ENVIRONMETRICS Environmetrics 2001; 12: 499±515 DOI: 10.1002/env.480) Wind speed prediction in a complex terrain D. G. T. Denison, 1,* P. Dellaportas 2 and B. K. Mallick 3 1 Dept. of Mathematics, Imperial College, 180 Queen's Gate, London, SW7 2BZ, UK 2 Dept. of Statistics, Athens University of Economics and Business, Patission 76, 104 34 Athens, Greece 3 Dept. of Statistics, Texas A&M University, College Station, TX 77843-3143, USA SUMMARY We present a novel method for analysing spatial data when response data is given at a ®nite number of locations and the aim is to predict the response at a new location, where only a short run of data is available. This is the type of dataset that is typically available when attempting to analyse wind velocity data. We demonstrate our method, and compare it to that introduced by Haslett and Raftery on a set of data collected from the island of Crete in Greece. Typically the distance between locations is used to de®ne the correlation matrix between responses at distinct locations even though this cannot always be justi®ed. The peculiarity presented in our data is that the sites are in a complex topography so differences in the local characteristics of the wind station the direction of the prevailing winds, and other unobserved covariates can all lead to unsuitable model ®tting. We use a nonlinear model to avoid these problems and demonstrate its predictive power in relation to the dataset under study. Copyright # 2001 John Wiley & Sons, Ltd. key word: Bayesian methods; Markov chain Monte Carlo; multivariate adaptive regression splines; nonlinear model 1. INTRODUCTION Recently, attention to the development of sophisticated methodologies which determine the wind energy resource has increased see, for example, Haslett & Raftery, 1989; Huang & Chalabi, 1995; Jamil, Parsa & Majidi, 1995). Interest usually lies in the understanding and prediction of wind turbines' performance which inevitably requires knowledge of the behaviour and structure of the wind itself. This knowledge is important, amongst other reasons, for the location design of wind parks as well as for the estimation of their potential to carry the imposed loads at an economical viable cost. This assessment requires estimates of wind speed and the problem becomes more serious in mountainous terrain sites where the most commonly used numerical models for wind ¯ow simulation fail to predict the wind regime. In such sites, where the wind ¯ow is characterised by serious terrain induced effects, the accurate estimation of wind potential is still a major research topic in wind engineering and has led to the development of various models and corresponding computer codes which vary in both complexity and accuracy. These models can be divided into numerical models and measure-correlate-predict MCP) techniques. An extensive review can be found in Derrick, Glinou, Ravey, Marti, Pahlke, Schwenk, Brand, Antoniou & Frandsen 1997). Although simple numerical Copyright # 2001 John Wiley & Sons, Ltd. Received 30 November 1999 Revised 18 November 2000  Correspondence to: D. G. T. Denison, Deptartment of Mathematics, Imperial College, 180 Queen's Gate, London, SW7 2BZ, UK. Tel.: 44 207 594 8586. E-mail: d.denison@ic.ac.uk