UMBC ebiquity
Artificial Intelligence for the Earth Systems
DUNE: A Machine Learning Deep UNET++ based ensemble Approach to Monthly, Seasonal and Annual Climate Forecasting
October 1, 2025
Capitalizing on the recent availability of ERA5 monthly averaged, long-term data records of mean atmospheric and climate fields derived from the high-resolution reanalysis, deep learning architectures provide an alternative to physics-based daily numerical weather predictions for subseasonal to seasonal (S2S) and annual forecasts. A novel deep U-Net++-based ensemble (DUNE) neural architecture is introduced, incorporating encoder–decoder structures with residual blocks. When initialized with data from the prior month, season, or year, this architecture delivers an artificial intelligence (AI)-based global forecasts for monthly, seasonal, and annual means of 2-m temperature (T2m) and sea surface temperature (SST). ERA5 monthly mean data are utilized as inputs for T2m over the land, SST over the oceans, and climatological monthly solar radiation at the top of the atmosphere, covering 40 years of data to train the model. Validation forecasts are conducted for another 2 years, followed by 5 years of forecast evaluations to capture natural annual variability. Rigorous testing was performed using a cross-validation approach with multiple validation and testing periods. The DUNE-trained inference weights enable forecasts to be generated within seconds. Performance metrics such as RMSE, anomaly correlation coefficient (ACC), and Heidke skill score (HSS) are analyzed globally and across specific regions. DUNE AI’s forecasts outperform persistence, climatology, and multiple linear regression across all domains. DUNE forecasts demonstrate comparable statistical accuracy to National Oceanic and Atmospheric Administration (NOAA)’s operational monthly outlooks for the United States but at significantly higher spatial resolutions. RMSE seasonal comparisons with NOAA’s North American Multimodel Ensemble (NMME) and European Centre for Medium-Range Weather Forecasts (ECMWF)’s fifth generation seasonal forecasts (SEAS5) show that DUNE is comparable or outperforms both in most seasons and captures major anomalies with finer spatial detail.
Significance Statement
This work introduces deep U-Net++-based ensemble (DUNE), a fast and high-resolution (0.25°) artificial intelligence (AI) model that forecasts global 2-m temperature (T2m) and sea surface temperature (SST). Accurate temperature predictions are crucial for agriculture, water management, energy planning, and drought monitoring. Evaluated over 2019–23, DUNE is comparable to or outperforms other AI models and delivers skill comparable to National Oceanic and Atmospheric Administration (NOAA)’s operational monthly forecasts. Relative to established seasonal systems [NOAA’s North American Multimodel Ensemble (NMME), European Centre for Medium-Range Weather Forecasts (ECMWF)’s fifth generation seasonal forecasts (SEAS5)], it is comparable or superior while running over 1000 times faster.
Article
AMS
AMS
4
4
Downloads: 139 downloads