A TRIAL OF APPLYING THE HYPER-SPECTRAL SENSING TECHNIQUE IN THE IDENTIFICATION OF SOME SOIL ATTRIBUTES

The soil is a heterogeneous system whose processes and mechanisms are complex and difficult to fully comprehend (Viscarra Rossel et al., 2006). Many conventional soil analytical techniques are used in an attempt to establish the relationship between soil physical and chemical properties and individual soil components, often disregarding their complex, multi-component interactions. Indeed, soil chemical extractions that alter the equilibrium between the phases may further complicate the interpretation of results. Historically our understanding of the soil system and assessment of its quality and function has been gained through this type of laboratory analysis. We need to further develop our analytical techniques to better understand the soil as a complete system and a resource so that we may make more efficient use of it and simultaneously preserve it for future generations. This is more important now than ever before since the acquisition of larger amounts of accurate soil data is essential if we are to manage our base resources sensibly to meet the food demands of future populations (Bilgili et al., 2010). Remotely sensed hyperspectral data have great potential for quantitative assessment of soil and vegetation parameter at spatial scale. The development of methods to map soil properties using optical remote sensing data in combination with field measurements has been the objective of several studies during the last decade. Also, it has been a challenge to find the most appropriate technique for studying soil properties from optical data and thus reducing the time and effort involved in field sampling and laboratory analysis. Hyperspectral sensors operate with more than hundreds of bands with good spatial and spectral resolution producing continuous spectra. With the progress and maturity of the technology, hyperspectral remote sensing has found a wide range of applications  in mapping soil types and quantifying soil constituents (Minu      et al., 2016). 

Infrared spectroscopic techniques are highly sensitive to both organic and inorganic phases of the soil, making their use in the agricultural and environmental sciences particularly relevant. Intense fundamental molecular frequencies related to soil components occur in the mid-infrared (MIR) between wavelengths 2500 and 25,000 nm. The visible and infrared portions of the electromagnetic spectrum are highlighted in Fig. 1. Weak overtones and combinations of these fundamental vibrations dominate the near-infrared (NIR) (700–2500 nm) and electronic transitions the visible (VIS) (400–700 nm) portions of the electromagnetic (EM) spectrum. Quantitative spectral analysis of soil using visible and infrared reflectance spectroscopy requires sophisticated statistical techniques to discern the response of soil attributes from spectral characteristics. Various methods have been used to relate soil spectra to soil attributes. For example, Elwan and Sivasamy (2013) used multiple regression and correlation analyses to relate specific bands in the NIR to a number of soil properties in a semi-arid area of India. Shibusawa et al. (2001) used stepwise multiple linear regression for the estimation of various soil properties from the NIR spectra of soil. Shepherd and Walsh (2002) used multivariate adaptive regression splines for the estimation of soil properties from soil spectral libraries.

Soil reflectance characteristics are determined over the entire visible (350–700 nm) and near-infrared (700–2500 nm) region with the use of a monochromator (Viscarra Rossel et al., 2006). Raw data, first-, and second-derivatives each provide valuable information that can be analyzed separately or combined using multivariate statistical methods or data mining techniques. Soil constituents have unique absorption features in these wavelength regions due to overtones related to stretching and bending vibrations in molecular bonds (Viscarra Rossel et al., 2006). Chang et al. (2001) predicted more than thirty soil properties simultaneously with variable levels of success using a principal component analysis method with cross-validation. They reported successful predictions (R²> 0.80) for total organic carbon and nitrogen (g kg^-1), gravimetric soil water content, soil water content, exchangeable calcium, cation exchange capacity (CEC) and silt and sand content. Brown et al. (2006) used over 4100 surface and subsurface soils from across the United States, Africa and Asia to evaluate the accuracy of VNIR empirical models for global soil characterization and reported strong predictability for kaolinite, montmorillonite, clay content, as well as CEC, soil organic carbon,      inorganic carbon, and extractable Fe. Others also used VNIR spectroscopy to

Fig. 1.  The electromagnetic (EM) spectrum highlighting the visible and infrared portions (after Viscarra Rossel et al., 2006).

successfully predict organic carbon and nitrogen (Reeves et al., 2002), Fe₂O₃, Al₂O₃, CaCO₃, potentially mineralizable nitrogen (Reeves and Van Kessel, 1999), heavy metals, micronutrients (Udelhoven et al., 2003), C:N ratio and soil biological properties (Ludwig et al., 2002). Additionally, the prediction of soil constituents that do not absorb within the VNIR range may be possible through their correlations with spectrally active constituents.

