4 minute read

By Alexandre Banquet (Alexandre.Banquet@oecd.org), Centre for Entrepreneurship, SMEs, Regions and Cities (OECD)

Recent years have brought with them an extraordinary shift in the urban landscape. As people seek new opportunities and a better quality of life, the population living in cities ¹ has more than doubled over the past 40 years, from 1.5 billion in 1975 to 3.5 billion in 2015. This trend is expected to continue and is projected to reach 5 billion by 2050, corresponding to more than half of the global population (OECD/EC, 2020). To accommodate population growth, cities tend to either expand or densify, which can have both economic and environmental impacts by increasing mobility demand, CO2 emissions, energy consumption, and cost of services.

Monitoring land use patterns across OECD metropolitan areas

Understanding how a city expands and comparing this with population trends is essential for sustainable urbanisation. Therefore, a timely monitoring of land management is crucial. In a recent OECD study (Banquet et al., 2022), we leveraged an innovative approach based on publicly available satellite imagery and deep learning to monitor land use in OECD metropolitan areas in near-real time.

The amount of land used for residential, commercial, or industrial purposes varies greatly across OECD countries. Residential use in cities ranges from about 40-60 m2 per inhabitant in Korea, Colombia and Türkiye to about 450 m2 per inhabitant in Finland, the United States and New Zealand. Differences across metropolitan areas within the same country can also be stark, which is the case in Chile, the United States, and Australia.

The map below shows the amount of residential area per inhabitant in 2021 across OECD metropolitan areas. This shows a clear geographical divide in residential use, particularly evident between the United States and Canada, compared to Mexico and Colombia, as well as between Spain, Türkiye, Estonia and the rest of the European continent.

The role of satellite imagery and AI

Satellite data are being made increasingly available to the public and now provide a wide range of information on what is happening on planet Earth at increasing levels of detail. At the same time, AI and cloud computing advances have allowed for a wider use of satellite data.

Until recently, the application of satellite imagery was limited. Governments and private companies sold medium (10-60 meters) and high-resolution (lower than 5 meters) images at high prices. In 2008, the United States marked the start of a new era by releasing Landsat images for free to the public. The Landsat Programme has been continuously collecting images of the Earth for more than 50 years. In 2014, the European Space Agency (ESA) started deploying the Sentinel satellites, enabling us to monitor the Earth at a higher resolution (up to 10 meters) and more frequently (every 5 to 6 days).

Processing and analysing these data require large computing power that, until recently, affordable computers could not provide. Cloud computing capacity has increased significantly in recent decades, costs have decreased, and AI research has facilitated the automation of tedious tasks. This allows, for example, to better monitor vegetation, to map land cover and land use at scale, and to track greenhouse gases and air pollutants in the atmosphere.

Observing cities from space

How did we leverage satellite imagery and AI in this OECD study? We gathered Sentinel-1 and -2 satellite images for all OECD metropolitan areas. Sentinel-1 satellites are equipped with radar systems actively emitting electromagnetic radiation and measuring the returning signal. Sentinel-2 satellites are equipped with multispectral sensors passively capturing electromagnetic radiation reflected by the Earth’s surface and located in both the visible and infrared parts of the spectrum. The figure below shows examples of images obtained with these satellites for the city of Luxembourg. In the first image, the colours for each pixel correspond to the red, green, and blue bands of the Sentinel-2 satellites and reflect what a human eye would see from space. The two other images contain information beyond the visual spectrum. These images capture infrared and radar signals, offering valuable information on moisture content, land surface temperatures, and topography that would otherwise be imperceptible to the human eye.

Figure 1: Information captured by satellite imagery for the city of Luxembourg

Source: ESA Sentinel-1 and Sentinel-2

To monitor land use across OECD metropolitan areas, we built an AI model able to identify land use patterns from any Sentinel satellite image automatically. This model can map the exact location and extent of residential areas, commercial and industrial complexes, transport infrastructure, croplands, open spaces, bodies of water, and wetlands. The figure below shows an example of predictions obtained with this AI model for the metropolitan area of Bogota (Colombia).

Figure 2: Land use prediction pipeline

Given that Sentinel satellites generate new images every 5 to 6 days, we can monitor land-use in near real-time and assess how fast metropolitan areas are expanding yearly. The figure below shows satellite images for 2018 and 2021 for the town of Naas, located in the metropolitan area of Dublin (Ireland), and urban expansion detected by the AI model during that period. The AI can detect the expansion of residential complexes in the South as well as of commercial and industrial areas in the surrounding area.

Figure 3: Example of built-up expansion detected for the town of Naas, Ireland

As cities continue to attract more and more people, AI and satellite imagery are very powerful tools that can provide near-real time insights on their development and expansion across the globe. The indicators derived from this study are already available on the OECD metropolitan database and will be regularly updated.

Read more in the paper here: Monitoring land use in cities using satellite imagery and deep learning.

References

Banquet, A., et al. (2022), “Monitoring land use in cities using satellite imagery and deep learning”, OECD Regional Development Papers, No. 28, OECD Publishing, Paris, https://doi.org/10.1787/dc8e85d5-en.

OECD/European Commission (2020), Cities in the World: A New Perspective on Urbanisation, OECD Urban Studies, OECD Publishing, Paris, https://doi.org/10.1787/d0efcbda-en.

Schiavina M., Freire S., Carioli A., MacManus K. (2023): GHS-POP R2023A – GHS population grid multitemporal (1975-2030). European Commission, Joint Research Centre (JRC), PID: http://data.europa.eu/89h/2ff68a52-5b5b-4a22-8f40-c41da8332cfe, doi:10.2905/2FF68A52-5B5B-4A22-8F40-C41DA8332CFE

1. Cities are here defined as high-density places of at least 50,000 inhabitants.↩