Research On Hail Recognition Based On Radar Reflectivity Cross Section

Hailstone is a destructive severe convective weather. It usually causes hugelosses to people’s lives and property. So, meteorological researchers pay a lot attentionto the recognition and forecast of hail.Radar reflectivity is base data of this paper. Radar reflectivity cross section wasobtained with interpolation algorithm. Features data was obtained with imageprocessing methods. Rules for hail and heavy rain recognition were acquired with datamining based on rough set. And automatic identification model for hail was built.The major work included:1、Radar reflectivity of hail or heavy rain was chosen on radar reflectivity figures.Cutting lines were determined using three methods. Then radar reflectivity crosssection was obtained with linear interpolation in vertical direction and nearestneighbor interpolation in azimuth and the radial distance.2、Weak echo region, bounded weak region and strong echo(over45dBZ) heightwere detected with image processing methods. Sections corresponding to three cuttinglines were compared with each other, the one with the most obvious characteristicswas selected as best section.3、To recognize the hail weather more accurately,1020radar data were analyzedto realize classification rules mining. These data was collected from52weatherprocesses (including29hail processes and23heavy rain processes). Feature data(height and width of weak echo region and bounded weak echo region,0℃and-20℃isothermal layer height as well as the height differences between them andstrong echo height)of these samples was processed with data mining method basedon rough set. Classification rules were obtained and automatic identification modelfor hail was built.4、Echo region on new radar reflectivity figure was found automatically. Itscharacteristics were extracted with proposed method. Feature data was matched withclassification rules. Then weather type of the sample was determined by voting.Hail automatic identification was realized using the proposed model.377hailsamples in28hail processes were tested with this model. The test results showed that:the percentage of accurately recognition is89.9%, and hail can be identified21.6 minutes earlier than it falling on ground. This provides an effective method for hailrecognition and short-time forecast. And it is helpful for reducing the loss caused byhail.