Therefore, the objectives of this study were to: (i) determine the efficiency of FPHR with comparison to traditional field and laboratory methods in enhancing soil horizon differentiation, (ii) determine whether FPHR spectroscopy can be used easily as a rapid, inexpensive alternative or supplement to traditional methods for measuring soil properties by correlation coefficients, and (iii) find the appropriate bands across the spectrum range that can properly predict the soil variable using the regression analysis.

MATERIALS AND METHODS

Pedon Sampling

   The study was conducted on three soil orders (Vertisols, Aridisols, and Entisols), which cover the major variations in soil types in Egypt. Standard eight pedons consisting of 34 samples (Fig. 2) were collected from different semi-arid sites in Lower Egypt. Two pedons were collected from Nile old deltaic plain (P1 and P2) and one pedon (P3) from bajada plain at wadi Al-Molak of East Delta; one pedon (P4) from Inshas, Sharkia Governorate; two pedons  (P5 and P6)  from  Abu-Soltan,  Ismailia Governorate; one pedon (P7) from Al-Tur, South Sinai; and one pedon (P8) from Al-Hamam area, Northeastern coast of Egypt. P1 was classified as Vertisols, P2 and P3 were classified as Aridisols, and P4 to P8 were classified as Entisols. Study areas have a semi-arid climate with a mean annual precipitation, evaporation, temperature and relative humidity of 13 mm, 881 mm, 22 ˚C and 55%, respectively.

  In the field, a range of soil features is generally used during the process of horizon description, including soil color, texture, and structure, which are essentially affected by the physical and chemical composition of the soil. For example, besides soil organic matter (SOM) and soil water content, Fe and Mn are the primary coloring agents for many soils. Morphological descriptions for all pedons were morphologically described and the horizons were differentiated based on visual examination and hand texturing according to Schoeneberger et al., (2012). Collected samples were transported to the laboratory in sealed plastic bags.

Laboratory Analyses        

In the laboratory, soil samples were air-dried and gently ground to pass through a 2 mm sieve, then subjected to standard soil characterization. The prepared samples were then scanned on the sample surface using FPHR (described below). Particle size analysis was accomplished via pipette method and sieved sands using a 63 μm sieve (Gee and Bauder, 1986). Soil organic carbon (SOC) was quantified via titration following the Walkley–Black dichromate oxidation method (Nelson and Sommers, 1996). Gypsum concentration was determined by the differential water loss method (Artieda et al., 2006). Carbonates were determined using a calcimeter (Kacar, 1994); soil pH with a 1:2 soil/water suspension using a glass electrode pH meter (McLean, 1982); electrical conductivity (EC) in soil extraction using a conductivity meter (Janzen, 1993). Furthermore, the soil samples were analyzed for soil available water (A.W.), SiO₂, Al₂O₃, and Fe₂O₃ following standard procedures (Jackson, 1973; Soil Survey Staff, 2014).  

Field Portable Hyperspectral Radiometer (FPHR)

   The GER 1500 model is a field portable hyperspectral radiometer (FPHR) covering the UV, Visible, and NIR wavelengths from 276 nm to 1093 nm (Fig. 3). It was used to scan each soil sample. The instrument uses a diffraction grating with a silicon diode array. The silicon array has 512 discrete detectors that provide

Fig. 3. Field portable hyperspectral radiometer (FPHR) sensor used in the current study.

the capability to read 512 spectral bands. The spectroradiometer includes memory for stand-alone operation as well as the capability for computer-assisted operation through its COM2, RS232 serial port. The spectral readings can be stored for subsequent downloading and analysis using a personal computer with a standard RS232 serial port and GER licensed operating software. Computers incorporating only USB serial ports may be connected to the GER 1500 by using the SVC ADP000015 USB Serial port adapter. An optional external GPS device may be connected via the instrument’s COM1 RS232 serial port. When connected, GER 1500 records the latitude, longitude, and time of each spectral reading. For all scanning, recalibration and verification with NIST standards were conducted every 20 scans and the aperture of the instrument was covered with a thin plastic wrap to prevent soil or dust from contaminating the aperture window. Data from the instrument was exported to MS Excel for analysis and display.

