Spatiotemporal associations and responses between the evolution of blue-green space landscape patterns and water quality in highly urbanized plain river-network regions: A case study of Suzhou City, China.

Under rapid urbanization, the blue-green space of plain river-network cities is continuously restructured, and its configuration may relate to water quality through dilution, hydrological connectivity, flushing-driven exchange, and hydraulic residence time. Yet the nonlinear, scale-, and stage-dependent associations between blue-green space pattern evolution and water quality remain poorly quantified. Using Suzhou-a representative plain river-network city in the Taihu Lake Basin-we integrated land-use data (2004-2024) with routine water-quality records from 66 monitoring sites, summarized using the Canadian Council of Ministers of the Environment Water Quality Index (CCME-WQI). Blue- and green-space landscape metrics were computed within multiscale circular buffers (500-2000 m), and an interpretable XGBoost-SHAP framework compared pattern-water-quality associations between two stages (2004-2012 and 2016-2024). The results showed the following: (1) water quality improved overall, with Poor and Marginal sections declining from ∼75% to ∼15% and Good-or-better sections rising from ∼10% to ∼45%, with a pronounced transition around 2016 and subsequent deceleration. (2) In both stages, LPI_BlueSpace outranked the compositional metric PLAND_BlueSpace in SHAP importance; given generally high blue-space coverage and conditional on the other variables, the size and dominance of the largest core water patch were more informative for discriminating spatial heterogeneity in water quality. ED_BlueSpace contributed negatively, consistent with disturbance along modified, complex water-land interfaces. Green-space metrics did not rank among the leading predictors; given the 30 m resolution and circular-buffer design, this indicates a weak independent statistical signal rather than limited green-space regulation of water quality. (3) The highest-importance scale was metric- and stage-dependent (LPI_BlueSpace 2000→1500 m, PLAND_BlueSpace 1000→500 m, ED_BlueSpace 500→2000 m); the lower explanatory power of the later-stage model indicates that the measured landscape patterns explained less of the later-stage variation; engineering and governance factors are plausible omitted variables but were not quantified. These findings support scale-specific, stage-differentiated blue-green space planning and water-environment governance in similar plain river-network cities.