One of the key components of any modern numerical weather prediction system is the array of in-situ and remotely sensed measurements used to represent the initial state of the atmosphere for a given forecast cycle. Especially important are the observations driving land surface models (LSMs) which determine the temporal evolution of the components of the surface energy balance and in turn, the distribution of near-surface temperature, moisture, and static stability. The Oklahoma Automated Surface Layer Instrumentation System (OASIS) provides such a framework from which observations of boundary layer fluxes can be quantified, and offers the additional advantage of manual and automatic quality assurance procedures and finer spatial resolution than the satellite-derived data currently used by the NCEP in the data assimilation systems for their suite of numerical models.

This investigation used data from nine of the ten OASIS Supersites to drive a one-year simulation using the latest version of the 1-D uncoupled version of the NCEP Noah LSM, valid for 1 January 2003 through 31 December 2003. By generating the initial conditions with a five-year PILPS protocol spinup run over the 2001 calendar year as opposed to site-specific climatology, the model was able to achieve accurate soil moisture initial states for each site in advance of the production run which used data from calendar year 2002. Forecasts of downwelling longwave radiation, surface temperature, and the four components of the surface energy balance were completed and compared to data collected at each of the Supersites during the validation period. Additionally, observations from the warm season and cold season were considered independently to document seasonal variation in model error.

Initial results, while promising, reveal that the accuracy of the Noah LSM is strongly impacted by spurious or missing observations in the forcing dataset. Errors were maximized during the cold season, most notably in surface temperature, downwelling longwave radiation, and ground heat flux. The mean diurnal cycle of net radiation also suffered from a low bias and a phase error on the order of an hour. During the warm season, prediction of the surface energy balance was more accurate with correlation coefficients between observed and predicted values exceeding 0.5 for all four components. This acted to offset some of the error in predicted downwelling longwave radiation and eliminated the observed bias in surface temperature. Additionally, bias in the mean diurnal cycle of net radiation decreased dramatically and the phase error was eliminated entirely.