Remote sensing has revolutionized how we measure and understand the Earth. We can now track deforestation across the globe, predict end-of-season crop yields and identify wildfires in near real-time. But exploration into its possibilities for urban areas has only just begun. Conventional land cover maps, which categorize the surface of the Earth into groupings like "forest" or "water" or "tundra," often lump urban areas into a single category, like "urban" or "built-up." This is useful for mapping urban extent generally, but does not capture the complexity of urban areas. Mapping land use, however, can reveal the patterns of a city, from road networks to housing availability to the distribution of commercial and industrial space — along with the jobs they represent — and much more.
Mapping urban land use over time can show what, where and when changes are happening: where the city is expanding or contracting; what patterns and densities characterize areas of growth or redevelopment; and which types of land use are being displaced. This information, for example, can highlight spatial inequality and other key issues for cities. Has housing kept pace with the population? Has infrastructure kept pace with development, and if not, where is it needed? Is growth encroaching on wetlands or farmlands that are critical to the city's well-being? Pressing questions like these abound for urban planners, officials and residents. Robust land use data can help provide the answers.
WRI, with support from the National Geographic Society, has developed methods and infrastructure to map urban land use in any city, providing a new tool to help cities manage their resources and improve quality of life. The results for cities in two countries, Mexico and India, illustrate what makes these new urban land use maps novel and how they can benefit cities.
Mapping Patterns of Life
More than half of the world's population now lives in cities, but there's still a lot we don't know about urban areas. This is especially true for rapidly growing cities in emerging economies, where expansion and redevelopment often outpace administrative capacity. From the city scale to the neighborhood scale, there is little rigorous, globally-consistent data on how cities are arranged, how they grow and change and how to improve them for the well-being of urban residents and the planet.
Detailed, global maps of land use within cities have, to date, simply not existed. Manyexistingproducts show land cover at the continental or global scale and are spatially coarse. Other, true land use remote products are based on expensive, relatively scarce high-resolution satellite imagery, and are thus limited to individual cities or even districts. Other land use maps are created by city planners through detailed, on-the-ground data collection to develop cadasters.
To address this information gap, we used data from the Atlas of Urban Expansion, remote sensing imagery, machine learning techniques and cloud computing tools — including Python libraries, Keras and the Descartes Labs Platform — to create computer vision models to categorize and map land use.
Our new urban mapping system is different: the resulting maps feature a detailed taxonomy of land use drawn at 5-meter resolution and require only medium-resolution satellite imagery, which is publicly available for the Earth’s entire surface and is constantly updated.
Thanks to the imagery archives of ESA’s Sentinel and NASA’s Landsat satellite constellations, we can retroactively map urban land use going back years. And because each data set captures the full globe at least a couple of times per month, our methodology can continually be applied to new imagery, creating an up-to-date and ever-richer dataset that can provide insights about how cities around the world are growing and changing. The outputs are publicly available so urban planners and other practitioners can start using them in research and to inform decisions immediately.
Insights from Hyderabad and Mexico City
Using these models, we mapped land use across multiple cities in Mexico and India, two rapidly urbanizing countries with major differences in their urban land use and development history. Despite having some of the largest cities globally, India is still mostly rural, with only 34% of its population living in urban areas. In contrast, 80% of Mexico’s population is considered urban.
Side-by-side comparisons of WRI’s land use maps (5 m spatial resolution) and the ESA CCI land cover maps (300 m resolution) illustrate the improved level of detail achieved in this new dataset. In the ESA land cover maps, both Hyderabad, India and Mexico City, Mexico look like giant blobs of “urban” across the landscape: they lack definition at the boundaries and offer little detail about the city’s interior landscape. Within the urban class, there is no information about areas with primarily residential versus primarily business activities, the location of transportation corridors or where infrastructure exists. All of these elements are critical to understanding the specific, local urbanization challenges faced by a city.
The finer-resolution maps developed through this model distinguish details in these “blobs,” revealing patterns and details invisible at coarser scales. For example, the largest non-residential regions in Mexico City are in the northern portion of the metropolitan area, where industrial uses and warehouses are common. Housing types that serve low-income people (informal, atomistic and residential projects) are also common near these non-residential districts in the north, as well as along the city’s periphery. Formal housing is predominant in the city center and south. The (in)famous sprawl of the Mexico City metropolitan area and the relative spatial segregation of land uses contributes to long commutes, especially for those living at the periphery, and is a significant contributor to the capital’s well-documented air pollution.
The pattern of informal settlements also distinctly differs between the two cities. In Mexico City, whose metropolitan population increased an average of 2.5% per year since 1990, rapid growth spawned both informal residential areas and “atomistic settlements” — areas that were not clearly subdivided and grew incrementally. These areas represent both planned developments to accommodate new residents (as evidenced in Chimalhuacan and surrounding municipalities, the large area of “informal subdivisions” in the southeast) and unplanned growth when housing efforts did not keep pace or meet residential needs (often atomistic settlements found primarily along the urban periphery and into the foothills of the mountains that surround the Valle de México). Many of these areas lack the basic infrastructure that other settlement areas enjoy, such as reliable water, sanitation, electricity and public transportation. People living in these areas also have more difficulty accessing critical urban services like jobs, education, food, recreation and health care. Traditional land cover maps do not reflect these distinctions in land use type, or the associated characteristics and problems.
Hyderabad is different: its unplanned settlements, many of which are long established, are distributed throughout the core of the city, along the periphery and in surrounding villages. The informal residential areas and non-residential areas that dot the region surrounding the city are typical of many South and Southeast Asian cities. These “desakota” regions have urban characteristics but are also highly reliant on agriculture. Many of these areas are already subsumed by the growing city, transformed into “urban villages,” with more areas likely to follow as the city grows, creating an even more sprawling megacity. As these unconnected dots grow into the more consolidated and contiguous city, monitoring land use change offers the opportunity to anticipate and manage these challenges.
There are numerous ways that fine-grained land use maps like these can be utilized, as researchers can compare patterns within cities and differences from city to city and country to country. These land use data can combine with other datasets — such as data on the socioeconomics of households, amenities and services, land values, economic activity, hazards and vulnerabilities — to derive new insights.
Based on these findings, policymakers, civil society and the private sector can make changes on the ground to improve peoples’ lives, including enabling better targeting of scarce resources based on evidence. And over time, these data can be used to measure the effectiveness of the U.N. Sustainable Development Goals and other development metrics.
We continue to develop and improve our methodology and aim, pending necessary resources, to ultimately produce equivalent maps for every city in the world every year. With high-fidelity land use data like these, researchers can better understand how urban areas develop over time, how land use changes along the urban periphery and which types of settlements are most affected by natural disasters such as flooding or landslides. These are just a few possible applications, but the potential for these data are even greater — and can only be fully realized by putting these maps and methods in the hands of users.