A weakly supervised framework for high-resolution crop yield forecasts

Predictor inputs and label data for crop yield forecasting are not always available at the same spatial resolution. We propose a deep learning framework that uses high resolution inputs (e.g. weather and soil) and low resolution labels (e.g. yield and crop area statistics) to produce crop yield forecasts for both spatial levels. The forecasting model is calibrated by weak supervision from low resolution crop area and yield statistics. We evaluated the framework by disaggregating regional yields in Europe from parent statistical regions to sub-regions for five countries (Germany, Spain, France, Hungary, Italy) and two crops (soft wheat and potatoes). Performance of the weakly supervised models was compared with linear trend models and Gradient-Boosted Decision Trees (GBDT). The weakly supervised (WS) models were statistically similar to GBDT models (p-values: 0.67 for soft wheat and 0.265 for potatoes), and generally better than the trend models. For soft wheat in France, combining forecasts from two versions of the WS model (with and without the yield trend of parent regions) provided good estimates of yields as well as spatial differences among sub-regions. Higher resolution crop yield forecasts are useful to policymakers and other stakeholders for local analysis and monitoring. Weakly supervised deep learning methods provide a way to produce such forecasts even in the absence of high resolution yield data.