A temporally harmonised gridded building-stock dataset for the Global South, 2016-2023

Background Rapid settlement change across the Global South is transforming the built environment, yet consistent spatio-temporal information on buildings remains limited in many low- and middle-income countries. Multi-temporal building datasets derived from satellite imagery can contain year-to-year inconsistencies caused by variation in image availability, acquisition conditions, and model performance, limiting their use for longitudinal analysis. Methods We developed a spatio-temporally harmonised gridded building-characteristics dataset for 130 countries in the Global South covering 2016–2023. The dataset is derived from annual gridded layers aggregated from the Google Open Buildings 2.5D Temporal dataset at a spatial resolution of 3 arc-seconds, approximately 100 metres at the equator. For each grid cell and year, three building metrics are provided: building count, building surface area, and building volume. To improve temporal consistency, we developed a deterministic adaptive forward-backward temporal smoother inspired by Kalman updating principles, rather than a formal Kalman filter or state-space model. The adaptive gain assigns less weight to annual observations that depart strongly from a grid cell’s multi-year trajectory relative to a spatially smoothed baseline deviation measure. Results Technical validation shows that the harmonisation reduces implausible year-to-year oscillations and relative zigzag amplitudes across the three building metrics. At the same time, high rank correlations between raw and harmonised annual layers indicate that the procedure preserves the broad spatial distribution of building characteristics. City-level examples further illustrate how the method handles intermittent missing observations, abrupt annual changes, and short-term oscillations while retaining longer-term temporal trajectories. Conclusions The resulting dataset provides a temporally consistent multi-year representation of building characteristics across data-scarce regions of the Global South. It is designed to support analyses of settlement expansion, population distribution, infrastructure planning, and environmental risk, while remaining best suited to multi-year trends rather than detection of short-term individual events.