# A CNN-RNN Framework for Crop Yield Prediction

**Authors:** Saeed Khaki, Lizhi Wang, Sotirios V. Archontoulis  
**arXiv:** [1911.09045](https://arxiv.org/abs/1911.09045)  
**Date:** 2019-11-20  
**Summary page:** https://www.summarizepaper.com/en/paper/1911.09045v2/

## Key points

- Crop yield prediction is a complex task that depends on various factors
- The authors propose a deep learning framework called CNN-RNN for crop yield prediction
- The CNN-RNN model outperformed other popular methods in predicting corn and soybean yields
- The model captures time dependencies and generalizes predictions to untested environments
- The model can reveal how weather conditions, soil conditions, and management practices affect crop yields
- This study presents a promising approach for accurate crop yield prediction using deep learning techniques

## Summary

Crop yield prediction is a complex task that relies on various factors, including crop genotype, environmental conditions, and management practices. In this study, the authors propose a deep learning framework called CNN-RNN for predicting crop yields based on environmental data and management practices. The CNN-RNN model is compared with other popular methods such as random forest, deep fully-connected neural networks, and LASSO. The researchers applied the CNN-RNN model to forecast corn and soybean yields across the entire Corn Belt in the United States for the years 2016, 2017, and 2018 using historical data. The results showed that the CNN-RNN model outperformed all other methods tested, achieving a root-mean-square-error (RMSE) of only 9% and 8% of their respective average yields. The CNN-RNN model has three key features that make it potentially useful for other crop yield prediction studies. Firstly, it captures the time dependencies of environmental factors and genetic improvements in seeds over time without requiring genotype information. Secondly, it demonstrates the ability to generalize yield predictions to untested environments without a significant drop in accuracy. Lastly, coupled with the backpropagation method, the model can reveal how weather conditions, accuracy of weather predictions, soil conditions and management practices contribute to variations in crop yields. Overall, this study presents a promising approach for accurate crop yield prediction using deep learning techniques. The findings have implications for improving agricultural planning and decision-making processes by providing more reliable forecasts of crop yields based on environmental data and management practices.