Remote sensing for collecting building inventory data for risk assessment
Funded by Willis Research Network, Cambridge Architectural Research Ltd.
Duration of project: 2007- Current

Background Aims of the study Case Study: Pylos, Greece (SEAHELLARC) Contact Project-Related Publications Please contact Dr. Keiko Saito with enquiries about this project

Figure 1. Aerial photograph/vector map of Basse Terre, Guadeloupe. Copyright Institut Geographique National (IGN), 2008.

The aim of the research at Cambridge University Centre for Risk in the Built Environment (CURBE), Department of Architecture, is to investigate the use of remotely sensed data to create building inventory data for vulnerability assessment. Knowledge on the location and type of (insured) exposure, particularly buildings, is a key element in assessing the risk from natural disasters. The current practice of collecting exposure data either relies on the data that the primary insurers collect, or the industry exposure database that is compiled and provided by the vendors of the commercially available catastrophe models. As a result the data availability and its quality can vary from country to country, region to region. During discussions with the insurance industry through the Willis Research Network meetings it has been identified that methods to create global exposure databases that are consistent in their quality and geographical scale would be valuable, which will allow the risks from perils to be compared across territories. The aim of the work carried out at the Department of Architecture, university of Cambridge is to streamline the processes involved in capturing data on exposure, with a particular focus on buildings, utilising the latest remotely sensed imagery and image processing techniques as well as available geospatial datasets.

Aim of the study

The required spatial resolution for the exposure datasets can vary according to the peril or the geographical scale that the user is interested in. These scales typically ranges from country-wide, regional to local. Remote Sensing provides an unparalelled data source that potentially allows the collection of data of consistent quality for anywhere in the world. In the past few years, the technology behind the satellite sensors have imporved significantly, allowing us to consider methods that focus on building level data, as well as medium scale data. However data availability is not guaranteed for everywhere, and the cost of acquiring the data can sometimes be prohibiting. The ultimate goal of the study is to produce a streamlined methodology that allow a consistent exposure dataset to be created using available geospatial datasets.

One of the key outputs from this project will be a methodology that allows the creation of building inventory data using a standardised dataset. Assuming that the four key physical attributes of a building that are mandatory for risk assessment are

• Footprint (area)
• Height
• Construction type
• Age

Another key output is a knowledge-base type decision tree that helps the user identify the optimum method of creating a building inventory dataset given the available geospatial data for the region of interest. Depending on the data availability, several options may exist in terms of creating building inventory data using geospatial datasets. For instance, building footprints may be available from the national mapping agency, or from LiDAR data or aerial photography. If all three are available, which would be the most appropriate dataset for the user? The decision tree will take into account issues such as the data available, level of accuracy required/achievable, the cost of acquiring and processing the data, processing time required, and select the optimum method for the user. Figure 2 shows the potentially available datasets and the building parameters that need to be extracted, as well as a short list of some potentially useful information that are likely to have an effect on the morphology of the built up areas in various countries.

Figure 2. Table showing potential remote sensing datasource against building parameters.

A case study has been conducted in Pylos, Greece, in conjunction with the EU funded project SEAHELLARC.

The height of the buildings were derived by measuring the façade length of the buildings in the image. A preliminary investigation using 90 buildings demonstrated that the height extracted using the façade length is more accurate than the length derived using the shadow length (Figure 3) . This may be due to the fact that the walls of the building s in Pylos are white and can be distinguished clearly from the roofs.

The footprints of the buildings were manually derived for all buildings in Pylos. The footprints as well as the height data for all buildings (900+) was compared against ground validation data collected as part of the SEAHELLARC project. A building-by-building ground survey was carried out during summer 2008 by Antonios Pomonis and Maria Gaspari (SEAHELLARC). The four key physical parameters of the buildings, as well as occupancy, proximity and condition were recorded. This dataset was used to validate the parameters derived using the image.

Figure 5 shows an example of a classification of the buildings in Pylos according to footprint area and height. By repeating this with other datasets of buildings around the world, and using more parameters such as roof shapes and irregularity of footprints, it is expected that a pattern will emerge that will allow inference rules to be made using certain building parameters to infer the construction type/age of buildings.