Evaluation of Machine Learning Models for Predicting Soil Texture Using Sentinel-1A SAR and Topographic Information

Applications such as agriculture, hydrology, and environmental management need the mapping of soil texture. In a research region near the Great Zab River in Iraq, this study assessed machine learning models for predicting important soil texture qualities using Sentinel-1A radar and digital elevation data. 75 soil samples in all were gathered, and their percentages of clay, silt, gravel, sand, and moisture content were determined. The models that were examined were artificial neural network (ANN), decision tree (DT), random forest (RF), support vector regression (SVR), and logistic regression (LR). Based on test data, results indicated that RF had the lowest root mean squared error (RMSE) in terms of forecasting clay (0.072 percent), specific gravity (0.011), gravel (10.736 percent), sand (10.213 percent), and silt (1.051 percent). Additionally, it had the greatest coefficient of determination (R2) values for clay (0.900), silt (0.883), sand (0.474), specific gravity (0.519), and gravel (0.568). When it came to predicting moisture content, ANN excelled (RMSE 2.515, R2 0.776). According to the RF feature significance scores, elevation was determined to be the most significant input variable. The study showed that precise maps of soil texture prediction may be obtained by utilizing RF machine learning in conjunction with Sentinel-1A data and digital elevation models. This provides an effective way for mapping soil properties in remote places with minimal effort.