气候变化对地球系统产生了深刻影响，导致极端气候事件频发、生态系统稳定性降低，严重影响人类的生产生活。气候变暖导致北半球温带植物春季展叶时间提前、秋季叶衰时间延迟，物候变化不仅会影响陆地生态系统的结构与功能，还会改变全球和区域碳水循环与能量平衡。森林是陆地生态系统最大的碳汇，对于维持生态系统稳定、调节气候具有重要作用，但是目前森林物候对气候变化的响应机制仍存在争议，物候模型模拟精度较低，严重限制了陆地生态系统响应气候变化的认识。因此，深刻理解森林物候对气候变化的响应机制、提高森林物候模型的模拟精度，对于有效认识和应对气候变化、促进森林生态系统的可持续经营与管理具有重要意义。

本研究以北半球中高纬度（>30°N）森林生态系统为研究对象，以揭示森林物候对气候变化响应及其潜在机制为目标，基于遥感反演植被物候数据、地面观测物候数据和室内气候变化模拟实验，利用数理统计和模型模拟等方法，分析了2001—2019年北半球中高纬度森林春秋季物候和生长状况的时空演变特征，揭示了森林春秋季物候对气候变化的响应机制，并基于所发现的机制改进了森林秋季物候模型，结合CMIP6未来不同气候变化情景数据，对未来森林秋季物候的演变特征进行了预测。本研究得到的主要结论如下：

（1）2001—2019年，北半球中高纬度超过75%的森林春季物候提前，春季物候初期（SOS15）平均每年显著提前1.45 ± 0.08天，春季物候中期（SOS50）平均每年显著提前0.83 ± 0.05天。秋季物候初期（EOS10）与中期（EOS50）的年际变化趋势相反，秋季物候中期平均每年推迟0.34 ± 0.07天，但是秋季物候初期平均每年提前0.30 ± 0.10天。由于不同物候阶段对气候变化的响应差异，春季物候和秋季物候的发生过程平均每年分别减慢了0.91 ± 0.11天和0.59 ± 0.09天，生长季长度平均每年延长了0.33~1.42天。研究区增强型植被指数（Enhanced Vegetation Index，EVI）的年内波动幅度和EVI面积逐渐增加，EVI最小值的时间变化趋势存在空间差异，低纬度区域EVI最小值增大，高纬度区域EVI最小值减小。

（2）季前温度是春季物候的主要决定因子，季前温度分别在78.4%（遥感数据）和78.2%（站点数据）的区域为春季物候的主导因子，前一年秋季物候对第二年春季物候也有影响，但是以秋季物候为主导因子的像元/站点所占比例很小（遥感：4.3%；站点：8.1%）。大尺度气候变化模拟实验发现，春季物候对温度响应受年均温的非线性影响，当年均温为12°C左右时，春季物候对温度响应最强，在更暖或更冷的年均温下春季物候对温度响应减弱，这主要是由于冬季冷激以及冷激和光周期对春季物候的限制作用导致的。构建的春季物候温度响应模型显示，当前春季物候对温度响应的敏感区主要集中在中国中部，2080—2100年春季物候对温度响应的敏感区将向北移动4°N，这一变化将促进树种分布区向高纬度扩张。

（3）季前温度对秋季物候有显著影响，分别在31.4%和30.5%的区域为秋季物候初期和中期的主导因子，所占比例最高。季前降水和辐射在20.0%~25.4%的区域为秋季物候初期和中期的主导因子。秋季物候除了受到季前温度、辐射和降水等环境因子的影响，还受到当年春季物候的控制，春季物候在24.4%的区域为秋季物候初期的主导因子，仅次于温度，在20%的区域为秋季物候中期的主导因子，春季物候提前会导致秋季物候提前。季前环境因子对秋季物候初期和中期的影响存在差异，生长季早期温度升高、辐射增强、降水增多会导致秋季物候初期提前，但是生长季后期温度升高、辐射增强会导致秋季物候中期延迟。

