Post-processing correction method for surface solar irradiance forecast data from the numerical weather model using geostationary satellite observation data

The numerical weather prediction (NWP) model still produces forecast errors. In this study, a correction method for the data from the NWP model is developed to obtain more skillful forecast data. The proposed method fuses information about surface global horizontal solar irradiance (GHI) from the NWP model and the satellite observations and gives corrected forecast data over the domain of the NWP model. There are three steps in the correction method algorithm. First, grid cells of the NWP model are grouped into clusters based on atmospheric conditions and regions. Second, cumulative distribution functions of the GHI from both datasets are inferred for each cluster separately. Last, correction functions are built using the statistical transformation. The correction effect is evaluated using satellite observation data as the reference. These analyses are performed using data for Japan. The correction function generally improves the forecast quality measured with the mean bias error (MBE), mean absolute error (MAE), and root mean square error (RMSE). When the forecast quality is measured with MBE, the improvement lasts for two forecast days, but when measured with MAE and RMSE, the improvement lasts for only one forecast day. The correction effects have seasonal and regional characteristics. The correction is effective from spring to autumn and the correction effect weakens in winter. Forecast quality at all ground observation stations in Japan, except for those in northern Japan, is improved owing to the correction. In contrast, the forecast quality at stations in northern Japan tends to be degraded.