Land Cover Change and Urban Growth
How much farmland became buildings? Annual 10 m land-cover maps of the whole world, 2017 to 2023, measured and compared for any city.
Choose a city or any area up to 100 km across and two years. The tool reads the annual global land-cover maps at 10 m, made from Sentinel-2 by a deep-learning model, and measures the area of water, trees, crops, built area, bare ground and rangeland in every year with the true ground area of each pixel. Between the two years it builds the full table of what became what, maps the main changes (farmland to buildings, other land to buildings, farmland lost, new farmland, water lost and gained, trees lost) and can count a change only when it holds in the following year, which removes most of the year-to-year flicker of the maps.
Overview
Iraqi cities have grown fast over their orchards and fields: Baghdad over the palm groves of its edges, Najaf and Karbala along their roads, Erbil and Sulaymaniyah over the plains. This tool puts a number on it, year by year, and shows where it happened.
It is useful to planners, agricultural departments, students of urban geography and anyone who wants to show, with a map and a figure, what has happened to the green belt of a city.
What you enter · What you get
What you enter
- A city from the list, a place found by name or an area drawn on the map (up to 100 × 100 km)
- First and last year (2017–2023)
- Stability test on or off
What you get
- Map of the main changes and land-cover maps of both years
- Built area, crops and water in km² with their change
- Farmland built over, in km² and hectares
- From-to transition table
- Area of every class in every year
How it works in the field
Method and equations
Maps
Annual global land use and land cover at 10 m (version 2), classified from the Sentinel-2 images of each year by a deep-learning model into water, trees, flooded vegetation, crops, built area, bare ground, snow and ice, clouds and rangeland. The tiles that cover the area are joined, and values are read without resampling.
Change
For every pixel the class of the first year and of the last are compared and added to a from-to matrix of areas. With the stability test, a pixel counts as changed only if its new class is also found in the year after the last one; otherwise it is treated as unchanged.
Accuracy and error limits
- The maps are about 85 % correct overall in their published validation; built area and water are the most reliable classes, while crops, rangeland and bare ground are confused most in dry lands. Areas of large changes are therefore reliable, small changes less so. Pixels are 10 m, so single houses and narrow roads may be missed or merged.
- What is not covered: years before 2017 and after 2023, the type of building or crop, and changes inside the built area (new floors, densification). A field left fallow in one year can look like farmland lost.
- To reduce the error: keep the stability test on, compare periods of several years rather than consecutive years, check the change map against the time-lapse tool or recent imagery, and read the trend chart rather than one year.
Who uses it
Standards and references
- Karra, K., Kontgis, C., Statman-Weil, Z., Mazzariello, J. C., Mathis, M. & Brumby, S. P. (2021). Global land use/land cover with Sentinel 2 and deep learning. IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 4704–4707.
- Impact Observatory, Microsoft & Esri (2023). 10 m Annual Land Use Land Cover (9-class), version 2.
- Pontius, R. G., Shusas, E. & McEachern, M. (2004). Detecting important categorical land changes while accounting for persistence. Agriculture, Ecosystems & Environment, 101, 251–268.
- Olofsson, P. et al. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42–57.
At a glance
- Category
- Remote sensing
- Runs on
- Phone, tablet and desktop browser
- Export
- Change map and land-cover maps as PNG, JPG and PDF; trend chart as PNG, JPG and PDF; areas and transition matrix as CSV; full PDF report
- Access
- Free
Run this survey on your next site
All 64 tools, report export and the AI assistant come with one subscription.