Abstract: Soil expansiveness is a potential geotechnical hazard presenting significant problems leading to serious economical consequences in construction sectors principally light weight infrastructure developments. Simple identification and ways of assessment of engineering behaviors of expansive soils is a continuous focus of research in geotechnical engineering, among which application of remote sensing techniques. Potential of laboratory spectroscopy in the spectral region beyond 2.5µm is investigated for identification and estimation of selected geotechnical characteristics of expansive soils. Spectra of bulk and fine fraction (passing the ASTM 0.075millimeter sieve aperture) are studied in the 3-5µm and 8-14µm atmospheric windows. While classification of soils based on spectral characteristics show statistically significant and strong correlations with soil plasticity classes; partial least squares regression analysis indicates that much of the variation in soil expansiveness can be accounted for by the spectral indices. Magnitudes of associations are found to vary for bulk and fine soil fractions, being stronger at the later. Similarly the strength of relationships between the 3-5µm and the 8-14µm spectral regions vary, being stronger at the former with better model performance indices (root mean square error of predictions, standard error of performances, bias and offsets). The proposed method can be practically applicable in geotechnical site investigations primarily at reconnaissance, feasibility and preliminary design phases of infrastructure development.
Get your grade
or your money back
using our Essay Writing Service!
Keywords: expansive soils, plasticity, geotechnical hazard, reflectance spectra, PLS
Presence of expansive soils that are susceptible to volume changes upon fluctuations of moisture contents is a major geotechnical concern. Soil expansion and shrinkage potential is important physical property of soils that largely determine their engineering use, either as a foundation material where structures are laid upon them or construction material where they can be used among other things as subgrades, embankment fills etc. This property is primarily due to presence of active clay minerals in soils (Al-Rawas, 1999; Allbrook, 1985; McCormack and Wilding, 1975; Morin and Parry, 1971; Pascal et al., 2004; Skempton, 1984; Thomas et al., 2000; Wan et al., 2002) that have got high affinity to water. Of importance in geotechnical engineering is estimation of the magnitude of volume change that soils are susceptible to following seasonal or artificial changes in moisture contents. The aim is to design structures that can withstand detrimental effects of considerable volume variations and associated changes in engineering behavior (e.g. lowering shear strength and bearing capacity) of soils which lead to distress in structures. Direct soil testing requires dense sampling, is expensive in terms of economy and time requirement apart from lack of accurate continuous representation of sites to be investigated. A great deal of effort has gone into investigating methods that can be supporting or alternate tools of estimating soil properties.
Research in soil science demonstrated potential applicability of the visible near infrared (VNIR) and short wave infrared (SWIR) regions of the electromagnetic spectrum for predicting soil physical, chemical and biological properties (Baumgardner et al., 1985; Ben-Dor and Banin, 1994; Ben-Dor et al., 2002; Selige et al., 2006; Shepherd et al., 2005; Shepherd and Walsh, 2002; Terhoeven-Urselmans et al., 2008; Udelhoven et al., 2003; Viscarra et al., 2006; Waiser et al., 2007; Zhang et al., 1992). The technique is thus becoming a norm than exception for indirect qualitative and quantitative characterization of soils and soil properties. Advances in remote sensing techniques have enabled discrimination of clay minerals that cause swelling and shrinkage in soils and mapping their abundances. The VNIR and SWIR spectral region is successfully used to identify (Chabrillat et al., 2002; Kariuki et al., 2004a; Van der Meer, 1999), characterize and map (Bourguisnon et al., 2007; Kariuki et al., 2003; Yitagesu et al., 2010; Yitagesu et al., 2009; Yitagesu et al., 2008) expansive soils.
