• North American autumn phenology showed no significant trend from 2000 to 2020. • Increased temperature delays senescence, while higher VPD advances it. • Future projections show stability, with delays in regions of significant change. Leaf Senescence Date (LSD) is a critical indicator of ecosystem response to climate change, yet its long-term trends and drivers in North America remain uncertain. In this study, we investigated the spatiotemporal dynamics of LSD across North America from 2000 to 2020 by integrating multi-source data, including ground-based and satellite observations. We employed an Interpretable Ensemble Machine Learning (IEML) model and Accumulated Local Effects (ALE) to quantify the influence of various pre-season meteorological factors. Furthermore, we projected future LSD until 2050 using a Gated Recurrent Unit (GRU) model under two climate scenarios (SSP245 and SSP585). Our findings reveal no significant overall trend in North American LSD over the past two decades, with marked spatial heterogeneity. The pre-season temperature, solar radiation, and Vapor Pressure Deficit (VPD) are important variables influencing autumn phenology. The ALE results further clarified that increases in pre-season temperature, rain, and its frequency promote a delay in LSD, while an increase in VPD leads to an advance in LSD. The mutually counteracting effects of these pre-season meteorological factors accounts for the absence of a significant overall LSD trend across North America. Future projections indicate that although the area exhibiting significant changes in LSD is approximately twice that observed during the historical period, more than 80% of the region is expected to maintain no significant trend in LSD. This research provides a comprehensive analysis of North American autumn phenology, highlighting the complex interplay of climatic drivers and further enhancing our understanding of vegetation responses to global climate change.
Sun et al. (Tue,) studied this question.