Surface observations are the main conventional observations. They are very important for weather forecasting. However, in modern numerical weather prediction, the use of the surface data (e.g., 2-m temperature and 10-m winds) remains a challenge. This study examines the impact of two unique issues in the assimilation of near surface observations: variable terrain and surface diurnal variations.

A series of numerical experiments is conducted with the community mesoscale Weather Research and Forecasting (WRF) model and its 3-dimensional data assimilation (3DVar) system and an ensemble Kalman filter (EnKF) method from NCAR Data Assimilation Research Testbed (DART). It is found that there is a fundamental limitation in the assimilation of surface observations over complex terrain using 3DVar. EnKF method can partly overcome the problems from 3DVar as it produces flow-dependent background covariance, although some issues still remain. With a simulated frontal case over the intermountain western US, the influence of representative errors in both model terrain and surface observations on analysis and forecast is also evaluated. Results show that the quality of assimilation results is sensitive to these representative errors. Compared with data rejection, proper treatments to the representative errors could improve the assimilation results. Finally, as errors are detected from the WRF model produced surface diurnal variations over complex terrain, potential effects of the errors in model produced diurnal variations on the surface data assimilation are also examined. Details will be discussed in the presentation.