Development of an integrated global sensitivity analysis strategy for evaluating process sensitivities across single- and multi-models

Evaluating the process sensitivities is critical for development and improvement of many process-based hydrologic models, yet this task remains challenging due to the diverse process conceptualizations. In this study, we developed an integrated global sensitivity analysis strategy tailored for process sensitivity analysis under different process conceptualizations, where a process can be characterized by (i) a single uncertainty parameter, (ii) multiple uncertainty parameters, both ignoring the process model uncertainty, and (iii) multiple alternative process models and parameters. To accommodate these scenarios, the new strategy integrates three methods: the traditional Sobol’s variance-based sensitivity analysis (Saltelli et al., 2007) for individual model parameters, the extended Sobol’s sensitivity analysis (Mai et al., 2020) for grouped parameters, and process sensitivity analysis for multi-models (Dai et al., 2017, Yang et al., 2022), facilitating the assessment of process sensitivities. We illustrated the application of this strategy to complex biogeochemical models of trichloroethylene (TCE) degradation, based on a laboratory experiment. Four interactive processes, namely, flow, electrolysis, transport, and cometabolic processes, were conceptualized to simulate the biodegradation of TCE. Process model uncertainty was accounted within the cometabolic process, where four alternative process models were postulated, resulting in generation of multi-models. Each process model of the four processes also contained a varying number of uncertainty parameters, ranging from 1 to 4. To reduce computational cost, deep learning-based multi-layer perceptron (MLP) surrogate models were employed. The new strategy provides a comprehensive and systematic paradigm for evaluating process sensitivities across single- and multi-models, thereby advancing our understanding of hydrological processes. [Download paper here]

Recommended citation: Jing Yang, Yujiao Liu, Heng Dai*, Songhu Yuan, Tian Jiao, Zhang Wen, Ming Ye. "Development of an integrated global sensitivity analysis strategy for evaluating process sensitivities across single- and multi-models." Journal of Hydrology. 2024, 643: 132014. https://www.sciencedirect.com/science/article/abs/pii/S0022169424014100’