Top-down or bottom-up in earthquake exposure modeling: a comparison of aggregated and building-by-building models

Earthquake building exposure models are descriptions of the type, monetary values and inhabitants existing in a determined geographical area. Building stock models, or aggregated exposure models, summarize these key values on a regional level and are an established part of the risk assessment chain. They exist on a continental (e.g. ESRM20 in Europe; SARA in South America) or a global scale (e.g. GEM Global Exposure Model; PAGER; GED4GEM). Such models are created from a combination of cadaster information, national statistics, built area proxies, census data and/or local expert knowledge. In each country the processing method (therefore the model) differs, based on the level and type of information available. The input information for aggregated exposure models may be outdated for regions that experience rapid developments, as national data collection as censuses take a large amount of effort and are only conducted every five to ten years.

The advent of global building footprint models through artificial intelligence (Open Buildings; Global ML Building Footprints) and the slow but steady increase of building footprint coverage in OpenStreetMap have provided opportunities to model key values from bottom-up. Such model is able to keep the global scale, but considers individual buildings rather than district totals. For each building, the maximum amount of information is gathered, based on the dataset itself and other global datasets containing relevant values (such as height or occupancy type). Structural, monetary and population values can be added based on the relative occurence of building types in the aggregated models. An example is the ’model of European buildings’.

Inevitably, a switch from a top-down to a bottom-up approach to exposure modeling brings advantages and disadvantages, apart from the obvious increase of resolution to the individual building scale that comes with building-by-building models. We have taken three case studies and compared the strengths and weaknesses of each of the approaches, such as building (type), population and monetary value distribution, recentness of the data and total floor space size. The findings help to identify future directions for exposure modeling and aim to find the best approach to capture the dynamic nature of the built environment.