A new dataset from Google shows features on the Earth’s surface in near real time, the company announced Thursday. The tool, called Dynamic World, uses deep learning and satellite imagery to develop a high-resolution land cover map that shows which pieces of land have features like trees, crops or water.
Land cover maps typically take a long time to produce, and there are large discrepancies between when images are taken and when data is released. They also often don’t have a detailed breakdown of what’s on the ground in a particular area – a city would be classed as “built up” (a designation for human-modified landscapes) even if there is large sections with parks, for example.
Dynamic World classifies the type of land cover for every 1,100 square feet, Google said. It shows how likely sections are to be covered by one of nine cover types: Water, Flooded Vegetation, Built-up Areas, Trees, Crops, Bare Ground, Grass, Shrub/Scrub, and Snow/Ice. Google detailed its system, developed with the World Resources Institute, in an article published in NatureEast Scientific data.
The above screenshot of New York City, for example, shows that most of the area is built up (red). But there are pockets of grass (green) and shrub/brush (yellow) for large city parks.
The Dynamic World model produces over 5,000 images per day and land cover data is continuously updated. This allows researchers and policymakers to quickly see the impacts of things like fires or hurricanes and help better respond to changes.
“If the world is to produce what is needed from the land, protect the nature that remains, and restore some of what has been lost, we need reliable, near real-time monitoring of every hectare of the planet,” said Craig Hanson, vice president of food, forests, water and ocean at the World Resources Institute, in the Google report. announcement.