Solar Radiation Forecasts from Large Eddy Simulations and Observations using Ensemble Kalman Filtering

The increase in renewable energy production demands forecasting of wind and solar radiation due to their greater variability compared to non-renewable energy sources. The variability in solar energy is primarily caused by clouds. Large Eddy Simulation (LES) proves effective for high-resolution solar radiation prediction, capturing clouds like stratocumulus where large-scale models cannot. LES uncertainty in clouds primarily stems from initial conditions taken from these large-scale models. In this study, an ensemble Kalman filter assimilates observations into LES initial conditions. A large ensemble is created from a limited amount of LES runs by taking advantage of their internal variability. From this ensemble and measurements from the Cabauw measurement site, improved initial conditions are calculated for a range of stratocumulus cases. These cases are simulated without further interference. The method shows a 60% reduction in Root Mean Square Error (RMSE) for shortwave down solar radiation at the initial condition over the unfiltered initial condition. This improvement persists at 45% after 3 hours of simulation, showing the lasting impact of assimilated observational data on predictive accuracy. The decrease can be accounted to a combination of microphysical processes, energy fluxes from the lower boundary condition and the advective tendencies in the model. In future work, possible improvements to these processes will be identified and the method will be evaluated for other sites and cloud conditions.