Abstract:

Subseasonal-to-seasonal （S2S） prediction lies between short-range weather forecasting and long-term climate prediction， and its forecast skill depends to a large extent on the initialization quality of climate system model. Unlike short-range weather forecasts， whose signals mainly come from initial conditions， and seasonal-to-interannual predictions， which are influenced by both initial condition and external forcing， the predictability of S2S forecasts arises from the interactions between fast atmospheric processes in the troposphere and slower processes such as those in the ocean， land surface， and stratosphere. How to account for processes operating on different timescales during initialization and to enhance the internal consistency of the coupled system is the key to improving S2S prediction capability. This paper reviews the recent progress in initialization methods for S2S prediction and classifies them into uncoupled， weakly coupled， and strongly coupled initialization approaches according to the degree of coupling. Based on an analysis of 18 forecasting systems from major operational centers worldwide， weakly coupled data assimilation currently achieves good performance in terms of physical consistency， numerical stability， and forecast skill， and has become the mainstream approach for operational initialization. Furthermore， starting from the major sources of S2S predictability， including the Madden-Julian Oscillation （MJO）， the El Niño-Southern Oscillation （ENSO）， stratosphere-troposphere coupling， land-surface memory， and sea ice， the study discusses the roles of different assimilation strategies and their influence on subseasonal prediction of extreme events. The methodological challenges facing strongly coupled data assimilation and possible future directions are also discussed.

Key words: subseasonal-to-seasonal （s2s） prediction, coupled data assimilation, initialization shock, sources of predictability, artificial intelligence