This research aims to accurately predict crop yield and select optimal genotypes using deep learning. The researchers created two novel convolutional neural network (CNN) architectures to analyze a dataset of soybean hybrids. The first model combines CNN and fully-connected neural networks, while the second model adds an LSTM layer for weather variables. A Generalized Ensemble Method was used to optimize the models’ accuracy. The models showed superior performance in soybean yield prediction compared to other baseline models, with a lower RMSE, reduced MAE, and higher correlation coefficient. The proposed approach proves valuable for genotype selection in scenarios with limited testing years. The study also highlights the importance of location in crop yield predictions.
Publication date: 25 Sep 2023
Project Page: https://arxiv.org/abs/2309.13021v1
Paper: https://arxiv.org/pdf/2309.13021
