چكيده لاتين :
In recent years, some areas of Khuzestan Province have experienced excessive drought and have become dust
production centers due to the decrease in soil resistance and loss of vegetation. Combination of these factors along
with poor management, has led the soil erosion resistance to reduce against winds. Among soil properties, texture
as an essential characteristic plays a crucial role in soil resistance to wind and rain erosive factors and affects
water movement and soil fertility (Hillel, 1980).
Since traditional methods of soil texture measurement are costly and time-consuming, in recent decades
researchers, have used novel methods such as Remote Sensing and reflectance spectroscopy to estimate soil
properties, especially for large areas (Ben-Dor et al., 2009; Curcio et al., 2013).
One of the limiting factors in the evaluation of soil properties by spectroscopy is the identification of preprocessing
methods in noise and error elimination and determining the appropriate regression model to estimate
these properties (Mohamed et al., 2017). Therefore, proper regression and pre-processing methods are required to
determine soil properties using soil reflectance. Xuemei and Jianshe (2013) used PLSR and LS-SVM models to
investigate soil properties. Based on their results, the LS-SVM model estimated the properties of organic matter,
nitrogen, phosphorus, and potassium with determination coefficients of 0.87, 0.82, 0.76, and 0.73, respectively.
Silva et al. (2016) used the PLSR model spectroscopy and second derivative methods to determine the soil texture
of southwestern Greece. Based on evaluation of coefficients of determination and root mean square error, the
results of their study showed the accuracy percentage of sand, silt and clay were 0.3, 0.59 and 0.69 and 5.47, 5.18,
5.39 respectively in 100 grams of soil. Wang et al. (2018) used partial least squares regression (PLSR) and
random forest (RF) methods to estimate soil salinity and based on their research results, the RF model had better
performance than the PLSR model.
In the context of considering the complicated relationship between soil properties and their reflectance, it is
necessary to use the spectroscopic method to determine the best statistical model for spectral analysis in different
regions. In this regard, the objectives of this study are 1-Estimation of clay, silt, and sand percentages
characteristics of productive dust soil of Khuzestan province by PLS-RF model, 2- Comparison of PLS-RF model
performance and accuracy in 6 spectral methods including: Main Spectrum, Savitzky-Golay filter, first derivative
with the Savitzky-Golay filter (FD-SG), second derivative with the Savitzky-Golay filter (FD-SG), Standard
Normalization Method (SNV) and Continuous Removal Method (CR), and 3-Determining the key wavelengths of
soil texture in these areas.
2-Material and methods
The area under study was in a geographical coordinate range between 30°24′ to 31°19′ and 49°21′ to 49°28′. The
area is located in south and southeast of Ahvaz toward the south of Khuzestan province which is about 110,000
hectares. One hundred forty-two soil samples were collected from 0 to 5 cm depth by the systematic-random
method. Hydrometer method was also used to determine soil texture.
Spectroscopy: ASD FieldSpec3 laboratory spectrometer was used to determine soil reflectance. At the laboratory,
a small amount of soil sample was transferred to a Petri dish with a diameter of 8 cm and a depth of 2 cm. Spectral
measurements were carried out using three separate detector types in the range 2500–3500 nm, which is the range
from visible to near-infrared. For each sample, ten reflectance spectra were measured, and ten replicates for each
soil sample were averaged using ViewSpect software then stored as a spectral library in the spectral library.
Spectrum analysis: Pre-processing was performed on the primary spectrum, including types of filters using
software and then modeling and estimation of soil texture by PLS-RF method.
3-Findings
The results of PLS-RF model showed that for clay, the highest accuracy belonged to the continuum removal
method and the least accuracy belonged to the primary spectrum. For sand percentage, second derivative and SG
methods had the highest and least accuracy, respectively. Finally, for the silt, the least accuracy was obtained
rather than clay and sand percentages. According to the results, the highest and least estimation accuracy was
observed in continuous removal (CR) and Savitzki-Golay straightening methods, respectively.
4-Conclusion
In this study, the performance of PLS-RF model in the primary spectrum and five pre-processing methods was
compared to estimate the silt, sand, and clay properties in the soil of dust center areas of Khuzestan province.
According to comparison with pre-processing methods in estimating soil properties, it was observed that
continuum removal method in both clay and silt percentage had the best performance and for the sand percentage,
the second derivative method showed the best accuracy estimation.
Based on the results of spectral correlation in this study, the key wavelengths for clay percentage were in the
wavelength range 1200-1210, 1800 and 2200 nm, for sand percentage in the wavelength range 1400-150, 1930-
1930, 2200 and 2220 nm and for soil silt percentage The wavelengths of 1320, 1615 and 2200 nm were observed.
Therefore, the use of PLS regression method and determination of significant components in each spectral group
resulted in reducing computational effort, increasing the speed of computational processing and finally obtained
the optimum performance in estimating soil properties.
References
Hillel, D., 1980. Applications of soil physics. Academic Press, San Diego.
Ben-Dor, E., Chabrillat, S., Demattê, J., Taylor, G., Hill, J., Whiting, M., Sommer, S., 2009. Using imaging spectroscopy to
study soil properties. Remote Sensing of Environment 113, 38-55.
Curcio, D., Ciraolo, G., D’Asaro, F., Minacapilli, M., 2013. Prediction of soil texture distributions using VNIR-SWIR
reflectance spectroscopy. Procedia Environmental Sciences 19, 494-503.
Mohamed, E., Saleh, A., Belal, A., Gad, A.A., 2017. Application of near-infrared reflectance for quantitative assessment of
soil properties. The Egyptian Journal of Remote Sensing and Space Science 21, 1-14.
Xuemei, L., Jianshe, L., 2013. Measurement of soil properties using visible and short wave-near infrared spectroscopy and
multivariate calibration. Measurement 46(10), 3808-3814.
Silva, E.B., Ten Caten, A., Dalmolin, R.S.D., Dotto, A.C., Silva, W.C., Giasson, E., 2016. Estimating Soil Texture from a
Limited Region of the Visible/Near-Infrared Spectrum. Digital Soil Morphometrics, Springer, pp. 73-87.
Wang, J., Ding, J., Abulimiti, A., Cai, L., 2018. Quantitative estimation of soil salinity by means of different modeling
methods and visible-near infrared (VIS–NIR) spectroscopy, Ebinur Lake Wetland, Northwest China. Peerj 6, 4703.
