Assessing the Performance of Earthquake Hazard Maps

Brooks, Edward Max

doi:https://doi.org/10.21985/n2-m1xp-rr94

Work

Assessing the Performance of Earthquake Hazard Maps

Public

Download PDF

Download All Files (.zip)

Despite advancing knowledge about the mechanics of earthquakes, earthquake prediction remains, and will likely remain, an unsolved problem. Hence in order to reduce the risk posed by earthquake shaking, seismologists have developed tools called earthquake hazard maps. Earthquake hazard maps communicate expected future shaking scenarios, and are used by engineers to develop building codes and build safer structures to minimize their chance of failure or collapse in an eventual earthquake. However, recent large earthquakes that caused great damage in areas predicted to be relatively safe illustrate the importance of criteria that assess how well earthquake hazard maps forecast shaking. This thesis defines metrics that measure the effects of over-prediction and under-prediction by quantifying overall performance, making it easy and straightforward to compare the performance of different maps. These metrics consider different aspects of performance, including the probabilistic nature of some maps, and spatial variations of predictions. Although no single metric alone fully characterizes map behavior, using several metrics can provide useful insight for comparing and improving hazard maps. I use these metrics to explore the performance of earthquake hazard maps made around the world. In Italy, I compare the performance of competing modeling procedures as a demonstration of how maps can be directly compared via the metrics. In Japan, I address criticisms of hazard mapping following the failure of the maps during the 2011 Tohoku earthquake by exploring the performance of simple, synthetic maps that describe uniform or randomized hazard. By the metric implicit in probabilistic seismic hazard assessment, the primary form of map-making worldwide, uniform maps are found to perform better than the original maps. This conclusion motivates additional research addressing over-parameterization, which finds that map performance may be improved by smoothing the predictions from the hazard models. In the United States, I assess the performance of one-year hazard maps made to describe the hazard from induced seismicity. The performance of the 2016 and 2017 maps prove to be better than any other map studied with the metrics I have defined, justifying the procedure for making short-term maps. Finally, I define a Monte Carlo simulation procedure to create artificial realizations of shaking scenarios and expand the utility of the metrics, outlining how they can be used as absolute, rather than relative, measures of performance.

Creator