Precipitation season forecasting is an important indicator for stabilizing water resources for all social activities, including economic, agricultural, and industrial activities. In recent years, in addition to forecasts based on general circulation models, forecasts based on deep learning models have been attracting attention for their ability to reduce the uncertainty of seasonal precipitation forecasting models. However, the forecasting capabilities of deep learning are black-boxed, posing a significant risk to the use of forecasts by policy makers. In this study, we evaluate variables that affect model performance for two deep learning-based precipitation season forecasting methods: CNN and Visual Transformer. By comparing the results of the CNN and Visual Transformer analyses, it is clear which of the two methods can reduce the risk to decision makers as an explanatory method for rainfall forecasting.
各種検索
サポート
ヘルプ