As the third pole of the Earth, the Tibetan Plateau (TP) is highly sensitive to climate change. Phenological changes in spring (i.e., the start of the growing season, SOS) over the TP are of much debate, while our understanding on the variation of autumn phenology (i.e., the end of the growing season, EOS) is still lacking. Given the significance of tracing the latest variation of EOS induced by the ongoing global change, we used the currently longest time series (1982–2018) of the normalized difference vegetation index of the Long Term Data Record from the Advanced Very High Resolution Radiometer (AVHRR LTDR NDVI) and investigated the responses of EOS to biological and climatic factors. We found that the EOS showed an insignificantly delayed trend at an average rate of 0.14 days·decade⁻¹ on the TP. This variation was found for all vegetation types except for alpine sparse vegetation. We further showed that spring SOS had the largest contribution to the interannual variations in EOS, where 46.5% of the vegetated areas showed significantly positive correlations between SOS and EOS (p < .05). Temperature (26.5%) and precipitation (18.8%) also had significant positive impacts on EOS, while the contribution of radiation was negative (11.3%). These correlations were regulated by local hydrothermal conditions, especially for precipitation gradients. Using the longest satellite data currently available, with the state‐of‐art data processing, we demonstrated a weakly delaying trend of EOS and the controlling role of spring phenology on the interannual variability of EOS, which is of great significance to improve our comprehension of the interactions between vegetation dynamics and climate change in the future.

Ecosystem models provide an effective approach to quantify the terrestrial carbon cycle, but the lack of accurate phenological information prevents them from better simulations of the physical processes. Compared with spring phenology (i.e. the start of the growing season, SOS), the vegetation phenology in autumn (the end of the growing season, EOS) is not well-simulated and it is challenging to incorporate vegetation phenology into ecosystem models. The simulation of EOS based on temperature and photoperiod was widely accepted, such as Delpierre et al (2009 Agric. For. Meteorol. 149 938–48)’s model (DM), yet its accuracy has not been fully discussed at a regional scale. Here, we developed a regional autumn phenological model (DMS) with inputs of temperature, photoperiod, and water availability for China’s terrestrial ecosystems. The new DMS model significantly improved the representation of EOS in terms of the lower root mean square error (RMSE), higher model efficiency, and a higher percentage of significant correlation with the referenced EOS. We observed widespread delaying trends of EOS with an average rate of 0.1 d yr⁻¹ for vegetated areas over 2001–2018. We further incorporated the improved EOS into the boreal ecosystem productivity simulator (BEPS) and found that the phenology-modified BEPS model had better performances in predicting annual gross primary productivity (GPP) with ∼28% lower RMSE than the original model when testing against GPP measurements from flux tower sites. From 2001 to 2017, the interannual GPP simulated by the modified BEPS model showed an increasing trend with a rate of 6.0 g C m⁻² yr⁻². In conclusion, our study proves that water availability is of great significance for modeling autumn phenology, and the incorporation of phenological dates into an ecosystem model is helpful for productivity simulation.