Irradiance Forecasts Verification Using Murphy–Winkler Factorization: Performance of Three American NWP Models in 2024

Irradiance forecasts issued by numerical weather prediction (NWP) models are essential to managing photovoltaic power over day-ahead horizons. Clearly then, it is important for the forecast users to understand how well the forecasts perform. Traditional verification relies heavily on a suite of error metrics, also known as performance indicators or accuracy measures, which have several known pitfalls that often lead to inconclusive or disagreeing outcomes. On this point, Murphy–Winkler factorization, which is a distribution-oriented verification framework, has been advocated. In a nutshell, this approach decomposes the joint distribution of forecast and observation into four distinct components, each representing one or more qualitative measures that are relevant to verification. In this work, Murphy– Winkler factorization is applied to 1–12-h forecasts produced by three American NWP models, namely, the North American Mesoscale model, the Rapid Refresh model, and the High-Resolution Rapid Refresh model, over 2024. It is confirmed that the three models possess superiority in different aspects. The importance of clarifying the directives before verification is thus emphasized.