The research team, led by Weina Zhen and Donghui Shi, focused on the BTH urban agglomeration, a sprawling region of roughly 218,000 square kilometers that encompasses the megacities of Beijing and Tianjin along with eleven prefecture-level cities in Hebei Province. The region is an ideal natural laboratory for studying urban water dynamics. It holds only about 0.84 percent of China’s total water resources despite its enormous population, with per-capita water availability of roughly 243 cubic meters per year. Annual precipitation ranges from about 370 to 730 millimeters, nearly 70 percent of which falls in summer, and drought ranks among the region’s most damaging natural hazards. At the same time, decades of intensive urbanization, massive inter-basin water transfers and ambitious ecological restoration programs have reshaped the hydrological landscape in ways that vary dramatically from city to city.

To map these changes, the researchers drew on the China Land Cover Dataset, a 30-meter-resolution land cover product built from 335,709 Landsat images processed on the Google Earth Engine platform. For every year from 2000 to 2023, they extracted and merged the water and wetland classes to define urban blue space. While the underlying dataset achieves an overall classification accuracy of 79.31 percent nationally, the team conducted their own regional validation for 2000, 2010 and 2020 using stratified random sampling and manual visual interpretation against high-resolution historical imagery from Google Earth. The resulting average overall accuracy of 94.33 percent gave them confidence that the observed trends reflect real landscape change rather than classification noise.

The headline finding is stark. Blue-space area across the BTH region declined at a statistically significant average rate of about 16.82 square kilometers per year, falling from 3,182.61 square kilometers in 2000 to a peak of 3,299.93 square kilometers in 2004 before sliding to 2,909.11 square kilometers by 2010 and then fluctuating between roughly 2,900 and 3,000 square kilometers through 2023. A Pettitt change-point test identified 2009 as a statistically significant structural breakpoint, with mean blue-space area dropping from 3,199.26 square kilometers in the 2000–2009 period to 2,941.21 square kilometers from 2010 onward. The timing coincides with a period of rapid land conversion and severe hydroclimatic stress, including the 2008–2009 drought that gripped northern China, while the subsequent stabilization aligns plausibly with strengthened water-resource management, ecological restoration and large-scale water-transfer programs, though the authors are careful to note these correspondences do not establish causation.

Beneath the regional average, however, lies a patchwork of divergent local stories. Nine of the thirteen cities recorded statistically significant increases in blue-space area, led by Baoding, which gained an average of 3.60 square kilometers per year, expanding from 101.95 square kilometers in 2000 to 156.89 square kilometers in 2023. Shijiazhuang, Qinhuangdao, Handan, Xingtai, Chengde and Langfang also expanded significantly. In sharp contrast, Tianjin lost the most water surface, shrinking from 1,246.01 square kilometers in 2000 to 905.38 square kilometers in 2023, a cumulative loss of 340.62 square kilometers, while Tangshan lost 169.49 square kilometers. Zhangjiakou and Hengshui showed no statistically significant monotonic trend. Any assessment based on regional averages alone would have buried these dramatic intra-regional differences.

The study’s most innovative contribution lies in its treatment of landscape configuration as a distinct dimension of blue-space health. Rather than relying on area alone, the researchers computed ten landscape metrics capturing quantity, patch size distribution, fragmentation, shape complexity and connectivity, including the number of patches, patch density, the largest patch index, the landscape shape index, the perimeter-area fractal dimension, patch cohesion, effective mesh size, the splitting index and the aggregation index. These were synthesized into a single Composite Landscape Index, or CI, using an entropy-weighted TOPSIS method that objectively weights each indicator and ranks observations by their distance from ideal values. Pooling 312 city-year observations ensured the index remained spatially and temporally comparable across the entire study period.

