Using Data Assimilation of TRMM Passive Microwave Observations to Improve Land Surface Modeling

It is critical to accurately represent the water and energy balances that exist at the land surface for improved weather forecasts, flood prediction, and agricultural management. A land surface model (LSM) can simulate these balances, but these models have not always been validated and are partially limited by the quality of the forcing data they receive. One project, which seeks to address this issue, is the Land Data Assimilation Scheme (LDAS) project.

The LDAS project is a multi-institutional research effort centered on the development of a data assimilation scheme suitable for near-real time and retrospective modeling. Through the use LSMs as well as terrestrial and space-based observations, this data assimilation scheme will reduce errors in surface fluxes and storage quantities that are often present in LSM simulations. The LDAS currently operates at a 1/8th-degree resolution over the continental United States and makes use of EDAS forcing data, NESDIS/GOES radiation data and Stage IV precipitation data. An LDAS web site -which features a real-time image generator and project information - has been created and is located at http://ldas.gsfc.nasa.gov.

The LSM, in this case Mosaic, is run in the LDAS framework, and at each timestep the model's conditions are translated into surface microwave brightness temperatures. Tropical Rainfall Measurement Mission (TRMM) observations of the surface brightness temperatures are then used in a data assimilation technique to develop an adjustment factor. This factor is used to nudge the model's surface soil moisture conditions towards the true land surface state (as observed by TRMM). This type of data assimilation, in addition to altering the surface soil moisture, can affect the amount of moisture in other layers, as well as the latent and sensible heat fluxes.

Validation of the data assimilation scheme centers on use of observations from various soil moisture observation stations in the United States. These data come from monitoring stations in the Oklahoma Mesonet, the Atmospheric Radiation Measurement (ARM) network and the Natural Resources Conservation Service (NRCS) network.

The control run (no data assimilation) does an adequate job with some discrepancies between the mean level of soil moisture in the model and at the observation sites. However, there appear to be occurrences where model and observed soil moisture records respond differently to individual precipitation events. Data assimilation, using the TRMM surface observations, improves upon the mean state of soil moisture and reduces discrepancies between modeled and observed soil moisture.

Additional runs, using degraded precipitation records, were performed to gauge the potential effectiveness of data assimilation against poor precipitation data. Also, for a select number of sites, addition runs were performed using precipitation data observed at the soil moisture observation sites.

Preliminary results show that data assimilation is able to correct for some errors in the precipitation (or other fields). However, the level of correction is limited by the confidence level of the remotely sensed observations. As this confidence level decreases with increasing vegetation complexity there is a limit to the positive influence of data assimilation of satellite remote sensing of surface brightness temperatures.