Operational framework to predict field level crop biomass using remote sensing and data driven models

Remote Sensing (RS) based monitoring provides opportunities to acquire timely and reliable information on crop growth at diverse scales. Crop yield forecasts can help decision makers formulate policies maintaining national food reserves, sustaining supply chains attaining security. Inter field heterogeneity arising from varied management agricultural practices necessitates this forecasting be done field-level. Such level aid farmers identify gaps in water farm take corrective actions, if needed. So far, no scalable tools are available predict yields that leverage the availability of open-access high-resolution RS data. This research implements an operational framework biomass by evaluating different regression algorithms develop data driven models, leveraging historical near real time high resolution optical satellite Sentinel-2, radar Sentinel-1 evapotranspiration (ETa) Net Primary Production (NPP) Food Agriculture Organization’s (FAO) Water Productivity through Open-access Remotely sensed (WaPOR) platform. NPP is used as a proxy for production/yield. Five most commonly were tested build data-driven model sugarcane prediction Wonji-Shoa estate, located Awash Basin, Ethiopia. The models Multivariate Linear Regression (MLR), Stepwise (SMLR), Boosted Trees (BRT), Support Vector (SVR), Random Forest (RFR). results revealed seasonal predictions, linear (MLR SMLR) yielded more accurate predictions than non-linear machine learning (BRT, SVR RFR) tested. highest accuracy was achieved MLR which estimates with 89% could made 4 months prior harvest accuracies 79% up 200 days (approx. 6.5 months) before harvest. however, not provide (accuracies < 61%). Cumulative vegetation indices (VIs) found have higher predictive power standard VIs predicting future NPP. Enhanced Vegetation Index (EVI) variable power, followed VH polarized Synthetic Aperture Radar WaPOR ETa. study shows usefulness level. methods presented here translated into automated towards system.