A Hybrid Deep Learning Method for Crop Yield Prediction Using Comprehensive Multi-Source Data

Abstract

Properly estimating crop yield is of great benefit to harvest management and food security, however, the numerous biological, technological and environmental factors that have direct influence on farmlands make predicting crop yield a challenging task. Recent research efforts have shown that crop yield prediction can benefit greatly from the use of remotely sensed data in conjunction with machine learning (ML) or deep learning (DL) techniques. Nonetheless, little attention is given to hybrid DL methods capable of finding and learning both temporal and spatial features effectively in available data. In this study, we present a hybrid long short-term memory (LSTM) and convolutional neural network (CNN) method for crop yield prediction. The LSTM block is first used to extract the temporal structure from the input data while the CNN component then learns the spatial features from the resulting representation. A comprehensive dataset comprising average yield, environmental variables, soil, management practice and moderate-resolution imaging spectroradiometer (MODIS) data is used in this research. Furthermore, we present a structured procedure for processing the MODIS data and also explain how the predictions are made. Our proposed LSTM-CNN framework together with other popular methods such as the CNN, LSTM and CNN-LSTM is used to forecast corn and soybean yield across three mid-western states in the United States (US). Experimental results show the superior performance of our hybrid LSTM-CNN approach in comparison to the other popular methods.

Keywords: convolutional neural network, corn yield, crop yield prediction, environmental data, long short-term memory, MODIS data, soybean yield