the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 16 Sep 2026
A global urban built-up area dataset for cities with populations exceeding 300,000 (2000–2025)
Abstract. Urban land-use efficiency (LUE), defined under Sustainable Development Goal (SDG) 11.3.1, is widely used to evaluate the coordination between urban expansion and population growth. However, its global assessment remains constrained by inconsistent definitions of urban boundaries and the lack of long-term, 30 m spatial resolution urban built-up area (UBA) datasets. Here we present the Global Urban Built-up Area Dataset (GUBAD), a multi-epoch dataset at 30 m spatial resolution covering 1,611 cities worldwide with populations exceeding 300,000 inhabitants from 2000 to 2025. GUBAD applied standardized UN-defined urban agglomeration boundaries using a spatial morphological framework integrating impervious surface data with population constraints. Impervious surface areas (ISA) were extracted using Random Forest classification of Landsat and Sentinel-1/2 imagery, combined with isotonic regression-based temporal correction. Validation results yielded a mean overall accuracy of 95.18 % and Kappa coefficient of 0.90 across all epochs. GUBAD shows strong agreement with GHSL-SMOD, DEGURBA, and UN-Habitat products while providing enhanced spatial detail. Unlike global urban boundary datasets that delineate administrative-functional city extents, GUBAD explicitly targets the physically built-up fabric derived from impervious surfaces under population-density and contiguity constraints. We further demonstrate the utility of GUBAD for estimating SDG indicator 11.3.1, revealing spatiotemporal variations in the relationship between land consumption rate and population growth rate (LCRPGR) at the global city scale. GUBAD provides a temporally consistent, city-scale UBA dataset based on standardized UN urban agglomeration definitions, enabling reproducible monitoring of urban expansion and SDG 11.3.1 assessment globally. GUBAD is available at Zenodo under CC BY 4.0 (Gunasekera et al., 2025) with DOI: https://doi.org/10.5281/zenodo.20051123. The dataset is released as version 1.0 with planned updates every 5 years.
- Preprint
(3003 KB) - Metadata XML
-
Supplement
(839 KB) - BibTeX
- EndNote
Status: open (until 20 Nov 2026)
-
CC1: 'Comment on essd-2026-381', Shengpeng Wang, 17 Sep 2026
reply
-
CC2: 'Reply on CC1', Zhongchang Sun, 17 Sep 2026
reply
Thank you very much for your positive and encouraging comment. We are glad that you recognize the contribution of our dataset in addressing data gaps in existing studies and its potential for supporting SDGs assessment, as well as future research on sustainable development and urban studies. We hope the dataset will prove useful to the broader research community, and we welcome any further discussion or feedback.
Citation: https://doi.org/10.5194/essd-2026-381-CC2
-
CC2: 'Reply on CC1', Zhongchang Sun, 17 Sep 2026
reply
-
RC1: 'Comment on essd-2026-381', Anonymous Referee #1, 11 Oct 2026
reply
In the manuscript titled “A global urban built-up area dataset for cities with populations exceeding 300,000 (2000–2025)”, the authors present the Global Urban Built-up Area Dataset (GUBAD), generating a 30 m multi-epoch (5-year steps from 2000 to 2025) dataset tracking built-up areas across 1,611 global cities with populations over 300,000. The workflow combines Random Forest classification of Landsat and Sentinel-1/2 imagery, temporal isotonic regression, and a post-processing morphological framework constrained by LandScan population grids.
While I can see this being a useful dataset, several methodological choices, sensor inconsistencies, and validation limitations weaken the reliability of both the dataset and the resulting physical interpretations. Major revisions are warranted before this work can be considered for publication.
Major Comments:
1. The feature stack shifts fundamentally between the 2000–2015 epochs (Landsat-only, spectral indices) and the 2020–2025 epochs (Sentinel-2 MSI at 10 m, Sentinel-1 SAR backscatter, and VIIRS nighttime lights). Adding radar backscatter and sub-pixel optical resolution substantially increases sensitivity to fragmented, peri-urban built fabric and structural edges. The authors observe a dramatic spike in land consumption rate and population growth rate (LCRPGR) during 2020–2025 and dismiss this as real-world divergence driven by post-2020 infrastructure investment. In all likelihood, this is an artifact of sensor transition and feature expansion. Applying the Pool-Adjacent-Violators Algorithm only enforces monotonicity; it does not correct baseline step-changes introduced by adding high-sensitivity sensors in 2020. A controlled sensitivity study holding sensor inputs constant (e.g., Landsat-only indices across all years) is necessary to prove this jump is physical rather than methodological.
