Meteorological Conditions during Compound Wind and Precipitation Extremes in Coastal Southeast Asia

Compound precipitation and wind (CWP) extreme events can bring a destructive impact to cities located along coastal areas. Total seasonal occurrence of CWP extreme events reaches its highest number of more than sixty events per year in several coastal cities of Southeast Asia (SEA) with a peak occurrence during summer (June-September). This study investigates nine meteorological variables to identify linkages between atmospheric conditions and CWP extreme events using the Coordinated Regional Climate Downscaling Experiment for Southeast Asia (CORDEX-SEA) dataset. These nine variables are chosen due to their importance as trigger factors to convections and wind gusts, e.g. equivalent potential temperature to represent moist enthalpy and atmospheric static stability as affecting wind gusts. Twelve coastal cities across Vietnam (five cities), the Philippines (three cities), Thailand (two cities), Cambodia (one city), and Myanmar (one city) are grouped into four groups with similar climatological patterns of the nine meteorological variables during the historical summer period (1975-2005). All groups imply the importance of their regional underlying zonal and meridional wind anomaly, outgoing longwave radiation (OLR) anomaly, and low-level moisture flux conditions during CWP extreme events days. CWP days for Group 1 (Cebu, Davao, and Metro Manila) are associated with low-level moisture convergence, negative OLR anomaly, and stronger zonal wind anomaly that enhances the precipitation intensity and wind gusts. The presence of a low-pressure system over the northern part of Metro Manila may also influences the CWP extremes for Group 1. Similarly, as a group that is prone to tropical cyclones, Group 2 (Da Nang, Hanoi, and Hai Phong) are also affected by similar dominant factors as Group 1 with an additional factor from the meridional wind anomaly. Located in between the South China Sea and the Indian Ocean, Group 3 (Yangon, Bangkok, and Chon Buri) is dominantly affected by low-level moisture convergence, zonal wind anomaly, and warm-moist transports from the Indian Ocean. Group 4 (Can Tho, Ho Chi Minh City, and Phnom Penh) shows a similar metrological pattern as Group 3 without notable changes in warm-moist transports. The regional means of these nine meteorological variables are further applied to train a Support Vector Machine (SVM) with an additional unbalanced data handling stage prior to the model training process. The best-trained SVM model results in the highest f1 score of 0.78 and 0.76 on the model’s testing set for Group 3 and 4. Further evaluation of the trained SVM model shows that the model’s predictions on a testing dataset fall within the 95% confidence interval. The best model is next used to predict the occurrence of CWP extreme events in the summer of 2006-2023. This model results in a predictive f1 score of 0.61 for Group 3 and 0.54 for Group 4, corresponding to a total of 98% and 97% correctly predicted (true positive), respectively.