Begin Main Content

Estimates of land surface conditions (including soil moisture, snow, and skin temperature) are available (i) from observational estimates, as retrieved, for example, from NASA satellites, and (ii) from land model integrations. The latter are forced with surface meteorological data (including precipitation and radiation) from observations, from atmospheric model integrations, or from an atmospheric data assimilation system. Both sources of land surface information. observations and model estimates. are subject to errors. Satellite retrievals are typically sparse in time and space and are available only under certain conditions. Considerable errors, including drifts and biases, are present in modeled land surface fields because land-atmosphere interactions are not perfectly parameterized in the model and because there are always errors in the meteorological forcing fields, regardless of their origin. A land data assimilation system produces optimal estimates of land surface conditions by correcting the modeled land surface fields toward the observational estimates, with the degree of correction determined by the levels of error associated with each.

Two assimilation strategies are currently pursued at the GMAO. The first strategy, "off-line" assimilation, involves driving a land surface model off-line (i.e., not coupled to an atmospheric model) with meteorological forcing data while assimilating scattered observations of land surface fields. In such a system, the land assimilation has no impact . no feedback . on the land surface forcing and the atmospheric state. For seasonal forecast initialization, off-line land assimilation is appropriate because atmospheric chaos drowns out the information from the atmospheric initialization after a few weeks.

Atmospheric and land initialization are, however, about equally important for sub-seasonal forecasts (2 weeks to 2 months lead). Atmospheric initialization is, of course, dominant for weather prediction. Therefore, the initial land and atmospheric states should be fully consistent with each other. Such consistency can be achieved by performing the land assimilation in conjunction with the atmospheric analysis during integrations of the coupled land-atmosphere system. In such "coupled" assimilation the assimilated land data then help improve the estimation of atmospheric states, and the assimilated atmospheric data help improve the estimation of land states. The final product is a joint set of land and atmospheric states that are fully consistent with each other . an optimal set of joint states for the initialization of the full prediction system. Such a fully coupled system is of course much more complex and computationally demanding than an off-line assimilation system.

The core of the GMAO land data assimilation system is based on the Ensemble Kalman filter (EnKF), a Monte-Carlo approach to the nonlinear filtering problem. The EnKF is based on the approximation of the conditional probability densities of interest by a finite number of randomly generated model trajectories. It is particularly well suited to the nonlinear and intermittent character of land surface processes (Reichle et al., 2002a) and was recently implemented at the GMAO for off-line land assimilation (Reichle et al., 2002b; Reichle and Koster, 2003).

Soil Moisture Assimilation combines meteorological forcing data with satellite soil moisture measurements to produce optimal soil moisture states.

Skin Temperature Assimilation combines meteorological forcing data with satellite skin temperature measurements to product optimal skin temperature states.

Snow Assimilation combines meteorological forcing data with satellite snow measurements to product optimal snow states.

References

Page author: Rolf Reichle
Email: reichle@janus.gsfc.nasa.gov

Content last updated: 28 January 2005