Corn Yield Prediction Using Remote Sensing Observations and Multi-source Unsupervised Domain Adaptation

Accurate corn yield prediction is of great importance to food security. With the recent advancement of satellite missions and artificial intelligence techniques, the supervised machine learning methods based on remote sensing observations have achieved impressive results in corn yield prediction. However, due to the domain shift between heterogeneous regions, supervised machine learning models tend to have poor spatial transferability. As a result, models trained with labeled data from one spatial region (i.e., source domain) often lose their validity when directly applied to another region (i.e., target domain). To address this issue, we proposed a Multi-source Maximum Predictor Discrepancy (MMPD) neural network, an unsupervised domain adaptation approach for corn yield prediction at the county level. The novelties of MMPD include 1) adopting the strategy of multi-source domain adaptation to avoid negative interference between labeled samples from different sources; 2) proposing to maximize the discrepancy between two source-specific yield predictors' outputs to detect unlabeled target samples that are far from the support of the source domain. A feature extractor is trained to align source and target domains by minimizing the predictor discrepancy. Case studies in the U.S. corn belt and Argentina demonstrated that the proposed model has effectively reduced domain shifts and outperformed several state-of-art deep learning and domain adaptation methods.