Short Time Prediction of Wind Speeds from Local Measurements

Abstract

Summary. We compare different schemes for the short time (few seconds) predic-tion of local wind speeds in terms of their performance. Special emphasis is laid onthe prediction of turbulent gusts, where data driven continuous state Markov chainsturn out to be quite successful. A test of their performance by ROC statistics isdiscussed in detail. Taking into account correlations of several measurement posi-tions in space enhances the predictability. As a striking result, stronger wind gustspossess a better predictability. 16.1 Wind Speed Predictions Whereas the three dimensional velocity field of the air in the atmospherecan be supposed to be described by the deterministic Navier–Stokes equa-tions, possibly augmented by equations for the temperature and humidity(and hence density of the air) and with suitable boundary conditions, a localmeasurementyieldsdatawhichappeartoberandom.Infact,deterministicbe-haviour of the local velocities is very unprobable, since (a) the Navier–Stokesequations contain already non-local interactions through the self-generatedpressure field, and (b) the local wind speed changes due to the drift of theglobal wind field across the measurement position.Data analysis of wind speed recordings yields results which are fully con-sistent with stochastic data. We use time series recorded in Lammefjord bythe Riso research centre [1]. These data are obtained from cup anemometersmounted on measurement masts at heights of 10, 20, and 30m above ground,taken with 8Hz sampling rate. The data sets report the absolute values of thewind speeds in the