Scholarlink

<p>Monthly rainfall forecasting in complex geophysical regions, such as the coastal areas of northern Iran, is essential for effective water resources management and natural hazard mitigation. This study develops an integrated forecasting framework that combines a tree algorithm with K-Means clustering for monthly rainfall forecasting in Mazandaran province, Iran. The analysis is based on a 20-year dataset (2005–2024) obtained from six synoptic meteorological stations. First, the forecasting results of three algorithms QUEST, CHAID, and CART on the rainfall amount for the following month were evaluated. According to the statistical indices, it was found that the CART model achieved a Correct of approximately 91% in the training part and 82% in the test part, which was selected as the best predictor among other algorithms. In the proposed framework, K-Means clustering was then applied to classify the rainfall values predicted by CART based on the most influential meteorological variables. The results show that monthly air temperature and wind speed consistently emerge as dominant predictors of rainfall variability. Distinct rainfall classes were associated with specific ranges of these variables, such that higher rainfall values were generally associated with lower temperature and wind speed categories, while lower rainfall levels were more associated with higher values of both variables. Overall, the integrated CART-K-Means framework enhances the interpretability of rainfall forecasts by revealing structured patterns between predicted rainfall classes and key climatic drivers. The proposed approach provides a transferable and data-driven tool that can support regional water resources planning and climate adaptation strategies in complex hydroclimatic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Integration of Decision Tree and K-Means for Enhanced Monthly Rainfall Forecasting and Variable Importance Ranking

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Seyed Hassan Mirhashemi,
  • Mohammad Reza Karimi

摘要

Monthly rainfall forecasting in complex geophysical regions, such as the coastal areas of northern Iran, is essential for effective water resources management and natural hazard mitigation. This study develops an integrated forecasting framework that combines a tree algorithm with K-Means clustering for monthly rainfall forecasting in Mazandaran province, Iran. The analysis is based on a 20-year dataset (2005–2024) obtained from six synoptic meteorological stations. First, the forecasting results of three algorithms QUEST, CHAID, and CART on the rainfall amount for the following month were evaluated. According to the statistical indices, it was found that the CART model achieved a Correct of approximately 91% in the training part and 82% in the test part, which was selected as the best predictor among other algorithms. In the proposed framework, K-Means clustering was then applied to classify the rainfall values predicted by CART based on the most influential meteorological variables. The results show that monthly air temperature and wind speed consistently emerge as dominant predictors of rainfall variability. Distinct rainfall classes were associated with specific ranges of these variables, such that higher rainfall values were generally associated with lower temperature and wind speed categories, while lower rainfall levels were more associated with higher values of both variables. Overall, the integrated CART-K-Means framework enhances the interpretability of rainfall forecasts by revealing structured patterns between predicted rainfall classes and key climatic drivers. The proposed approach provides a transferable and data-driven tool that can support regional water resources planning and climate adaptation strategies in complex hydroclimatic environments.