Digital mapping of soil-texture classes in Batifa, Kurdistan Region of Iraq, using machine-learning models

Soil texture is a key physical property that has a large effect on several other soil properties that are important for managing and planning agriculture. In the Batifa region (Iraq), there is currently no digital soil map on a moderate scale available. In this regard, remotely sensed data can aid in mapping soil-texture fractions. The purpose of this study was to evaluate the performance of three different machine-learning (ML) models (random forest [RF], support vector regression [SVR], and extreme gradient boosting [XGBoost]) to spatially estimate soil-texture classes using Landsat 8 and a digital elevation model (DEM). To this end, 96 soil samples with a surface layer 0–30-cm deep were collected to estimate their soil-texture fractions using 19 variables. These comprised Landsat Spectral Bands 1–7, 10, and 11, the Normalized Difference Soil Index, Enhanced Vegetation Index, Simple Soil Ratio Clay Index, Brightness Index, Grain Size Index, Normalized Difference Vegetation Index, Normalized Difference Sand Index, Soil Adjusted Vegetation Index, Landsat Bareness Index, and DEM. High coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) values were used to select the best ML model. The results of this study showed that the XGBoost model estimated the soil-texture fractions (clay: R² = 90%, silt: R² = 85%, and sand: R² = 91%) better than the RF (clay: R² = 86%, silt: R² = 66%, and sand: R² = 86%) and SVR (clay: R² = 72%, silt: R² = 56%, and sand: R² = 86%) models. The best estimators of soil-texture fractions were the GSI for clay, the DEM for silt, and Band 10 for sand, followed by other variables based on satellite data. In general, however, the DEM is considered a good estimator for all soil-texture fractions. These findings will help support techniques for managing soil in places where the surface soil has different textures.