Land cover classification is the assignment of each pixel or area of an image to a class describing the observed physical cover of the earth’s surface, which yields a thematic map 12.

Digital classification uses the spectral information in one or more bands to classify each pixel, and the analyst must match spectral classes to the information classes of interest, a match that is rarely simple or one to one 1. FAO defines land cover as the observed (bio)physical cover on the earth’s surface, and USGS gives water and pavement as examples of physical features 23.

Land cover classification is narrower than land use classification, because FAO warns that many existing legends confuse cover with use 2. NASA treats results of analyses of lower-level data as Level 4 output, although its page does not name land cover classification 4. ESA notes that AI applications in Earth observation have mainly used computer vision techniques for interpreting satellite imagery 5.

Across the retained sources, there are no accuracy figures, no standard legend and no ranking of classification algorithms.

Footnotes

  1. Natural Resources Canada, “Image Classification and Analysis”. https://natural-resources.canada.ca/maps-tools-publications/satellite-elevation-air-photos/image-classification-analysis ↩ ↩2

  2. FAO, “Land Cover Classification System - Classification concepts and user manual”. https://www.fao.org/4/y7220e/y7220e06.htm ↩ ↩2 ↩3

  3. USGS, “An interagency perspective on improving consistency and transparency of land use and land cover mapping”. https://pubs.usgs.gov/publication/cir1549/full ↩

  4. NASA Earthdata, “Data Processing Levels”. https://www.earthdata.nasa.gov/learn/earth-observation-data-basics/data-processing-levels ↩

  5. ESA Φ-lab, “Focus on AI4EO”. https://philab.esa.int/flagship-programmes/focus-on-ai4eo/ ↩