The spectral region from 2.5-14µm on the other hand is reported (Boyd and Petitcolin, 2004; Bras and Erard, 2003; Farmer and Russell, 1964; Frost et al., 2001; Salisbury and D'Aria, 1992b; Salisbury and D'Aria, 1994; Salisbury and Eastes, 1985) to be promising for compositional analysis of geologic materials where its potential is demonstrated at a laboratory scale. Better performance of the mid infrared spectral region over the visible and near infrared regions is reported (Viscarra et al., 2006) where they explored suitability of this spectral region for quantitative characterization of soil properties (pH, organic carbon, cation exchange capacity, sand, silt and clay contents etc). They attributed the superior performance of mid infrared spectral region with the high intensity and specific nature of spectral signals of materials in this region. Recently Yitagesu et al., (2011?) demonstrated potential of laboratory spectroscopy in this spectral region for differentiating clay minerals that are responsible for soil expansiveness and quantifying their relative abundances in prepared mixtures. Naturally occurring expansive soils are compositionally heterogeneous (contain different minerals apart from the clay minerals discussed in their paper) and might produce complex spectra than that of the pure clay minerals and their mixtures. However, their laboratory experimental data on pure clay minerals and their mixtures provides with essential background knowledge on spectral characteristics of active clay minerals whose presence dictates soil expansiveness. This on the other hand can help in establishing meaningful relationship between spectral characteristics and compositional variations, hence reasonable estimations of clay mineralogical content of natural soils as well as their geotechnical properties.
Always on Time
Marked to Standard
The current paper, as a continuation of our previous work (Yitagesu et al., 2011??) on pure clay minerals focuses on spectral characterization of expansive soils, determination of their mineralogy and estimation of their plasticity from their respective spectra in the 3-5µm and 8-14µm spectral regions of the electromagnetic spectrum. Spectra of soils are best acquired preferably under similar sample conditions as those under which remote sensing observations are to be made or subject to similar extent of processing under which their geotechnical parameters are determined, hence bulk and fine fraction portions of soil samples respectively.
Materials and Methods
2.1 Study site
The study was carried out in an area located in the central part of Ethiopia (Figure 1), in the upper valley of the Awash River which drains the northern part of the Rift Valley. Topography ranges from a relatively flat to hilly, undulating and steep mountainous terrain. Elevation ranges from 1500 meters to 2500 meters above sea level. Conical-shaped isolated hills of scoria formed during the late stages of volcanism are common in the study area. Climate is moderate to wet with mean annual rainfall of 1200 millimeters in Addis Ababa and areas close by, and 870 millimeters around the town of Nazret. Temperature ranges from 8 to 25 five degree centigrade. While topography controls the ease with which the soils are drained, heavy rainy periods followed by prolonged dry periods leading to major moisture changes in the soils contribute towards the susceptibility of soils to pronounced volume changes.
Geology (Abebe et al., 1999) around TuluDimtu (Figure 1 shows names of major towns along the existing highway connecting Addis Ababa with the town of Nazret) consists of Tertiary to Quaternary volcanic formations which include alkaline basalts, spatter and cinder cones, ignimbrites, rhyolitic flows and domes, and trachyte. Near Debre Zeyt, alluvial and lacustrine deposits dominate which include sand, silt and clay. From Debre Zeyt to Modjo town lacustrine deposits, and after Modjo town fall and poorly welded pyroclastic deposits dominate with ryolitic and trachytic formations intercalations. The geology of the study area implies ranges possible parent materials from alkaline to intermediate and siliceous varieties that the soils are possibly derived from.
Soils in the study area can be classified into vertisols, luvisols, leptosols, phaeozems and andosols (Figure 1). According to FAO (1998) vertisols are clay rich (principally smectitic) expanding soils that swell and shrink with fluctuation in moisture content. Luvisols are common soil types in flat or gently sloping land, derived from a variety of unconsolidated material including alluvial, colluvial and eolian deposits. Luvisols usually exhibit high cation exchange capacity (CEC) and water retention potential which is associated with accumulation of active clay minerals (Gray and Murphy, 2002). Leptosols are very shallow soils over hard rock or in unconsolidated gravely material that are common in mountainous areas. Phaeozems are soils that are predominantly derived from basic parent material and are rich in organic matter. Andosols are young soils in volcanic regions that are usually associated with pyroclastic parent materials. Andosols generally show low CEC and water retention capacity (Gray and Murphy, 2002) unless allophane or immogolite is involved (Gray and Allbrook, 2002; Parfitt and Hemni, 1980; Wan et al., 2002). From engineering perspective, soils that are mostly black and contain highly expansive clay minerals (which are vertisol family soils commonly termed as black cotton soils) are found in the section from Addis Ababa to Modjo town covering an extensive area. According to Abebe et al., (1999), the black cotton soils are of alluvial, lacustrine and colluvial origin. The hilly and mountainous terrains are mainly covered with fresh to partially weathered basalts.
