Disaster damage estimation models: Data needs vs. ground reality

Integrated assessment models are being used extensively in the field of disaster damage estimation and assessment. However, there is a great deal of uncertainty involved with the use of these models---not only because of the uncertainty of predicting the occurrence of hazards but also because of the quality of data that are input into these models. The use of these models for real-world decision-making is limited by the data. Poor quality data can lead to poor decisions, particularly at a local level of analysis. This dissertation looks at the issue of model-data interaction and the uncertainty inherent due to the lack of good quality data. The above interaction is researched using the HAZUS(TM) model (a state-of-the-art damage estimation model) and focusing on building inventory data for two cities: City of Seattle, WA and City of Long Beach, CA. It assesses how the local level building inventory data compares with default building inventory data in HAZUS(TM) for the two cities above. Finally it looks at how changes in the building inventory data lead to changes in the damage estimation from HAZUS(TM). In order to understand patterns of variation, both of the above are analyzed at the full city level and at the level of census tracts comprising the cities. The dissertation finds that although a lot of basic GIS data exist for large cities at the local level, the building inventory data are severely lacking in some required information, accuracy and completeness. Where good data exist, the results show that there is a large variation in building inventory in the default data which leads to an even larger variation in damage estimation. All occupancy classes excepting residential are significantly underestimated and much of the underestimation is concentrated in the commercial, industrial, education and institutional classes. There is even large variation for downtown census tracts and single use census tracts such as ones with universities, etc. Where good data do not exist (as in the case of City of Long Beach), the use of local data is difficult and requires significant expertise and assumptions. In such cases, the use of HAZUS(TM) should be with a great deal of caution.