Hail damage assessment in tomato crops using UAV-based vegetation indices and machine learning-based regression
Computers and Electronics in Agriculture, vol.250, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 250
- Publication Date: 2026
- Doi Number: 10.1016/j.compag.2026.111883
- Journal Name: Computers and Electronics in Agriculture
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, BIOSIS, Compendex, Environment Index, Geobase, INSPEC
- Keywords: Feature selection, Multispectral imaging, Precision agriculture, Random forest regression, Recursive feature elimination, Yield loss estimation
- Erciyes University Affiliated: Yes
Abstract
Hailstorms are extreme weather events that severely damage plants and reduce crop yields. Traditional hail simulations often use leaf defoliation to assess plant damage, but hail damage is typically classified into several categories. In this study, we simulated eight hail scenarios on tomato plants using water-absorbed and frozen pumice stones of various sizes and quantities to replicate real hail. These simulations were conducted at four developmental stages of the tomato plant. We calculated 62 vegetation indices before and after each simulation. The percentage of hail damage was then estimated using six machine learning regression models and five feature selection methods, including correlation, ReliefF, LASSO, Random Forest Feature Importance (RFFI), and Recursive Feature Elimination (RFE). The combination of RFE and the Random Forest model provided the most accurate estimates, with a Pearson correlation coefficient (R) of 0.9019, a mean absolute error (MAE) of 4.7287, and a root mean square error (RMSE) of 5.6991.