Predictability of hydrological variables and their seasonal prediction in the GFDL climate model

A set of 22-year (1979-2000) AMIP-type integrations with multiple ensembles were performed with one recent version of the GFDL climate model to investigate the predictability of hydrological variables, especially precipitation, at seasonal timescale. The variability of the annual and seasonal precipitation are estimated with these integrations. Then a suite of integrations, all with a similar setup, were esigned and carried out as to decompose the total variability into different components, which are related to variations of the essential elements of the climate system, namely, the ocean, the land surface and the atmosphere. It is found that ocean (sea surface temperature) plays the dominant role in determining the precipitation variability at seasonal to interannual timescales in the tropics and much of the mid-latitudes. The land surface, has some contribution to the precipitation variability, and it contributes to the potential predictability of the hydrological cycle over the continents. Therefore, we speculate that a soil moisture initialization will also help improve the seasonal prediction of the hydrological cycle over land. This speculation is tested with seasonal hindcasts for several selected years using the GFDL climate model. Different land surface initial conditions are used to make different sets of ensemble forecast. We show that realistic soil moisture initial conditions are able to improve the skill of seasonal forecast.