Error Analysis And Correction Of Model Forecast Surface Wind Speed Over Complex Terrain

Under the influence of many factors on the underlying surface,surface wind speed is highly unstable,and there are problems in numerical weather prediction models,such as insufficient space resolution and imperfect parameterization schemes,which cause obvious errors in surface wind speed over complex terrain and bring huge challenges to many fields,like disaster prevention and mitigation,wind resource assessment and so on.Taking SMS-WARR(Shanghai Meteorological Service-WRF ADAS Rapid Refresh System)as the representative,this paper analyzes the error characteristics of the surface wind speed forecasted by the current numerical weather prediction model.By studying the relationship between the wind speed error and few terrain characteristic parameters,the possible causes of the error related with complex terrain are discussed,and the problem of strong wind speed prediction during the periods of typhoon influence is especially concerned.On this basis,the statistical correction models of the surface wind speed are further established,and the evaluation shows that the accuracy of wind speed prediction can be significantly improved.The specific research contents and main conclusions are as follows:Firstly,this research evaluates the forecast wind speed of SMS-WARR using wind speed observation data of more than 5500 observation stations in Jiangsu,Zhejiang,Shanghai and Fujian in nearly one year.The results of bias,root mean square error(RMSE)and statistical distribution characteristics show that the SMS-WARR model overestimates the wind weaker than 6 grade and underestimates the wind stronger than or equal to 6 grade.The bias will increase and decrease with the increase of model forecasting length respectively.Forecast and observational wind speed have similar distribution between 5%?90% quartiles.During the period of typhoon influence,the area of strong breeze(6 grade)is much larger than the area where the strong breeze was actually recorded and the area of large deviation is mainly distributed in the area of forecast strong breeze.In order to understand the characteristics of wind speed error over complex terrain,terrain height(dh),slope angle(sa),slope length(sl),relative position of the slope(rps),standard deviation(std?3km)and subgrid scale orographic standard deviation(std?sso)are introduced to quantitatively describe the terrain characteristics,and discusses the influence of terrain on wind speed error is further.The results indicate that there is a significant positive correlation between dh and the wind speed bias;sa is negatively correlated with bias.Wind speed bias will decrease with the increase of sa,and the influence of sa is more obvious with the increase of the forecast wind speed.rps also has a significant effect on bias: The bias continues to increase from the bottom to the top of the slope while going uphill,the bias increases wholly from the bottom to the top of the slope while going downhill,and only decreases slightly from the back of the slope to the top of the slope;sl and std?sso have no obvious effect.Those error was caused by the obvious underestimation of high terrain(such as mountain peaks)and imperfect surface layer parameterization scheme in the model.Under the influence of typhoon,the results of site analysis show that the relation between dh and bias,sa and bias are more obvious outside the ten-level wind circle,and is similar to the results of overall error analysis above.Finally,the above terrain factors are introduced into the statistical correction models to compare the effect of stepwise regression correction methods and neural network correction methods for SMS-WARR forecast wind speed,and discusses the applicability of the correction model under the condition of the typhoon.The results indicate that RMSE of forecast wind speed is reduced by more than 40% and bias is close to zero through liner regression correction methods,using stepwise regression correction model after classifying wind speed has the best performance.The RMSE of BP neural network method has further reduced by about 2%,and there is no significant difference whether to use genetic algorithm to optimize neural network.During the period of typhoon impact,the stepwise regression correction model of wind speed classification can improve the shortcoming that the affected area of typhoon gale is too large,but after the correction,it can significantly reduce the maximum values.Therefore,it is necessary to further improve this correction model in the later period to be suitable for typhoon and other disaster weather.