Advances on the Global Human Settlement Layer by joint assessment of Earth Observation and population survey data

Abstract

The Global Human Settlement Layer (GHSL) project fosters an enhanced, public understanding of the human presence on Earth. A decade after its inception in the Digital Earth 2020 vision, GHSL is an established project of the European Commission’s Joint Research Centre and an integral part of the Copernicus Emergency Management Service. The 2023 GHSL edition, a result of rigorous research on Earth Observation data and population censuses, contributes significantly to understanding worldwide human settlements. It introduces new elements like 10-m-resolution, sub-pixel estimation of built-up surfaces, global building height and volume estimates, and a classification of residential and non-residential areas, improving population density grids. This paper evaluates the key components of the GHSL, including the Symbolic Machine Learning approach, using novel reference data. These data enable a comparative assessment of GHSL model predictions on the evolution of built-up surface, building heights, and resident population. Empirical evidence suggests that GHSL estimates are the most accurate in the public domain today, e.g. achieving an IoU of 0.98 for the water class, 0.92 for the built-up class, and 0.8 for the non-residential class at 10 m resolution. At 100 m resolution, we find that the MAE of built-up surface estimates corresponds to 6% of the grid cell area, the MAE for the building height estimates is 2.27 m, and we find a total allocation accuracy of 83% for resident population. This paper consolidates the theoretical foundation of the GHSL and highlights its innovative features for transparent Artificial Intelligence, facilitating international decision-making processes.

Keywords

• GHSL
• geospatial XAI
• built-up surface
• population grids
• degree of urbanization
• building height