作者：Sullivan, C. & Almeida, L.

Temperate lakes are experiencing intensifying heatwaves, or prolonged periods of anomalously warm water, with potentially severe ecological consequences. Previous studies have predominantly used linear trend analyses to understand heatwave dynamics, potentially masking temporal acceleration patterns and shifts in variance. Here, we analyzed long-term, nonlinear trends in heatwave frequency, intensity, and timing across 4391 New York lakes using simulated daily surface water temperatures from an Entity-Aware Long Short-Term Memory deep learning model (validated root mean square error = 1.82 degrees C; 1980-2020). We applied generalized additive mixed models to quantify complex temporal patterns while accounting for lake-specific effects and environmental characteristics (lake surface area, elevation, and latitude). All heatwave metrics demonstrated nonlinear and dynamic temporal trends. Heatwave frequency showed early increases in the number of events (0.05-1.32 events during the 1980s) followed by accelerated growth after 2000; other metrics exhibited more complex decadal patterns. Temporal variance increased significantly for both heatwave frequency (tau = 0.73, p < 0.001) and intensity (tau = 0.33, p = 0.003). Lake characteristics drove some heatwave dynamics; smaller lakes experienced higher intensities and complex elevation-latitude interactions varying by heatwave metric. Spatial patterns revealed increasing heterogeneity for heatwave intensity and frequency while heatwave timing homogenized to earlier dates across the landscape. Our findings reveal nonlinear heatwave dynamics with increasing variability, which may reduce predictability and intensify ecological vulnerabilities beyond current projections.

温带湖泊正经历日益加剧的热浪——即水温长期处于异常偏高的状态，其生态后果可能十分严重。既往研究大多采用线性趋势分析来理解湖泊热浪动态，这可能掩盖时间上的加速格局及方差的变化。研究利用实体感知长短期记忆网络（Entity-Aware Long Short-Term Memory，EA-LSTM）深度学习模型模拟的逐日表层水温数据（验证均方根误差 = 1.82 °C；1980—2020年），分析了纽约州4391个湖泊热浪频率、强度与发生时间的长期非线性趋势。采用广义可加混合模型（GAMM）量化复杂的时间格局，同时控制湖泊特异性效应及环境特征（湖面面积、海拔与纬度）。结果表明：所有热浪指标均呈非线性、动态变化的时间趋势。热浪频率在20世纪80年代呈早期增长（每十年0.05—1.32次事件），2000年后增长加速；其他指标则表现出更为复杂的年代际格局。热浪频率（tau = 0.73，p < 0.001）与强度（tau = 0.33，p = 0.003）的时间方差均显著增大。湖泊特征驱动了部分热浪动态：小型湖泊热浪强度更高，且海拔—纬度交互作用复杂，因热浪指标而异。空间格局显示，热浪强度与频率的空间异质性不断增强，而热浪发生时间则在区域尺度上趋于同质化、整体提前。研究发现表明，湖泊热浪呈变异性不断增强的非线性动态，这可能降低其可预测性，并使生态脆弱性的加剧程度超出当前预估。

（来源：Limnology and Oceanography 2026 Issue 6  DOI: 10.1002/lno.70429）