Improving Coastal and Valley Fog Forecasts by Assimilating Boundary Layer Observations

Accurate prediction of the duration of fog in California's Central Valley and of incusions of marine stratus along California's coast are necessary for skillful prediction of aggregate solar generation in the State. In work sponsored by the California Energy Commission, we have performed a series of experiments with a forecast system consisting of the WRF weather prediction model, the WRFDA data assimilation 3DVAR scheme, and a set of surface and boundary-layer observations including RASS, Sodar, LIDAR, and machine-learning postprocessing to predict GHI observations at stations not used in the data assimilations process. We introduced a tuneable parameter into the cloud-top entrainment scheme of the YSU boundary-layer parameterization, allowing ensemble forecasts with ensemble members exhibiting a range of fog layer persistence behaviors. Forecasts were run for several months of 2018, including all seasons.

We present analysis of the impact of the assimilated data on the behavior of the model, and its dependence on initial conditions, and on the impact of assumptions made about the background model error covariance. The impact varies strongly from forecast to forecast, but is generally stronger for Central Valley fog prediction than for coastal fog prediction. Finally, we discuss the implications of our results for the cost-benefit analysis of adding in-situ observational capabilities in order to reduce forecast errors for electrical grid operators.