Scholarlink

<p>Extreme events involving compound wind and precipitation extremes (CWPE) can cause severe impacts in coastal cities located in northern Southeast Asia (SEA), with occurrences typically peaking during the summer (June to September). This study investigates the key atmospheric conditions during CWPE days for 12 SEA cities using data from the Coordinated Regional Climate Downscaling Experiment for Southeast Asia models under the RCP 4.5 scenario. The analysis is performed for four geographical domains, each consisting of three nearby cities sharing similar climatology during CWPE days. Domains 1 and 2 cover the Philippines and northern Vietnam cities that are prone to tropical cyclone activity, while a summer monsoon pattern is exhibited in Domains 3 and 4 that span the western and southern parts of continental SEA. Common atmospheric conditions in each domain indicate the potential role of local anomalous zonal wind and moisture convergence in CWPE occurrence. Using nine meteorological variables as inputs, Support Vector Machines (SVMs) trained under historical climate (1975–2005) for CWPE prediction have all four domains showing predictive f1 scores ranging 0.65 to 0.75 under present climate (2006–2023), indicating well-performing SVMs. Domain 1 has the least-performing SVM, possibly due to a smaller CWPE sample size. SVM model interpretation using SHapley Additive exPlanations (SHAP) analysis indicates that the two most dominant predictors are the horizontal wind anomaly and atmospheric river-related variables. With an increase in zonal wind anomalies under future climate (2026–2056) across all domains, except for Domain 2 that has different dominant predictors, CWPE likelihood occurrence is expected to increase in the future.</p> Graphical Abstract <p></p> <p>The graphical abstract illustrates the three components of this study on compound wind and precipitation extremes (CWPE). 12 large coastal cities (population &gt; 1&#xa0;million) in northern Southeast Asia as grouped into 4 geographical domains were analysed to identify common atmospheric conditions occurring over each domain during CWPE days. Data from the CORDEX-SEA models were used. Nine meteorological variables as representing these atmospheric conditions are used to develop a Support Vector Machine (SVM) for predictions on CWPE occurrence in each domain. The SVMs were trained and tested on the unbalanced historical climate (1975–2005) dataset with good f1 scores achieved. They further predicted CWPE days under present climate (2006–2023) with similar accuracy wherein all the data were available for testing. A SHAP analysis revealed the top common predictors identified as consistent with the underlying atmospheric conditions and the role of zonal wind anomaly and atmospheric-river variables. These top predictors point to changes in CWPE occurrence under future (2026–2056) climate with increases in three of the four domains studied.</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

Compound Wind and Precipitation Extremes – Atmospheric Conditions and SVM Predictions for Northern Southeast Asia Coastal Cities

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • D. V. Lestari,
  • W. Jian,
  • E. Y. LO

摘要

Extreme events involving compound wind and precipitation extremes (CWPE) can cause severe impacts in coastal cities located in northern Southeast Asia (SEA), with occurrences typically peaking during the summer (June to September). This study investigates the key atmospheric conditions during CWPE days for 12 SEA cities using data from the Coordinated Regional Climate Downscaling Experiment for Southeast Asia models under the RCP 4.5 scenario. The analysis is performed for four geographical domains, each consisting of three nearby cities sharing similar climatology during CWPE days. Domains 1 and 2 cover the Philippines and northern Vietnam cities that are prone to tropical cyclone activity, while a summer monsoon pattern is exhibited in Domains 3 and 4 that span the western and southern parts of continental SEA. Common atmospheric conditions in each domain indicate the potential role of local anomalous zonal wind and moisture convergence in CWPE occurrence. Using nine meteorological variables as inputs, Support Vector Machines (SVMs) trained under historical climate (1975–2005) for CWPE prediction have all four domains showing predictive f1 scores ranging 0.65 to 0.75 under present climate (2006–2023), indicating well-performing SVMs. Domain 1 has the least-performing SVM, possibly due to a smaller CWPE sample size. SVM model interpretation using SHapley Additive exPlanations (SHAP) analysis indicates that the two most dominant predictors are the horizontal wind anomaly and atmospheric river-related variables. With an increase in zonal wind anomalies under future climate (2026–2056) across all domains, except for Domain 2 that has different dominant predictors, CWPE likelihood occurrence is expected to increase in the future.

Graphical Abstract

The graphical abstract illustrates the three components of this study on compound wind and precipitation extremes (CWPE). 12 large coastal cities (population > 1 million) in northern Southeast Asia as grouped into 4 geographical domains were analysed to identify common atmospheric conditions occurring over each domain during CWPE days. Data from the CORDEX-SEA models were used. Nine meteorological variables as representing these atmospheric conditions are used to develop a Support Vector Machine (SVM) for predictions on CWPE occurrence in each domain. The SVMs were trained and tested on the unbalanced historical climate (1975–2005) dataset with good f1 scores achieved. They further predicted CWPE days under present climate (2006–2023) with similar accuracy wherein all the data were available for testing. A SHAP analysis revealed the top common predictors identified as consistent with the underlying atmospheric conditions and the role of zonal wind anomaly and atmospheric-river variables. These top predictors point to changes in CWPE occurrence under future (2026–2056) climate with increases in three of the four domains studied.