From expert knowledge to data-driven landscape classification: Mapping ecological site groups across climatic and edaphic gradients

Authors: Meles, Menberu; GUERTIN, D. PHILLIP - University Of Arizona; BURNS, I. SHEA - University Of Arizona; Goodrich, David; HERNANDEZ, MARIANO - University Of Arizona; METZ, LORETTA - Natural Resources Conservation Service (NRCS, USDA); SAEEDIMOGHADDAM, MAHMOUD - University Of California, Davis; Ponce Campos, Guillermo; NEARING, GREY - Google; Williams, Christopher; BARKER, STEVE - University Of Arizona; HOUDESHELL, CARRIE-ANN - Natural Resources Conservation Service (NRCS, USDA); ARCHER, STEVEN - University Of Arizona

Submitted to: Landscape Ecology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/10/2026
Publication Date: 6/3/2026
Citation: Meles, M.B., Guertin, D., Burns, I., Goodrich, D.C., Hernandez, M., Metz, L.J., Saeedimoghaddam, M., Ponce Campos, G.E., Nearing, G., Williams, C.J., Barker, S., Houdeshell, C., Archer, S.R. 2026. From expert knowledge to data-driven landscape classification: Mapping ecological site groups across climatic and edaphic gradients. Landscape Ecology. 41. Article 105. https://doi.org/10.1007/s10980-026-02385-0.
DOI: https://doi.org/10.1007/s10980-026-02385-0

Interpretive Summary: The study used a machine learning technique to predict Ecological Site Groups (ESGs) based on soil and climate data associated with USDA-Natural Resources Conservation Service (NRCS) National Resources Inventory (NRI) points in MLRA 65 and 69. We built a Random Forest model that predicted expert assigned ESGs at NRI points with high accuracy in both MLRAs. The model was then used to extend point-based predictions to the entire Major Land Resource Area (MLRA) using soil map polygons and their associated properties. This approach provides a valuable tool for studying Conservation Effects Assessment Projects on grazing lands (CEAP-GL) and improving conservation planning on a larger scale. However, further data collection is required to analyze prediction errors. The study demonstrates the potential of machine learning techniques for an attribute-based ecosystem classification of the components of complex, heterogeneous landscapes and for expanding point-based attributes to broader spatial scales.

Technical Abstract: An Ecological Site Group (ESG) is a concept applied in rangeland and ecosystem management to categorize and describe similar Ecological Sites (Ess) in sustainable land management, conservation, and adaptation to changing environmental conditions towards improved stewardship of natural resources. We used a machine learning technique to predict expert-assigned Ecological Site Groups (ESGs) at National Resources Inventory (NRI) points in Major Land Resource Area (MLRA; 2006, v4.2) 65 and 69, spanning the states of Nebraska and Colorado. The XGBoost algorithm in the Random Forest model was applied to create a predictive model based on soil and climate data associated with NRI points. The resulting model predicted the expert-assigned ESGs at NRI points with 96 and 99% accuracies in MLRAs 65 and 69, respectively. We then used the model to extend point-based predictions to the entire MLRA using SSURGO soil map units and associated attributes. This approach provides a valuable tool for studying Conservation Effects Assessment Projects on grazing lands (CEAP-GL), improving conservation planning, identifying areas that need specific management practices and extrapolating efforts to a broad landscape scale. However, scaling up of point based ESGs to landscapes require a comprehensive analysis of prediction errors due to the inherent variably across landscapes, a key next step which demands further data collection.