An AI-based geospatial framework for farmland suitability assessment in the Omo sub-basin, Ethiopia.

This study presents an AI-assisted geospatial framework for farmland suitability assessment in the Omo sub-basin of Ethiopia, within the Sustainable Land Management Program (SLMP). Leveraging Google Earth Engine (GEE), twelve key biophysical variables were extracted from multi-temporal satellite data and processed using LLM-based code generation scripts, refinement and validation. Soil erosion was quantified using the Revised Universal Soil Loss Equation (RUSLE), which integrates topography, rainfall, vegetation cover, and soil datasets. Multi-criteria decision analysis was implemented using the Analytic Hierarchy Process (AHP), which achieved a consistency ratio of 8.1%, confirming the reliability of expert judgments. Among the variables, precipitation (24.1%), soil moisture (17.4%), and soil organic carbon (15.2%) emerged as the most influential determinants of agricultural potential. The resulting Farmland Suitability Index (FSI), computed via a Weighted Linear Combination, ranged from 0.16 to 0.697. Spatial classification revealed that 26.93% of the study area was of low suitability, 33.75% moderately suitable, 28.68% highly suitable, and only 10.65% very highly suitable. The hybrid methodology integrates AI-enabled automation with expert-in-the-loop refinement to facilitate more reproducibility. Findings underscore the pivotal role of water availability and soil health in shaping agroecological sustainability under predominantly rainfed systems. By providing a spatially explicit suitability map, this framework offers a practical decision support tool for policymakers, planners, and resource managers to optimise land allocation, guide climate-smart agricultural investments, and prioritise land restoration. More broadly, the study illustrates the transformative potential of AI-augmented Earth observation for precision land evaluation in data-scarce, climate-vulnerable regions.