| Sumario: | This first pass implementation of the AgWise Fertilizer Recommendation Module for rice in Cambodia assembles fertilizer response trial dataset, calibrates both mechanistic and machine learning models, and generates preliminary maps of indigenous nutrient supply along with prototype fertilizer recommendations. A total of 1,909 georeferenced fertilizer response trials across 17 provinces were harmonized with soil, climate, and terrain covariates following the AgWise framework. Analyses reveal higher and more stable yields in the dry season than in the wet season; strong yield responses to nitrogen, a narrow -optimum for phosphorus, and potassium responses that peak at intermediate rates. Reverse and forward -QUEFTS, together with -gradient boosting and random forest models, reproduce observed yields with moderate to good predictive skill and provide -apparent soil N, P, and K supply estimates that can be mapped across the rice domain. These maps show uniformly low P, moderate but heterogeneous N, and high, highly variable K, providing clear evidence for a transition from current fertilizer recommendation toward site- and season--specific nutrient management.- These are explicitly treated as preliminary, pending refinement through incorporation of additional datasets and model re calibration.
|
|---|