Presentations

Abstract

Validation of horizontal flow models by cross-prediction for 39 sites Accurate horizontal flow modelling is central to bankable wind resource assessment and energy yield prediction. This thesis is a collaboration between Delft University of Technology and RWE, with the aim is benchmark three widely used modelling systems against a diverse set of real-world observations to quantify accuracy, bias structure, and practical suitability across site types. The study analyses three state-of-the-art horizontal flow models: a coupled mesoscale-mass consistent model, a standalone mesoscale, a meso-LES model. They are evaluated by using a common validation framework applied to 39 onshore sites spanning simple to complex terrain, a range of roughness regimes, and varied directional climatologies. Each site has quality-controlled measurements from masts or lidar, harmonized to common heights and averaging periods. The methodology. First, a long-term reference climate is established for each site using measure correlate predict with either reanalysis or mesoscale drivers. Second, each model produces site specific wind resource grids (WRG). Third, comparisons are carried out by projecting the WRG to a measurement location and then doing a cross-prediction to benchmark the model’s site performance. Statistical metrics include mean bias, mean absolute error, and root mean square error (RMSE) for wind speed and direction, and sectoral Weibull parameter deviations. Terrain complexity is characterized with relief-based indicators, terrain ruggedness index, and forestation so that performance can be stratified by site class. Results show that all three models can reproduce the overall climatology in simple terrain with modest and largely correctable biases. In moderate and complex terrain, differences become material. The large eddy approach delivers smaller mean bias and RMSE, and narrower confidence interval, especially in sectors where the atmosphere tends to be increasingly unstable. Linear or engineering components embedded in mesoscale model and meso-mass consistent model remain effective for gentle terrain and for early screening, but they tend to underrepresent terrain induced flows in complex areas, which translates to inaccurate wind speed predictions. Across the full set of 39 sites, error reductions from the large eddy approach are most pronounced in the regions with increased terrain complexity. LES also benefitted from performing a cross-prediction instead of comparing to single measurements, because the errors occurring on synoptic, and mesoscale are mitigated The thesis concludes with phase specific guidance. For prospecting and portfolio screening, the coupled mesoscale mass consistent model and standalone mesoscale model yield adequate results, in some cases with measurement based bias correction. At sites with simple to moderate terrain complexity, there is no significant difference in performance of the mesoscale and meso-mass consistent models and the meso-LES model. For complex projects where topography, roughness transitions, and turbulence are of increased significance, it is recommended to adopt the meso-LES model. It more adequately incorporates the complex atmospheric flows and turbulent effects that occur when terrain gets more complex. It is also recommended to implement the meso-LES model in the later stages of wind resource assessment, whereas the other models can also be used for the pre-FEED stage and initial site scoping.