Reconciling solar forecasts: Probabilistic forecast reconciliation in a nonparametric framework

No forecast is complete without understanding it probabilistically. Previously in “Reconciling solar forecasts: Geographical hierarchy” [Sol. Energy 146 (2017) 276–286], four different point forecast reconciliation techniques were demonstrated using simulated data from 318 photovoltaic systems in California. In this paper, I show how to extend those techniques to probabilistic solar forecasting. More specifically, probabilistic forecast reconciliation is performed in a nonparametric framework through block bootstrapping. As compared to the parametric framework, which requires the base forecasts to be characterized by elliptical distributions, the nonparametric framework is not limited by such assumptions. Probabilistic forecast reconciliation not only provides a description of forecast uncertainty, it could also issue optimal point forecasts based on a directive in the form of a statistical functional. In this regard, there is very little reason to favor point forecast reconciliation, or any point forecasting for that matter, in solar energy meteorology. And probabilistic forecast reconciliation, or more generally, probabilistic solar forecasting, should be made default.