Modelling selected soil chemical properties using TerrSet and Random Forest: a case of Hangadi watershed, Oromia, Ethiopia

Hangadi watershed hasn’t been research subject on Land Use/Land Cover (LULC) change despite having the only natural forest in mid-highland of Guji zone. Objective of this study was to examine LULC patterns, Map soil parameters using Random Forest algorithm. Changes in LULC over a 30-year period were evaluated (1988, 2008 and 2018), with 90.9% overall classification accuracy and 0.8827 kappa statistics indicated the consent of satellite imagery with ground truth. Random Forest in RStudio platform was applied to identify covariates and map TN, CEC, pH, AvK, AvP and EC. TerrSet 2020 and satellite images from 1988, 2008 and 2018 were used to track changes in LULC. Results showed that agroforestry increased by 41.2%. Forested land declined by 47.2%. 11 covariates were identified for unsampled locations prediction. In contrast to CEC (high = 33.70, low = 22.44), higher TN (high = 0.586, low = 0.215) and pH were recorded at the upper slope that could be attributed to household wastes, nitrogen-fixing trees, liming and minimal tillage, respectively. Forest land use had higher AvK (high = 300.7, low = 47) and EC (high = 0.4797, low = 0.1176) values. Agroforestry has relatively higher AvP (high = 9.03, low = 2.29) than forest that could be attributed to soil management and parent material. Application of organic fertilizer and composting is crucial because OC serves as a substitute for many minerals. A larger number of predictors ought to be included in soil properties prediction to address issue of ecosystem service trade-offs, ecological restoration/rehabilitation in the watershed.