Crop Yield Estimation and Interpretability With Gaussian Processes

This work introduces the use of Gaussian processes (GPs) for estimation and understanding crop development yield using multisensor satellite observations meteorological data. The proposed methodology combines synergistic information on canopy greenness, biomass, soil, plant water content from optical microwave sensors with atmospheric variables typically measured at stations. A composite covariance is used in GP model to account varying scales, nonstationary, nonlinear processes. reports noticeable gains terms accuracy respect other machine learning approaches corn, wheat, soybean yields consistently four years data across continental U.S. (CONUS). Sparse GPs allow obtaining fast compact solutions up a limit, where heavy sparsity compromises credibility confidence intervals. We further study interpretability by sensitivity analysis, which reveals that remote sensing parameters accounting soil moisture greenness mainly drive predictions. finally us identify climate extremes anomalies impacting productivity their associated drivers.