Keywords
Scenario Discovery, Deep Uncertainty, Robust Decision Making, Water Resources, Water Resources Planning and Management
Start Date
15-9-2020 6:00 PM
End Date
15-9-2020 6:20 PM
Abstract
Scenario Discovery (SD) is a vulnerability analysis method for identifying regions of system input parameters corresponding to system outputs of interest. SD samples plausible future States of the World (SOW), then evaluates system performance across these SOW. A space-partitioning algorithm explores the multi-dimensional input space of deeply uncertain factors in a modeled system, bounding regions, called boxes, where a system performance metric transgresses an acceptable threshold. SD benefits the broader decision-making process by objectively identifying what uncertain system conditions are important for system performance following exploratory modeling, as opposed to creating scenarios assumed relevant. However, use of SD for real-world water systems raises three implementational challenges. 1) Uncertainty in water systems often include time-variant inputs like monthly demand or streamflow hydrographs, but SD requires scalar values. 2) Numerous SD algorithms are available, but selecting an application-appropriate algorithm is non-trivial. 3) Candidate scenarios should be evaluated for interpretability, significance, and, more broadly, ability to inform decision makers. However, the available evaluation metrics often disagree on what uncertainty parameters are most significant, and guidelines for what constitutes a ‘useful’ scenario are largely non-existent. This research investigates methods of summarizing time-variant deep uncertainties into scalar values, reviews the available Scenario Discovery and evaluation methodologies, and, based on the review, applies SD to a reservoir operation policy on the Colorado River Basin, USA. We expect the results of this research to provide recommendations on choice of SD algorithm dependent on application, identify statistical characterizations of streamflow hydrographs significant in box definitions, and interpret conflicting scenario evaluation metrics to select decision-maker-relevant scenario boxes.
Evaluation of Scenario Discovery Methods for Multi-Reservoir System Planning
Scenario Discovery (SD) is a vulnerability analysis method for identifying regions of system input parameters corresponding to system outputs of interest. SD samples plausible future States of the World (SOW), then evaluates system performance across these SOW. A space-partitioning algorithm explores the multi-dimensional input space of deeply uncertain factors in a modeled system, bounding regions, called boxes, where a system performance metric transgresses an acceptable threshold. SD benefits the broader decision-making process by objectively identifying what uncertain system conditions are important for system performance following exploratory modeling, as opposed to creating scenarios assumed relevant. However, use of SD for real-world water systems raises three implementational challenges. 1) Uncertainty in water systems often include time-variant inputs like monthly demand or streamflow hydrographs, but SD requires scalar values. 2) Numerous SD algorithms are available, but selecting an application-appropriate algorithm is non-trivial. 3) Candidate scenarios should be evaluated for interpretability, significance, and, more broadly, ability to inform decision makers. However, the available evaluation metrics often disagree on what uncertainty parameters are most significant, and guidelines for what constitutes a ‘useful’ scenario are largely non-existent. This research investigates methods of summarizing time-variant deep uncertainties into scalar values, reviews the available Scenario Discovery and evaluation methodologies, and, based on the review, applies SD to a reservoir operation policy on the Colorado River Basin, USA. We expect the results of this research to provide recommendations on choice of SD algorithm dependent on application, identify statistical characterizations of streamflow hydrographs significant in box definitions, and interpret conflicting scenario evaluation metrics to select decision-maker-relevant scenario boxes.
Stream and Session
false