Keywords: deep learning, CNN, RNN, LSTM, clustering
Description:
Crop yield prediction is a challenging and intriguing topic in agricultural science. Recently
more researchers would like to find machine learning and deep learning techniques to make
contributions to crop yield predictions. In this project, we use different machine learning techniques, including benchmark models such as linear regressions and regression trees, to make predictions. We also use clustering methods (Kmeans) to distinguish multiple clusters for various environmental types according to their effect on yield. After that, following the instructions and work from Khaki et al. (2020), we reproduce the architecture of CNN-RNN framework to make predictions for soybean yield. This architecture outperforms than all benchmark models and captures the time dependency of environmental features, including their differences revealed by the clustering methods. It also includes all effects from weather, soil condition, management practice factors, and even genotype information even not given genotype data. However, we also find that this model does not capture effect from geological factors and point out the potential factors from geological information.
Report: Please mail to me (hw2894@columbia.edu) or through the mail symbol at the right bottom.