Hyperspectral reflectance and data processing

Thirty-four soil samples of selected eight pedons were air-dried, crushed, and sieved (2 mm); then the samples were scanned using FPHR in the laboratory condition. Soil samples were individually spread on a white paper (30 × 42 cm diameter) forming a layer of 1.8 cm (1.5 cm is considered as optically infinitely thick for soil). The sample surface was scraped plane with a ruler, as pressing can affect the porosity of the soil and result in a false measurement. The absolute reflectance of samples was recorded for 276–1093 nm at 1.5 nm spectral resolution, yielding a total of 512 data channels per spectrum. Reflectance spectra were measured mid noon in between 11.30 am to 12.30 pm, for allowing good sunlight. The zenith angle of the FPHR was set to 45^o by pointing the instrument at a distance of 30 cm above the soil surface. A standard panel coated with barium sulphate (BaSO₄) was used as a reference for the reflectance calibration before each set of measurements. Each reflectance measurement produced a single spectrum. Reference measurements were taken before the first measurements set and after every five minutes onwards to adapt the changing atmospheric conditions. The percent reflectance spectrum was calculated as the ratio between the reflected spectra from the target (soil sample) and the incident spectra from the panel (reference) using the following formula.

The spectral reflectance data, both absolute and percent reflectance values, were transferred from the FPHR to a personal computer as ASCI files with .asc extension utilizing a specific software supplied with the instrument. These files were later opened in a spreadsheet programme and further analyses were carried out. In the current study, raw spectra were tested both separately and jointly in predicting soil horizonation. Furthermore, the reflectances in blue, green, red, and near-infrared (NIR) bands were selected due to their most sensitive wavelengths to soil components and calculated for each sample by taking mean of reflectance values in the wavelengths ranges of 450-520 nm, 520-600 nm, 630-690 nm and 760-900 nm, respectively, to match the bands in the Landsat Thematic Mapper (TM) sensors (Fig. 1).

Statistical analysis

  Reflectance data were translated from binary to ASCII and exported in batches using ViewspecPro (Analytical Spectral Devices, Inc., Boulder, CO, 80301). The ASCII files were later opened in a spreadsheet programme and further analyses were carried out. The sequential percent spectral reflectance readings obtained from each sample (512 values) at approximately 1.5 nm bandwidth interval across the spectrum wavelengths were performed using Microsoft Excel spreadsheet software, producing a master data file with one representative reflectance spectrum per soil sample. Calibrations between soil reflectance and soil parameters were performed using both correlation (r) and multiple regression models for selecting bandwidth that is the best for prediction of soil properties.

   Correlation between each band and each soil property were worked out separately for spectral datasets and evaluated the relationship of correlation of reflectance with soil properties with the change in bandwidth. Multiple regression models for each soil property were developed using each spectral data sets. Model predictability (Model R²) was evaluated for selecting the best bandwidth for prediction of soil properties. Optimum bandwidth found was to be used in the study for prediction of soil properties. Regression equations were fitted and plotted as graphs to quantify the soil constituents using reflectance in the band in which the highest correlation was registered, as suggested by Chang et al., (2001). Bivariate correlations analysis was done between soil properties and spectral data sets using SPSS software. Correlation analysis was performed for each soil property with each band. Best correlated bands from each reflectance related datasets were selected separately for each soil property, considering the absolute values of correlation coefficients.

   The prediction model was developed for each soil properties considering all the bands as a variable. Model predictability (R²) was used for evaluating the spectral data sets for prediction of soil properties. The spectral dataset with the highest R² was selected for model development for each soil property. The correlation with each soil properties and reflectance data at different bandwidth was computed and plotted against wavelength. Correlation between soil properties and reflectance at different wavelength for spectral data sets was evaluated for all soil properties. Multiple linear regression is a common multivariate tool which, at its simplest level, forms a model that specifies the relationship between a response variable (Y) and a set of dependent variables (X). The soil property was considered the dependent variable, and the band reflectance was the independent variables. After a choice of the number of bands, multiple linear regression was carried out for each soil attribute and best-correlated bands from each spectral dataset were selected. Best dataset and optimum number of bands to be included in the model have been selected based on the highest R² value.

RESULTS AND DISCUSSION

Field morphological horizonation

   The selected morphological characters of studied pedons are furnished in Table 1 and visualized in Fig. 2. Moreover, the textural classes across all landscapes are widely varied from sand to clay, indicating the heterogeneity of soil forming processes and variety of parent materials. Sand fraction concentration ranged from 9.3% in P1 to 95.4 in P 8 while clay content varied from 1.2% in the subsurface layer of P8 to 62% in the deepest horizon of P1 (Table 1). Most pedons selected for this study were derived from alluvium as it is the dominant parent material of Lower Egypt. Pedons P1 and P2 were developed on Nile old deltaic plain at toeslope landscape position in wadi Al-Molak, East Delta while P3 was developed on bajada plain at toeslope of wadi Al-Molak. Pedon P4 was derived from ferruginous sandstone on the alluvial plain of Inshas area while pedons (P5 and P6) from alluvial plain at Abu-Soltan region. Furthermore, pedon P7 was formed on alluvial fans and outwash plain at Al-Tur area while pedon P8 were formed locally from weathered marine limestone on inland portion of the northeastern coastal region.

