Research On Predictability Of Surface Wind Vector Based On TIGGE Data

Wind forecasts have close affiliation with people’s life and industrial production. Transfer and diffusion of pollutants, destruction of buildings, transportation and power equipment caused by strong winds, severe disasters resulting from typhoons, and even the rising demand of clean energy due to the excess usage of fossil fuels have all contributed to the urgency of wind vector forecasting. Ensemble forecast is one efficient method of surface wind prediction. So far, the study of ensemble forecast, especially research on TIGGE data set of THORPEX plan, mainly focuses on the forecasts of circulation field, temperature or precipitation, while less attention has been paid to the surface wind prediction.Based on the information above, this study takes the 0-240h forecast of surface wind with a time interval of 24h as the database. The data is collected from China Meteorological Association (CMA), ranging from 2009 to 2013. We discussed the predictability of surface wind vectors and linear/nonlinear characteristics of forecast error via rank correlation coefficient and MVL diagrams. At last, Post data-analysis method is proposed to improve the prediction of surface wind and several conclusions are drawn as follows.1. Within a short period of 24 hours’validity, strong predictability could be found between ensemble parameters:zonal wind velocity (U), radial wind velocity (V), speed (S), speed direction (D), 1000hPa geopotential height (GH). Generally, parameters with obvious forecast errors within 24h have larger prediction uncertainty after 24h. This correlation could be maintained even with a prediction interval of 120 hours. Seasonally, this correlation of forecast error is stronger in summer and autumn compared with that in spring and winter. All above shows that the quality of predictions of different parameters at the early stage could be used as different weights to improve the level of post ensemble prediction.2) Annually, within a period of 120 hours’ validity, linear correlation between perturbation and perturbation variance is discovered for zonal and radial wind velocity, which is also true for speed within a period of 240h and GH within a period of 24 hours’validity. Seasonally, speed, zonal and radial wind velocity show linear correlation within 120h validity in all seasons except speed in winter. Geopotential height presents the linear correlation within 48h validity in all seasons except spring. The perturbations of the four meteorological parameters above increase with the prediction time while the increasing rate becomes smaller except for geopotential height. Linear correlation between perturbation and perturbation variance of all four meteorological parameters within short time validity is thus demonstrated.3) It is also found that compared with mean value method, the forecast accuracy of U, V, S by PLS method first increase and then decrease, and peaks at 48h on an annual scale. With the increase of prediction time, the prediction accuracy decreases, both for PLS and mean value method. On a seasonal scale, the prediction errors for U, V, S within period of 120 hours’ validity, descend in the order of autumn, summer, spring and winter. While after 120 hours, the seasonal characteristics become fuzzy. Compared with ensemble mean, PLS regression prompts the largest enhancement to the forecast performance in summer within 240 hours’ validity, but least in autumn.4) On an annual scale, forecast errors of PLS method and mean value method increase with the period of validity for GH. PLS method performs better within the first 48 hours. On a seasonal scale, whatever forecast method we choose, best performance is achieved for summer, followed by winter. And both forecast methods perform worse when time extends. Compared with ensemble mean, PLS regression improves the forecast performance best in spring. And in summer, autumn and winter, PLS regression performs better only within 48 hour. On both annual and seasonal scale, the line charts of precision improvement ratio showed similar trends with MVL diagrams, which indicates the greater influence of the pertubation variance V (T) on the improvement of accuracy ratio.Finally, based on the conclusions above, this dissertation proposes a post-treatment method for ensemble prediction of surface wind vectors:utilize PLS regression forecast within a short period of 120 hours’ validity and use ensemble mean values directly after 120 hours, for the prediction value of U, V, S in spring, autumn and winter. While in summer, PLS regression values for wind vector prediction are encouraged within a period of 240 hours’ validity. In the 1000 hPa geopotential height field, we discover better performances of PLS regression values during a short period of 48 hours’ validity, and for the rest period of validity ensemble mean value dominates.