A Spatial Modeling Framework for Analyzing Potential Earthquake Damage: an Application to Memphis

Identifying options and making decisions to reduce natural hazard losses from large earthquakes rely on vulnerability and hazard assessments for a parcel of land, neighborhood, community, or region. These assessments are based on probabilistic estimates of damages; however, the damage probabilities and projections of economic losses contain uncertainties large enough to complicate mitigation policy decisions. The modeling framework developed here extends the use of scientific information for damage estimation by incorporating spatial uncertainties about the earthquake source and local geologic conditions. A well established methodology exists for evaluating probabilistic damage and loss estimates at a site. But what is less understood is how to aggregate site information into a risk analysis for a portfolio of sites (study area). The most straightforward approach is to assume spatially independent damages and losses. However, empirical scientific evidence indicates that levels of shaking and liquefaction at sites close to one another tend to exhibit some degree of positive correlation. While such correlations do not affect the expected losses in a given study area, they do increase the variance (uncertainty) of these losses. We demonstrate these effects in a neighborhood of more than 1,200 land parcels in Memphis, TN, by comparing simulated spatially independent and dependent losses resulting from a 7.7 magnitude earthquake. Repeated simulations on a regional scale yield a sampling distribution of total realized losses provide maximum-likelihood estimates for exceedance probability (EP) functions of loss. A value at risk (VaR) assessment based on these exceedance probabilities is then used to illustrate the effect of spatial uncertainty on a hypothetical mitigation plan decision in a Memphis neighborhood. A comparison of these assessment results for both spatially dependent and independent scenarios shows how a failure to account for unobserved spatial dependencies can lead to an underestimate of potential extreme damage and loss conditions. * Department of Electrical and Systems Engineering, University of Pennsylvania ** Western Geographic Science Center, US Geological Survey