Figure Location map of the study area with kilometer markers on the new alignment showing its length in five kilometer intervals. Locations of major towns with names are indicated along the existing highway.
2.2 Sampling and Testing
Much of the new Addis Ababa - Nazret expressway route traverses on expansive soils mainly black cotton soil. Figure 1 also shows that majority of the route passes on vertisol soil classes. Possible swell-shrink potential of soils should be predicted to eliminate or minimize its detrimental effect on the highway subgrade and associated adverse impacts on the adjoining environment.
Soil samples were collected along the new expressway route from its starting point near TuluDimtu to its end at the town of Nazret. The sampling was part of a comprehensive geotechnical investigation and testing scheme for assessing suitability and quality of subgrade materials. Samples were recovered from shallow trial pits of 1m depth which is commonly the depth at which shallowly founded structures are laid. Samples were taken at every 500 m intervals. Additional deep trial pits of about 3m depth were dug between 3-5 km intervals with the aim of determining vertical extent of potentially expansive soils. A subset of these samples (forty) randomly selected from whole is used for this particular study.
Determination of geotechnical characteristics
This Essay is
a Student's Work
This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.Examples of our work
Identification and prediction of soil expansiveness was based consistency limits. This has an advantage of using relatively easy to measure parameters. Consistency limits namely liquid limits (LL), plasticity limits (PL) and plasticity indices (PI) were determined in accordance with ASTM D4318-05 standard test method.
Particle size distributions tests were conducted in accordance with ASTM D6913-04e1 standard test method using sieve analysis (for the fraction passing through two millimeter, 0.425millimeter and 0.075millimeter ASTM sieve openings). Grading of soils finer than 0.075 millimeter ASTM sieve is determined by hydrometer analysis in accordance with ASTM D422-63(2007) standard test method.
Spectral data acquisition and processing
Spectra of soil samples are acquired using the Bruker vertex 70 Fourier transform infrared spectrometer, in the ~1.7-14µm spectral region (of which the 1.7-2.5µm spectral region is included for the sake of comparison). The spectrometer is equipped with an integrating sphere coated inside with a diffusely reflecting gold surface attached to its external port, hence enabling directional hemispherical spectral reflectance measurement possible. In this case Kirchhoff's law can then be used to derive directional spectral emissivity (Johnson et al., 1998; Salisbury and D'Aria, 1994). The spectrometer was configured to provide spectral reflectance curves at a 4cm-1 spectral resolution with 512 scans per each measurement and eight measurements per individual specimen which are later averaged for a better signal to noise ratio. The spectrometer was continuously purged with nitrogen gas to remove any water vapor and carbon dioxide from the system, and liquid nitrogen was used for cooling the detector. OPUS software in a desktop system which is integrated with the spectrometer is used for parameter setting and visualization of the acquired spectra.
Soil samples were prepared in two ways bulk, and fine fraction passing the ASTM 0.75milimeter sieve aperture with the later being the smallest practical sieve for sieve analysis of soils in engineering applications. Bulk soil samples provide relatively similar conditions that the soils exhibit in-situ. The fine fractions often consist of clay particles (Carter and Bentley, 1991) that have large specific surface area (Yong and Warkentin, 1975), hence significantly influence much of soil physiochemical characteristics (Thomas et al., 2000). The fine fractions represent the extent of processing that soils are subjected to in soil mechanics laboratories for determining their geotechnical characteristics.
Spectral interpretations were based on characteristic absorption features where visual recognition of diagnostic features and spectral matching technique is applied. Better differences appear on a continuum removed spectra. Absorption in a spectrum has two components, a continuum and individual features (Clark and Roush, 1984; Van der Meer, 2004). The continuum is a convex hull fitted on top of a spectrum using straight line segments that connect local spectral maxima. Continuum removal enhances individual absorption features (Clark and Roush, 1984) by removing the background or overall albedo of the reflectance curve (Van der Meer, 2004).
