1
1 : Météo-France Direction de la Climatologie 42 Av. Coriolis 31057 Toulouse Cedex – France. [email protected] Mean wind speed and wind gust interpolation from the french observing network and Arome modelisation. Introduction : Validation : Conclusions and perspectives : References : Many studies on spatial analysis of meteorological parameters exist. Most of them refer to precipitation and temperature parameters. Here we try to adapt such a method to wind. The quality of the result is function of: - density and quality of the network, - choice of the statistical technique, - use of extra information at a grid mesh smaller than the observing network (from teledetection and/or meteorological models). th A method for spatial interpolation of 10 mn mean wind speed (mws) and wind gust (wg) was tested at Meteo-France/Direction de la Climatologie after the Xynthia storm struck France on february 28 , 2010. This method is used now for a daily production of interpolated data with special interest when strong winds prevail. The selected method comprises two steps: first a multiple linear model, then a kriging of residues from the linear model. We start with hourly observations of mean wind speed and wind gust from synoptic and automatic stations over continental France. After severe QC'ing (withdrawal of poor quality data and unrepresentative stations), around 500 points of measurement are kept with the current network with a mean distance of 35 km between stations. The candidates predictors are coming from data of the Arome model and from a Digital Terrain Model (DTM). The target grid is the one of Arome model operational with a 0,025° lat/lon grid-point system. The DTM is that of 1 SRTM and is adapted to the 0,025° lat/lon grid. The predictors finally selected from Arome for the linear model are the following: mean wind speed at 10 m, gust, wind stress, mean wind speed at 850 hPa and 500 hPa, MSLP. The predictors coming from the DTM are: elevation, zonal and meridian components of the slope vector, coefficient of concavity/convexity, standard deviation of elevation of neighboring points. After processing of the linear model, we have a residue at each point of observation. We realize an ordinary kriging of these residues onto the 0,025° lat/lon grid. Then we add the part coming from the linear model and the part coming from the kriging. 1 srtm.csi.cgiar.org The linear model results for 24 hours during Xynthia storm for mean wind speed are the following: Residual standard error: 2.125 m/s Multiple R-squared: 0.7795 For wind gust: Residual standard error: 3.373 m/s Multiple R-squared: 0.8202 We show biplots of associations between Arome parameters and mean wind speed or wind gust. We show also variograms. Hourly and daily fields of mean wind speed and wind gust are now produced. A time series covering three years is available. Monthly max wind gust and monthly average of 10 mn mean wind speed are also computed. To validate the method, we drive the process with 90% of the observation (learning sample) and compute the estimation on the remaining 10% stations (test sample). The RMSE scores for the day of the Xynthia storm (24 hourly maps) vary around 1,5 m/s for mean wind speed and 2,5 m/s for the gust. The mean relative error is around 20 to 30% of the observed measurement if we take all values but it decreases around 10 to 20% while focusing on the observation above 16m/s (about 60 km/h). This result is interesting but leaves an important uncertainty. To strengthen the validation, we have processed a set of days in 2010 for th th th th th which we have strong winds over France: 30 of March, 4 of May, 14 of July, 8 and 10 of November. Data and Method : Results : The result seems to be the best possible when we combine regression+kriging, considering the data density available from the observing network. The data from the Arome model is a major contribution to the quality of the result. This is particularly true when data at synoptic hours are assimilated in Arome model. This product may help climate monitoring, local estimate of exposure to extreme events, data quality control from additional observing network and even production of climatological high resolution maps when the time series are long enough. [Baillargeon, 2005] Le krigeage : revue de la théorie et application à l'interpolation spatiale de données de précipitations S. Baillargeon 2005 Université Laval – Québec [Benichou, 1986] Prise en compte de la topographie pour la cartographie des champs pluviométriques statistiques P. Benichou O. LeBreton 1986 Direction de la Météorologie Nationale [COST-79, 1997] Seminar on Data spatial distribution in meteorology and climatology – Volterra 28 september – 3 october 1997 European Union COST ACTION 79 – 1997 Ed. M. Bindi B. Gozzini [Daly, 1993] A statistical-topographic model for mapping climatological precipitation over mountainous terrain C. Daly R.P. Neilson D.L. Philips 1993 Journal of Applied Meteorology [DeGaetano, 2007] Spatial interpolation of daily maximum and minimum air temperature based on meteorological model analyses and independent observations A. DeGaetano B.N. Belcher 2007 American Meteorological Society [Gratton, 2002] Le krigeage: laméthode optimale d'interpolation spatiale Y. Gratton 2002 Institut d'Analyse Géographique [Joly, 2010] Temperature interpolation based on local information: the exemple of France D. Joly T. Brossard H Cardot J. Cavailhes M Hilal P. Wavresky 2010 International Journal of Climatology [Souyri, 2007] Adaptations statistiques de vent à 10 mètres (force et direction) : Etude de faisabilité I. Souyri 2007 Météo-France Results for 24 hours february, 28 th 2010: rmse mws in m/s rmse mws in % rmse mws > 16 m/s rmse mws > 16 m/s in% 1.946 28.5 3.413 18.5 rmse wg in m/s rmse wg in % rmse wg > 16 m/s rmse wg > 16 m/s in % 2.805 20.5 3.361 14.7 Results for 24 hours marsh, 30 th 2010: rmse mws in m/s rmse mws in % rmse mws > 16 m/s rmse mws > 16 m/s in % 1.915 31.9 non significative sample rmse wg in m/s rmse wg in % rmse wg > 16 m/s rmse wg > 16 m/s in % 2.769 23.0 3.583 18.0 Results for 24 hours may, 4 th 2010: rmse mws in m/s rmse mws in % rmse mws > 16 m/s rmse mws > 16 m/s in % 1.935 29.9 non significative sample rmse wg in m/s rmse wg in % rmse wg > 16 m/s rmse wg > 16 m/s in % 2.386 19.2 3.086 16.5 Results for 24 hours july, 14 th 2010: rmse mws in m/s rmse mws in % rmse mws > 16 m/s rmse mws > 16 m/s in % 1.294 33.6 non significative sample rmse wg in m/s rmse wg in % rmse wg > 16 m/s rmse wg > 16 m/s in % 1.847 23.6 3.560 19.6 Results for 24 hours november, 11 th 2010: rmse mws in m/s rmse mws in % rmse mws > 16 m/s rmse mws > 16 m/s in % 1.597 29.6 5.184 27.4 rmse wg in m/s rmse wg in % rmse wg > 16 m/s rmse wg > 16 m/s in % 2.048 20.6 2.517 13.1 2 D Réalisation: Création DSO/COM Toujours un temps d’avance 1 Authors: Pierre Lassègues , Jean-Michel Veysseire, Cécile Marie-Luce. Météo-France Direction de la Climatologie.
