Data-driven similar response units for agricultural technology targeting: An example from Ethiopia

Abstract Ethiopia has heterogeneous topographic, climatic and socio-ecological systems. Recommendations of agricultural inputs management practices based on coarse domains such as agro-ecological zones (AEZ) may not lead to accurate targeting, mainly due large intra-zone variations. The lack well-targeted recommendations contribute the underperformance promising technologies. Therefore, there is a need define units where similar environmental biophysical features prevail, which specific can be made for response (SRUs). We used unsupervised machine learning algorithms identify areas high similarity or homogeneous called ‘SRUs’ that guide targeting SRUs are landscape entities defined by integrating relevant covariates with intention responses. Using spatial data layers edaphic ecological variables delineation SRUs, we applied K- X-means clustering techniques generate various granular levels zonation similarity. results were validated through expert consultation comparison an existing operational AEZ map Ethiopia. also augmented validation heterogeneity using field-based crop fertiliser application experimental data. highlighted provide improved interventions. Comparison indicated same number captured better less within-cluster variability former. In addition, show lower optimal compared AEZs classes. This implies refined input technology targeting. work in this study developed framework users deploy fetch from cloud their interest.