The CI results reveal a systemic quality decline that mirrors, and in some cases contradicts, the area story. Regionally, the CI fell significantly, sharing the same 2009 breakpoint as the area series. Fragmentation metrics rose: more patches, higher patch density, greater subdivision and increasingly irregular patch shapes, while aggregation, cohesion and effective mesh size all declined, signaling weakened connectivity that hinders the movement of material, energy and organisms between water bodies. At the city level, the picture fractured further. Six cities, including Beijing, Shijiazhuang, Baoding, Cangzhou, Qinhuangdao and Langfang, showed significant CI increases, with Cangzhou improving fastest. Five cities, including Tianjin, Tangshan, Zhangjiakou, Chengde and Hengshui, declined significantly. Beijing achieved a structure-led improvement, with stable area but rising CI, while Chengde showed the opposite pattern, gaining water surface that was evidently more dispersed, artificial or seasonal, producing higher fragmentation and lower connectivity. Tianjin and Tangshan suffered simultaneous declines in both extent and configuration, marking them as priority targets for management intervention.

To identify what drives these changes, the team built separate XGBoost regression models for blue-space area and CI, interpreting them through SHAP (SHapley Additive exPlanations) values, a technique borrowed from cooperative game theory that quantifies each variable’s contribution to individual predictions. Eight candidate drivers passed multicollinearity screening: precipitation, temperature, the normalized difference vegetation index, fractional vegetation coverage, population, GDP, urbanization rate and environmental regulation intensity, the last proxied by the share of environment-related sentences in municipal government work reports. XGBoost outperformed Random Forest and Ridge regression baselines, achieving a test R-squared of 0.8984 for area and 0.8740 for CI, and feature-importance rankings proved robust across model architectures.

The SHAP analysis delivered a striking result: population, fractional vegetation coverage, NDVI and urbanization rate together accounted for 83.8 percent of the explanatory importance in the area model and 86.8 percent in the CI model, dwarfing the contributions of climate and policy variables. Population ranked first for area at 28.0 percent, with higher-population cities generally coinciding with larger blue spaces, while vegetation metrics dominated the CI model, with FVC contributing 27.9 percent. Notably, higher vegetation cover showed predominantly negative associations with both variables, which the authors interpret cautiously as a possible spatial competition between blue and green spaces under constrained land availability, or a reflection of vegetation-dominated landscapes where water patches remain small and fragmented. These are statistical associations conditional on city scale and geographic context, not evidence that urbanization benefits water or that vegetation harms it. Climate variables and environmental regulation each contributed roughly 10 percent or less, suggesting that within this dataset, human activity and surface ecological status, rather than macro-climate or governance signals, carry most of the explanatory weight.

The broader implications reach well beyond northern China. The findings align with documented blue-space losses in Delhi, Kolkata and Khulna, yet also echo evidence from the United States and elsewhere that urban water change is not a universal consequence of urbanization but depends on hydroclimatic background, development pressure, water engineering and management choices. The study’s central methodological message is that area and configuration must be assessed as separate dimensions; a city can gain water while losing connectivity, or hold its area constant while its landscape structure quietly improves or collapses. The authors acknowledge limitations, including mixed-pixel effects at 30-meter resolution, the smoothing inherent in city-level aggregation, and the crude proxy used for environmental regulation, and they call for finer spatial units and richer institutional data in future work. For planners in rapidly urbanizing regions worldwide, however, the lesson is already clear: protecting urban water means protecting not just how much there is, but how it is arranged.

Subject of Research: Long-term remote sensing analysis of urban blue space extent, landscape configuration and their drivers in the Beijing–Tianjin–Hebei region

Article Title: Quantifying the spatiotemporal dynamics and drivers of urban blue spaces in the Beijing–Tianjin–Hebei region using remote sensing

Article References: Quantifying the spatiotemporal dynamics and drivers of urban blue spaces in the Beijing–Tianjin–Hebei region using remote sensing. (n.d.). https://doi.org/10.1016/j.indic.2026.101525

Image Credits: AI Generated

DOI: 10.1016/j.indic.2026.101525

Keywords: urban blue spaces, remote sensing, Beijing–Tianjin–Hebei, landscape fragmentation, XGBoost, SHAP, water resources, urbanization, land cover change, ecosystem connectivity, China, sustainable urban planning