Probabilistic Estimation of Irradiance Components: A State Space Framework to Integrate Empirical Models, Forecasts, and Imperfect Sensor Measurements

Abstract

This work explores the general problem of combining an arbitrary set of noisy ground sensor measurements (horizontal, tilted, diffuse, direct, satellite-derived, etc. all for the same point) together with empirical and physical models (separation, transposition, and forecasting), to estimate a discrete set of irradiance components (in the simplest case direct and isotropic diffuse). We propose formulating the problem in a recursive Bayesian state estimation framework: calculating the joint probability distribution of a state vector of irradiance components, given the measurements, and considering distribution priors based on the models. Variations on a simple model are tested on data for two sites and compared against a deterministic benchmark.