Forecasts of the Sea of fog by a Machine Learning of Gradient Boosting Algorithm

大橋唯太

In this study, forecasts of the sea of fog were executed by using a machine learning algorithm of the lightGBM for high seasons of the occurrence at the Miyoshi basin in 2018 to 2021. This study has the advantage that the forecasts require only the AMeDAS meteorological data before 21 local time (LT) of the previous day. The SHAP analyses revealed that a temperature drop from 18 LT to 21 LT near the ground surface was the most important factor for forecasting the sea of fog in the following morning, and the model performance was improved by adding the mountainous AMeDAS data to the basin bottom data. The accuracy rate including both the occurrence and nonoccurrence of the sea of fog was 76.7% in the four-year average. In addition, probability forecasts were introduced as a method serving decision-making by a user.

環境情報科学学術研究論文集36

環境情報科学センター

36

112

117