Improvement of land surface skin temperature in NCEP operational NWP models and its impact on satellite data assimilation

Land surface skin temperature (LST) is considered as one of key parameters in land-atmosphere interactions, and particularly important to remote sensing and data assimilation. NCEP operational Global Forecast System (GFS) forecast needs fields from the land surface such as albedo, upward long-wave flux and surface heat fluxes, which largely depend on LST predicted by the model. Satellite observed brightness temperature (Tb) in various spectral channels is assimilated through the JCSDA's Community Radiative Transfer Model (CRTM) on the NCEP Gridpoint Statistical Interpolation (GSI), and LST is a critical factor to determine Tb simulation for satellite surface sensitive channels.

The amount of satellite data assimilated over land in the GSI/CRTM is far less than over ocean. One of the chief reasons is the much larger error in GFS predicted LST as compared to the global analyses of sea surface temperature (SST) used by the GFS. The large bias in GFS predicted LST results in large errors in CRTM simulated satellite brightness temperatures over land, especially for surface sensitive satellite channels.

Investigation of GFS tests has revealed a major cause of cold daytime LST bias in the treatment for roughness length for heat (Zot) in the physics of surface turbulent heat transfer. Alternative formulation of thermal roughness length developed by Zeng et al. was tested. The new Zot formulation was found to very substantially reduce the large GFS daytime LST cold bias over desert and arid regions in the warm season. Moreover, LST beyond arid regions as well as other fields such as surface air temperature, whose biases were small in current operational GFS, were not much affected. The reduction of LST cold biases resulted in larger amounts of satellite data being accepted in the data assimilation over land in the GSI/CRTM.