5.2 Building footprint maps

Keywords: Building footprint, element at risk, cadastral map, high resolution images, OpenStreetMap

Authors: Dr. Manzul Kumar Hazarika, Syams Nashrrullah, Mujeeb Alam, Lixia Chen and Cees van Westen, Adityo Dwijananto, Ma Louisen Roxas

Links: Methodology Book section 5 2: Characterization of assets

Introduction

Buildings are one of the most important elements-at-risk for risk assessment. A building houses both assets and people, and the behavior and response of a building to a specific hazard determines the potential damages to be incurred as well as the number of people who might get injured or killed. A building footprint provides the outline of a building drawn along the exterior walls, with a description of the exact size, shape, and location of its foundation. Building footprint is the most basic information necessary for evaluating the vulnerabilities of a building for a specific hazard. It represents the total area of a building and provides a better description of its spatial characteristics compared to a point representation in terms of spatial location, form, distribution, floor space ratio, and relationship between buildings and other objects (topological, orientation, proximity, etc.). Once building footprints are available, attribute information such as building type, number of floors, use, occupancy, etc. can be added and used for vulnerability and risk analysis. The characterization of building attributes focusing on the impacts of flood and landslide can be found in Chapter 5.2 of Methodology Book. In the following sections, different sources and methods used for generating building footprint maps are explained. Examples of generating building footprint maps and attribute characterization for the Caribbean countries are also described here.

Objectives

  • Explain different ways and sources to obtain building footprint maps
  • Explain the methodology for generating building footprint maps using remote sensing data
  • Explain the methodology for generating building footprint maps using an open-source platform

Description

There are several ways to obtain building footprint maps, either collecting from the available dataset such as a cadastral map, creating a new dataset from ground survey, digitizing using remote sensing data, or digitizing using an open-source platform. Cadastral maps provide detailed information about real properties or parcels of land in a specific area usually for purposes of taxation, thus it may also contain information about building footprint. A ground-based surveying method is a common technique to generate cadastral maps. In some countries, a Land Information System (LIS) is developed as a GIS tool for cadastral maps, which consists of an accurate, current, and reliable land record detailing its associated attributes and spatial data (building footprint). This can be a very good source of high quality building footprint data. However, in many countries, the cadastral maps may not be available in the correct GIS format. They might be in a data format like AutoCad DXF, which does not have a topology and a complete segment around each polygon. Editing this data for several polygons is very difficult and time consuming. Furthermore, the cadastral map might be outdated, and detailed information about building characteristics (e.g. building type, building use, building materials, roof type, etc.) are often not included. It is also important to note that sometimes there are restrictions to access cadastral data due to complicated processes involving legal issues and security concerns against abuses especially on sensitive data such as taxes.

High resolution optical satellite images and aerial photographs are typical data sources to generate building footprint maps. Unlike cadastral maps, building footprints from satellite data are generally extracted from the ‘footprint’ of the roof which may be larger than the building itself. Consequently, the building footprint may have some errors and the exact footprint of the building itself is not known. High-resolution satellite images such as QuickBird, IKONOS, WorldView, and GeoEye are relatively costly. However, it is now possible to have free high-resolution satellite images from the Bing Map of Microsoft under the OpenStreetMap (OSM) project. In recent decades, other data sources such as Light Detection and Ranging (LiDAR) and Synthetic Aperture Radar (SAR) have made it possible to obtain building height information bringing in new opportunities and challenges on building footprint extraction.

Methods for generation of building footprints from remote sensing data

Building footprints can be delineated using manual, semi-automatic, or automatic methods. Manual on-screen digitizing can be a very time-consuming and labor intensive work. While there is no high level of expertise required to delineate building footprints, the level of experience of the operator does affect the speed of the digitization. Although it can be applied over a wide city area, the time necessary for the digitization makes it mostly appropriate at only over neighborhood/town level. A possible way to improve the scalability is to distribute the work among many GIS operators and collate the results in the end. This method provides great control over the results and the digitized building footprints can be of the highest accuracy.

As previously mentioned, generating building footprint maps for a large area manually is very costly and time consuming. Therefore, the development of semi-automatic or automatic building extraction methods has been explored in many studies. However, the automatic method for building extraction is not an easy task and with the current existing method, it is still impossible to achieve 100% accuracy. The complex shape of buildings and various compositional materials of roofs are the main reasons for these difficulties. In many cases, most simple rectangular building roofs can be detected correctly, while the extraction of building footprints with complex shapes are not satisfactory. The extraction of individual building footprints is even more complicated in urban areas where spaces between buildings are very close (e.g. slum areas, business districts) and there are many other objects in close proximity such as trees andpower lines may occlude the buildings’ rooftops.

