학술행사
- 발표자 송창훈 박사(서울대학교)
- 개최일시 2025-05-28 14:00-16:00
- 장소 국가수리과학연구소 산업수학혁신센터(판교)
-
일시: 2025년 5월 28일(수), 14:00-16:00/
-
장소: 판교 테크노밸리 산업수학혁신센터 세미나실/
-
경기 성남시 수정구 대왕판교로 815, 기업지원허브 231호 국가수리과학연구소(무료주차는 2시간 지원)/
-
발표자: 송창훈 박사(서울대학교)/
-
주제: Radar-Based Precipitation Nowcasting: From Existing Approaches to Practical Challenges and Lessons Learned/
Abstract: Accurate weather forecasting is essential for mitigating the impact of severe weather events. In this talk, we begin with a brief overview of representative deep learning models and publicly available datasets for weather prediction. We then shift focus to radar-based precipitation nowcasting and examine recent advances in this domain. The latter half of the presentation centers on our experience working with domestic radar data, where we share the practical challenges, failures, and insights gained during model development. These include issues such as data imbalance, capturing localized rainfall structures, and stabilizing training for high-resolution predictions. Finally, we introduce our Swin Transformer-based model, SwinPreCast, and present its results on both public and domestic datasets. We conclude with a discussion on future directions, including generalization across regions and integration with physics-based forecasting models.
-
일시: 2025년 5월 28일(수), 14:00-16:00/
-
장소: 판교 테크노밸리 산업수학혁신센터 세미나실/
-
경기 성남시 수정구 대왕판교로 815, 기업지원허브 231호 국가수리과학연구소(무료주차는 2시간 지원)/
-
발표자: 송창훈 박사(서울대학교)/
-
주제: Radar-Based Precipitation Nowcasting: From Existing Approaches to Practical Challenges and Lessons Learned/
Abstract: Accurate weather forecasting is essential for mitigating the impact of severe weather events. In this talk, we begin with a brief overview of representative deep learning models and publicly available datasets for weather prediction. We then shift focus to radar-based precipitation nowcasting and examine recent advances in this domain. The latter half of the presentation centers on our experience working with domestic radar data, where we share the practical challenges, failures, and insights gained during model development. These include issues such as data imbalance, capturing localized rainfall structures, and stabilizing training for high-resolution predictions. Finally, we introduce our Swin Transformer-based model, SwinPreCast, and present its results on both public and domestic datasets. We conclude with a discussion on future directions, including generalization across regions and integration with physics-based forecasting models.
