@CONFERENCE\{IMM2001-0682,
    author       = "I. Marti and T. S. Nielsen and H. Madsen and J. Navarro and C. Barquero",
    title        = "Prediction models in complex terrain.",
    year         = "2001",
    month        = "jul",
    keywords     = "Forecasting methods, Statistics, {HIRLAM,} Downscaling",
    edition      = "Peter Helm",
    pages        = "1248",
    booktitle    = "European Wind Energy Conference 2001, Copenhagen",
    volume       = "1",
    series       = "",
    editor       = "",
    publisher    = "WIP-Renewable Energies/ETA",
    organization = "",
    address      = "Sylvensteinstr.{2,} {D-}81369 Munchen/Piazza Savonarola, 10, {I-}50132 FLorence",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/682-full.html",
    abstract     = "The objective of the work is to investigatethe performance of {HIRLAM} in complex terrain when used as input to energy production forecasting models, and to develop a statistical model to adapt {HIRLAM} prediction to the wind farm. The features of the terrain, specially the topography, influence the performance of {HIRLAM} in particular with respect to wind predictions. To estimate the performance of the model two spatial resolutions (0,5 Deg. and 0.2 Deg.) and different sets of {HIRLAM} variables were used to predict wind speed and energy production. The predictions of energy production for the wind farms are calculated using on-line measurements of power production as well as {HIRLAM} predictions as input thus taking advantage of the auto-correlation, which is present in the power production for shorter pediction horizons. Statistical models are used to discribe the relationship between observed energy production and {HIRLAM} predictions. The statistical models belong to the class of conditional parametric models. The models are estimated using local polynomial regression, but the estimation method is here extended to be adaptive in order to allow for slow changes in the system e.g. caused by the annual variations of the climate. The results show that {HIRLAM} wind speed predictions can be improved by considering other {HIRLAM} variables that wind speed e.g. pressure gradients, and increasing the spatial resolution of the {HIRLAM} model.",
    isbn_issn    = "88-900442-9-2/3-936338-09-4"
}