Spatio-temporal Kriging of Lower Caribbean Wind Data

Abstract

Planning of a wind farm location requires significant data. However, wind speed data sets in the lower Caribbean are usually incomplete. This paper considers imputation by spatio-temporal kriging using data from neighbouring locations. Temporal basis functions with spatial covariates are used to model diurnal wind speed cyclicity. The residual set of our spatio-temporal model is modelled as a Gaussian spatial random field. Fitted models may be used for spatial prediction as well as imputation. Examples of predictions are illustrated using two months of hourly data from eight Caribbean locations with prediction accuracy being assessed by cross validation and residuals.