High Spatial Resolution Building Characteristics for the Global South: Insights from the Google Open Buildings...

The need for detailed built-up area data for applications such as population modelling, urban planning, and environmental research is growing due to the pace of global population changes, particularly in the Global South, where existing datasets have limitations.

Here, we processed the Google Open Buildings Temporal (OBT) dataset to derive six 100-m spatial resolution datasets per year on building characteristics. The characteristics include building count, total perimeter, total area, total volume, height variance, and mean distance to the nearest building edges. These were calculated using arithmetic operations, convolutions, and spatial aggregation. The derived data was validated against a set of existing largescale open spatial datasets on buildings and human settlement extents for single timepoints. Additionally, temporal consistency was assessed, with polynomial fitting explored to test suitability for smoothing the data where significant fluctuations were seen.

The new dataset strongly correlated with the Google Open Buildings Polygons dataset (e.g., building count: r = 0.88; building area: r = 0.90) but showed systematic perimeter underestimation in dense areas due to blending effects. Weaker correlations were found with other datasets due to methodological differences. Internally, building height variance correlated moderately with total volume ( r = 0.47). A strong positive correlation ( r > 0.8) existed between building count, area, volume, and population. Temporal analysis revealed significant fluctuations in most characteristics, especially height-related metrics, with second-order polynomial fitting proving optimal for smoothing.

A validated 100-m resolution building characteristics dataset for the Global South, covering each year from 2016 to 2023, derived from Google OBT, was produced. While showing consistency with similar largescale spatial datasets, temporal fluctuations indicate a need for further processing for time-series applications.