A soil horizon is defined as a layer of soil or soil material approximately parallel to the land surface and differing from adjacent genetically related layers in physical, chemical, and biological properties or characteristics such as color, structure, texture, consistency, kinds and number of organisms present, or degree of acidity or alkalinity (Soil Science Society of America, 2017). As the formation of horizons is

a function of a variety of physical, chemical, geological, and biological processes associated with the landscape and climate over long time periods, the differentiation of soil horizons is essential for the understanding and classification of soil (Schaetzl and Anderson, 2005). The process of field horizonation to some degree is a process of subjective approximation of soil features by field soil surveyors. Surveyors use all the tools available to differentiate soil horizons and establish minimal within-horizon variability, considering a variety of soil properties. As such, significant variations of soil properties should occur between soil horizons in a given pedon. Clearly, the most important part of horizonation is the identification of differences between soil horizons. Pedon P1 has highly developed horizonation sequence: Apzg-Btg-Bssz-Btk-Btkm and classified as Aquic Salitorrerts, which characterized by anthraquic condition (APzg and Btg), salic horizon (Bssz). Likewise, the horizon sequences of Aridisols pedons are Ap-E-Btn-Btk-Bt for P2 (Petronodic Natrargids) and Ap-Btnz-Bq-C for P3 (Calcic Haplosalids). On the other hand, the layer sequences in the Entisols pedons are: Ap-C-2CK-3C1-3C2 for P4; C-2C-3C-4C for P5; Ap-C-2C1-2C2 for P6; C-2C1-2C2-3C for P7; and C-2Cqy1-2Cqy2 for P8 (Fig. 2).

Table 1. Soil characterization data for thirty-four soil samples of eight pedons from different areas across Lower Egypt

The boundary between two layers was clearly identified in the field based on color, texture, structure, hardness, or other features which largely influenced by the pedogenic process. Boundaries between the horizons in studied pedons are varied in both distinctiveness and topography. Abrupt smooth boundaries were the dominant within the most investigated pedons. The occurrence of smooth and abrupt to diffuse boundaries in studied pedons is often cited as field evidence for a lithologic discontinuity (LD) (Schaetzl and Anderson, 2005). In P4, the soil exhibited abrupt smooth C-2Ck and 2Ck-3C1 boundaries in P4 indicating two unlike parent material with different modes of deposition occurring on more stable surfaces, however, the wavy boundary was found between Ap-C horizons indicating same materials (Fig. 2).

Prediction of soil horizonation

Basic soil constituents affecting soil reflectance characteristics are anthraquic features of Vertisols, soil water, clay content, organic matter, and Fe–Al oxides (Bowers and Hanks, 1965). Fig. 4 shows raw reflectance spectra and their prediction in pedon horizonation (Fig. 2). Soil reflectance was generally lower in the visible range of blue and green bands (450-600 nm) and higher in the red and near-infrared range (630-900 nm). With regard to FPHR analysis of pedon horizonation, the studied pedons qualitatively showed good alignment with field-established horizonation. Absorption peaks for Vertisols were higher and the percent soil reflectance was low (16-20%) while for Aridisols it was slightly low (20-25%) in P2 and slightly high (32-40%) in P3. By contrast,  the soil reflectances of Entisols pedons were higher compared to Vertisols and Aridisols and varied from 35% in C layer of P4 to 75% in P8 (2Cqy2). Horizonation was easily identified based on the reflectance characters for each soil sample (horizon or layer) (Fig. 4).

Given the raw reflectance data from FPHR (Fig. 4) vs. morphological horizonation (Table 1) from field survey, it is suggested that FPHR could be used as a tool to assist in field morphological horizon differentiation. For example, FPHR could be used to identify multiple argillic horizons within a Vertisol pedon and lithologic discontinuity in P4 at depths of 55 and 105 cm. In summary, the results concluded that the data afforded by the use of FPHR sensor offer pedologists unique insights into predicted differences between soil horizons-differences that may be indicative of lithologic discontinuities and soil horizonation.