Al Fahad, A., Molod, A., Menemenlis, D., Trayanov, A.I., and Zhang, H. (2025)
Presented at:
AGU Annual Meeting 2025Effective initialization of coupled models is a primary source of prediction skill for numerical weather prediction and subseasonal-to-seasonal (S2S) climate forecasting. However, initializing models with component states from independent reanalyses can introduce significant thermodynamic and dynamic imbalances, leading to an "initialization shock" that degrades forecast quality. This study investigates the impact of different initialization strategies on this shock and subsequent prediction skill using the GEOS-MITgcm. We present multi-ensemble experiments to contrast dynamically imbalanced and balanced initialization approaches. The "imbalanced" experiment is initialized directly using fields from MERRA-2 (atmosphere) and ECCOv4 (ocean). The "balanced" experiment uses a 3-month atmospheric "replay" to generate a dynamically balanced initial state prior to the forecast. Analysis of the dynamically imbalanced experiment reveals a severe initialization shock, driven by a large initial air-sea temperature difference. This disequilibrium forces a persistent, anomalously high heat flux, progressively destabilizing the lower troposphere. Convective Available Potential Energy (CAPE) builds steadily, culminating in a strong convective event marked by intense vertical ascent, heat flux, and precipitation. This event triggers significant atmospheric feedbacks, including the formation of optically thick clouds and precipitation-driven downdrafts that cool the near-surface layer. In contrast, the dynamically balanced initialization experiment substantially mitigates this spin-up, leading to more stable and physically consistent initial states.