There are various methods to generate building footprint maps automatically from high resolution satellite images based on the characteristics and geometric structure of buildings, including supervised and unsupervised classifications, edge detection methods such as Canny algorithm, Hough transformation, and object-based classification with image segmentation techniques. Other methods such as active contour model (snakes) and energy function have also been introduced to improve the extraction of complex buildings in urban areas. With the advancements of LiDAR technology, capable of generating 3D terrain data, buildings can be identified based on height, size, and shape information from point clouds. Methods such as plane fitting and region growing algorithm segmentation use LiDAR points to identify building roofs, trace building boundaries and regularize said boundaries.

The current trend of building extraction methods isa mixture of different data sources and various algorithms. High-resolution optical imagery and LiDAR are integrated for more accurate building extraction. Optical imagery provides spectral information, while LiDAR data contains height and intensity information. In some studies, LiDAR data is used to extract non-ground features, then removing the vegetation based on the Normalized Difference Vegetation Index (NDVI) derived from optical images. In some other studies, object-based classification of high resolution images is used to extract built-up areas and then the LiDAR point clouds are analyzed to shape roofs within the built-up areas. Although the time required for automatic building footprint extraction methods to generate the results is relatively less compared to the manual digitizing method, this method should be implemented by an expert with the necessary technical skills. Furthermore, there is still a lot of manual editing needed to generate satisfactory results and currently, the technique is not yet a substitute for manual interpretation.

Methods for generation of building footprints from OpenStreetMap

OpenStreetMap(OSM) is an open-source collaborative project to create free and editable maps of the world. This means that everyone can contribute and improve the map, and similarly, users can also make use of the data in OSM for whatever purpose for free. Its open-source nature allows users to modify and manipulate the downloaded data as needed. For example, you want to make a tourist objects map in your area, you can download OpenStreetMap data and make a tourist objects map out of it.

OSM is particularly useful in generating building footprints map because the data in OSM are relatively richer and more updated compared to other map providers such as Google. This is possible because every OpenStreetMap user can add, edit, or delete OpenStreetMap data anywhere anytime. If there are newly built/demolished buildings in an area, a user can easily add/delete the building in OSM. However, data availability and data updates heavily depend on OpenStreetMap users therefore the richness and completeness of OSM data in various places might also vary depending on the focus of user contributions.

OpenStreetMap can, and has, been used for a wide variety of purposes - from disaster response to commercial use. The first organized use of OSM in disaster response was following the 2010 Haiti Earthquake. As high-resolution imagery of the affected area was made available to the public, over 600 individuals from the global OSM community began digitizing the imagery and tracing roads and other infrastructure. They made what quickly became the most detailed map of Port-au-Prince in existence, which was then used by search and rescue teams to help route supplies around the devastated capital and to coordinate many other aspects of the response and reconstruction effort.
In OpenStreetMap, you map using satellite imagery as your background map. OpenStreetMap is supported by several satellite imagery providers such as Microsoft Bing, ESRI, Mapbox, and Maxar. This support from satellite imagery providers has given OpenStreetMap’s community to keep growing until now.

Nevertheless, it is really easy to contribute to OpenStreetMap.You just need to create an account in the OpenStreetMap website first (www.openstreetmap.org), and then you can start to add data in OpenStreetMap through several ways: (i) contributing directly in the OpenStreetMap website using its built-in editor (iD Editor); (ii) using mobile applications available to iOS and Android users, and; (iii)by using the HOT tasking manager, a collaborative platform for the OSM community developed by the Humanitarian OpenStreetMap Team (HOT) — an international organization dedicated to humanitarian action and community development through open mapping.

If you want to contribute directly in the OpenStreetMap website, you just need to zoom into your area of interest and start editing your area using any of the several editors provided by OSM, such as iD editor and Potlatch 2 (in-browser editor) or Java OpenStreetMap (JOSM, desktop application). However, the downside in contributing directly to the OpenStreetMap website is not knowing whether there are other users who are editing in the same area that you are currently working on. When that happens, the chance of having conflicting data in OSM is very high, especially if you add/edit large features such as roads or land use. Luckily, there is a solution to prevent this problem, by using the collaborative platform of HOT Tasking Manager.

