Understanding how ecosystems respond to climate extremes like severe droughts and unusually wet events is key to predicting future carbon emissions. Ecosystem respiration (ER), the release of carbon dioxide from plants and soils, is a major land carbon flux. Scientists use models to estimate ER, but most models are built using data with normal climate. In this study, we tested whether these models can still work during extreme climate events. We analyzed over 13 years of data from a grassland in northern China and found that models trained with normal-year data performed poorly during extreme droughts and wet years. We then adjusted the model parameters specifically for extreme conditions and found that they changed a lot, meaning that the way ecosystems function during extremes is quite different. This shows that using fixed model settings from normal years is not enough. To improve our ability to predict how ecosystems will behave as extreme climate events become more common, we need flexible models that can adjust their behavior depending on the climate conditions.