Abstract:
Sub-seasonal and seasonal prediction of summer precipitation and screen-level temperature over mid-latitude land must rely on predictability associated with the slowly varying land boundary conditions, in particular root zone soil moisture. Our goal is therefore to derive a realistic representation of global land surface conditions in order to initialize the General Circulation Model that produces the seasonal forecast. This may be partly achieved by forcing a land surface model with observed fields of precipitation and radiation (rather than with the typically poor surface meteorology produced by the GCM) up to the start time of the forecast, and transforming the resulting land surface initial conditions to be consistent with the GCM's climatology for forecast initialization. Such a system is now in place for the seasonal forecasts produced at the NASA Global Modeling and Assimilation Office (GMAO).

In addition to observations of land surface forcing fields (such as precipitation and radiation), there are limited satellite observations of land surface states, notably surface soil moisture. Such state observations can be retrieved from satellite measurements of microwave radiation emitted by the land surface. Land data assimilation techniques can be used to merge these satellite observations of land surface states with model-derived states, where the latter incorporate the information contained in the observed land surface forcings as well as our best knowledge of land surface dynamics as formulated in the land model. Estimates of land surface conditions produced by land data assimilation should, if derived properly, optimally combine all available information. In practice, strong biases and large uncertainties in models and observations pose severe challenges to our ability to derive such estimates. We demonstrate that despite the complications, estimates derived from data assimilation can improve our knowledge of soil moisture and have the potential to improve seasonal forecasts.