Figure 1. Mapping in OpenStreetMap using in-browser editor (iD editor).

The HOT Tasking Manager is a mapping tool designed and built for the Humanitarian OpenStreetMap Team’s (HOT) collaborative mapping process in OpenStreetMap. The purpose of the tool is to divide up a mapping project into smaller tasks that can be completed rapidly with many people working on the same overall area. It shows which areas need to be mapped and which of the areas that have been mapped needs to be validated. Imagine if you have a disaster-stricken area that needs to be mapped urgently. During your mapping, there are 30 mappers who want to help map the area, the HOTTasking manager will divide your area into smaller squares/grids to allow all 30 mappers to contribute to the map without worrying about having conflicts caused by mapping in the same area at the same time. In order to create a task for your area of interest, you need to become a project manager or contact HOT to help create the task. For information on Managing and Creating Projects on the HOT Tasking Manager, fill out a form here, http://bit.ly/TaskManagers, to receive training and permissions to become a TM Project Manager.

Figure 2: HOT Tasking Manager for collaborative mapping in OpenStreetMap

Example 1: Generation of a building footprint map for Dominica

We only had a building footprint map for the Roseau area. In order to be able to calculate building exposure, we applied a method for the generation of a building footprint map for the entire country. We used the satellite data as indicated in Table below, and we used a thresholding method for the three spectral bands, separating areas with high reflection.

Table 1: Satellite images used for building detection in Dominica

We combined the three masked areas into one single map, which was still overestimating the number of buildings and contained also other high reflection areas, like bare surfaces, road, quarries etc. Next, we converted this raster map into a polygon map, and the polygon map was again converted into a point map. This resulted in a large number of points, many of which didn’t represent buildings but other features. Therefore, we analyzed the point visually on top of a color composite image of the satellite image. The resulting building map was developed as a point map, and was carefully checked. The Figure below shows an example for a part of the country.

Figure 3: Building points derived through semi-automatic image classification and visual editing. Left: portion of Roseau which had building footprint polygons, together with the points of the buildings generated in this project for the whole island. Right: example from Pointe Michel of the building points on a high resolution image.

Example 2: Building attribute characterization for Grenada

The physical planning unit of Grenada provided building footprints for the whole country. However, no attribute information was available to determine, whether a certain building is a dwelling, market, hospital, a school, or some other structure. Without any attribute information, the usefulness of such data becomes limited. To solve this problem, a procedure was used to characterize buildings by classifying them based on their possible use. It was important to separate residential buildings from all other buildings so that population information can be attached only with residential buildings. Since no attribute information was linked with building footprints to make this distinction, the only possible option was to make use of the latest available satellite imagery and Google Earth, using visual image interpretation. High resolution imagery of Pléiades satellite was available for the whole island. The resolution of the multi-spectral image is 2 meters whereas panchromatic is 0.5 meters. Both images were fused to get the highest possible resolution with color. This provided a good quality data that could be used together with the vector building footprints to characterize the buildings. Building footprints were overlaid on the Pléiades satellite imagery in ArcGIS for visual interpretation. Two attributes ‘Use type’ and ‘occupancy’ were added in the buildings attribute table. Similarly, the building footprint map was exported to KML (Keyhole Markup Language) format to view using Google Earth.

Obviously, it was not easy to distinguish residential buildings from other buildings even on very high resolution imagery. Hence, the strategy was to identify and isolate large buildings, which could potentially be hotels, industries, schools, churches, offices, business centers, or supermarkets, etc. Snapshots of some examples of buildings identified using Google Earth are presented in Figure 4.

Figure 4. Snapshots of examples of various buildings identified: 1) supermarket & mall 2) beach hotel 3) University of Grenada 4) industries 5) hospital 6) school 7) national cricket stadium 8) beach resorts 9) sea port area, and 10) church (left) and cemetery (right) Source: Google Earth

After identifying these visible and known structures, the remaining buildings were considered to be residential. In Grenada, over 85% of residential houses are separate houses, therefore; one cannot expect a large population living in big buildings or apartments. Through visual inspection all large buildings and other obvious buildings like schools, forts, churches, etc. were identified and characterized (Table 2) manually. The remaining buildings were considered as residential houses and attributed ‘residential’.

