Toimprovetheaccuracyofnowcasting,anewextrapolationtechniquecalledparticlefilterblendingwasconfiguredinthisstudyandappliedtoexperimentalnowcasting.Radarechoextrapolationwasperformedbyusingtheradarmosaicatanaltitudeof2.5kmobtainedfromtheradarimagesof12S-bandradarsinGuangdongProvince,China.Thefirstbilateralfilterwasappliedinthequalitycontroloftheradardata;anopticalflowmethodbasedontheLucas–KanadealgorithmandtheHarriscornerdetectionalgorithmwereusedtotrackradarechoesandretrievetheechomotionvectors;then,themotionvectorswereblendedwiththeparticlefilterblendingalgorithmtoestimatetheoptimalmotionvectorofthetrueechomotions;finally,semi-Lagrangianextrapolationwasusedforradarechoextrapolationbasedontheobtainedmotionvectorfield.Acomparativestudyoftheextrapolatedforecastsoffourprecipitationeventsin2016inGuangdongwasconducted.Theresultsindicatethattheparticlefilterblendingalgorithmcouldrealisticallyreproducethespatialpattern,echointensity,andecholocationat30-and60-minforecastleadtimes.Theforecastsagreedwellwithobservations,andtheresultswereofoperationalsignificance.Quantitativeevaluationoftheforecastsindicatesthattheparticlefilterblendingalgorithmperformedbetterthanthecross-correlationmethodandtheopticalflowmethod.Therefore,theparticlefilterblendingmethodisprovedtobesuperiortothetraditionalforecastingmethodsanditcanbeusedtoenhancetheabilityofnowcastinginoperationalweatherforecasts.
