Integrating remote sensing and soil features for enhanced machine learning-based corn yield prediction in the Southern U.S.

Authors: SARKAR, SAYANTAN - Texas Agrilife Research; OSORIO LEYTON, JAVIER - Texas Agrilife Research; NOA-YARASCA, EFRAIN - Texas Agrilife Research; Smith, Douglas; MEKI, MANYOWA NORMAN - Texas Agrilife Research

Submitted to: Sensors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/15/2025
Publication Date: 1/18/2025
Citation: Sarkar, S., Osorio Leyton, J., Noa-Yarasca, E., Smith, D.R., Meki, M. 2025. Integrating remote sensing and soil features for enhanced machine learning-based corn yield prediction in the Southern U.S.. Sensors. https://doi.org/10.3390/s25020543.
DOI: https://doi.org/10.3390/s25020543

Interpretive Summary: Corn breeders as well as agricultural producers continue to seek fast, efficient ways to predict yield during the growing season. This study was conducted to determine the most effective ways to use machine learning to estimate corn yield during the growing season. High resolution aerial imagery was used and 55 variables were used in an attempt to predict corn yield during 11 discrete corn growth stages. More then 6,800 yield values were used to train the machine learning algorithms, which were validated with more than 1,700 data points. In order, reproductive stages 1 and 4 and vegetative stage 5 were the growth stages with the greatest ability for the machine learning algorithms to predict yield. This information may help producers and plant breeders more efficiently select corn varieties for desired traits.

Technical Abstract: Fast and efficient corn yield prediction is important for varietal selection by plant breeders and growers for management decision-making. Therefore, the objective of our study was to identify the most effective machine learning model for corn yield estimation, determine the best vegetation indices for predicting corn yield, and identify the most suitable corn growth stage for yield estimation using machine learning. Our study used high-resolution (6 cm) aerial multispectral imagery. Fifty-five different predictors including spectral bands, vegetation indices, color space indices, and wavelengths were derived from the multispectral data collected at the 11 growth stages of corn. Four regression and machine learning algorithms were evaluated for yield prediction, including Linear Regression, Random Forest, XGBoost, and Extra Tree Regressor. A total of 6865 yield values were used for model training and 1716 for validation. Results show that Vegetation Termination/Reproductive stage 1 to Reproductive stage 4 had the best yield prediction followed by Vegetative stage 4 and Vegetative stage 5 (Root Mean Square Error of 0.65 and 0.69, respectively for a mean yield of 10.19 Mg/Ha) using the Extra Tree Regressor. Best performing vegetation indices for yield prediction included many variables which reflect key physiological traits of corn. This information may help producers and plant breeders more efficiently select corn varieties for desired traits.