Table 2: Key for the characterization of building footprints

Figure 3 shows the resulting building footprint layer with added attribute information. Of course, the attribute information should be improved through surveys on the ground. However, this is time consuming and generally requires a large team of people to do this. Such an attempt is currently underway in Saint Lucia. The building footprint maps with additional attribute information serve many purposes other than disaster mitigation and preparedness, and their collection and management should be an inter-departmental effort, also involving the statistics office. Eventually, a link with census information might also be obtained, as illustrated in Use Case 7.5.

Figure 5: Building footprint layer for Grenada with added attributes.

Example 3: Generating building data for Hanover, Jamaica using OpenStreetMap and HOT Tasking Manager

The 2020 Atlantic Hurricane Season began June 1st 2020. HOT has been requested by disaster preparedness and response actors to map buildings in Caribbean countries and other surrounding countries impacted by the hurricane season and the ongoing COVID-19 Pandemic. The data mapped by the community will help these actors to have a complete building layer for their common operational dataset. The goal of the task is to digitize buildings using Bing Imagery which was chosen for its high-quality imagery of the area.

Figure 6. Task project for Jamaica in support of the COVID-19 and Atlantic Hurricane disaster response. HOT created the task to complete the mapping of buildings in Jamaica.

Usually when mapping in OpenStreetMap, one maps for general features such as buildings, roads, and rivers. But, if an individual has good knowledge of the mapping area, they can map more than general features. In every task in the tasking manager platform, one can find the feature type that to map. In this task for Hanover, contributors only needed to map buildings. Every task in the tasking manager also has specific instructions about what imagery one needs to use, how to map the object, and other task-specific instructions which can be accessed within the task.

Figure 7. In-browser editor (iD editor) in the tasking manager website helps you map faster.

To start mapping in the tasking manager, begin by selecting the preferred grid. The tasking manager platform also already integrates with iD editor, meaning there is no need to switch tabs in the browser. After selecting the grid, start mapping with an imagery as a background. Every object drawn in OpenStreetMap needs some attribute as their identifier, in OpenStreetMap mappers usually tag each general building as building=yes this is further explained in Wiki OpenStreetMap (https://wiki.openstreetmap.org/wiki/Key:building). Other things to note: mappers need to square each building they map. Using an in-browser editor has advantages because when a mapper saves changes by clicking the Upload button, edits will appear in OpenStreetMap within a few minutes. With the help of the OpenStreetMap community, all the buildings in Hanover, Jamaica have been fully mapped. Of course, if one wants to improve the attribute for every building in Hanover, they can start conducting a field survey to collect the information from the ground.

Figure 8. A portion of the city of Hanover, Jamaica already fully mapped in OpenStreetMap. This came from the OpenStreetMap community who helped map the area using the HOT tasking manager.

With this completeness, users can start downloading the OpenStreetMap data including building footprints by using another platform called HOT Export Tool (https://export.hotosm.org/en/v3/). Users will have to use their OSM user credentials to log-in to the platform. Using this tool, one can set an area of interest and choose the type of data they would like to export in the data format that they prefer (e.g. shapefiles, mbtiles, geopackage, etc.). Depending on the breadth of the requested data, the export process can take several minutes hence the user will have to wait for an email notification that the data has been exported and is ready to be downloaded.

Figure 9. Exporting building footprints data from Hanover, Jamaica using HOT Export Tool.

References

Feng, T., Zhao, J. (2009). Review and comparison: Building extraction methods using high-resolution images. Second International Symposium on Information Science and Engineering.

Zeng, C., Wang, J., Lehrbass, L. (2013). An evaluation system for building footprint extraction from remotely sensed data. Applied Earth Observations and Remote Sensing Vol 6 (3), pp 1640-1652.

Vicini, A., Bevington, J., Esquivias, G. Iannelli, G-C., Wieland, M. (2014). User guide: Building footprint extraction and definition of homogeneous zone extraction from imagery. Global Earthquake Model (GEM) Technical Report.

Bhadauria, A.S., Bhadauria, H.S., Kumar, A. (2013). Building extraction from satellite images. IOSR Journal of Computer Engineering Vol 12 (2), pp 76.81.

van Westen, C.J., Alkema, D., Damen, M.C.J., Kerle N., Kingma, N.C. (2011). Multi-hazard risk assessment: Distance education course Guide book. United Nations University – ITC School on Disaster Geo-information Management (UNU-ITC DGIM).

www.openstreetmap.org 

https://tasks.hotosm.org/ 

Last Update: 22-01-2021