Calibrating a Runoff Generation Scheme at Global Scale
Accurately simulating the runoff process is challenging for Earth system models (ESM). Model calibration generally seeks to improve model performance, but high computational costs make calibrating ESMs at large scales impractical. In this work, researchers used an uncertainty quantification framework to efficiently calibrate the runoff generation processes in the Energy Exascale Earth System Model (E3SM) at a global scale. The calibrated model performed better than the model using default parameters. Additionally, the uncertainty associated with the simplified mathematical representations in models, known as parameterizations, is significantly constrained.
Runoff is an essential source of fresh water, and its variability has profound socio-economic impacts. However, the uncertainty in modeling different variables makes it difficult for current ESMs to accurately represent runoff. This study uses a novel framework to reduce runoff-related uncertainty in E3SM. The calibrated variables can be used to project changes in runoff caused by climate change with high confidence. This novel framework can be further applied to calibrate other processes of ESMs at large scales, which is not computationally feasible with traditional methods.
Runoff is a critical component of the terrestrial water cycle and ESMs are essential tools for studying its spatio-temporal variability. Runoff schemes in ESMs typically include many parameters, so model calibration is necessary for improving the accuracy of simulated runoff. However, runoff calibration at a global scale is challenging due to its high computational cost and a lack of reliable observational datasets. In this study, researchers calibrated 11 runoff-relevant parameters in the E3SM Land Model using a surrogate-assisted Bayesian framework. The results show that model performance is significantly improved when the inferred parameter values from the calibration are used. Although the parametric uncertainty of simulated runoff is reduced after parameter inference, it remains comparable to the multi-model ensemble uncertainty represented by other commonly used global hydrological models. Additionally, the annual global runoff trend observed during the simulation period is not well constrained by the inferred parameter values. This suggests the importance of including parametric uncertainty in future runoff projections.