Crop Yield Prediction Using Deep Neural Networks.

Area of Science:

• Agricultural Science
• Computational Biology
• Genetics

Background:

• Crop yield is a complex trait influenced by genotype, environment, and their interactions.
• Accurate yield prediction necessitates understanding these relationships through comprehensive data and advanced algorithms.

Purpose of the Study:

• To develop a superior maize hybrid yield prediction model using deep neural networks (DNNs).
• To analyze the relative importance of genotype versus environmental factors in determining crop yield.

Main Methods:

• Utilized a deep neural network (DNN) approach on large-scale maize genotype and historical yield performance datasets (2008-2016).
• Employed state-of-the-art modeling techniques for prediction and performed feature selection on the trained DNN model.
• Compared DNN performance against Lasso, shallow neural networks (SNN), and regression trees (RT).

Main Results:

• The DNN model achieved superior prediction accuracy, with a root-mean-square-error (RMSE) of 12% of average yield using predicted weather data.
• With perfect weather data, RMSE decreased to 11% of average yield.
• Feature selection reduced input dimensionality without significant accuracy loss, and environmental factors showed a greater impact on yield than genotype.

Conclusions:

• Deep neural networks offer a powerful and accurate method for maize yield prediction.
• Environmental factors play a more critical role in influencing maize crop yield compared to genotype.
• The developed DNN model outperforms traditional machine learning methods for this task.