A data-mining approach for developing site-specific fertilizer response functions across the wheat-growing environments in Ethiopia

Summary The use of chemical fertilizers is among the main innovations brought by 1960s Green Revolution. In Ethiopia, fertilizer application during last four decades has led to significant yield gains, yet remains below its potential across much country. One challenges responsible for low response been ‘blanket’ recommendations, whereby no tailoring amount and frequency done based on soil requirements. As a result, applied ranges widely, can be either sub- or supra-optimal. There thus an increasing need site-specific recommendations which take into account site characteristics such as climate variables (temperature, rainfall, solar radiation); factors (soil organic carbon, moisture, pH, texture, cation exchange capacity, level macro- micronutrients); topographic position indices. This article reports data-mining approach we developed large dataset 6585 wheat ( Triticum aestivum ) field trials. includes detailed, biophysical create nutrient functions that guide optimal application. used machine-learning model (random forest) capture relationship between nutrients – nitrogen (N), phosphorous (P), potassium (K), sulfur (S) yield. explained about 83, 82, 47, 69% variances N, P, K, S omission, respectively, with consistent performance training testing datasets. Expectedly, N P omission data, most important explanatory are rate, followed carbon pH. For K S, however, climatic played role alongside rates. yield–fertilizer curves derived from our highly variable location location, they affected climatic, soil, conditions site. Importantly, using principal component analysis, showed shape result multiple environmental (including topography, climate) at play given site, rather than specific dominant one. research output expected respond national policy demands sound method identify rate increase economic returns investments utilization one step further.