17:15 〜 18:45
[HCG22-P02] Deep Learning Statistical Downscaling of Compound Events drivers during the West Africa summer monsoon: Validation and exploring internal Sensitivity behavior of Model
キーワード：Deep Learning, Downscaling, Perfect prognosis approach, Compound Event
Deep learning has recently emerged in the climate science community as a new method for downscaling large-scale atmospheric fields to regional scales. Our research focuses on the benefits of using a coupled-monsoon trained deep learning architecture to downscale the drivers of Compound heatwave and drought (temperature and precipitation) during the passage of the West African monsoon flow. In this study, we compare deep learning frameworks with previous traditional downscaling models using the perfect prognosis approach, with a primary focus on exploring each model's transferability ability in warm and cold settings. Our findings show that deep learning coupled-monsoon architecture adds great benefit to high-elevation regions, particularly in the central Sahel region. Conclusively by carrying out a sensitivity test, we want to increase the accuracy of climate information for current and future compound event risks in the West African domain.