Built-Up Area Change

A probability-based sample analysis is the recommended good practice approach to estimating land cover and land use extent and change. The global surface built-up land map enables a higher sampling efficiency through stratification and can be extended to report on regional and national dynamics. Current validation results inform only the global scale.

Small features, such as isolated farm houses or huts, are generally not captured by this prototype map. Future versions will employ reference data to refine the definition at the sub-pixel scale.

Small roads, especially with dirt or gravel surfaces, are inconsistently characterized and result in both omission and commission errors in areas with linear features, for example agriculture lands.

Deep learning parameterization, for example normalization, can lead to boundaries along our moving windows. Future iterations will address this issue, hoping to avoid any artifacts from the algorithm itself.