Skip to main content
ARS Home » Research » Publications at this Location » Publication #137150
Title: ARTIFICIAL NEURAL NETWORKS FOR CORN AND SOYBEAN YELD PREDICTION

Author
item Kaul, Monisha
item HILL, R - UNIVERSITY OF MD
item Walthall, Charles

Submitted to: Agronomy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/28/2004
Publication Date: N/A
Citation: N/A

Interpretive Summary: Tools to manage crop nutrient applications are needed. Artificial neural networks (ANN) were tested as a tool to predict corn and soybean growth using field-specific rainfall data and the USDA-NRCS Soil Rating for Plant Growth (SRPG) for Maryland Coastal Plain and Piedmont locations. ANN corn yield models for Maryland explained 77% of the variation of yield data versus 42% for linear regression. ANN soybean yield models for Maryland explained 81% of the yield variation versus 46% for linear regression. Although more time consuming to develop than linear regression models, ANN models proved to be superior for accurately predicting corn and soybean yields under typical Maryland climatic conditions. ANN models will provide producers with a means of optimizing nutrient applications, thus leading to reduced excess chemical loss to waterways.

Technical Abstract: Crop yield models can be used to quantify nutrient requirements for nutrient management. The objectives of this study were to investigate the effectiveness of artificial neural networks (ANN) for predicting Maryland corn and soybean yields under typical climatic conditions; compare the prediction capabilities of models at state, regional, and local scales; evaluate ANN model performance relative to variations of developmental parameters; and compare the effectiveness of multiple linear regression models to ANN models. Field-specific rainfall data and the USDA-NRCS Soil Rating for Plant Growth (SRPG) for multiple Coastal Plain and Piedmont locations were used. Effective corn and soybean yield predictions required SRPG and weekly rainfall means. Adjusting ANN learning rate and number of hidden nodes resulted in more accurately predicted yield with optimal learning rates between 0.7 and 1.0. Using smaller data sets resulted in fewer hidden nodes and lower learning rates during model optimization. ANN models consistently produced more accurate yield predictions than regression models particularly for smaller geographic areas. ANN corn yield models for Maryland resulted in r2's of 0.77 versus 0.42 for linear regression. ANN soybean yield models for Maryland resulted in r2's of 0.81 versus 0.46 for linear regression. Although more time consuming to develop than linear regression models, ANN models proved to be superior for accurately predicting corn and soybean yields under typical Maryland climatic conditions. ANN models will provide producers with a means of optimizing nutrient applications, thus leading to reduced excess chemical loss to waterways.