predict wind conditions, thus wind power, only a week ahead dueto the difficulties of forecasting the state of the atmosphere beyond10 days. Nevertheless, any reference for the future state ofwind conditions is crucial since it will provide suppliers with vitalinformation with regard to the optimum purchasing of base load
. Moreover, it will contribute to the maintenance schedule forwind farms, the whole operation of which can be time consumingand expensive. Also, wind power forecasting will allow SOs toevaluate the potential wind energy production and schedule powersystems in a more effective way
. The use of wind indices hasbeen proposed in Ref.
since they provide suppliers/generatorsand SOs with estimates regarding the long-term production fromwind farms
. In one research project, three parsimonious methodsfor estimating monthly wind energy were compared. Fromthis comparison, it was shown that an hourly based approach anda linear regression method performed similarly. This conclusionestablished the linear regression method, despite its simplicity, foruse in estimating wind energy at monthly time scales
. In Ref.
, it was shown that it was feasible to predict to a reasonabledegree the monthly electricity production from wind farms in LaVenta, Oaxaca, Mexico. The authors compared two methods: autoregressiveintegrated moving average
and artificialneural network
models. The results indicated in this casethat the seasonal ARIMA
models performed betterthan the ANN models. In another study, a probabilistic model wasproposed for use in long-term energy resource planning
. Duringthe study, both autoregressive
and autoregressive movingaverage
models were tested for generating windspeed time series. The wind speed was converted into power, andby employing a spatial smoothing technique, the individual poweroutput from a wind turbine was extrapolated across geographicallydisperse wind farm sites. However, some other studies doubtthe efficacy of ARIMA models when compared with the ANNmodel. In Ref.
, feed forward and recurrent networks havebeen developed for predicting wind speed at a monthly time scaleand have been compared with ARIMA models. This comparisonshowed that ANN models performed better. Also, in Ref.
, theauthors developed an ANN model and compared it with an ARmodel. The results again indicated the superiority of ANN modelsfor predicting wind speed on a monthly basis. Similarly, Ref.
proposed two multilayered network architectures for predictingmonthly mean wind speed with satisfactory results, while in Ref.
a feed forward, back propagation network was developed forthe same purpose. It is clear from the above discussion that thereis not yet consensus as to which approach
nonlinear or stochasticmodel
produces the better result.This paper presents a statistical analysis of wind speeds over aperiod of 52 years for the UK. Based on autoregressive techniquesand by using the seasonal patterns identified in the time series,monthly wind speed forecasts have been generated. The modelsdeveloped in this study are compared, and it is found that the bestforecasts are made when the seasonal component is taken intoaccount.2 Data Collection and AnalysisObservations of several meteorological parameters, such aswind speed, have been retrieved from the Met Office IntegratedDatabase Archive System
and have been collectedfrom a wide range of locations across the UK, including synopticand climatological stations. The data used in this paper include1Corresponding author.