Classification of Heavy Rainfall Types and Detailed Characteristics Analysis in the Korean Peninsula Using Surface Observation Data.

The Korean Peninsula (KP), located on the eastern side of the East Asian continent, has experienced significant spatiotemporal variability in precipitation due to the influences of both the continent and the ocean, as well as its complex topography. The meteorological agency continuously adds AWS and ASOS stations each year to comprehensively understand the spatiotemporal variability in precipitation distribution. Recent research indicates that the frequency and intensity of concentrated heavy rainfall (Hev_Ran) events in the KP have been changing due to the impact of global warming. In this study, we utilized AWS and ASOS observational rainfall data from the past decade (2013-2022) to classify the types of Hev_Ran occurrences on the KP and analyze their detailed characteristics. After a simple quality control process to address missing and abnormal data, approximately 400 stations were selected from the monsoon (May-September), ensuring a missing rate of 15% or less for each month. The selected 400 stations were then investigated for the frequency of exceeding the Korea Meteorological Administration concentrated Hev_Ran rainfall warning and alert criteria on a monthly basis. The constructed dataset includes a total of 30 variables, considering time (3sets: 1/3/12 hours), rainfall intensity and frequency (2sets), and monsoon months (5sets: 5-6/7/8/9). These variables were normalized using the Robust transformation based on their deviation from the median. Additionally, due to the very low frequencies of exceeding alert criteria at most locations, the analysis was performed for the entire summer season rather than on a monthly basis, and for warnings, the frequencies in May and June were combined due to the lower occurrence. Furthermore, considering the influence of input variables on clustering results, the variable group with the highest Explained Cluster Variance (ECV) was selected for adjustment, resulting in a reduction to five input variables. Three commonly used clustering methods K-Means, Self Organizing Map, and Hierarchical Clustering were employed. The number of clusters was determined as six through ECV analysis. After clustering with these three methods, the results were compared, and since there was little difference between the clusters, the K-Means clustering result with the highest ECV was presented as the central outcome. Cluster-1, characterized by overall lower rainfall frequency, peaks in August, and is mainly located inland, excluding the around Seoul areas. Cluster-2 corresponds to the western of the inland region with higher rainfall frequencies in July-August. Cluster-3 covered the eastern and southern coastal areas, including parts of Jeju, experiencing increasing rainfall frequencies from May/June to September.  Cluster-4, located inland, demonstrates concentrated Hev_Ran, especially in August. In the southern coastal areas and some parts of Jeju, there is a moderate and relatively similar frequency of Hev_Ran occurrences on a monthly basis, with a peak observed in July (Cluster-5). Finally, Cluster-6, encompassing Jeju and Geoje, consistently displays high rainfall frequencies, especially in August and September, recording the highest number of Hev_Ran warnings. The presentation will focus on detailed characteristics, including daily variations, of concentrated Hev_Ran occurrences for each cluster.