Accurately determining crop growth progress and crop yields at field-scale can help farmers estimate their net profit and enable insurance companies to ascertain payouts, ultimately bolstering food security. At field scales, the trifecta of management practices, soil health and weather conditions combine to impact crop growth progress, and this progress can be monitored in-season using satellite data. Here, we use satellite derived metrics from both optical and radar satellites as well as machine learning models to model field-scale crop yields for over 3,000 soybean and wheat farms in Argentina. When we compare several machine learning models, our results show the promise of combining mixed effect models with non-parametric models in improving yield modeling capabilities. We also demonstrate the utility of specific satellite derived metrics and extracted features in improving model performance. Overall, our approach can explain greater than 80% of the variation in yields while remaining generalizable across crops and agro-ecological zones.

In order to address the issue of cloud cover limiting satellite data acquisition, we also use double-bounce parameters derived from Sentinel-1 integrated with Difference Vegetation Indices (DVI) derived from Landsat-8 for prediction of soybean yield at the field level over central Argentina. The Artificial Neural Network (ANN) model was trained using time series of Synthetic Aperture Radar (SAR) and optical features during the growing season. For comparison of SAR versus optical versus their integration for soybean yield prediction, the ANN model was trained and tested for three scenarios of SAR-only, optical-only and SAR-optical integration. Accuracies of yield prediction including correlation of determination (R2), root mean square error (RMSE) and mean absolute error (MAE) are 0.80, 0.589 t/ha, 0.445 t/ha for SAR-only; 0.65, 0.800 t/ha, 0.546 t/ha for optical-only; and 0.85, 0.554 t/ha, 0.389 t/ha for SAR-optical integration scenarios, respectively. These accuracies demonstrate high potential of SAR and SAR-optical integration for soybean yield prediction at field level.