Land surface data assimilation into an atmospheric forecast model using an ensemble Kalman filter approach

Previous studies have shown that the state of the land surface measured in terms of variables such as soil temperature, soil moisture, fractional vegetation coverage, leaf area index, and surface heat fluxes strongly influences the values of near-surface atmospheric variables. The 2 m temperature fields and 10 m wind fields are both affected by changes in the land surface, as are precipitation and convective initiation. Improved initialization of land surface variables by numerical forecast models should greatly improve forecasts of near-surface atmospheric variables by these same models.

To test this hypothesis, we will assimilate land surface observations such as soil temperature, soil moisture, and sensible and latent heat fluxes into the NOAH land surface model (LSM) coupled to the MM5 model. An ensemble Kalman filter (EnKF) method of data assimilation was chosen to incorporate the land surface observations into MM5. The EnKF procedure involves generating an ensemble of forecasts and estimating the background error covariance from that ensemble. The EnKF approach has shown improvement over 3DVAR, especially when the differences between the background field and observations are flow-dependent.

Land surface observations will come from the OASIS (Oklahoma Atmospheric Surface-layer Instrumentation System) network and the Oklahoma Mesonet, mesoscale networks of observing stations throughout the state of Oklahoma operated by the Oklahoma Climatological Survey (OCS). In each test case, MM5 will be initialized using Eta analyses and control runs will be carried out using this basic initialization. Short-term (6-24 h) forecasts made with the land surface observations included will be compared to the control forecasts, and the impact of the observations will be assessed.