PreprintArticleVersion 1Preserved in Portico This version is not peer-reviewed
Generalizability of a Random Forest-Based Model of Maize Lodging Built with Satellite Image Data and Its Application to Monitoring and Evaluating Maize Lodging Risks
Version 1
: Received: 29 May 2023 / Approved: 2 June 2023 / Online: 2 June 2023 (04:08:42 CEST)
How to cite:
Guo, H.; Ming, B.; Nie, C.; Zhang, G.; Yang, H.; Gao, S.; Xue, B.; Xin, J.; Feng, D.; Jia, B.; Hou, P.; Xue, J.; Xie, R.; Wang, K.; Li, S. Generalizability of a Random Forest-Based Model of Maize Lodging Built with Satellite Image Data and Its Application to Monitoring and Evaluating Maize Lodging Risks. Preprints2023, 2023060123. https://doi.org/10.20944/preprints202306.0123.v1
Guo, H.; Ming, B.; Nie, C.; Zhang, G.; Yang, H.; Gao, S.; Xue, B.; Xin, J.; Feng, D.; Jia, B.; Hou, P.; Xue, J.; Xie, R.; Wang, K.; Li, S. Generalizability of a Random Forest-Based Model of Maize Lodging Built with Satellite Image Data and Its Application to Monitoring and Evaluating Maize Lodging Risks. Preprints 2023, 2023060123. https://doi.org/10.20944/preprints202306.0123.v1
Guo, H.; Ming, B.; Nie, C.; Zhang, G.; Yang, H.; Gao, S.; Xue, B.; Xin, J.; Feng, D.; Jia, B.; Hou, P.; Xue, J.; Xie, R.; Wang, K.; Li, S. Generalizability of a Random Forest-Based Model of Maize Lodging Built with Satellite Image Data and Its Application to Monitoring and Evaluating Maize Lodging Risks. Preprints2023, 2023060123. https://doi.org/10.20944/preprints202306.0123.v1
APA Style
Guo, H., Ming, B., Nie, C., Zhang, G., Yang, H., Gao, S., Xue, B., Xin, J., Feng, D., Jia, B., Hou, P., Xue, J., Xie, R., Wang, K., & Li, S. (2023). Generalizability of a Random Forest-Based Model of Maize Lodging Built with Satellite Image Data and Its Application to Monitoring and Evaluating Maize Lodging Risks. Preprints. https://doi.org/10.20944/preprints202306.0123.v1
Chicago/Turabian Style
Guo, H., Keru Wang and Shaokun Li. 2023 "Generalizability of a Random Forest-Based Model of Maize Lodging Built with Satellite Image Data and Its Application to Monitoring and Evaluating Maize Lodging Risks" Preprints. https://doi.org/10.20944/preprints202306.0123.v1
Abstract
Lodging is a common problem in maize production that seriously impacts yield, quality, and the capacity for mechanical harvesting. Evaluation of site-specific lodging risks requires establishment of a method for multi-year monitoring. In this study, spectral images collected by the Sentinel-2 satellite were processed to obtain three types of data: gray-level co-occurrence matrix texture (GLCM), vegetation indices (VIs), and spectral reflectance (SR). Lodging classification models were then established with Random Forest (RF) using each of the three data types separately (the GLCM, VI, and SR models) and in combination (SR+VI model, SR+GLCM model, VI+GLCM mod-el, and SR+VI+GLCM model). By gradually removing features with low importance scores from the SR+VI+GLCM model and analyzing the changes in the overall accuracy (OA), the optimal set of predictive variables was identified and used to construct the optimal model. A model built us-ing data from a single timepoint in 2021 was tested on data collected at a similar timepoint in 2019 and vice versa to assess interannual model generalizability. The results of this study demon-strate that for monitoring maize lodging, models constructed with a single feature type, the GLCM model had significantly lower accuracy compared to the VI and SR models. During certain growth stages, the model constructed with combined features had significantly higher accuracy in monitoring maize lodging compared to models constructed with a single feature. During the pro-cess of selecting the optimal predictive variables, it was found that the accuracy of the model did not increase as the number of predictive variables increased. The results show that the positive and negative validation models had an accuracy of 96.55% and 95.18%, with kappa values of 0.93 and 0.83, respectively. This indicates that the model has strong generality for the same repro-ductive stage between years. This study provides a detailed method for large-scale maize lodging monitoring, allowing for identification of optimal planting practices to reduce the probability of lodging and ultimately improving regional maize yield and quality.
Keywords
Sentinel-2 multispectral data; Maize lodging; Random Forest classification; Predictive variables; Model generalizability
Subject
Biology and Life Sciences, Agricultural Science and Agronomy
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.