Under CopyrightGeiger, DavidBirge, FerhudaFerhudaBirge2025-05-202025-05-202025https://doi.org/10.24406/publica-4692https://publica.fraunhofer.de/handle/publica/48774510.24406/publica-4692This study evaluates the performance of three reanalysis models-ERA5, CERRA, and NEWA-in predicting wind power generation from 2016 to 2021, using actual energy productiondata from the SMARD dataset. The study focuses on both onshore and offshorewind power and employs statistical methods, including Bias, Pearson Correlation, MAE,and RMSE, to assess the accuracy and reliability of the reanalysis models. Wind turbinedata from the MaStR dataset, combined with wind speed data retrieved via FraunhoferAPI, were used to estimate power generation through the power curve model. The resultsshow that ERA5 consistently provides the most accurate and reliable predictions for bothonshore and offshore wind power, with strong Pearson correlations (above 0.98) and lowRMSE values (below 1.6 for onshore and 0.7 for offshore). CERRA performs well, particularlyin offshore generation, but shows more variability in onshore predictions. NEWAdemonstrates weaker performance, especially for onshore wind, with higher biases andlarger errors, such as RMSE values up to 5.2 for onshore power generation. The findingshighlight the strengths and limitations of each model, offering valuable insights for selectingthe appropriate model for wind power applications. The study is limited by the use ofthe SMARD dataset, which aggregates energy production at various time resolutions, necessitatingthe optimization of the MaStR dataset to match SMARD’s installed capacity.Future research should focus on obtaining more detailed data, refining the optimizationprocess, performing bias correction, and exploring additional models and machine learningtechniques to enhance prediction accuracy.enwind powerweather datareanalysismaster thesis