【摘要】 利用自组织神经网络算法(SOMs),对1961—2017年宁夏55次典型暴雨过程的500 h Pa环流场进行分型,总结每种类型的环流特征,并结合实况,计算暴雨发生概率。结果表明,SOMs算法可以成功实现宁夏暴雨的分型。影响宁夏的暴雨基本可分为蒙古冷涡型、平直气流型、河西低槽型和两高夹击型4种类型。其中,蒙古冷涡型最易产生暴雨,两高夹击型不易产生暴雨。在4种分型中,蒙古冷涡型和平直气流型的分型效果较好。对2016—2017年汛期13次典型暴雨反算分型,发现其分型准确率达84. 6%。更多还原

【Abstract】 Based on the daily precipitation at 25 national weather stations during 1961-2017 and 220 regional automatic weather stations during 2009-2017 in Ningxia,and 500 h Pa geopotential height fields of NCEP reanalysis data with 2. 5° spatial resolution,the rainstorm processes were selected during 1961-2017 in Ningxia,and 500 h Pa circulation pattern of rainstorm processes were classified by using self-organizing maps( SOMs) methods. And on this basis the circulation characteristics of rainstorm were summarized. Moreover,combined with the real-time observation,the occurrence probability of rainstorm was calculated. The results show that SOMs method could successfully classify 500 h Pa circulation pattern of rainstorm processes in Ningxia and describe their circulation characteristics clearly. The circulation pattern of rainstorm processes in Ningxia were classified into four types,which were Mongolia cold vortex pattern,straight zonal circulation pattern,Hexi corridor trough pattern and two high surrounding pattern,respectively. Mongolia cold vortex pattern was easier to produce the rainstorm in Ningxia,and the probability of rainstorm was close to 44%,while two high surrounding pattern wasn’t easy to produce the rainstorm with 7% probability. The effect of classification for Mongolia cold vortex and straight zonal circulation patterns was better. The circulations pattern of rainstorms in flood period during 2016-2017 were calculated in reverse,the rainstorm processes with 84. 6% proportion were classified successfully,but the rate of vacancy prediction by using SOMs method was higher,the hit rate was 20%. SOMs method could successfully classify the circulation pattern of rainstorm in Ningxia,the results could be used as a basic condition of rainstorm forecast.更多还原