| Soil loss caused by wind erosion has made a huge contribution to land degradation and desertification,which in turn could make the problem of wind erosion more serious.The wind erosion mass is one of the core contents of soil wind erosion research.The wind erosion models are the main technical mean to obtain the regional wind erosion mass.The accurate estimation of wind erosion is inseparable from the Wind data with high temporal resolution,such as hourly wind speed data.However,most wind erosion models(e.g.,WEPS,RWEQ)required for hourly or more detailed wind speed data in some areas are not always available.These regions may only have daily wind statistics(e.g.,daily average and maximum wind speed).Therefore,the simulation research on hourly wind speed data has become an effective way to solve this problem.This study is based on the hourly wind speed data of 191 climates stations of arid and semi-arid of China during 1998(or 2004,or 2005)through 2009 to statistically evaluate the simulation effect of the Guo(1/π)model,Guo(1/2)model,WINDGEN model,Ephrath model,Debele model and the Goudriaan model in arid and semi-arid China,and analyzed the change of typical diurnal pattern of observed wind speed with clustering method,explore the optimization effect of parameter changes in the Guo(1/π)model and Guo(1/2)model,further study the characteristics of wind speed changes and its impact on wind speed model under sandstorm and non-sandstorm conditions.The main conclusions are as follows:(1)Linear regression analysis implied that the hourly wind speed simulation effect of the six wind speed generation models is: The performances of the Guo(1/π),Guo(1/2)and WINDGEN models were similar with the values of R2(RMSE)were 0.5172(1.0929 m s-1),0.5273(0.9703 m s-1)and 0.5699(0.8497 m s-1),and better than that of the Ephrath,Debele and Goudriaan models with the values of R2(RMSE)were 0.4185(1.7238 m s-1),0.4224(0.8943 m s-1)and 0.2909(1.6256 m s-1).The cumulative distribution and the probability distributed histogram show the overall agreement between measured and simulated cumulative wind speed probability distribution of Guo(1/π),Guo(1/2)and WINDGEN models appeared to be satisfactory,for erosive wind speeds(the wind speeds greater than 5 m s-1),the agreement also appeared better for Guo(1/π),Guo(1/2)and WINDGEN models.(2)Based on daily AWPD values with non-zero values calculated from the measured compared with generated daily AWPD of the 6 models were used to test the models’ performances.The statistical characteristics for the good fit data was Guo(1/π)model(R2=0.9686,RMSE =19.2667 m s-1).But the values of R2 were smaller and RMSE were bigger for the other 5 models,the values of R2(RMSE)were 0.9400(31.3282 m s-1),0.9345(25.2392 m s-1),0.7553(144.4350 m s-1),0.9336(45.7505 m s-1)and 0.8891(31.3418 m s-1)for Guo(1/2),WINDGEN,Ephrath,Debele and Goudriaan models,respectively.Overall,Guo(1/π)model was the best for the wind erosivity(AWPD)estimation.Linear regression analysis implied that Guo(1/π),Guo(1/2)and Ephrath models tend to overestimate AWPD but WINDGEN,Debele,Goudriaan models underestimate AWPD.(3)Seven typical diurnal patterns are obtained by the K-means clustering method,and find diurnal patterns of hourly wind speed variation dictated the performances of the wind speed generation models.The results show that the performances of the Guo(1/π),Guo(1/2)and WINDGEN models were similar and better than that of the Ephrath,Debele and Goudriaan models,and they can better simulate the typical daily patterns that meet this characteristic of reaches the maximum in the afternoon and the minimum during the night.It seemed that Ephrath,Debele and Goudriaan models can well depict the diurnal variation pattern showing a quasi-symmetric structure with the maximum wind speed around CST 12:00.(4)For the results of parameter adjustment for the GUO(1/π)model and GUO(1/2)model,the parameters of the GUO(1/π)model and the GUO(1/2)model are adjusted within a certain range,The two models will appear parameter with the value of the maximum R2 and the small RMSE,and the accuracy of the AWPD by Guo(1/π)and Guo(1/2)models under this parameter value will be significantly improved,which could further influence the accuracy of the wind erosion modeling.(5)The result for the characteristics of wind speed changes under sandstorm and nonsandstorm conditions show that we gain seven typical diurnal patterns under sandstorm and non-sandstorm conditions by the K-means clustering method,respectively.The results show that the probability of the typical daily variables that meets this characteristic of reaches the maximum in the afternoon and the minimum during the night are 79.66% under sandstorm condition and 88.67% under non-sandstorm condition.The performances of the WINDGEN model under sandstorm condition(The values of R2(RMSE)were 0.5191(1.5793 m s-1))were worse than the model under non-sandstorm condition(the values of R2(RMSE)were 0.5426(0.8934 m s-1)). |