X-ray diffraction (XRD) analysis
Mineralogy of the soils specimen was analyzed by X-ray diffraction (XRD) techniques. Siemens D5000 X-ray diffractometer was used. The analysis was conducted on bulk and clay fractions to determine the overall constituents as well as to quantify major, minor and trace clay species in the soil samples. For the semi-quantitative determination of clay minerals in the clay fraction, soil samples were treated initially to remove organic matter, iron oxides and carbonates. Then oriented slides were prepared in four different ways; untreated, treated with ethylene glycol vapor, treated with ethylene glycol vapor plus heated at 400 OC and 550 OC respectively. X-ray fluorescence (XRF) analysis was used to identify the oxides present in the soil samples and quantify organic matter content on loss on ignition.
Statistical analysis techniques are useful to explore structure of spectral data and extract meaningful information that will be translated in to furthering our understanding of spectral characteristics of soils and their relationship with soil properties (Cloutis, 1996; Haaland and Thomas, 1988).
Cluster analysis and cross tabulation
Cluster analysis is a procedure that attempt to identify and cluster homogeneous groups based on distance from specified or computed cluster centers (Kaufman and Rousseeuw, 1990). K-means cluster analysis based on Euclidian distance where the cluster means are iteratively approximated and variables are assigned to a cluster for which their distance to the cluster mean is the smallest (Kaufman and Rousseeuw, 1990), is used. In k-means clustering the user specifies the number of clusters and the algorithm starts with initial sets of means and classifies variables based on their distances to these means. New sets of means will then be computed using variables that are assigned to each cluster which will be followed by classification based on the new sets of means. This will be repeated N times until cluster means do not vary much between successive steps, and the cluster means will be computed once more followed by assignments of variables to their final clusters.
Cross tabulation (contingency table) is a fundamental statistical tool to compare and relate categorical variables. The strength of relationship or lack thereof, and its statistical meaning as well as significance can be tested by means of computing Pearson chi-square, likelihood ratio, asymptotic significance, contingency coefficient etc (Conover, 1999; Quade and Salama, 1975; Wilson, 1989).
The spectral information was processed using k-means clustering to find six groups of homogenous characteristics. These groups are then compared with the six plasticity classes of soils from plasticity chart using cross tabulation techniques. Bulk and fine soil samples spectra were treated independently.
Multivariate regression analysis
Multivariate regression analysis is important statistical tool for exploring possible relationships between soil spectra and soil properties (Ben-Dor et al., 2002; Cloutis, 1996; Kariuki et al., 2004b; Rainey et al., 2003; Shepherd and Walsh, 2002; Viscarra et al., 2006; Waiser et al., 2007; Yitagesu et al., 2009). In multiple linear regressions (MLR) the explanatory variables are assumed to be linearly independent. If the explanatory variables are significantly interdependent (as in the case in spectroscopic measurements), problem of multicollinearity will arise which can lead to numerically unstable and spurious estimates of regression coefficients and over fitting (Hair et al., 1987; Martens and Naes, 1989; Yeniay and Goktas, 2002).
Principal component regression (PCR) analysis decomposes set of explanatory variables into eigen vectors and scores that are orthogonal to each other, hence overcome collinearity problems. After achieving optimal projection of the explanatory variables in few principal components, it regress them against the response variable in a separate step. Choice of relevant number of principal components can be complex in cases where relevant underlying effects are small in comparison with noise (Brereton, 2000).
Partial least squares (PLS) regression on the other hand decomposes both explanatory variables and the responses simultaneously to capture their common structure, which will be projected into a small number of mutually independent factors. Explanatory variables are decomposed into new coordinates called 'T-scores' that are computed in such a way that they capture part of the structure in the explanatory variables that is relevant to the response. 'U-scores' on the other hand summarize part of the structure in the response which is explained by the explanatory variables with a given number of principal components. Decomposition and regression is a single step, through fewer principal components than that required by PCR (Brereton, 2000; Martens and Naes, 1989; Wold et al., 2001; Yeniay and Goktas, 2002). Principal components (PLS factors) are extracted in decreasing order of relevance, hence choice of optimal number of principal components is not a problem. PLS regression analysis has got a wide range of application in various disciplines, among which is estimation of soil properties from their reflectance spectral data (Cloutis, 1996; Rainey et al., 2003; Shepherd et al., 2005; Waiser et al., 2007; Yitagesu et al., 2009). PLS regression analysis is used to establish relationships between spectral indices and clay mineralogical content. Distributions of variables were checked and appropriate transformations were carried out on variables that showed skewed distribution before the regression analysis. As described in Martens and Naes (1989) and (Wold et al., (2001) data were mean centered and scaled to unit variance prior to calibration. This enhances variance in the explanatory data in addition to removing any systematic bias. A full cross validation method where one sample is left out at a time is used to validate the model results.
