Due to the diverse circulation backgrounds of South China heatwaves, this study employed circulation-based classification and machine learning to improve seasonal prediction skill of heatwaves. Specifically, spectral clustering was first used to objectively classify circulation patterns of South China heatwaves during 1981–2024 into two categories: high-pressure-dominated heatwaves (HP-HW) and tropical cyclone-driven heatwaves (TC-HW). The total day numbers of HP-HW and all heatwaves (ALL-HW, i.e., unclassified heatwaves) were selected as two predictands to examine whether circulation-based classification enhances seasonal prediction skill. Results indicate that the HP-HW is a dominant heatwave type in South China, accounting for 75.1% of the total heatwave days. In the independent prediction period, HP-HW predictions achieve a mean coefficient of determination (R 2 ) of 0.42 and a correlation coefficient (CC) of 0.78, exceeding those of ALL-HW (R 2 = 0.23; CC = 0.69) by 19% and 13%. Mechanistic analysis shows that large-scale precursor signals, such as Indian Ocean SST anomalies, consistently reinforce a seasonal high-pressure background over South China, thereby enhancing the seasonal predictability of HP-HW. These results demonstrate that circulation-based classification can effectively isolate predictable physical pathways and offers a practical framework for improving seasonal forecasts of regional climate extremes. 针对华南热浪环流背景多样的特点, 采用谱聚类将1981-2024年热浪分为高压型(HP-HW)与气旋型两类.利用机器学习方法选取HP-HW与全部热浪(ALL-HW)作为预测对象开展季节预测试验.结果表明, HP-HW为华南主导热浪类型, 占总日数的75.1%;独立预测期HP-HW平均R²达0.42, 相关系数0.78, 较ALL-HW分别提升19%与13%.机制分析表明, 印度洋海温异常等大尺度前兆信号通过强化华南高压背景, 显著增强了HP-HW的可预测性.本研究表明环流分类能有效分离可预测的物理途径, 为区域极端气候季节预报提供了新的技术框架.
Niu et al. (Sun,) studied this question.