In‐season crop phenology using remote sensing and model‐guided machine learning

Accurate in-season crop phenology estimation (CPE) using remote sensing (RS)-based machine-learning methods is challenging because of limited ground-truth data. In this study, a biophysical model was used to guide neural network (NN)-based, CPE. Using the Decision Support System for Agrotechnology Transfer (DSSAT), we conducted uncalibrated simulations corn (Zea mays L.) across Iowa and Illinois in U.S. Midwest with weather historical information planting harvest. We investigated guiding NN CPE method growth stage (GSTD) water stress factor (WSF) outputs from these simulations. Results show that guided NNs are able estimate onset progression phenological stages more accurately than an unguided baseline model-only method. GSTD guidance improved during seasons when progress deviated regional average temperature but detrimental delayed WSF harvest were by heavy rainfall performed less well grainfill mature stages. Neural network-based both provided most accurate estimates pre-emergence, emerged, silking, as lower RMSE median transition date reported three full-season studies. An RS estimating could link DSSAT current window improve upon results. This model-guided approach can be extended other crops regions unlock risk assessments directly linked phenology.