[A-TT30] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT30-06] Estimation of past and future snow depth over Japan using machine learning techniques

★Invited Papers

*石崎 紀子¹、佐々木 秀孝¹、Damiani Alessandro¹ (1.国立環境研究所)

キーワード：積雪深、気候シナリオ、機械学習、水平分布

Snow is not only important as a water resource and for the water cycle, but also for alpine ecosystems and organisms that prefer cold regions. Therefore, it is important to understand the regional distribution of snow cover and future changes for its conservation. However, snow cover is redistributed by wind and affected by slight topographical undulations, making it difficult to observe spatial distribution of snow depth. Therefore, attempts have been made to estimate snow cover depth from other meteorological variables. In this study, machine learning was used to estimate snow cover depth using historical observations and topographical factors. Regional climate model NHRCM with a grid spacing of 20 km were used as predictors for machine learning, and the method was repeated 22 times, learning for 21 years and applying to one year. Machine learning estimates successfully reproduced both the maximum snow depth and the duration of snow cover compared to the Degree-day method using daily temperature and precipitation (Fig. 1). Sensitivities to the predictors are also investigated, and methods for application to past and future scenarios are discussed.