Probabilistic forecasting of the clear-sky index using Markov-chain mixture distribution and copula models
Joakim Munkhammar, Dennis van der Meer, Joakim Widén
Department of Engineering Sciences, Uppsala University, Uppsala, Sweden

Two probabilistic forecasting models for the clear-sky index, based on the Markov-chain mixture distribution (MCM) and copula clear-sky index generators, are presented and evaluated. In terms of performance, these models are compared with two benchmark models: a Quantile Regression (QR) model and the Persistence Ensemble (PerEn). The models are tested on clear-sky index data, which was estimated from irradiance data for two different climatic regions: Hawaii, USA and Norrk¨oping, Sweden. Results show that the copula model generally outperforms the PerEn, while MCM and QR are superior in all tested aspects. Comparing MCM and QR reliability, the QR is superior, while the MCM is superior in mean CRPS and skill score. The MCM and copula models are also evaluated as potential probabilistic forecasting benchmark candidates.