（4）秋季温度和辐射耦合影响秋季物候，温度升高导致秋季物候延迟（r = 0.37），辐射增强导致秋季物候提前（r = -0.23）。低辐射强度下，秋季物候随温度升高而推迟，随辐射强度增加，秋季物候的温度敏感性降低，在高辐射条件下，秋季物候随温度升高而提前。实验研究进一步证实了该结论，辐射增强导致幼树秋季物候提前12.1~14.8天，结构方程模型发现，辐射增强不仅可以直接导致秋季物候提前，还可以通过降低植物叶片的叶绿素含量间接提前秋季物候。

（5）基于温度和辐射对秋季物候的影响机制，在秋季物候寒冷度日模型（CDD）和积冷-光周期模型（CDDP）中考虑了秋季辐射和温度对秋季物候的耦合作用，构建了积冷-辐射模型（CDDR）和温光-辐射模型（CDDPR）。改进后的CDDR模型和CDDPR模型模拟精度均显著提升，内部验证RMSE分别为8.6天和7.1天，比改进前的CDD模型和CDDP模型分别降低了3.7天和1.1天，外部验证模型模拟精度也显著提升，RMSE分别降低了1.8天和3.5天。模型真实值和模拟值的相关系数整体分别提升了0.16和0.38。基于最优的CDDPR秋季物候模型，结合未来不同排放情景气候数据预测了未来秋季物候的演变特征，发现中度和高度排放情景下秋季物候会继续延迟，平均每年延迟0.10天和0.21天。

本研究揭示了北半球中高纬度森林春秋季物候对气候变化的响应机制，研究结果为理解气候变化背景下森林的动态响应提供了重要理论支撑，对于优化森林生态系统管理，提高森林生态系统碳吸收能力具有重要意义。

Climate change has caused profound impacts on the Earth's systems, leading to frequent extreme climate events and declines in the stability of terrestrial ecosystems, seriously affecting human living. Climate warming led to earlier spring leaf unfolding and delayed autumn leaf senescence for temperate plants in the Northern Hemisphere. Phenological shifts not only influence the structure and function of terrestrial ecosystems, but also affect global and regional carbon and water cycles, as well as energy balance. Forests are the largest carbon sinks of terrestrial ecosystems, playing an important role in maintaining ecosystem stability and climate regulation, however, how forest phenology responds to climate change is still under debate. The limited precision in current phenology models hinders our understanding of how terrestrial ecosystems respond to climate change. Therefore, understanding the response mechanisms of forest phenology to climate change and improving the accuracy of phenology models are crucial for effectively identifying and mitigating climate change, as well as promoting sustainable forest ecosystem management.

Based on forest ecosystems in the mid-to-high latitudes (>30°N) of the Northern Hemisphere, this study aims to reveal the response of forest phenology to climate change and its mechanisms. Using satellite-derived vegetation phenology data, ground phenological observation data, and climate-controlled experiments, we investigated the spatiotemporal shifting characteristics of spring and autumn phenology in the forests of the Northern Hemisphere from 2001 to 2019 by employing mathematical statistics and model simulations. We unveiled the response mechanisms of spring and autumn forest phenology to climate change, improved the autumn phenology models, and utilized CMIP6 datasets to forecast future changes in autumn phenology. The main conclusions of this study are:

1. From 2001 to 2019, spring phenology in over 75% of forests at the middle and high latitudes of the Northern Hemisphere significantly advanced. The onset of spring phenology (SOS15) advanced by an average of 1.45 ± 0.08 days per year, the mid-spring phenology (SOS50) advanced by 0.83 ± 0.05 days per year. The onset of autumn phenology (EOS10) and mid-autumn phenology (EOS50) exhibited contrasting temporal trends: EOS50 delayed by 0.34 ± 0.07 days per year, but EOS10 advanced by 0.30 ± 0.10 days per year. Owing to varied responses to climate change at different phenological stages, the rate of progression of spring and autumn phenology decelerated by 0.91 ± 0.11 days per year and 0.59 ± 0.09 days per year, respectively. The growing season length extended by an average of 0.33~1.42 days per year. The intra-annual fluctuation amplitude of the Enhanced Vegetation Index (EVI) and EVI area progressively increased. The temporal trend of the minimum EVI varied spatially, with an increase in lower latitude regions and a decrease in higher latitude regions. The temporal trend of the minimum EVI varies by latitudes, with increases in low-latitude areas and decreases in high-latitude regions.

