSE error values very close to zero indicate the usability of ordinary kriging estimates in the generated maps [78,79,80,81,86].

4.1 Model evaluation

The swelling potential map generated by consolidating the common areas of the studies used was employed to reassess the soil index properties of swelling potential limit values (Table 9). However, it should be noted that this table also incorporates the USCS soil classification system. Upon applying the limit values of the soil parameters obtained in this study, it was determined that all samples, except for M-2, M-3, and M-8 in the CH and MH soil classes, exhibited varying swelling potential. In the ranges of soil parameters provided in Table 9, the M-2 (CH) and M-3-8 (MH) specimens displayed very high potential for swelling across all parameters. The study’s determination of soil properties’ swelling potential value ranges suggests that they apply to soils with high plasticity (CH or MH). However, although swelling soils maintain plasticity across a diverse range of water content, indicating a high PI, it cannot be concluded that all high plasticity soils are swelling soils. Although M-1-7-9 soil samples were in MH soil type, they did not exhibit a consistent swelling potential in all parameters considering soil properties. Upon evaluation of the obtained value ranges for cohesive soils with high plasticity, it is evident that this study is particularly relevant to the boundary area with a high potential for swelling.

The dataset was split randomly into 80% training and 20% validation. Interpolation maps were created using training samples for each soil parameter. The validation samples were placed on the map and the kriging prediction values were calculated at their location. The RMSSE values were calculated to see how close the actual and predicted values were. In the prepared maps, it is understood that the closer the RMSSE value of the estimate is to 1, the more accurate the map is [78,95,96]. In other words, the final map has acceptable and good RMSSE values for soil index parameters (Table 10).