This map illustrates several types of surface coverings present on Earth, including water, forested areas, flooded vegetation, agricultural land, buildings, bare soil, snow and ice, clouds and rangeland. The term "land cover" refers to the observable features occupying Earth's surface. While it’s possible to document some of this kind of information in person, the most effective tool is overhead imagery from Earth observation satellites.

The map layer was created using data from the Copernicus Sentinel-1 and 2 Earth observation satellites collected by the European Space Agency from 2017-2022. This map displays data collected in 2022. The data is a collection of grid cells covering the globe. Each square is about 10 square meters (approximately 108 square feet). Unlike polygons and lines that are the correct shape of countries and rivers, this data is an equal grid that appears like individual pixels when zooming closer to the surface. Using artificial intelligence (AI), we can analyze the content of each square to categorize it. This helps us answer the questions “what would we see on this land?” and “how do people use this land?”

Using machine learning, the land cover was sorted by a computer into nine categories: water, trees, flooded vegetation, crops, buildings, bare ground, snow/ice, clouds and rangeland. While we easily understand categories like water, trees, crops, buildings, snow/ice and clouds, there are other types that might be a unique combination of plant and water. Flooded vegetation includes swamps, marshes and wetlands. Rangelands are lands that are not forests or planted farms. They include grasslands, shrublands, savannas, deserts and tundra, covering roughly 46%–50% of Earth's land surface. Rangelands are sometimes used for grazing animals like cows or goats.

Machine learning is part of artificial intelligence. The computer processes enormous amounts of information as it learns and adapts based on training data. A person helps train the computer and then let it use the examples to match similar information. It learns from a small sample that forests are green based on the imagery from the satellite. Therefore, all the green spaces might be categorized as trees. The learning model can identify other types of land and structures. Key applications include tracking land change, modeling flood risks, analyzing habitats and guiding sustainable development.

Explorer Narumasa Tsutsumida - Bringing Human Perspectives to AI

Narumasa Tsutsumida is a geographer, information scientist and National Geographic Explorer working in Japan. Dr. Tsutsumida’s work focuses on the technology of categorizing imagery data with a specialized form of machine learning called deep learning. Deep learning is used to analyze more complex pictures using more computing power to get a more detailed sense of what an image shows. This detail can be used to help a computer do more complex tasks. Some examples of deep learning use are image recognition, natural language processing and self-driving cars.

Dr. Tsutsumida is researching additional ways to improve image-recognition maps like the Land Cover map in MapMaker. In some cases, his technology uses street-level views to see what satellites cannot. For example, a computer might have identified an area as a forest based on satellite images, but from the ground, it’s a parking lot surrounded by mature trees. Only from the street view can it be confirmed as a parking lot. The picture from the satellite only shows trees.

His geography tools can also tell us more about the kinds of plants in a given area or how people understand and use land.

Traveling across Earth reveals a variety of landscapes shaped by humans for cities, farming and conservation. Satellites provide broad views of land cover and human impact. Some land uses pose risks, such as urban heat islands, poor air quality and wildlife habitat loss. Satellite images help track these changes, while socioeconomic data offers a comprehensive look at how people manage land.

Thinking Geographically

Explore what the map can tell us about our global landscapes. Use the Booksmarks tool to zoom to various locations around the globe. What’s unique about each place?

Open map layers and use the swipe tool to compare the land cover category with the satellite images in the base map for a specific location. What do you see in the image? What is the category on the map? Does that seem accurate to you?

Zoom to your community by using the search tool. Open the layers and then use the swipe tool to compare the land cover layer’s category with the basemap. What differences do you see in the land use between the 2022 land cover layer and the more current image on the basemap? Do you notice any positive or negative impacts of the changes on your community or the environment?