Spatial variability of soil properties in Kiltie watershed, Upper Blue Nile Basin, Northwestern, Ethiopia

Spatial analysis of soil properties is essential for making site-specific decisions on soil management and environmental modelling. Interpolation techniques are widely utilized in soil sciences for mapping processes to estimate soil property values at unsampled sites. The study aimed to explore the spatial variability of soil properties in Kiltie watershed, Ethiopia, for site-specific soil management. Specifically, this study evaluated and compared four spatial interpolation methods using leave one out cross validation (LOOCV) and backward elimination regression (BER) analysis methods, aimed at generating accurate soil property maps. For this, 19 georeferanced soil samples at a depth of 0–20 cm were collected using grid sampling techniques and analyzed in the laboratory. Descriptive and geospatial analysis were performed using R and ArcGIS software, respectively. Results show that soil properties varied widely with low (0.04 to 0.06) to moderate (0.12 to 0.35) coefficients of variation. The semivariogram for sand and moisture content (MC) was best fitted by the Gaussian model; silt, STI and pH were best fitted by Spherical model, whereas clay, BD and OC were best fitted by the Exponential model. The nugget/sill ratio indicated a weak spatial dependence for sand, silt, clay BD and MC (0.94 to 1); moderate spatial dependence for OC (0.28) and a strong spatial dependence for structural stability index (STI) (0.20) and pH (0.12). The study evaluated interpolation methods using LOOCV based on R² and RMSE alone did not consistently identify the optimal method therefore, backward elimination regression was incorporated to strengthen model selection.