Land surface modeling, critical for accurate weather and climate forecasts, incorporates elements from both the water and energy balances at the earth's surface. Unfortunately, land surface modeling is limited by potential errors in the meteorological forcing variables, parameter values characterizing the land surface, and the land surface model (LSM) parameterization. However, use of data assimilation can incorporate additional observations to improve the value of the land surface prediction.

One such observation that can be used is from the Tropical Rainfall Measuring Mission (TRMM). The 10 GHz Brightness Temperatures, from TRMM, which have a spatial resolution of approximately 45 kilometers, contain information about the moisture in the top centimeters of the land surface. Although this frequency is not optimum for soil moisture measurement, it has the advantage of being currently available, as well as having a store of multiple years of data (since Dec. 1997).

In order to best use the TRMM data, we have taken multiple approaches to how to perform the data assimilation. The first approach is to use the Kalman filter to incorporate soil moisture values into the LSM. The second approach is to transform the land surface description contained in the LSM into a brightness temperature signal and then compare that value with an observed brightness temperature value. Each approach's advantages and disadvantages will be discussed.

Due to the complexities involved when combining real observations with real world conditions we have first performed synthetic fraternal twin experiments. In this manner, we use values from one LSM, such as Mosaic, to produce our observations, either of soil moisture or brightness temperature, and then we incorporate these into a running simulation using a second LSM, such as the Common Land Model (CLM). This permits knowledge about what factors were critical in generating the value of the simulated remote sensing observation.

The second phase of our research is using actual observations from TRMM. Again, we have use both approaches, i.e. doing the data assimilation in soil moisture space or in brightness temperature space. Again, both seem to have advantages and disadvantages which will be discussed, however for certain conditions we are able to identify improvement in the land surface predictions because of the data assimilation of the TRMM data.