Terrestrial vegetation plays an important role in shaping the Earth’s climate due to its control on the global carbon cycle. Understanding and predicting vegetation phenology and biomass turnover into soil organic matter is therefore of great importance for our understanding and quantification of carbon exchange with the atmosphere, which varies seasonally. In the past, models of tree phenology have been developed extensively, but equivalent models for herbaceous ecosystems, which cover a significant area of the terrestrial land surface and provide many ecosystem services to humans, have been much more poorly developed for both the start and the end of season. These limitations may be due to spatially and temporally sparse observational data in grasslands, but more importantly, their distribution across a large range of climatic and environmental conditions, as well as a lack of understanding of underlying processes. It follows that a refined autumn phenology model for grasslands is a necessary component of land surface models (LSMs). Here we present a novel approach to grassland autumn phenology by introducing a general, dynamic leaf turnover model controlled by environmental conditions into the QUINCY (QUantifying Interactions between terrestrial Nutrient CYcles and the climate system) LSM and show that decoupling leaf senescence from growing season triggers improves site-level carbon dynamics in herbaceous systems globally. We tested the model at 59 sites with differing climates and show that our model was able to reduce errors in gross primary productivity (GPP) predictions as well as in the timing of the onset of leaf senescence, especially in seasonally dry and very cold sites. Our model is able to reduce the root mean square error (RMSE) of daily GPP at a seasonally dry site from 1.25 to 0.76 <em>g C m<sup>−2</sup> d<sup>−1</sup></em>. At a seasonally cold and light-limited site, RMSE decreased from 0.6 to 0.46 <em>g C m<sup>−2</sup> d<sup>−1</sup></em> and at a temperate, oceanic site, from 1.56 to 1.20 <em>g C m<sup>−2</sup> d<sup>−1</sup></em>. Our study provides a way forward towards general, non PFT or site-specific autumn phenology modules in LSMs, as well as improving predictions of carbon fluxes in grassland ecosystems globally.