Building Footprints Tutorial (ArcGIS Pro Version)

Runyu Gao (rgao18@jhu.edu), Xuechun Li (xli359@jhu.edu), Susu Xu (sxu83@jhu.edu)

Overview

This tutorial outlines the process of working with open-sourced building footprints, including downloading, converting, preprocessing, and merging data. Additionally, it covers generating building footprints in regions without open-source data, finding historical world basemap imagery, and rasterizing building footprints. The workflow included in this paper is initially built for the DisasterNet model (https://github.com/SusuXu/DisasterNet) Blue blocks indicate the geoprocessing tool to be used in ArcGIS, and yellow blocks represent the parameters for each tool. The underlined steps provide detailed guidance within this tutorial.

Performance Improvement

In our study region in Morocco, 186,715 pixels have buildings in the old OSM building footprint map, in the merged new building footprint map, 410,607 pixels have buildings.​ 237,452 pixels have buildings in the old OSM building footprint map but do not have buildings in the new merged building footprint map. The coverge increases about 119.9% in pixels in our study region in Morocco. This statistics suggests the great improvement of our workflow in ​the coverage of building footprint map. This tutorial is based on Google Open Building Dataset, The dataset contains 1.8 billion building detections, across an inference area of 58M km2 within Africa, South Asia, South-East Asia, Latin America and the Caribbean.​ It is more complete than OpenStreetMap Buildings and Microsoft Buildings in the covered areas as mentioned above.

Below is an example of the old and new merge building footprint example. The left one is the original OSM data; the middle one is the original Google data, and the right one is the merged result. As we can see, the accuracy and the coverage are much better

​

Tutorial Steps

1. Download and Convert Building Footprints

• Microsoft Building Footprints: Tutorial Link
• Google Open Buildings: Tutorial Link
• Convert the downloaded data to GeoJSON format.

2. Convert GeoJSON to Shapefile

• In ArcGIS Pro, use the JSON To Features Tool. Set the Geometry Type Parameter to Polygon.

3. Preprocess Google Open Buildings Data

• Filter by confidence score. Like extracting building footprints with a confidence score threshold of 0.85.

4. Merge Building Footprints from Different Sources

• Use the Join Feature tool to find intersections of different sources.
• Apply the Pairwise Erase tool to filter target sources by the intersection.
• Utilize the Merge tool to combine filtered sources into one.

5. Eliminate Overlaps in Merged Dataset

• To address overlaps, find the union of overlapping building footprints.
• First, apply the Repair Geometry Tool, then use the Pairwise Dissolving Tool to avoid errors.

6. Generate Building Footprints in Uncovered Regions

• Install the Deep Learning Library from Esri's GitHub.
• Download the Deep Learning Tool Packages:
  • USA: Package Link
  • China: Package Link
  • Africa: Package Link
  • Australia: Package Link
• Use the Detect Objects Using Deep Learning tool with specific parameters for input layer, suppression, target region coordinates, and cell size.
• Post-generation, apply the Regularize Building Footprint tool for polygonization, selecting the Right Angle only method as appropriate.

7. Find Historical World Basemap Imagery

• Access historical basemap imagery via Living Atlas.
• Download map tiles or open directly on Living Atlas for your target region and timestamps.

8. Rasterize Building Footprints

• Utilize the Polygon to Raster Tool.
• Clip to target extent and use the Con tool to generate a binary map.