2. Validating 1,611 complex urban environments with only 10 validation points per city (5 urban, 5 non-urban) is statistically insufficient for local error characterization. While the authors aggregate across all points to claim a narrow global confidence interval, a 10-point sample per city cannot meaningfully capture classification accuracy, edge commission/omission errors, or morphological variance across heterogeneous urban landscapes. Stratified sampling should be scaled proportionally with city footprint, and accuracy must be reported across different urban morphology tiers and climate/biophysical zones (say, Koppen climate zones).
3. The conversion from impervious surface area to urban built area explicitly incorporates 1 km LandScan population density thresholds to delineate the urban footprint. When this footprint is subsequently combined with UN population growth rates to estimate LCRPGR, the analysis borders on circular reasoning. This is because population data are used both to constrain whether land is included in the built-up footprint and as the denominator against which that footprint's growth is evaluated. The authors must clarify and bound how demographic assumptions in LandScan bias the resulting land consumption rate calculations.
4. The pixel-level PAVA isotonic regression strictly enforces monotonic urban expansion, explicitly forbidding transitions from built-up to non-built-up land. While global urbanization dominates, this hard constraint artificially masks real urban abandonment, structural retreat, conflict-driven displacement, and brownfield re-greening. The manuscript needs to quantify how much true land-use change or sensor noise is overridden by this monotonic assumption.
5. Aggregating 10 m Sentinel-2 ISA down to 30 m using a simple 0.5 area-weighted majority threshold introduces clear scale bias compared to native Landsat 30 m mixed pixels. A native 30 m pixel integrates reflectance over a 900 m2 instantaneous field of view, whereas a binary threshold on 10 m sub-pixels systematically eliminates sub-pixel impervious elements below 50% fraction (e.g., suburban residential streets or sparse informal settlements). A sub-pixel impervious fraction comparison or a soft-aggregation scheme should be explored.
6. As a final note, it is unclear to me why this dataset is more relevant for estimating LCRPGR than the multiple other time series of impervious, or built, or urban land that currently exist in the literature, which already do not agree on the degree of urbanization globally and regionally (Chakraborty et al. 2024). Just because new datasets can be developed does not need mean that they need to be developed.
Minor Comments
Figure 8: The scatterplots comparing GUBAD with WUP show considerable dispersion and systematic bias for high-density metropolitan areas (R2 dropping to ~0.54–0.60 in later epochs). A more thorough breakdown of why these discrepancies scale with city size is needed.
References:
Chakraborty, T. C., Venter, Z. S., Demuzere, M., Zhan, W., Gao, J., Zhao, L., & Qian, Y. (2024). Large disagreements in estimates of urban land across scales and their implications. Nature communications, 15(1), 9165.
Citation: https://doi.org/10.5194/essd-2026-381-RC1 -
RC2: 'Comment on essd-2026-381', Anonymous Referee #2, 21 Oct 2026
reply
- Section 5.2: Consider replacing “overestimate urban extents” with “delineate larger urban extents,” as some differences may arise from the definitions adopted by each dataset.
- Section 5.3: Briefly discuss how hole filling and small-patch removal may affect enclosed open spaces and fragmented peri-urban settlements.
- Lines 395–399: Please correct the explanation of extreme LCRPGR values: near-zero population growth decreases the denominator’s absolute value, potentially producing extreme ratios.
- Section 4.3: Please clarify that “global UBA” refers to the combined built-up area of the 1,611 study cities, rather than all urban areas worldwide.
Citation: https://doi.org/10.5194/essd-2026-381-RC2
Data sets
Global Urban Built-up Area Dataset (GUBAD) for Cities with Populations Exceeding 300,000 (2000–2025) (1.0) D. Gunasekera et al. https://doi.org/10.5281/zenodo.20051123
Model code and software
GUBAD: Global Urban Built-up Area Dataset — Processing Code D. Gunasekera et al. https://github.com/dinoora/GUBADv1.0
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 595 | 264 | 108 | 967 | 81 | 106 | 111 |
- HTML: 595
- PDF: 264
- XML: 108
- Total: 967
- Supplement: 81
- BibTeX: 106
- EndNote: 111
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This work helps address the lack of data in existing studies and achieves a high level of overall accuracy. It can make an important contribution to the assessment of the Sustainable Development Goals and may also be further applied in future research on sustainable development and urban studies.