2. Preseason temperature emerged as the key driver of spring phenology, being the dominant factor in 78.4% of regions according to satellite data and in 78.2% per ground observations. Additionally, the autumn phenology of the previous year also affected spring phenology, though to a lesser extent, serving as the dominant factor in 4.3% (satellite) and 8.1% (ground) of the study area. Large-scale climate control experiments indicated that the responsiveness of spring phenology to temperature was nonlinearly affected by mean annual temperature. The highest response rate was at a mean annual temperature of ~12°C and lower response rates were at warmer and colder temperatures, mainly due to the winter chilling accumulation and the limiting effects of chilling and photoperiod on spring phenology. The spring phenological responsiveness model indicated that trees are currently most responsive in central China, which corresponds to the species’ main distribution area. Under a high-emission scenario, a 4-degree latitude shift in the responsiveness maximum toward higher latitudes was predicted over the rest of the century.

3. Preseason temperature significantly affected autumn phenology, being the dominant factor for 31.4% and 30.5% of the regions for EOS10 and EOS50, respectively. Preseason precipitation and radiation dominated in 20.0%-25.4% of the regions for EOS10 and EOS50. Beyond preseason temperature, radiation, and precipitation, autumn phenology was also affected by spring phenology. In 24.4% of regions, spring phenology was the dominant driver for EOS10 -second only to temperature- and it was the dominant factor for EOS50 in 20% of regions, indicating that earlier spring phenology leads to earlier autumn phenology. The effect of preseason environmental factors on EOS10 and EOS50 varies: while the increase in temperature, radiation, and precipitation during the early growing season advances the onset of autumn phenology, the increase in temperature and radiation in the later growing season results in a delayed mid-autumn phenology.

4. Autumn temperature and radiation interactively affect autumn phenology. Increasing temperature delayed autumn phenology (r = 0.37), whereas increased radiation advanced autumn phenology (r = -0.23). Under low radiation conditions, warming delayed autumn phenology, yet the temperature sensitivity of autumn phenology decreased with increasing radiation, and under high radiation conditions, autumn phenology even advanced with warming. Experiments yielded consistent results, increasing light intensity advanced autumn phenology of saplings by 12.1-14.8 days. Structural equation modeling revealed that increased radiation could advance autumn phenology both directly and indirectly by reducing chlorophyll content in plant leaves.

5. Utilizing insights on how temperature and radiation affect autumn phenology, the cold degree days (CDD) and cold degree day-photoperiod (CDDP) models were improved by incorporating the interactive effects of autumn radiation and temperature. This led to improved cold degree day-radiation (CDDR) and cold degree day-photoperiod-radiation (CDDPR) models. The CDDR and CDDPR models significantly improved the accuracy of autumn phenology predictions. In comparison to the CDD and CDDP models, their internal validation Root Mean Square Errors (RMSEs) were reduced by 3.7 days and 1.1 days, reaching 8.6 and 7.1 days, respectively. External validation also revealed substantial enhancements in simulation precision, with RMSE reducing by 1.8 days and 3.5 days. Moreover, the correlation between the observed and simulated EOS increased by 0.16 and 0.38, respectively. Leveraging the advanced CDDPR model with future climate datasets, it was predicted that autumn phenology will continue to delay under both moderate and high-emission scenarios, with an average delay of 0.10 and 0.21 days per year, respectively.

This study revealed the response mechanisms of spring and autumn forest phenology in the mid-to-high latitudes of the Northern Hemisphere to climate change. These findings provide crucial theoretical support for understanding the dynamic responses of forests under climate change conditions. This understanding is vital for optimizing forest ecosystem management and enhancing the carbon sequestration capacity of forest ecosystems.