Short-term wind energy prediction system by using mesoscale/microscale modelling with model output statistics on various terrain types

World's energy demands, along with developing technology and industry, are being headed towards sustainable energy sources such as hydroelectric, nuclear and wind energies. Despite the discontinuous behavior of wind field, which makes energy production much difficult, wind energy is the fastest growing energy sector amongst all other renewable energy sources since it is clean, natural, and free for the customers' point of view. Due to the harm that fossil fuels cause in the environment and human health, the area of use is getting narrower and this gap in the market is being replaced by wind energy. When looking at the energy consumption rate in the world, there has been an increase of about 6% in the last decade, and about 5% of this increment is in wind energy. In addition, depending on the consumption considering the electricity production, it is obvious that wind energy estimation and prediction systems play a vital role. That is why scientists have tried to develop methods for estimation of wind energy. The right guess brings more money, time and cost savings together with it. Thanks to more accurate predictions, the economic gains are increasing on the industry basis whereas costs are decreasing. Two types of prediction systems have been developed in order to predict wind energy, namely Long-term (multiple-days-ahead) and Short-term (day-ahead) systems. The short-term wind energy prediction system (WEPS) estimates the power of wind farm on the 48 or 72 hour. On the other hand, forecasts which are for active horizon of up to several days are provided by the long-term WEPS. The methodologies of Short-term WEPS around the world vary as a result of different terrain and prevailing meteorological conditions such like topography, surface roughness, ground cover, temperature inversions, foliage, gravity waves, and low-level jets. Turkey, especially Northwestern region, which is located in the transitional zone of the local winds and atmospheric systems, holds an important place in terms of wind energy potential. The facts that the three sides of Turkey are surrounded by the sea, the presence of an inland sea and the majority of the topography is being high plateaus are affecting the local winds. In addition, Turkey's wind potential is increased by the fact that its location is in the passage zone of different synoptic scale pressure patterns. In this thesis, a short-term WEPS is established for the studies of Anemos and Turkey at a resolution of a few kilometers and with 3 days' lead-time. The thesis also shows parallelism with the Research Project of the Scientific and Technological Research Council of Turkey (TUBITAK) conducted by Ş. Sibel Menteş, and also with the Anemos Project. In the Anemos study, data from five different wind farms, which are located in Denmark, Spain and Germany, are used. In the Turkey part of the study, data are collected from Soma (Manisa) and Çatalca (Istanbul) wind farms. The study involves nine stages: 1) Numerical Weather Prediction (NWP) data from European Centre for Medium-Range Weather Forecasts (ECMWF) downloaded for Turkey cases for one year (2010). For Anemos cases data downloaded from Dec 1998 to May 2003. 2) Three meso-scale numerical weather prediction models are used. Because models such as the High Resolution Limited Area Model (HIRLAM) could not produce accurate estimates for Turkey, Weather Research and Forecasting Model (WRF) is selected for Turkish cases. WRF has so many such as having three dimensional variation (3DVAR) data assimilation system, being non-hydrostatic model, having advances in physics, numeric and data assimilation, including atmospheric physical processes, and possessing topography with vertical coordinates and so on. In the Anemos cases, HIRLAM and Lokalmodell of Deutscher Wetterdienst (LM) models are used. HIRLAM is short-term multi-model limited area prediction system, which is used for the synoptic scale motions. Within Anemos project two different grid solution and lead-time data, namely Spain-HIRLAM and Denmark-HIRLAM's, are examined. LM is non-hydrostatic weather forecasting model. This model is preferred for non-hydrostatic scales such as short-term severe weather events. 3) The Prediktor is a tool to obtain short-term wind energy estimates. It is a system that uses NWP outputs, Model Output Statistics (MOS) and parameterizations of local effects by using Wind Atlas Application and Analysis Program (WAsP). 4) WAsP, which is a PC program for predicting wind climates, wind resources and power productions from wind turbines and wind farms, models micro-scale wind. 5) MOS minimizes the error between NWP and observation data. It is used for statistical improvement. Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are calculated for each MOS modules. Downhill simplex method is used for optimization. 6) The programming language is Jython. Jython is designed as a combination of JAVA and Python. Jython script in this study is written by Gregor Giebel, and is run for each seven wind farms. 7) Matrix Laboratory (MATLAB) program is used for plotting graphs of MOS output. 8) For Turkey cases Spectral analysis is applied. 9) In order to obtain reliable range of prediction, simulations and local observations are compared. To run the WRF model at high speed, whose results are required for the Prediktor, a workstation account is opened at National Center for High Performance Computing (NCHPC)-Istanbul Technical University (ITU) and necessary calculations are done on that machine.