Offline development and initialization of coupled models; a perspective from the 12km North American Land Data Assimilation System (NLDASE) project

Brian Cosgrove

NASA/GSFC

Abstract:

The 12km North American Land Data Assimilation System (NLDASE) project builds on the uncoupled land surface modeling approach pioneered in the NLDASE collaborative research effort to investigate two issues which confront land surface modelers: initialization of coupled NWP model land surface states, and offline development of coupled land surface models (LSMs). LDAS systems have sought to reduce the biases which can accumulate in the land surface stores of coupled modeling systems, and thus are attractive sources of initial land surface conditions for NWP models. These uncoupled modeling systems are also much less computationally expensive to run than their coupled counterparts, and so are used extensively for LSM model development. This talk will focus on the questions which arise from these uses--how does such LDAS-based initialization impact subsequent NWP model forecasts, and how closely can uncoupled modeling systems mimic coupled systems when used for LSM development activities? These issues were investigated using NCEP's coupled workstation Eta model and NASA's uncoupled Land Information System (LIS). The workstation Eta model was configured to mirror the operational Eta model, and both the workstation Eta and LIS systems used the same version of the Noah LSM, and were executed on the standard operational 12km Eta Arakawa E grid.

Initial results indicate that use of offline LIS land surface conditions to initialize the workstation Eta model results in large improvements in the RMSE and bias of 2m temperature and relative humidity forecasts over the CONUS out to 84 hours. The impacts on surface wind, precipitation, and upper air forecasts are more mixed. Examination of the coupled/uncoupled LSM development issue has also yielded several key conclusions. Chief among these is that uncoupled LSM output can significantly diverge from that of the parallel coupled system if the offline model surface exchange coefficient for heat (Ch), surface roughness, and forcing type/interval are not matched as closely as possible to that of the coupled system.