Application of a nonlinear Kalman filter for post-processing of solar irradiance prediction with a numerical weather model

Abstract A numerical weather prediction (NWP) model is operated for forecasting solar irradiance to predict the output of photovoltaic systems. The model tends to estimate the irradiance larger than the observation. Kalman filters are introduced for post-processing of the model to adjust the overestimation. A linear Kalman filter is introduced first. The filter improves the output of NWP. However, it over-adjusts the overestimations of the irradiance estimated NWP. A nonlinear Kalman filter is designed and introduced. The filter works appropriately, and when applied, the over-adjustment of the linear Kalman filter is resolved. A more detailed analysis shows that the filter improves the output of NWP, especially in high-irradiance range of more than 800 W m−2. In this range, the irradiance is large and hence the photovoltaic (PV) output is also large, and it is important to predict PV output. This confirms the validity of the Kalman filter for post-processing of NWP for forecasting solar irradiance.