3.1 Geotechnical characteristics
The soil samples are characterized by high plasticity, are predominantly fine grained with high clay fractions. Plasticity chart showing distribution of soil samples according to their potential expansiveness is illustrated in Figure…
Figure Distribution of soil samples on plasticity chart showing their expansion potentials, majority of the soil samples fall above the 'A' line which indicates that they are dominated by active clay minerals.
X-ray diffraction analysis on selected soil samples indicate presence of active clay minerals mainly montmorillonite, nontronite, interstratified montmorillonite-illite, and illite clay minerals in the soils. The results show presence of these minerals in majority of tested soil samples as major constituents, comprising more than 30 % by weight. Illite/montmorillonite mixed clay minerals, illite, kaolinite were also found in the soils ranging from major (>30 %) to moderate (10-30m %), minor (2-10 %) and trace (<2 %) amounts. Presence of quartz, feldspars and goethite are also indicated from the mineralogical assemblage analysis. Calcite was also found in trace amount in some of the soil samples. Formations of these minerals are favored by the geology of the study area, coupled with the tropical climatic conditions and topographic setting. XRD patterns and a table summarizing clay mineral and chemical oxide constituents of the soil samples are presented.
Figure XRD patterns of soil samples on bulk and clay fractions.
3.3 Spectral characterization of expansive soils
Differences appear in overall spectral contrast and reflectivity among the bulk and fine fraction soil spectra, as well as between the two atmospheric windows 3-5µm and 8-14µm spectral regions respectively. Reflectance in the 3-5µm is generally higher in spectra of finer fraction soil samples than in the spectra of similar bulk soil specimens. On the other hand, in the 8-14µm spectral region (particularly between ~8.6µm and ~10.5µm with peak near 9.4µm) spectra of bulk soil samples exhibit reflectance higher than their fine fraction varieties.
Grouping spectra of soil samples by plasticity classes and computing mean spectra of each class indicates statistical differences among the six groups. Figure … presents plots of continuum removed mean spectra of bulk and fine fraction soil specimens in the 3-5µm spectral region of the six plasticity classes depicting characteristic differences in spectral properties. Each mean spectrum is statistically significantly different from one another.
Variations in spectral features appear to show larger similarity in pattern among the spectra of bulk and fine fraction specimens. The absorption feature annotated with number 1 which is ascribed to presence and relative abundances of montmorillonite and illite clay minerals (Yitagesu et al., …) shows decrease in depth intensity, width and area as plasticity gradually changes from extreme to very high, high, medium, low and non-plastic (Figure .. and …) nature. In both bulk and fine fractions, appearance of features of illite and kaolinite becomes pronounced with decreasing plasticity (note the strong kaolinite and quartz feature near 4.2µm). The absorption deep centered ~3.9µm also show variations in depth intensity which appear to correspond with changes in plasticity (increasing as plasticity gets higher). Shape of spectral curves in the ~4.3µm to 5µm spectral region which appears as broad rounded trough in the dominance of montmorillonite or illite (deeper in the former); it tend to be flat or shallow in the domination of kaolinite.
CR3-5_Meanbulk per plastcity_annotated.tif
CR3-5_MeanFine per plastcity_annotated.tif
Figure Mean continuum removed spectra of: NP (non plastic), Low (exhibiting low plasticity), M (showing medium plasticity), H (showing high plasticity), VH (showing very high plasticity) and Ext (showing extreme plasticity) characteristics of spectra of bulk soil samples (A) and fine fraction soil samples (B); Where 1) absorption feature due to presence of montmorillonite or illite, 2 and 3) illite, 4) either illite/montmorillonite or illite/kaolinite interstratified clay species, 5) kaolinite and 6) montmorillonite or illite or kaolinite.
