Bayesian statistical analysis applied to solar radiation modelling

This paper proposes to use a rather new statistical approach in the realm of solar radiation modelling namely Bayesian inference. In this work, the theory of Bayesian inference will be presented at length. The Bayesian analysis consists in two levels. The first one is related to the parameter estimation while the second one concerns the model selection problem. As an illustration, a Bayesian parameter estimation method is used to derive a logistic hourly solar diffuse fraction model. A major difference between Bayesian and frequentist (or classical) methods is that the Bayesian inference offers a framework (through the use of prior information) to continuously update our posterior beliefs. In other words, all previous work is not wasted as the preceding model’s parameters can be used as prior information for the derivation of the parameters estimates of the next (new) model. For this particular application, it is also shown that the use of Bayesian methods instead of classical statistical techniques lead to a less biased model.