A new prediction model for tropical storm frequency over the western North Pacific using observed winter-spring precipitation and geopotential height at 500 hPa
A new prediction model for tropical storm frequency over the western North Pacific using observed winter-spring precipitation and geopotential height at 500 hPa
复制标题
DOI：
10.1007/s13351-011-0302-6
复制
发表时间：
2011-07
影响因子：
--
通讯作者：
Huijun Wang
中科院分区：
文献类型：
--
作者：
Huijun Wang
利用前期1 - 2月（JF）和4 - 5月（AM）格点资料，建立了一个新的西北太平洋热带风暴年数季节预报模式。利用北方半球JF、AM平均降水量和AM平均500 hPa高度场，以及南半球JF平均500 hPa高度场，采用逐步多元线性回归技术，建立了ATSN预报模式。所有JF和AM平均数据均限于东半球。利用ERA 40再分析资料和NCEP再分析资料，结合实测降水资料，建立了两个ATSN经验预报模型。模型的性能通过交叉验证进行了验证。通过比较两个模型的回顾性预测和1979年至2002年观测到的ATSN，获得了0.78和0.74的异常相关系数（ACC）。两个模型的多年平均绝对预测误差分别为3.0和3.2，约为平均ATSN的10%。在实践中，最终的预测是通过平均两个模型的ATSN预测。这导致了更高的分数，ACC进一步增加到0.88，平均绝对误差降低到1.92，或平均ATSN的6.13%。
A new seasonal prediction model for annual tropical storm numbers (ATSNs) over the western North Pacific was developed using the preceding January-February (JF) and April-May (AM) grid-point data at a resolution of 2.5° × 2.5°. The JF and AM mean precipitation and the AM mean 500-hPa geopotential height in the Northern Hemisphere, together with the JF mean 500-hPa geopotential height in the Southern Hemisphere, were employed to compose the ATSN forecast model via the stepwise multiple linear regression technique. All JF and AM mean data were confined to the Eastern Hemisphere. We established two empirical prediction models for ATSN using the ERA40 reanalysis and NCEP reanalysis datasets, respectively, together with the observed precipitation. The performance of the models was verified by cross-validation. Anomaly correlation coefficients (ACC) at 0.78 and 0.74 were obtained via comparison of the retrospective predictions of the two models and the observed ATSNs from 1979 to 2002. The multi-year mean absolute prediction errors were 3.0 and 3.2 for the two models respectively, or roughly 10% of the average ATSN. In practice, the final prediction was made by averaging the ATSN predictions of the two models. This resulted in a higher score, with ACC being further increased to 0.88, and the mean absolute error reduced to 1.92, or 6.13% of the average ATSN.