基于K-均值聚类方法的大气环流模式IAP AGCM4.1对西北太平洋热带气旋的模拟评估

IAP AGCM4.1（Institute of Atmospheric Physics Atmospheric General Circulation Model, version 4.1）是中国科学院大气物理研究所自主研发的大气环流模式，也是中科院地球系统模式CAS-ESM1（Chinese Academy of Sciences Earth System Model, version 1）的大气分量模式。本文利用极端气候分析软件TECA（Toolkit for Extreme Climate Analysis），对IAP AGCM4.1模拟的1979～2012年西北太平洋热带气旋（TC）进行了识别与评估。结果表明IAP AGCM4.1模拟的TC空间分布、路径走向与生成源地与观测基本一致，但模拟的TC个数有所低估，仅为观测的36%。基于K-均值聚类方法的分类评估显示，这种低估主要体现在模式对于西北行转向类和西行类TC没有模拟能力。对于近海西—西北行类、西转向类和东转向类TC，模式模拟的个数可分别达到观测的39%，48%和85%，模拟的季节变化与观测的相关系数在0.89～0.91之间，周期误差在1～2天。就TC路径而言，模式对于近海西—西北行类和东转向类TC模拟效果较好，质心经度误差、质心纬度误差和经纬向标准差的模拟误差分别为1%～5%、4%～16%和5～15%。此外，环流合成分析表明模式很好地再现了东转向类TC发生、发展期间环境流场的演变以及副热带高压的变化情况，模拟的副热带高压强度和面积指数与观测的相关系数可达0.89。模式对西北行转向类和西行类TC模拟能力较差的原因可能与模式对副热带高压的模拟偏差有关。

Evaluation of Simulated Tropical Cyclones over the Western North Pacific with IAP AGCM4.1 Based on K-Means Method

As the atmospheric component of CAS-ESM1 (Chinese Academy of Sciences Earth System Model, version 1), IAP-AGCM4.1 (Institute of Atmospheric Physics Atmospheric General Circulation Model, version 4.1) is being developed independently by Institute of Atmospheric Physics. In this study, the authors used TECA (Toolkit for Extreme Climate Analysis) to identify and evaluate tropical cyclones (TC) over the western North Pacific simulated by IAP AGCM4.1 from 1979 to 2012. The results show that IAP AGCM4.1 can reproduce the spatial distribution, track, and source of TC reasonably well compared to observation data, but it underestimates the number of TC, with only 36% of the observed tropical cyclones over the western North Pacific simulated. Further analysis using K-means clustering revealed that this underestimation is mostly due to the model’s inability to reproduce northwestward-turning and westward TC. For TC with westward–northwestward, westward-turning, and eastward-turning tracks, the numbers simulated are approximately 39%, 48%, and 85% of those observed, respectively. Moreover, the correlation coefficients of the seasonal variations between simulated and observed TC can reach 0.91, with duration biases of roughly 1–2 d. IAP AGCM4.1 performs well in simulating the tracks of the westward–northwestward and eastward-turning TC, with relative biases ranging between 1%–5% for the longitude of the centroid, 4%–16% for the latitude of the centroid, and 5%–15% for the latitudinal and meridional standard deviations. In addition, IAP AGCM4.1 reproduces the evolutions of environmental circulation and subtropical highs quite well during the lifetime of eastward-turning TC, with the simulated strength and area indexes of the subtropical highs highly correlated with the observations (the correlation coefficient is 0.89). The poor simulations of northwestward-turning and westward TC are likely due to simulated biases in the subtropical high.