Spatial Statistics 2015: Emerging Patterns A spatial model for the instantaneous estimation of wind power at a large number of unobserved sites

Abstract We propose a hierarchical Bayesian spatial model to obtain predictive densities of wind power at a set of un-monitored locations. The model consists of a mixture of Gamma density for the non-zero values and degenerated distributions at zero. The spatial dependence is described through a common Gaussian random field with a Matern covariance. For inference and prediction, we use the GMRF-SPDE approximation implemented in the R-INLA package. We showcase the method outlined here on data for 336 wind farms located in Denmark. We test the predictions derived from our method with model-diagnostic tools and show that it is calibrated. © 2015 The Authors. Published by Elsevier B.V. Peer-review under responsibility of Spatial Statistics 2015: Emerging Patterns committee. Keywords: Wind power prediction; Bayesian hierarchical models; integrated nested Laplace approximation 1. Introduction Worldwide, wind power is rapidly increasing its representation on newly installed electricity capacity. Denmark has the largest proportion of wind energy capacity compared to the size of the electricity consumption; it represented 32.7% of the total energy consumption in 2013 and increased to 39.1% in 2014. The heavy investments that have been placed in this area ask for advanced forecast methodologies to address issues related to intermittency and limited predictability of power generation. Increasing the accuracy of wind energy forecasts is not only important for efficient management of power systems