Study On The Characteristics And 24 H Prediction Of Thunderstorm In East Region Of Northwest China

Severe convective weather such as thunderstorm is one of the main meteorological disasters in east region of Northwest China. Thunderstorm can cause different degree of damages to air transportation, electric power facilities, communication equipment and the buildings, sometimes even can cause serious casualties. East region of Northwest China has various kinds of landscapes and terrains. Therefore, the difficulty degree of forecasting thunderstorm is relatively high. It has important scientific significance and practical values to conduct the forecast of thunderstorm in east region of Northwest China. Based on conventional observation data of 60 sites in east region of Northwest China from 1980 to 2013 and NCEP/NCAR 1°×1° data from 2008 to 2013, analyze temporal and spatial distribution of thunderstorm in east region of Northwest China using climatic statistical method, then divide the thunderstorm subregions. Define the concept of regional thunderstorm day of east region of Northwest China, then, analyze the synoptic situation when regional thunderstorm day occurred to summarize several main kinds of weather circumfluence types of regional thunderstorm day. Design automatic recognition and intensity index calculation methods of each weather circumfluence type to classify the weather circumfluence type automatically. Applying physical diagnostic analysis to all samples conformed to weather circumfluence types to decrease false alarm ratio, establish the 24 h forecast flowchart of east region of Northwest China regional thunderstorm combining weather circumfluence classification with physical diagnostic analysis. Establish single site thunderstorm 24 h forecast model of east region of Northwest China using Linear Stepwise Regression, Logistic Regression and BP Neural Network, test and compare the forecast effect of each model, choose the best model for each site. The main conclusions are as follows:(1) The average annual thunderstorm days of 60 sites in east region of Northwest China are 15.6 d, regional annual thunderstorm days are 70.7 d. Thunderstorm spatial distribution characteristics show that: thunderstorm occurrence is most frequent in Gannan plateau, secondly in Southern Qin Mountains and northern Shaanxi, then in Loess plateau and Gobi Desert, while Central Shaanxi Plain have fewest amounts of thunderstorms. According to different underlying surface characteristics, thunderstorm is divided into five subregions: Gobi Desert region, Loess Plateau region, Plateau and Mountain region, Central Shaanxi Plain region and Southern Qin Mountains region. Thunderstorm inter-annual variation characteristics show that: thunderstorm occurrence is frequent in middle 1980 s and early 2000 s, and infrequent from late 1980 s to early 1990 s and after 2004. Thunderstorm seasonal variation characteristics show that: thunderstorm amounts are mainly concentrated in summer(The peak appears in July), average annual thunderstorm days of 60 sites and regional annual thunderstorm days are accounted for 67.3% and 66.4% of total amounts of a year, thunderstorm amounts are fewer in spring and autumn while are fewest in winter. Thunderstorm diurnal variation characteristics show that: thunderstorms mainly occur from 14:00 to 02:00, and the amounts of thunderstorm occurred in the afternoon are more than which occurred in the first half of the night. The amounts occurred from 02:00 to 14:00 are relatively fewer.(2) The main weather circumfluence types of east region of Northwest China thunderstorm are divided into four kinds: upper vortex, upper trough, northwest airflow and southwest airflow. The occurrence times of Northwest airflow and upper trough are most, followed by southwest airflow, and which of upper vortex are fewest. The atmospheric temperature configuration style of upper vortex, upper trough and northwest airflow is: lower temperature and humidity at the upper atmosphere, higher temperature and humidity at the lower atmosphere, but the configuration style of southwest airflow is: high temperature and humidity at whole layer of atmosphere. Summarize the corresponding quantitative criteria of these four weather circumfluence types, discriminate all April to October thunderstorm weather circumfluence types from 2008 to 2012 automatically using these criteria, 1030 d which are conformed to weather circumfluence types are attained. The Critical Success Index(CSI) of automatic classification is 33.6%, and the False Alarm Ratio(FAR) is 66.3%.(3) Conduct further physical diagnostic analysis to the 1030 d attained in last step, the back-substitution test results show that: 433 d of empty samples are eliminated, the CSI increases to 54.1% and the FAR decreases to 45.5%. Forecast 2013 thunderstorm from April to October combining weather circumfluence automatic classification with physical diagnostic analysis, the results show that: 129 d are forecasted to be regional thunderstorm, 85 d are eliminated, the CSI is 51.8%, and the FAR is 44.2%. The attempt forecast results are close to back-substitution forecast results, suggests that the method combining weather circumfluence classification with physical diagnostic analysis is stable and useful.(4) Compare average forecast effect of Linear Stepwise Regression model, Logistic Regression model and BP Neural Network model, the results show that: Logistic Regression model is stablest, the forecast effect is the best, followed by Linear Stepwise Regression model, BP Neural Network model is most unstable, the forecast effect of this model is common. Logistic Regression model and Linear Stepwise Regression model have better forecast effect on sites which have more thunderstorm occurrence times, BP Neural Network model has better forecast effect on sites which have fewer thunderstorm occurrence times. The subregion that has highest average CSI for each forecast model is the Plateau and Mountain region which has highest thunderstorm climatic probability. The Central Shaanxi Plain region which has lowest thunderstorm climatic probability has lowest average CSI for each model. Gathering the attempt forecast result of the best forecast model for each site, the average best forecast CSI of all 60 sites reaches to 26.6%, and the FAR decreases to 49.6%. These forecast results are reasonable.In conclusion, the regional thunderstorm 24 h forecast flowchart of east region of Northwest China is established combining weather circumfluence type classification with physical diagnostic analysis, the single site thunderstorm 24 h forecast model of east region of Northwest China is established applying Linear Stepwise Regression, Logistic Regression and BP Neural Network. These works can provide references and gists for the prediction and early warning of thunderstorm in this region.