Short-term wind speed forecasting using regime-switching spatio-temporal covariance models
Date
2024-06-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Spatial Statistics
Abstract
This paper presents a methodology for forecasting short-term wind speed over a broad geographical area using regime-switching covariance models. We establish a theoretical framework for a regime-switching covariance model where the predominant (large-scale) wind speed and direction alter the space-time asymmetry in the covariances of the wind speed observations. Using a Markovian framework in time, we define regime-specific dynamics in such a way that the long-term statistics in each regime agree with a particular parametric covariance function. The predominant wind speed and direction are incorporated into the dynamics of the wind speed process via a new form of the Lagrangian covariance function. Analyzing the convergence of regime-specific long-term statistics of the process, we provide a rationale to adopt the weighted least squares method to estimate the parameters of the candidate covariance functions. Furthermore, by working with a discrete domain in space, we validate the positive definiteness constraint of the covariance functions numerically, which widens our choices for candidate functions since we are not restricted to those that are theoretically established to be positive definite. We apply our methodology to predict wind speed for up to 6 hours at 131 weather stations simultaneously across Alberta. Two publicly available datasets are utilized: the ERA5 reanalysis dataset is used for identifying atmospheric regimes, and wind speed data from Alberta weather stations is used for training and validation of the regime-switching covariance models. The results on the 2-year test dataset suggest that considering prevailing wind speed and direction is beneficial for areas with consistent prevailing winds. For areas with more complicated wind patterns, using a symmetric model with relaxed parameter constraints is preferable.
Description
This is a preprint article submitted to Spatial Statistics for peer review.
Keywords
wind speed modeling, regime-switching models, covariance models, atmospheric clustering