Abstract Background An Ecological Site Group (ESG) is a classifi cation framework used in rangeland management to group Ecological Sites(ESs) that exhibit similar responses to land use, disturbance, conservation practices, and environmental change. Methods In this study, we applied supervised machine learning techniques to predict expert-assigned ESGs for rangelandpoints from the USDA Natural Resources Conservation Service (NRCS) National Resources Inventory (NRI) dataset.The analysis focused on Major Land Resource Areas (MLRAs) 65 and 69, located in Nebraska and Colorado, USA. Wedeveloped a predictive ESG classifi cation model using the XGBoost algorithm and compared it to the performancesof benchmark models: Random Forest and Decision Tree. Predictor variables included soil properties derived fromgSSURGO, climate variables from PRISM, and vegetation classes from the GAP land cover dataset. Model evaluationwas conducted using an 80/20 stratifi ed train–test split. Results The XGBoost model achieved overall test accuracies of 0.84 and 0.95 for MLRAs 65 and 69, respectively,outperforming the benchmark models. Point-based predictions were then scaled to the landscape level usingSSURGO soil map units and associated environmental attributes, generating ESG classifi cation maps covering 53,467km² in MLRA 65 and 30,710 km² in MLRA 69. Landscape-scale validation against expert-assigned NRI points showedstrong spatial agreement between predicted and observed ESGs in the map units, with 78% and 92% agreement inMLRAs 65 and 69, respectively. Conclusions These results demonstrate that ensemble boosting methods such as XGBoost can eff ectively translate point-basedESG observations into reproducible, landscape-scale ecological classifi cations. The resulting maps provide a scalable,data-driven framework to support conservation planning, targeted management, and assessment of rangelandresponses across heterogeneous environmental gradients.
Meles et al. (Wed,) studied this question.