Winter warming events (WWEs) are short-lasting events of unusually warm weather, occasionally combined with rainfall, which can cause severe ecosystem impacts by altering ground temperatures and water fluxes. These impacts are generally overlooked in large-scale ecosystem models. The frequency and intensity of WWEs will likely increase further in the future. We used an ecosystem model, LPJ-GUESS, to investigate the responses of four subarctic ecosystems to different levels of predicted WWEs, and identify model gaps hindering accurate estimates of these responses. In response to WWEs, the model simulated substantial ground cooling (up to 2 °C in winter) in contrast to the observed warming, leading to changes in biogeochemical fluxes often comparable in magnitude to those from altered winter climatologies. The mismatch between the modelled and the observed ground temperature changes may be due to the 1) lacking surface energy balance, 2) daily timestep, and 3) simplistic water retention scheme in LPJ-GUESS.

The Soil and Water Assessment Tool (SWAT) model has been widely applied for simulating the water cycle and quantifying the influence of climate change and anthropogenic activities on hydrological processes. A major uncertainty of SWAT stems from poor representation of vegetation dynamics due to the use of a simplistic vegetation growth and development module. Using long-term remote sensing-based phenological data, we improved the SWAT model’s vegetation module by adding a dynamic growth start date and the dynamic heat requirement for vegetation growth rather than using constant values. We verified the new SWAT model in the Han River basin, China, and found its performance was much improved in comparison with that of the original SWAT model. Specifically, the accuracy of the leaf area index (LAI) simulation improved notably (coefficient of determination (R2) increased by 0.193, Nash–Sutcliffe Efficiency (NSE) increased by 0.846, and percent bias decreased by 42.18%), and that of runoff simulation improved modestly (R2 increased by 0.05 and NSE was similar). Additionally, we found that the original SWAT model substantially underestimated evapotranspiration (Penman–Monteith method) in comparison with the new SWAT model (65.09 mm (or 22.17%) for forests, 92.27 mm (or 32%) for orchards, and 96.16 mm (or 36.4 %) for farmland), primarily due to the inaccurate representation of LAI dynamics. Our results suggest that accurate representation of phenological dates in the vegetation growth module is important for improving the SWAT model performance in terms of estimating terrestrial water and energy balance.