Study On Refined Forecast Methods Of Cloud Cover Around Hetao

Clouds are common weather phenomenon.Minor changes in the amount of clouds may have a greater impact on the development and evolution of weather systems.Clouds also directly affect aerospace activities,which has always been one of the weather phenomena that the Air Force and Civil Aviation Department has paid close attention to.As an important industrial and military area in China,studing and mastering the changing characteristics and local laws of clouds,providing fine cloud forecasting,and improving the accuracy of cloud forecasting are necessary for people’s daily life,agricultural production,and military activities have very important practical significance.By using the ground message data in the format of Integrated Meteorological Information Analysis and Processing System（MICAPS）,the reanalysis data of the ECMWF’s（European Centre for Medium-Range Weather Forecasts）Interim reanalysis（ERA-Interim）,and the Global Weather Forecast System（GFS）,the statistics characteristics of cloud of 39 stations in the surrounding area of Hetao on different time scales have been analisized in this paper.The data of the GFS forecast field is used to construct five types of forecast factors of the cloud cover around Hetao from the perspective of the mechanism of cloud generation.The main factors affecting total cloud cover（TCC）,low cloud cover（LCC）,and convective cloud cover（CONV）were analyzed.Time-refined forecast models of total cloud cover,low cloud cover,and convective cloud cover were established by using multivariate stepwise regression forecasting method,and dynamic time-varying parameter methods,namely adaptive least squares regression and adaptive recursive Kalman filtering methods,were used for stepwise regression equation and established a dynamic regression coefficient.In this paper,by using back propagation（BP）neural network forecast algorithm,least squares support vector machine（LSSVM）regression forecast algorithm and Elman recursive neural network algorithm cloud cover forecsting were studied.Finally,based on several forecast models,taking the total cloud amount as an example,an ensemble forecasting model was established using Ensemble Mean（EMN）,Bias-Removed Ensemble Mean（BREM）,Weighted Bias-Removed Ensemble Mean（WBREM）,and Superensemble（SUP）.The main conclusions are as follows:（1）The annual change characteristics of cloud cover around Hetao are as follows:from 1979 to 2013,the total cloud cover,low cloud cover observations,and ERA-Interim are all distributed high in the south and low in the north.The observations of the distribution of low cloudiness are more consistent with ERA-Interim.The cloudiness gradually increases from northwest to southeast,and the values of ERA-Interim are lower than the observed values.The ERA-Interim cloud cover Sen’s trend distribution characteristics from 1979 to 2018 are:the western region has increased,especially in the southwestern region,the trend of total cloud cover and low cloud cover Sen’s trend is more than 10^-1/decade（%）,and the trend of low cloud cover increase is greater than the total cloud cover;while the southeast performance is reduced by the total cloud cover and low cloud cover,which is-15×10^-2/decade（%）.The trend is that the amount of clouds in the central and eastern regions is slightly reduced.（2）The forecast factors that affect the amount of cloud at each forecast time are mainly water vapor forecast factors,cloud forecast factors directly output from the GFS model,and atmospheric instability factors.The correlation between convective cloud cover and forecasting factors is weaker than the total cloud cover,and low cloud cover.The 39-sites stepwise regression forecast model also introduced the highest frequency of water vapor forecast factors and cloud forecast factors directly output from the GFS model,the most frequently introduced is the relative humidity of the whole layer,and more than 200 times.（3）Through the interpretation of the GFS forecasting field,the total cloud amount forecast value obtained by the stepwise regression forecast method has significantly improved the forecast accuracy rate of the direct output of the model,and the improvement effect of low cloud amount is the greatest,the average correction capacity of the northwest region is more than 20%.The adaptive linear least squares（LS）forecast method is better than the adaptive recursive Kalman filtering method in the forecast of total cloud cover,low cloud cover and convective cloud.（4）There is little difference between the three non-linear forecasting methods after the principal component extraction,and the average absolute error of the substitution and the forecast gradually increases with the forecast timeliness,the average convection error and forecast error of convective clouds are the smallest,basically below 10%.The LSSVM forecasting method has a better fit to the cloud back generation than the other two neural network forecasting methods.The three non-linear forecasting methods have large spatial and temporal differences in convective cloud forecast.The correlations between the predicted values and live values of the three non-linear forecasting methods are significantly stronger than the multiple stepwise regression method,and the correlations between the predicted values of the total cloud cover,low cloud cover,and convective cloud cover and the actual situation are significantly enhanced.（5）All the forecasting methods can obviously improve the hit rate of the predicted value under the condition of few clouds and cloudy sky,and the adaptive LS method is the best（on average,the hit rate of few clouds is increased by 24%,and the hit rate of cloudy is increased by 34%）.The adaptive recursive Kalman filtering method has the highest hit rate under cloudy conditions.（6）Compared with the linear forecast methods,non-linear forecast method,and the four ensemble forecast methods,the total cloud cover forecast results have different forecast capabilities for the total cloud cover.The adaptive linear LS regression method has higher forecast skills,the average absolute error is about 20%.And adaptive The recursive Kalman filtering method has the worst performance.After ensemble forecasting,the four ensemble forecasting models are significantly improved than the original forecast method,and the SUP method forecast technique is the best in the ensemble forecasting methods.In short,in this study,we discussed the detailed cloud area forecast in the surrounding area of Hetao.The results can provide a scientific reference for those who are engaged in the refinement and release of numerical forecast products and a reference for local business personnel who are engaged in weather services.