MLP 기반 서울지역 WBGT 예측 성능 향상 평가 = Evaluation of an MLP-Based Approach for Enhanced WBGT Prediction in the Seoul Region

다국어 초록 (Multilingual Abstract)

The Wet-Bulb Globe Temperature (WBGT) is widely used to assess heat stress in military and industrial settings, however, its operational use is limited by the scarcity of globe temperature observations. The empirical WBGT model of the Korea Meteorological Administration (KMA2016), based on air temperature and humidity, provides a practical alternative but tends to underestimate high-WBGT conditions. This study developed a multilayer perceptron (MLP) model trained on the observed temperature and humidity, and its performance was evaluated against KMA2016. Using summer observations from Seoul’s Songwol-dong station during 2021-2024 for training and independent data from 11 stations in the summer of 2025 for validation, the MLP reduced the RMSE by approximately 10% for all-hour predictions and nearly 30% for daily maximum WBGT. The MLP also improved WBGT flag classification accuracy (89.4%) compared with KMA2016 (87.9%) and reduced the number of critical misclassification cases (±2 flag levels). When driven by short-term weather forecasts, the MLP maintained superior +24 h and +48 h WBGT forecast performance. These results indicate that the MLP provides a more accurate and operationally robust WBGT prediction framework for Seoul’s summer heat-stress environments.