简介：有在天气和气候的可预测性问题的三种普通类型，它各包含不同抑制非线性的优化问题：最大的可预言的时间，最大的预言错误的上面的界限，和最大的许可的起始的错误和参数错误的更低的界限的更低的界限。高度有效的算法被开发了解决第二个优化问题。并且这个优化问题能在现实主义的模型被使用让天气和气候学习最大的预言错误的上面的界限。尽管过滤策略被采用了解决另外的二个问题，直接答案甚至为一个很简单的模型是很费时间的，它因此在现实主义的模型限制这二个可预测性问题的适用性。在这份报纸，新策略被设计解决这些问题，包含存在的使用为第二个可预测性问题的高度有效的算法特别地。而且，在更旧的过滤策略之间的一系列比较和新方法被执行。这被表明新策略不仅输出象旧的一样的结果，而且也是更计算地有效的。这将建议学习在天气或气候的现实主义的预报模型与这二个非线性的优化问题联系的可预测性问题是可能的。关键词抑制了非线性的优化问题-可预测性-算法

简介：Aconvection-allowingensembleforecastexperimentonasqualllinewasconductedbasedonthebreedinggrowthmode(BGM).Meanwhile,theprobabilitymatchedmean(PMM)andneighborhoodensembleprobability(NEP)methodswereusedtooptimizetheassociatedprecipitationforecast.Theensembleforecastpredictedtheprecipitationtendencyaccurately,whichwasclosertotheobservationthaninthecontrolforecast.Forheavyrainfall,theprecipitationcenterproducedbytheensembleforecastwasalsobetter.TheFractionsSkillScore(FSS)resultsindicatedthattheensemblemeanwasskillfulinlightrainfall,whilethePMMproducedbetterprobabilitydistributionofprecipitationforheavyrainfall.Preliminaryresultsdemonstratedthatconvection-allowingensembleforecastcouldimproveprecipitationforecastskillthroughprovidingvaluableprobabilityforecasts.Itisnecessarytoemploynewmethods,suchasthePMMandNEP,togenerateprecipitationprobabilityforecasts.Nonetheless,thelackofspreadandtheoverpredictionofprecipitationbytheensemblemembersarestillproblemsthatneedtobesolved.

简介：Basedonthetropicalcyclone(TC)observationsinthewesternNorthPacificfrom2000to2008,thispaperadoptstheparticleswarmoptimization(PSO)algorithmofevolutionarycomputationtooptimizeonecomprehensiveclassificationrule,andapplytheoptimizedclassificationruletotheforecastingofTCintensitychange.Intheprocessoftheoptimization,thestrategyofhierarchicalpruninghasbeenadoptedinthePSOalgorithmtonarrowthesearcharea,andthustoenhancethelocalsearchability,i.e.hierarchicalPSOalgorithm.TheTCintensityclassificationruleinvolvescoreattributesincluding12-HMWS,MPI,andRainratewhichplayvitalrolesinTCintensitychange.ThetestingaccuracyusingthenewminedrulebyhierarchicalPSOalgorithmreaches89.6%.ThecurrentstudyshowsthatthenovelclassificationmethodforTCintensitychangeanalysisbasedonhierarchicPSOalgorithmisnotonlyeasytoexplainthesourceofrulecoreattributes,butalsohasgreatpotentialtoimprovetheforecastingofTCintensitychange.