Justifying the application of deep learning on detecting weather fronts for coarse-scale General Circulation Models
Yiwen Mao, Tomohito J. Yamada
Received 18 November, 2024
Accepted 22 April, 2025
Published online 23 August, 2025
Yiwen Mao1), Tomohito J. Yamada2)
1) Graduate School of Engineering, Hokkaido University, Japan
2) Faculty of Engineering, Hokkaido University, Japan
A U-shaped convolutional network (U-NET) based deep learning model is developed to predict weather fronts over Japan and the surrounding seas in summer (June, July, and August) using upscaled gridded frontal datasets from observations between 1980 and 2020 and reanalysis data corresponding to the same period. We justify the applicability of the deep learning model for predicting weather fronts in summer based on outputs from coarse-scale General Circulation Models (i.e. GCMs) from two perspectives. First, the coarse resolution of GCMs (e.g. 1.25 degrees) can capture the general morphological features of weather fronts. Second, models trained in a colder climate are applied to predict fronts in a warmer climate with some decrease in predicted peak frequency of fronts, but the general features of the spatial distribution of fronts can be represented by the deep-learning model predictions. By applying the deep learning models to predict weather fronts of multiple ensemble members from past and future climate experiments of GCMs, we can see that the locations of peak frequency tend to move slightly more southwesterly in a slant zone within the belt region between 25°N to 40°N as climate warms in the future.
Copyright (c) 2025 The Author(s) CC-BY 4.0