Ensemble data assimilation to quantify land-atmosphere coupling errors in numerical weather prediction model formulation

Experiments to quantify daytime land-atmosphere coupling errors in similarity theory-based parameterizations typical of mesoscale numerical weather prediction models are constructed and analyzed. Ensemble data assimilation provides a context for simple and unambiguous interpretation of model error by providing a distribution of initial model states that is consistent with chaotic prediction errors and observation uncertainty. When the ensemble accurately predicts future observations, so that analysis increments from the assimilation are small, over-fitting the observations is ruled out. Systematic increments, computed as the time and ensemble-mean increments, and can be interpreted as resulting from errors in model formulation. The structure of the increments offers clues for identifying the source of the errors.

Here the atmospheric surface layer is constrained with ensemble data assimilation of shelter and anemometer-height observations (2-m temperature, 2-m water vapor mixing ratio, and 10-m winds) in a single column model implementation of the Weather Research and Forecast (WRF) mesoscale numerical weather prediction model. Results show that 30-minute predictions of those observations are accurate, and that increments to lowest-layer model temperature, wind, and humidity from the assimilation are systematically small. Errors in the surface-layer profile are then interpreted in the context of surface flux and flux-profile errors.