Estimating Crop Phenology from Satellite Data using Machine Learning

Monitoring crop growth and development is important for agricultural management and policy interventions enhancing food security worldwide. Traditional methods of examining crop phenology (the timing of growth stages in plants) at large scales often are not sufficiently accurate to make informed decisions about crops. In this study, we proposed an approach that uses a satellite data fusion and Machine Learning (ML) modeling framework to predict crop phenology for eight major crops at field scales (30 meter) across all of Germany. The observed phenology used in this study is based on the citizen science data set of phenological observations covering all of Germany. By fusing satellite data from Landsat and Sentinel-2 images with radar data from Sentinel-1, our method effectively captures information from each publicly available Remote Sensing data source, resulting in precise estimations of phenology timing. Through a fusion analysis, results indicated that combining optical and radar images improves ML model ability to predict phenology with high accuracies with R² > 0.95 and a mean absolute error of less than 2 days for all the crops. Further analysis of uncertainties confirmed that adding radar data together with optical images improves the modeling reliability of satellite-based predictions of crop phenology. These improvements are expected to be useful for crop model calibrations, facilitate informed agricultural decisions, and contribute to sustainable food production to address the increasing global food demand.