Continuum removed spectra of bulk and fine fraction soil specimens in the 3-5µm spectral region are illustrated in Figure … showing spectral differences within each plasticity class. Each pair of spectra of soil samples in Figure … belong to the same category of plasticity, hence same expansion potential exhibiting only subtle differences in their liquid limit. Relative abundances of dominant clay minerals can be qualitatively perceived based on spectral indices of characteristic spectral features established as diagnostic of the three clay mineral species dictating soil expansiveness (Yitagesu et al., 2011?). The Figures indicate presence of montmorillonite, illite and kaolinite clay minerals as well as quartz in the soil samples. Diagnostic spectral features are indicated by annotations. The montmorillonite clay species appear to dominate those soils with extremely high and very high plasticity. Presence of illite appear to range from extremely plastic varieties where it shows signatures of interstratified type with montmorillonite, to non-plastic soil where it appears again as interlayer type with kaolinite clays. Spectral signatures of kaolinite are seen in soil showing plasticity of medium and low, as well as in non-plastic soils.
Figure Soil samples obtained A (bulk) and B (fine fraction): from kilometer 2.5 (dotted line) and 8.5 (solid line) exhibiting liquid limits of 90 and 98 percent respectively. Both fall in the extremely high plasticity portion of the plasticity chart. The bulk soil spectra appear to be dominated by montmorillonite (A), and the fine fraction specimen spectra (B) are spectrally interpreted as dominated by illite-montmorillonite interstratified clay minerals. While the ~3.1µm absorption feature shows differences in depth intensity which probably indicates higher fractional abundance of montmorillonite in the soil specimen recovered from kilometer 8.5, the feature ~3.9µm suggest higher fractional abundances of illite in the soil sample from kilometer 2.5.
C (bulk) and D (fine fraction): from kilometer 42 (solid line) and 46 (dotted line) showing liquid limit of 84 and 85 percent respectively and fall in the very high plasticity portion of the plasticity chart. Spectrally both samples appear to be dominated by illite (owing to the presence of clearly defined doublet), pronounced depth near 3.1 suggests presence of montmorillonite as well.
E (bulk) and F (fine fraction): from kilometer 56 (solid line) and 59.5 (dotted line) showing liquid limit of 52 and 50 percent respectively. The two samples fall in the high plasticity portion of the plasticity chart, are spectrally interpreted as dominated by illite clay mineral. The spectra show prominent differences in the doublet feature that is typical of illite. While spectrum of sample 56 exhibits a deeper doublet, this same feature shows lower depth intensity in the spectrum of sample 59.5 correspondingly the two samples show variation in their liquid limit values which can be attributed to differences in illite concentration.
G (bulk) and H (fine fraction): from kilometer 50.4 and 64 exhibiting liquid limit of 44 and 39 percent respectively, Spectra show subtle differences in depth of the ~3.1µm feature and the doublet feature near 3.9µm which is attributed to correspond with differences in illite/kaolinite contents.
I (bulk) and J (fine fraction): from kilometer 68.5 (solid line) and 79.5 (dotted line) showing liquid limit of 30 and 33 percent respectively. Note the differences in depth intensity of the feature near 3.1µm as well as the absorption deep near 3.9µm which correspond with differences in concentration of illite and kaolinite in the samples.
K (bulk) and L (fine fraction): from kilometer 7 (solid line) and 62.5 (dotted line) both non-plastic soils. Presence of illite and kaolinite is evidenced by the presence of features near 3.4µm (illite), 3.6µm (kaolinite) and another one near 3.9µm which also indicates presence of intermixed illite-kaolinite.
3.4 Empirical relationships between plasticity and spectral reflectance of soils
Statistical relationships of plasticity classes with the six clusters generated from k-means clustering where distance to mean is used as a measure of determining similarity or dissimilarity among samples are presented in contingency tables Table …. A and B. Figure… graphically illustrates which plasticity class occurs in which k-means cluster.
The obtained chi-squared test statistic are 86.76 and 147.71 respectively with associated two sided asymptotic significance of p <0.001 and large likelihood ratios for bulk and fine fraction soil samples respectively. This signifies that the classes from k-means clustering and plasticity categories are associated than is expected due to chance. Further examination of the pattern on the number of joint occurrences of classes in the contingency tables and bar charts (Figure…) indicates for instance, all non-plastic soils fall in the same cluster, majority of extremely plastic soils fall in one cluster and so on. Symmetric measures (phi, Cramer's V and contingency coefficient) that are based on the chi-squared statistic are used to determine the strength of relationship between the classes. All the three measures indicate statistically significant (note the associated significance levels of p<0.001) also strong relationship between the classes.
