Analysis of the Fruit Drop Rate Caused by Typhoons Using Meteorological Data

Abstract

Typhoons, which are a common natural disaster in Korea, have seen a rapid increase in annual economic losses over the past decade. The objective of this study was to utilize historical crop insurance records to predict fruit drop rates caused by typhoons from 2016 to 2021. A total of 1848 datasets for the fruit drop rate were generated based on the impact of 24 typhoons on 77 cities with typhoon damage histories. Three different types of measures—the average value, the maximum or minimum value, and the value at a specific point during the typhoon—were applied to four meteorological factors, yielding a total of twelve variables used as model inputs. The predictive performance of the proposed models was compared using five evaluation metrics, and SHAP analysis was employed to assess the contribution of predictor variables to the model output. The most significant variable in explaining the vulnerability to typhoons was found to be the maximum wind speed. The categorical boosting model outperformed the other models in all evaluation metrics, except for the mean absolute error. The proposed model will assist in estimating the potential crop loss caused by typhoons, thereby aiding in the establishment of mitigation strategies for the main crop-producing areas.

Keywords

• crop insurance data
• fruit drop rate
• machine learning
• typhoon