With the extension of drought areas, decreasing soil moisture (SM) induces plant water stress and limits land evapotranspiration, and the sensitivity of land ecological systems may change. The changing climate and corresponding vegetation responses challenge the simulation and forecasting capabilities of Land Surface Models (LSMs). We combine FLUXNET data and the Community Land Model (CLM5.0) to analyse energy fluxes and gross primary productivity (GPP) at sites experiencing long-term and flash droughts across diverse climate and vegetation types. Overall, latent heat flux (LE) was underestimated, whereas GPP was underestimated at arid sites and overestimated at wet sites. The plant hydraulic stress module in CLM5.0 underestimates the soil water stress threshold in arid areas and overestimates the threshold in humid areas. On this basis, we derive a quantitative function. The model performed worst for annual LE, H and GPP at arid sites because of cumulative soil moisture errors. Further analysis revealed that LE and GPP were underestimated during the middle of the growing season because of exaggerated midday stomatal closure. Among the various vegetation types, the model performed best for evergreen broadleaf forests and worst for croplands. Contrasting responses to drought were observed across different vegetation types. We used power functions to quantify this process. Forests exhibited greater resilience due to stable physiological traits while grasslands and croplands exhibited greater sensitivity. The capacity for post-drought recovery varies substantially across vegetation types-a critical dynamic that remains poorly captured by the model, particularly for grasslands. This finding was consistently validated across multiple climate scenarios, including SSP1-2.6, SSP2-4.5, and SSP3-7.0. These findings highlight the critical role of land surface processes and provides insights for enhancing model parameterizations to better capture the complexities of plant-atmosphere interactions.