Research And System Implementation Of Radar-based Storm Forecast Integration Algorithm

Storms often cause extreme weather phenomena such as strong winds,heavy precipitation,hail and tornadoes,which affect the national economy and threaten personal safety directly or indirectly.Storms have the characteristics of small spatial scale,short life history,sudden occurrence and rapid development,which are the key and difficult points of the now-casting service(mainly for 0～2 hours).The extrapolation technology based on Doppler weather radar is one of the most important technologies for now-casting to use radar data to identify,track and extrapolate the position or shape of storms at the next moment.At present,the now-casting service system used in operation is usually based on a certain extrapolation technique,so only one extrapolation result is provided for now-casting,which is is not ideal for some storms’ forecast and has limited value for reference.Based on reflectivity factor volume scan data,this paper studies and designs a storm now-casting system integrating advantages of three extrapolation techniques,which can not only obtain the attributes information and structural composition of storm,but also effectively extrapolate the centroid position,regional profile and reflectivity factor of storm.The research focus and conclusion of this paper is as follows.(1)Studying the SCIT algorithm to identify,track and extrapolate the centroid position of storm cell within one hour,and four groups of measured data with different typical weather characteristics are used for analysis.By comparing the storm attributes with the reflectivity factor image’ radar characteristics to analyze the accuracy of the algorithm on identification,the results show that the algorithm can identify the high reflectivity factor region greatly.The prediction accuracy is analyzed by comparing the difference between the predicted cell centroid position and the real one.The results show that the algorithm has higher prediction accuracy for single storms with independent distribution and higher intensity,while the extrapolation error is larger for multi-cell storm and squall line with concentrated cells distribution.(2)Studying the storm identification,matching and extrapolation algorithm based on tree structure which is established based on the decision tree algorithm theory to realize the classification identification of different convection regions in the storm system and the extrapolation of regions within one hour.And uses three groups of radar measured data of different storm types for analysis.Extracting the contour of each recognition area and comparing it with the radar reflectivity factor image,it shows that the classification identification results of the algorithm are credible.Comparing the extrapolated contour image with real radar reflectivity image at the corresponding time,the results show that the algorithm is practicable to various storm,and has a stabler extrapolation effect for storms with larger area while large errors are easier to occur in smaller area,which causes separation of some regions.(3)Studying and implementing the tracking and extrapolation now-casting algorithm for radar echoes based on COTREC theory.Three sets of radar measured data with different typical weather characteristics are used for analysis.And the storm extrapolation effect is verified by calculating the area overlap ratio of the extrapolated echo and the actual radar echo above 30 d BZ.The results show that the algorithm is suitable for storms with larger areas.The comparison of images shows that the algorithm is also suitable for stratiform cloud precipitations with a certain area.The extrapolation image generated from the algorithm has a "hole",which is more obvious for echo with larger intensity gradient.So the algorithm is not suitable for small and strong storms.(4)Designing a storm now-casting system based on VS2015 and Qt,which integrates three radar extrapolation now-casting technologies including SCIT,decision tree and COTREC.It can realize the comprehensive display of storm cell identification,storm centroid,region and echo extrapolation results.Analyzing the extrapolation results of radar data with different typical weather characteristics,it can be found that the three extrapolation results provided by the system can complement each other and forecast different types of storms effectively.To sum up,the system integrates the advantages of extrapolation methods,realizes automatic storm identification and extrapolation prediction,improves the universality of storm types,and has more reference value for storm monitoring and prediction.However,there are still many deficiencies in the research work.Firstly,the region separation problem of Decision Tree algorithm and the "hole" problem of COTREC still need to be improved in the follow-up work by correcting the motion vector and filling the smoothing method.And the layout and functional design of the system should be further improved.In addition,the extrapolation results provided by the system now require forecasters to rely on their own judgment of storm types for selection and reference.If the adaptive algorithm results can be automatically selected and presented according to different storm types in the future to form a comprehensive extrapolation result,the intelligence of the system will be greatly improved and the workload of forecasters will be reduced.