engleski

Scale-dependent evaluation of mesoscale low-level winds obtained with ALADIN MNWP model

In the complex terrain of the eastern Adriatic where wind climate is governed by regional and local winds, it is beneficial to utilize a chain of numerical models to refine wind predictions. Verification of these mesoscale flows is a challenging task for which adequate approaches still need to be revised and defined. Traditionally used statistical scores (MBIAS, RMSE, MAE, etc.) seem to be insufficient for that purpose since small spatial and temporal errors of generally well simulated phenomena can profoundly change the verification results. Therefore, besides moment-based verification, we used spectral analysis as a supplementary verification method to provide a scale dependent measure of model performance, with emphasis on understanding the sources of prediction errors through relating various aspects of physical and spectral verification measures. The verification was performed on wind forecasts obtained by Aire Limitée Adaptation dynamique Développement International (ALADIN) MNWP model with 8 km horizontal grid spacing in period 2010-2012. These forecasts were further refined to 2 km grid spacing using: i) full- physics based model and ii) so-called dynamical adaptation method (DADA) over subdomain that covers broader area around Croatia. Major objectives of this study were to: i) determine whether an increase of model horizontal resolution and/or complexity improves the accuracy and ii) quantify the importance of different sources of errors and inspect the potential correlation of moment-based measures with various spectral scores. Based on variety of statistical and spectral scores, it is suggested that the results generally improve with increasing the model horizontal resolution and complexity. The largest portion of root-mean square errors (RMSE) can be attributed to phase errors (PHE), while the most significant increase of accuracy was found for diurnal and sub-diurnal periods of motions. Furthermore, we found a significant correlation between the share of the bias of the standard deviation (BSD) in RMSE and model ability to properly simulate the ratio of modeled and measured power in individual spectral range. For bora dominated coastal stations a considerable correlation exists between the share of measured power within individual spectral ranges and the share of individual sources of errors (PHE, BSD and the bias of the mean-BM). Finally, it was shown that simplified DADA forecasts were valuable in forecasting wind properties on a majority of stations, except for those near the very coast and steep terrain where full-physics based model was more successful in representing regional/local wind systems.

complex terrain; wind forecast; horizontal resolution; moment-based verification; spectral scores

IPA projekt WILL4WIND, IPA2007/HR/16IPO/001-040507