夏季影响海南的热带气旋频数预测

FREQUENCY PREDICTION OF TROPICAL CYCLONES AFFECTING HAINAN IN SUMMER

利用1976-2013年夏季影响海南的热带气旋频数(SumTCs)资料、NCEP/NCAR的500 hPa高度场、高低层纬向风切变场、SLP场、OLR场和海温场等再分析资料,研究了前期不同环境场因子对SumTCs的影响,提取有一定物理意义的高相关预测因子群,经因子降维处理后建立SumTCs的模糊神经网络(FNN)预测模型。研究结果表明:1976-2007年SumTCs的交叉检验预测结果与实况SumTCs的相关系数为0.71,平均绝对误差为0.65个。该预测模型在2008-2013年6年的独立样本检验中平均绝对误差为1.5个,评分为76.7分,优于相同因子的逐步回归预测模型和仅基于海温场和500 hPa高度场因子的FNN预测模型。该模型可以投入海南省热带气旋频数的季节预测业务参考使用。

Based on the data of the frequency of tropical cyclones affecting Hainan in summer （SumTCs） and the NCEP/NCAR reanalysis data of 500 hPa height field, wind vertical shear field, SLP field, OLR field and SST field from 1976 to 2013, the effects of different environmental factors on SumTCs are analyzed. Some high correlation predictors are extracted by reducing the dimension of the factor group, and the Fuzzy Neural Network （FNN） prediction model Of SumTCs is further established. The results indicated that the correlation coefficient between the observation and cross-validation predicted frequency of SumTCs is 0. 71 and the average absolute error is 0. 65 during 1976-2007. An independent sample validation shows that the average absolute error and the prediction score of the prediction model are 1.5 and 76. 7 respectively from 2008 to 2013. The prediction model is not only better than the stepwise regression prediction model based on the same factors but also better than the FNN prediction model based on the factors of SST field and 500hPa height field. It can be used in the seasonal prediction operation of SumTCs.