Public Abstract

DE-SC0019438: Multi-model and Multi-scale Global Sensitivity Analysis for Identifying Controlling Processes of Complex Systems

Award Status: Inactive

Institution: Florida State University, Tallahassee, FL
UEI: JF2BLNN4PJC3
DUNS: 790877419

Most Recent Award Date: 07/10/2019
Number of Support Periods: 1
PM: Hnilo, Justin

Current Budget Period: 09/15/2018 - 07/31/2020
Current Project Period: 09/15/2018 - 07/31/2020
PI: Ye, Ming

Supplement Budget Period: N/A

Public Abstract

Multi-model and Multi-scale Global Sensitivity Analysis for Identifying Controlling Processes of Complex Systems Ming Ye, Florida State University, (Co-Investigators), Gary P. Curtis, USGS, Li Li, Pennsylvania State University Subsurface biogeochemical systems are open and complex, involving a large number of physical, chemical, and biological processes and their interactions at multiple scales in space and time. It is difficult, if not impossible, to understand all the processes and their interactions. On the other hand, since the dynamics of subsurface biogeochemical systems are determined mainly by controlling processes, understanding the controlling processes can lead to predictive understanding of subsurface biogeochemical systems. Therefore, identifying controlling processes is always the first step for gaining predictive understanding. The identification, however, is challenging because of uncertainty inherent in system processes. For example, it is always uncertain how to quantitatively represent a system process, as a process may be represented by several plausible conceptual-mathematical models. For a given conceptual-mathematical model of a process, its parameters that characterize the process are always not known deterministically. These uncertainties can be reduced to certain extent by collecting more data and gaining more knowledge, but cannot be fully removed due to system complexity and limitation on quantity and quality of collected data. Therefore, a research question of immense importance is how to identify the controlling processes of subsurface biogeochemical systems under uncertainty. This project aims at developing a new approach to identify controlling processes of complex systems such as groundwater-surface water transition zones, in which intricate hydrologic, microbiologic, and geochemical processes occur and interact to affect the hydro-biogeochemical behaviors at multiple scales (e.g., from local to reaches and to reach networks). The identification of controlling processes is necessary for the development and improvement of mechanistically-based hydro-biogeochemical models. Taking spatial variability of hydrofacies (a geological process) and river stage dynamics (a hydrological process) as examples, if the former is more important than the latter, limited research resources (money and time) should be spent to better characterize spatial variability of hydrofacies. The identification of controlling processes faces the theoretical and computational challenges as follows: (1) How to take into account of the inherent uncertainty in conceptualizing and modeling individual processes? (2) How to quantitatively and explicitly measure the relative importance of a large number of individual processes? (3) How to identify controlling processes in a computationally efficient manner for large-scale, computationally expensive models? The goal of this project is to address these three challenges by developing a new method of global sensitivity analysis for identifying controlling processes at multiple scales to support development and improvement of mechanistically-based models. The overarching scientific question to be answered in this project is as follows: If we are not certain about the choice of process models and model parameters, can we correctly identify the controlling processes of a complex system? To answer this question, this project introduces the concept of multiple working hypotheses into the identification of controlling processes to explicitly take into account the uncertainty in conceptualizing and simulating individual processes. By using the methods of global sensitivity analysis, this project will define a new process sensitivity index as a summary measure of relative process importance for individual processes. The new index is evaluated using the state-of-the-art methods of sparse-grid collocation approaches that have been demonstrated to be computationally efficient for sensitivity analysis. The new index lays out a foundation for integrated model-and-data analysis in future studies. By collaborating with scientists at the Pacific Northwest National Laboratory (PNNL), this project will be focused on answering scientific questions at the PNNL scientific focus area (SFA). By collaborating with scientists at U.S. Geological Survey working at the Naturita Site in Colorado (a DOE UMTRA Title I site) and scientists at the Pennsylvania State University working at the Shaver’s Creek Site in Pennsylvania (the extended site of the NSF Susquehanna Shale Hills Critical Zone Observatory), this project will also provide broader perspectives for other sites of importance to DOE and the nation. This project is complementary to the previous and on-going research conducted at the PNNL SFA, and the research results of this project can be used to confirm existing insights and to gain new insights for advanced scientific understanding of groundwater-surface water transition zones. The proposed methods of global sensitivity analysis is mathematically general, and are expected to be applicable to a wide range of subsurface biogeochemical systems.

