Predicting Ecosystem Respiration Under Climate Extremes Requires Varying Parameters

Ecosystem respiration (ER) is the second-largest terrestrial carbon flux, yet ecosystem models often fail to capture its variability under climatic extremes. The increasing frequency and severity of precipitation and drought extremes pose substantial challenges to accurately predicting ER. It remains unclear whether parameters calibrated under normal climates can reliably predict ER under extreme events. Here, we used long-term eddy covariance data from a semi-arid grassland to investigate the predictability of conventional linear and non-linear microbial models. Both models were parameterized using a Monte Carlo Markov Chain assimilation approach based on data from normal climatic years. However, both exhibited poor performance in simulating daily ER during extreme drought and wet years, due to significant parameter divergence between normal and extreme years. We derived model parameters for extreme drought and wet years, revealing pronounced divergence: all parameters in the linear and microbial model varied significantly between normal and extreme years, with ∼29% displaying high variability (coefficient of variation >0.3). Furthermore, principal component analysis revealed substantial parameter divergence among hydrological regimes. Sensitivity analysis showed 93% of parameters exhibit asymmetric responses in extreme drought and wet years. These results indicate that fixed parameters calibrated under normal climatic conditions cannot represent the emergent properties of ecosystems during extreme events. Our findings highlight that varying parameters are not merely a technical adjustment but a fundamental requirement for improving the predictability of ER under climate extremes.