Oral

IAMAS

JMCP18 - Sub-seasonal to Decadal Prediction (S2S-S2D)

Deep-learning-based prediction of Heatwave events over Asia linked to tropical and extratropical intraseasonal oscillations

Heatwave events profoundly impact human and ecosystem health, socio-economic stability, and the environment worldwide. While greenhouse gas warming increases the frequency and intensity of heatwave events, the timing and duration of those events are modulated mainly by internal variability, particularly on intraseasonal timescales. Accurately predicting extreme temperature events is crucial for enhancing preparedness and mitigating the associated risks. However, state-of-the-art numerical weather prediction models often struggle with significant biases, particularly in forecasting extreme events associated with intraseasonal oscillations. This study aims to improve extreme temperature events over Asia and explore their connection to tropical and extratropical intraseasonal oscillations using an advanced AI-based global weather model, KIST��s Atmospheric Rhythm with Integrated Neural Algorithms (KARINA). Trained on daily-mean atmospheric variables from the ECMWF ERA5 reanalysis dataset at a 250-km horizontal resolution from 1979 to 2015, KARINA provides global forecasts up to 30 days initiated every day from 2016 to 2022. The model demonstrated high predictive skills for extreme temperature events up to several weeks ahead over Asia, underscoring the importance of better capturing intraseasonal oscillation for predicting the extremes. These results highlight KARINA��s potential as a powerful AI-driven tool for improving temperature extreme predictions, ultimately enhancing preparedness and mitigation strategies across affected regions.