SCEC Award Number 13145 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Trimming of the UCERF 3 Logic Tree: Assessing the Value of Hazard Information with Expected Annualized Loss and Portfolio Loss Exceedance Curves
Investigator(s)
Name Organization
Keith Porter University of Colorado, Boulder
Other Participants Thomas Nelson
SCEC Priorities SCEC Groups WGCEP, EEII
Report Due Date 03/15/2014 Date Report Submitted N/A
Project Abstract
 This study summarizes an effort to quantify the sensitivity of societal risk to branches in the logic tree of the Uniform California Earthquake Rupture Forecast version 3. It uses a deterministic sensitivity study sometimes called a tornado-diagram analysis to identify logic-tree branches that matter most to the probability distribution of expected annualized loss (EAL) to California woodframe single-family dwellings. The value of EAL varies between branches of UCERF3. Each branch of the logic tree has an associated marginal probability (in the Bayesian sense) or weight (in the language of frequentists), so EAL has a probability distribution. We identified the branches that matter most to that distribution: probability model, ground motion prediction equation (not an element of UCERF3, but important to EAL), total magnitude-5 rate, and choice of off-fault spatial seismicity probability density function. One can fix the other 5 parameters at arbitrary baseline values without a large bias in CDF of EAL. Varying only these 4 rather than 9 parameters, one can reduce by 99.6% the computational effort required to calculate the distribution of EAL without a substantial difference in the results.
Intellectual Merit Field et al. (2014) thoroughly quantified uncertainty in UCERF3. But there are situations where the computational burden can be too high to quantify risk in light of UCERF3's full logic tree. It can be valuable therefore to reduce the complexity of the UCERF3 logic tree, but necessary to do so without either biasing the probability distribution of the measure of risk. We found in previous tree-trimming work on UCERF2 that it is practical to reduce the computational burden by 2 orders of magnitude without introducing significant bias in statewide EAL or significantly underestimating uncertainty. The present project repeated that study for UCERF3. We found that one can eliminate 99.6% of the computational effort with a minor increase (4%) in the mean value of a measure of statewide risk and almost no change to the coefficient of variation (a 1% increase).
Broader Impacts Prior work showcased the value of tornado diagram analyses, and PI has seen others within SCEC using this simple but powerful tool. PI expects one of the benefits to society to be a better understanding of UCERF3's many uncertainties, especially understanding of which ones matter from a social and economic perspective. This knowledge will facilitate uptake of UCERF3 within the loss-estimation community.
Exemplary Figure Figure 5