Data Accounting in Agriculture: Satellite, Station & IoT Fusion for Reliable Forecasts

A structured assessment and weighting of heterogeneous sensing streams of data accounting enable more reliable agricultural forecasts by exposing and correcting biases, gaps and redundancy across satellite, station and IoT inputs. This paper presents a reproducible pipeline that integrates an explicit data-accounting layer with a physically consistent fusion backbone (ESM–GAMS) and a learned ensemble-weight (EW) operator implemented as an LSTM. This study defines a Data-Value Index (DVI) and a Data-Value Ratio (DVR) to quantify per-source contribution and used those diagnostics inside both the fusion step and a crop-model ensemble merge. Weather-map forecasts were generated with a physics-led downscaling ensemble and a convolutional residual corrector. Crop yields were simulated with a process model and weighted by LSTM-derived EWs that depend on historical and recent data-quality diagnostics. Experiments over three southern India districts for 2010–2022 showed that data accounting reduced precipitation RMSE by 8–16% and decreased crop-model yield bias by 6– 12% relative to naïve fusion.

Hali tarjima qilinmagan