Table k-means cluster classes from spectra of the bulk (A) and fine fraction (B) soil samples in the 3-5µm spectral region and soil plasticity classes cross tabulation; and symmetric measures showing strength of the relationship between the k-means cluster classes and plasticity classes of bulk (C) and fine fraction (D) soil samples.
bar_bulk vs fine_kmeans vs plasticty.tif
Figure Clustered bar charts illustrating the relationships between k-means clustering classes from spectra of the A) bulk soil and B) fine fraction soil samples in the 3-5µm spectral region and the six soil plasticity classes. Those samples of comparable plasticity (e.g. Ext and VH) appear to be spectrally similar hence fall in similar cluster; while those exhibiting plasticity of exaggerated difference (e.g. Ext and L) fall in distinct clusters.
It appears that the strength of the relationship between plasticity classes and k-means clusters (Table …) obtained from spectra of fine fraction soil samples are higher than those of the bulk soil samples. This higher strength of relationship can be ascribed with presence of more clay particles and clay minerals in the finer soils than in the bulk soils, and their influence on the spectral behaviors of soils. The results are suggestive of the possibility that soil spectral and plasticity characteristics are influenced by similar causative factors.
Box plots summarizing the relationship between liquid limit and the k-means clusters from bulk and fine soil fractions are presented in Figure… The means of each k-means cluster class in the fine fraction soils are markedly different from one other with each class exhibiting narrower variability in liquid limit values except for cluster 4 which shows larger variability than is apparent in the other clusters. On the other hand, the means of each clusters in the bulk soil samples do not differ pronouncedly particularly among clusters 2, 3 and 4. Although the extreme cases seem separable the ranges of variation in liquid limit values of member samples in four of the clusters appear to be larger. Altogether the k-means clustering on spectra of fine fraction soil specimens well reproduce or show better agreement with the classification from plasticity chart.
Figure Box plots showing ranges of distribution of liquid limit values of soils samples grouped by the k-means classes A) of bulk soil samples B) of fine fraction soil samples; Mean of each cluster in the case of fine fraction soils exhibit appreciable differences amongst each other with clusters showing smaller variability in liquid limit. Whereas in the case of bulk soil samples particularly the means of clusters 2, 3 and 4 appear to be not as distinctly different as in the clusters from spectra of the fine fraction soils with wider spread in values of liquid limit.
3.4.1 Partial least squares regression analysis
Partial least squares regression analysis conducted on spectra of bulk and fine fraction soils is presented in this section.
A score plot of the first two PLS factors that summarize most of the spectral information which account for much of the variation in the liquid limit of the soil samples is shown in Figure…Samples are coded by plasticity classes reported in the plasticity chart presented in Figure…. The plot shows no strong grouping pattern in the dataset, but clear differences among samples manifesting extreme and lower plasticity characteristics. While those samples of extremely high and very high plasticity appear more similar in their loadings, those with medium and low plasticity also fall close to each other. On the other hand, samples exhibiting high plasticity characteristics plot in between. The loading conform to the spatial dependence of nature of soil geotechnical characteristics and the fact that soil plasticity is rather a fuzzy nature than is crisp. Hence the spectral properties of the soils exhibiting varying amount of plasticity although unique, vary subtly. Effect of other factors and contribution of varying amount of clay concentration apart from mineralogy can be attributed to the rather mixed appearances of few samples where they fall apart from their own species.
Exploring the data structure using T and U score plots where the degree of linear or non-linear correlations and tendency of data clustering can be visualized (Esbensen and Geladi, 2010) indicated strong data structure with correlations between the X (explanatory) and Y (response) spaces are large in all cases.
Figure score plot showing distribution of soil samples in the first two PLS factors which summarize much of the variance in the data with samples classified according to their plasticity and labeled accordingly. Although no strong grouping pattern is apparent, loadings appear progressively different among the samples belonging to the extreme and lower plasticity classes signifying their dissimilarity.
Fine 3-5 and 8-14 reg overview.tif
Figure relationship of the measured and predicted liquid limit of fine fraction soil samples in the 3-5µm (A) and 8-14µm (B) spectral regions.
