Identifying the Important Contributors to Model Variability in a Multiprocess Model

Researchers define a new sensitivity index to quantify the uncertainty contribution from each process under model structural uncertainty.

The Science

Earth system models consist of multiple processes, each of them being a submodel in the integrated system model. A research team, including scientists at Florida State University, Pacific Northwest National Laboratory, and Oak Ridge National Laboratory, derived a new process sensitivity index to rank the importance of each process in a system model with multiple choices of each process model.

The Impact

The new process sensitivity index tackles the model uncertainty in a rigorous mathematical way, which has not been dealt with in conventional sensitivity analyses. Accounting for model structural uncertainty in complex multiphysics, multiprocess models has been a long-recognized need in the modeling community.


Most of the processes in a multiprocess model could be conceptualized in multiple ways, leading to multiple alternative models of a system. One question often asked is which process contributed to the most variability or uncertainty in the system model outputs. Global sensitivity analysis methods are an important and often used venue for quantifying such contributions and identifying the targets for efficient uncertainty reduction. However, existing methods of global sensitivity analysis only consider variability in the model parameters and are not capable of handling variability that arises from conceptualization of one or more processes. This research developed a new method to isolate the contribution of each process to the overall variability in model outputs by integrating model averaging concepts with a variance-based global sensitivity analysis. The researchers derived a process sensitivity index as a measure of relative process importance, which accounts for variability caused by both process models and their parameters. They demonstrated the new method with a hypothetical groundwater reactive transport modeling case that considers alternative physical heterogeneity and surface recharge submodels. However, the new process sensitivity index is generally applicable to a wide range of problems in hydrologic and biogeochemical problems in Earth system models. This research offers an advanced systematic approach to prioritizing model inspired experiments.

Principal Investigator(s)

Ming Ye
Florida State University

Xingyuan Chen
Pacific Northwest National Laboratory (PNNL)


This work was supported by the U.S. Department of Energy, Office of Science, Office of Biological Research, Early Career Award and PNNL Subsurface Science Research Scientific Focus Area and ORNL Terrestrial Ecosystem Science Scientific Focus Area.


Dai, H., M. Ye, A. P. Walker, and X. Chen. 2017. “A New Process Sensitivity Index to Identify Important System Processes Under Process Model Uncertainty and Parametric Uncertainty,” Water Resources Research 53(4), 3746-90. [DOI: 10.1002/2016WR019715].