Multi-model and Multi-scale Global Sensitivity Analysis for Identifying Controlling Processes of Complex Systems

Ming Ye, Florida State University,

(Co-Investigators), Gary P. Curtis, USGS, Li Li, Pennsylvania State University

Subsurface biogeochemical systems are open and complex, involving a large number of physical, chemical, and biological processes and their interactions at multiple scales in space and time. It is difficult, if not impossible, to understand all the processes and their interactions. On the other hand, since the dynamics of subsurface biogeochemical systems are determined mainly by controlling processes, understanding the controlling processes can lead to predictive understanding of subsurface biogeochemical systems. Therefore, identifying controlling processes is always the first step for gaining predictive understanding. The identification, however, is challenging because of uncertainty inherent in system processes. For example, it is always uncertain how to quantitatively represent a system process, as a process may be represented by several plausible conceptual-mathematical models. For a given conceptual-mathematical model of a process, its parameters that characterize the process are always not known deterministically. These uncertainties can be reduced to certain extent by collecting more data and gaining more knowledge, but cannot be fully removed due to system complexity and limitation on quantity and quality of collected data. Therefore, a research question of immense importance is how to identify the controlling processes of subsurface biogeochemical systems under uncertainty.

This project aims at developing a new approach to identify controlling processes of complex systems such as groundwater-surface water transition zones, in which intricate hydrologic, microbiologic, and geochemical processes occur and interact to affect the hydro-biogeochemical behaviors at multiple scales (e.g., from local to reaches and to reach networks). The identification of controlling processes is necessary for the development and improvement of mechanistically-based hydro-biogeochemical models. Taking spatial variability of hydrofacies (a geological process) and river stage dynamics (a hydrological process) as examples, if the former is more important than the latter, limited research resources (money and time) should be spent to better characterize spatial variability of hydrofacies. The identification of controlling processes faces the theoretical and computational challenges as follows: (1) How to take into account of the inherent uncertainty in conceptualizing and modeling individual processes? (2) How to quantitatively and explicitly measure the relative importance of a large number of individual processes? (3) How to identify controlling processes in a computationally efficient manner for large-scale, computationally expensive models? The goal of this project is to address these three challenges by developing a new method of global sensitivity analysis for identifying controlling processes at multiple scales to support development and improvement of mechanistically-based models.

The overarching scientific question to be answered in this project is as follows: If we are not certain about the choice of process models and model parameters, can we correctly identify the controlling processes of a complex system? To answer this question, this project introduces the concept of multiple working hypotheses into the identification of controlling processes to explicitly take into account the uncertainty in conceptualizing and simulating individual processes. By using the methods of global sensitivity analysis, this project will define a new process sensitivity index as a summary measure of relative process importance for individual processes. The new index is evaluated using the state-of-the-art methods of sparse-grid collocation approaches that have been demonstrated to be computationally efficient for sensitivity analysis. The new index lays out a foundation for integrated model-and-data analysis in future studies. By collaborating with scientists at the Pacific Northwest National Laboratory (PNNL), this project will be focused on answering scientific questions at the PNNL scientific focus area (SFA). By collaborating with scientists at U.S. Geological Survey working at the Naturita Site in Colorado (a DOE UMTRA Title I site) and scientists at the Pennsylvania State University working at the Shaver’s Creek Site in Pennsylvania (the extended site of the NSF Susquehanna Shale Hills Critical Zone Observatory), this project will also provide broader perspectives for other sites of importance to DOE and the nation. This project is complementary to the previous and on-going research conducted at the PNNL SFA, and the research results of this project can be used to confirm existing insights and to gain new insights for advanced scientific understanding of groundwater-surface water transition zones. The proposed methods of global sensitivity analysis is mathematically general, and are expected to be applicable to a wide range of subsurface biogeochemical systems.