A research paper by the APEC Climate Center (APCC) team, led by Dr. Uran Chung, titled "Advancing Sub-seasonal to Seasonal (S2S) Multi-Model Ensemble Precipitation Prediction in East Asia: Deep Learning-Based Post-Processing for Improved Accuracy," has been published online in the internationally renowned journal Heliyon.
(Online Paper Link: https://www.sciencedirect.com/science/article/pii/S2405844024119640)

S2S Climate variability, which spans several weeks, is closely interconnected across vast regions covering thousands to tens of thousands of kilometers. It is influenced by various components of the Earth system, including the atmosphere, hydrosphere, cryosphere, lithosphere, and biosphere. While traditional climate prediction models have primarily focused on atmospheric simulations, recent advancements have expanded their scope to include the ocean, land surface, sea ice, and vegetation.

However, for forecasts extending beyond one to two weeks, the reliability of these models significantly decreases due to the rapid decline in the influence of initial conditions. This limitation presents challenges in making practical use of climate predictions, particularly for accurately forecasting precipitation amounts and frequency at the S2S  scale.

To address these challenges, the APCC research team has developed a deep learning-based post-processing technique to improve the reliability of two- to four-week S2S multi-model ensemble precipitation forecasts in East Asia. Deep learning-based post-processing involves training AI models on accumulated sub-seasonal forecast data to learn and anticipate long-term weather patterns, enabling sequential prediction of daily weather conditions.

The study’s findings allow for the evaluation of precipitation forecast accuracy within climate prediction models by comparing machine learning or deep learning-based post-processed forecasts. As a result, it is now possible to select region-specific climate prediction models with superior performance for East Asia, thereby contributing to the generation of more reliable climate forecasts.

Dr. Uran Chung stated, "This study enhances the reliability of precipitation forecasts, which are crucial for climate disaster management. By enabling the effective use of climate information in climate-sensitive sectors such as agriculture, it supports informed decision-making and helps mitigate human and property losses caused by climate-related disasters."