基于正则化最优指纹法的华南极端温度变化归因研究

Toward Regularized Optimal Fingerprinting in Detection and Attribution of Extreme Temperature Changes in South China

摘要: 在全球变暖背景下，明确人类活动对区域极端气温演变的影响对于气候风险评估至关重要。本文以华南地区为研究对象，基于1961至2014年ERA5-Land高分辨率观测数据与14个CMIP6全球气候模型的多模型集成(MME)，对四个核心极端温度指数(TXx, TXn, TNx, TNn)进行了时空特征分析与检测归因。针对区域尺度下空间格点数大于时间样本量所导致的协方差矩阵病态难题，本研究引入基于L2范数的收缩估计，构建了正则化最优指纹法(ROF)及估计方程(EE)模型。研究结果表明，华南极端温度在1990年后出现显著的变化，且空间上呈现“趋势–噪音错位”现象。在归因分析中，人类活动信号在极端最高温TXx与白昼极端低温TXn中被成功检测，且模型模拟的响应幅度与实际观测高度匹配。同时，人类活动显著推动了夜间极端高温TNx的上升，但CMIP6模型存在明显的高估偏差，这可能源于模型对局地人为气溶胶冷却效应的模拟不足；而极端最低温TNn由于受到气候模型内部变率的影响，未能通过统计学信号检测。本研究不仅证实了人类活动对华南极端温度的主导作用，更凸显了考察物理强迫在区域气候归因中的核心价值。

Abstract: In the context of global warming, clarifying the impact of human activities on the evolution of regional extreme temperatures is crucial for climate risk assessment. Focusing on South China, this study conducts a spatio-temporal characteristic analysis and detection and attribution for four core extreme temperature indices (TXx, TXn, TNx, TNn), based on high-resolution ERA5-Land observational data and a Multi-Model Ensemble (MME) of 14 CMIP6 global climate models from 1961 to 2014. Addressing the ill-conditioned covariance matrix problem caused by the number of spatial grid points exceeding the time sample size at the regional scale, this study introduces an L2 norm-based shrinkage estimation to construct a Regularized Optimal Fingerprinting (ROF) and Estimating Equations (EE) model. The results indicate that extreme temperatures in South China have undergone significant changes since 1990, exhibiting a “trend-noise mismatch” phenomenon spatially. In the attribution analysis, human-induced signals were successfully detected in the extreme maximum temperature (TXx) and daytime extreme minimum temperature (TXn), with the model-simulated response magnitudes highly matching the actual observations. Meanwhile, human activities have significantly driven the rise of the nighttime extreme maximum temperature (TNx), but the CMIP6 models exhibit a distinct overestimation bias, which may stem from the models’ inadequate simulation of the local anthropogenic aerosol cooling effect. Conversely, the extreme minimum temperature (TNn) failed to pass the statistical signal detection due to the profound influence of internal climate variability. This study not only confirms the dominant role of human activities on extreme temperatures in South China but also highlights the core value of investigating physical forcings in regional climate attribution.