Framework of Forecast Verification of Surface Solar Irradiance From a Numerical Weather Prediction Model Using Classification With a Gaussian Mixture Model

A clustering and classification method using a Gaussian mixture model (GMM) is used to summarize and simplify meteorological data from a numerical weather prediction (NWP) model. Each horizontal grid in the integration domain of the NWP model is characterized by a feature vector, which consists of a multivariable with multiple pressure levels. All horizontal grids at every forecast time are classified based on the GMM clustering. The classification results show that grids are clustered into air masses or disturbances with the same meteorological characteristics. This paper describes application of the proposed classification method as a framework to verify the forecast of surface solar irradiance from the NWP model. Satellite observation data are used as the reference so that verification can be performed over the integration domain of the NWP model for each air mass or disturbance that moves and changes shape over time. The mean square error (MSE) is decomposed into the square of the mean error and the MSE between variables centered on zero, the square root of which is called the centered root mean square error (CRMSE). The analyses are performed for forecast data over a 2 day forecast horizon. The change in mean error is not significant until the second day, whereas the CRMSE is maintained only during the first day. Each air mass has a different forecast error structure. The proposed framework clarifies the structure of the forecast error of the surface solar irradiance.