Evaluating crop yield prediction models in illinois using aquacrop, semi-physical model and artificial neural networks

TL;DR

This study compares different models for predicting corn and soybean yields in Illinois, finding that artificial neural networks perform best.

Contribution

The study demonstrates that artificial neural networks outperform semi-physical and AquaCrop models in predicting crop yields in Illinois.

Findings

ANN models achieved the highest accuracy (R² = 0.96) in predicting soybean yields.

ANN models outperformed semi-physical and AquaCrop models for both corn and soybean yield prediction.

Meteorological data from 2000 to 2024 were used to train and evaluate the models.

Abstract

Crop yield is important for agricultural productivity and the country’s economy. While crop yield estimation is an essential aspect of modern agriculture, it continues to be one of the most challenging tasks to manage effectively. Corn and soybean are the important crops in Illinois, USA, considerably enhancing the region’s agricultural output and economy. The present study integrates semi-physical model, AquaCrop and Artificial Neural Network (ANN) Models for estimating corn and soybean yields. Data of different meteorological parameters including precipitation, maximum and minimum temperature, relative humidity, wind speed, solar radiation, photosynthetically active radiation and fraction of photosynthetically active radiation, land surface water index were collected for a period of 25 years from 2000 to 2024 from NASA POWER, USDA and NASS. The observed yield of soybean and corn was ranges from 2.49 to 4.37 ton/ha and 7.06 to 14.66 ton/ha. The predicted corn yield using the AquaCrop, semi-physical, and ANN models ranged from 7.60 to 14.42 ton/ha, 9.01 to 13.42 ton/ha, and 6.81 to 15.63 ton/ha, respectively. For soybean, the predicted yield ranged from 2.80 to 4.34 ton/ha, 2.92 to 3.84 ton/ha, and 2.45 to 4.43 ton/ha, respectively. The ANN model achieves the highest coefficient of determination (R² = 0.96) in predicting soybean yield, while the semi-physical model records the lowest R² value of 0.42, indicating the superior predictive capability of the ANN model. For both corn and soybean yields, the ANN model showed the highest prediction accuracy among the other models. Thus, the study underscores the significance of employing the ANN model for crop yield estimation, particularly in the regions that share similar physiographic and meteorological conditions with Illinois.

Figures