In some spectra (probably illite-montmorillonite and illite-kaolinite interstratified varieties as also verified by the XRD analysis on selected soil samples) apart from the ~6.1µm water absorption feature there appears another feature near 6.9µm. In such spectra the doublet feature which is typical in the presence of illite (Yitagesu et al., 2011?) is resolved as a single absorption deep (Figure…). Weather the clay mineral involved is illite-montmorillonite or illite-kaolinite interlayer type can be identified by additional features for instance presence of absorption minimums at ~3.84µm and ~3.98µm in illite that tend to lower in depth intensity as mineralogy is dominated by kaolinite (Yitagesu et al., 2011??).
Quartz is the most abundant and common soil forming mineral (Fitzpatrick, 1980) on earth's surface. Although presence of quartz is evident in the soil specimens as demonstrated in the XRD analysis (Figure … and Table…), no quartz features appear in the spectra of the soil samples particularly in the 8-14µm spectral region where soil spectra is reported to be dominated by quartz reststrahlen bands (Salisbury and D'Aria, 1992a). This lack of prominence of quartz spectral features is probably due to coating of individual quartz grains by clay minerals as explained in Yong (1975) where it is described that the smallest particles in soils do not spread evenly, but often arranged in parallel orientations forming coating around larger grains. However, a spectral feature near 4.3µm exhibited in spectra of sandy soils (mostly exhibiting lower plasticity) might be related with quartz as described by Salisbury and D'Aria (1994) where they assigned the weaker overtone or combination tone absorption bands of quartz to appear in the 3-5µm spectral region. They noted the strongest of these bands near 4.5µm and 4.7µm, while soil spectra presented here shows similar features at shorter wavelengths near 4.3µm.
In spectra of bulk soil samples, high reflectance peak at ~9.4µm is observed, showing comparable maxima as the one at ~3.9µm.
Model performances in terms of magnitude of relationship as well as error terms are found to be better in the finer soil fractions than in the bulk. This can be attributed to minor differences in mineralogy, possibly due to the fact that higher concentrations of clay minerals which are responsible for soil plasticity in the finer fraction than in the bulk soil where other inert minerals as well as physical characteristics might be influencing.
Again model performance in terms of magnitude of relationship as well as error terms are found to be better in the 3-5µm spectral region than in the 8-14µm. Organic matter is reported (Salisbury and D'Aria, 1992a) to be highly absorbing in the 8-14µm spectral region, hence can have a masking effect on clay minerals spectral features. According to Gaffey (1986)…
Considerable scatters in the plots depicting the relations between the measured and predicted liquid limit values indicate other sources of interferences contributing to the spectral variability of soils, yet to be discovered. Similar effects are assumed to be responsible for the larger variability in liquid limit exhibited by k-means cluster classes of bulk soil samples shown in Figure….
A major challenge in using spectroscopy in the spectral region beyond 2.5µm appears to be lack of standard in sample preparations. As demonstrated in this paper and established by a domain of experts in the field (Bras and Erard, 2003; Gaffey, 1986; Salisbury and D'Aria, 1992a, 1994) different factors are identified to induce variation in spectral characteristics unrelated to differences in mineralogy.
Conclusions and recommendations
Alternate techniques of quantitative estimation of soil expansion potential are required. This is to aid in optimal design of structures and formulation of proper mitigation measures for formulating measures in order to counteract problems associated with volume change characteristics of these soils. This study demonstrated the potential use of the spectral region between 3-5µm and 8-14µm for estimating geotechnical properties of expansive soils showing subtle variation in their liquid limit values. As presented in the results differences in spectral signatures of the soils are able show subtle variations in their plasticity. Of particular importance are differences in absorption features near 3.1µm and ~3.9µm that appear to vary with smectite and illite contents respectively.
The partial least squares regression analysis show good prediction ability with strong coefficients of determination, indicating spectral variability of soils are able to account for much of the variation in their expansiveness. It can be concluded from this study that reflectance spectra of soils in the two atmospheric windows (3-5µm and 8-14µm) are rich in information content that can be related with soil geotechnical characteristics more so those related with their potential expansiveness. We proposed a method that identifies statistically significant wavelengths for estimating soil geotechnical characteristics and established empirical relationship between soil expansiveness and their respective spectral characteristics whereby liquid limit of the soil samples are estimated.