Supplementary information files for "High-performance modelling of urban non-point-source pollutant dynamics: a full-process approach"

Supplementary files for article "High-performance modelling of urban non-point-source pollutant dynamics: a full-process approach"

Non-point-source pollutants (NPSPs) are a major cause of urban water quality degradation. Effective management requires an understanding of pollutant dynamics across spatial scales. Existing storm water quality models are often based on empirical or hydrological approaches, which have limited capacity to represent flow-pollutant interactions and particle-facilitated dynamics in complex urban environments. Physically based models provide a more realistic description, but their applications are constrained by high computational costs and detailed data requirements. This study presents a high-performance, GPU-accelerated hydrodynamic-particle-based water quality model to simulate the full dynamics of wash-off, deposition, and transport of urban particulate-based NPSPs at high spatial resolution. The initial pollutant mass and particle size distribution (PSD) fields were derived from a physics-informed Random Forest build-up model trained on literature-reported data. The model was validated using two monitored events in a road catchment near Paris, achieving NSEs of 0.86 for runoff and 0.67 for pollutant fluxes. Sensitivity analyses revealed a strong dependence on the single particle mass (Pm), with simulation accuracy becoming stabilised beyond 50 particles per grid. Application to a real-world urban case study confirmed the framework's efficacy in reproducing flood inundation and NPSP propagation. The analysis further underscores that a resolution finer than 5 m is necessary for reliable simulations in complex urban settings. The particle-tracking capability enabled spatio-temporal pollutant source identification. This framework presents a valuable tool for scientists, policymakers, and environmental practitioners to advance urban water quality management.

© The Author(s), CC BY 4.0

论文《城市面源污染物动态过程全流程高性能建模》补充材料

面源污染物（Non-point-source Pollutants, NPSPs）是引发城市水质退化的核心诱因。有效的治理工作需要明晰不同空间尺度下的污染物动态演化过程。现有雨水水质模型多基于经验方法或水文学路径，在复杂城市环境中，其对径流-污染物交互过程以及颗粒物介导的动态过程的表征能力较为有限。基于物理机制的模型能够提供更贴合实际的描述，但其应用却受限于高昂的计算成本与严苛的精细化数据需求。

本研究提出一款高性能、GPU加速的水动力-颗粒物耦合水质模型，可在高空间分辨率下模拟城市颗粒物介导面源污染物的冲刷、沉积与输移全动态过程。初始污染物质量场与粒径分布（Particle Size Distribution, PSD）场，源自基于文献报道数据训练的物理引导随机森林污染物累积模型。

该模型通过巴黎近郊道路汇水区的两场实测降雨事件进行验证，径流模拟的纳什效率系数（Nash-Sutcliffe Efficiency, NSE）达0.86，污染物通量模拟的NSE达0.67。敏感性分析结果显示，模型对单颗粒物质量（Single Particle Mass, Pm）具有较强依赖性；当每个网格内的颗粒物数量超过50个时，模拟精度趋于稳定。

将该框架应用于真实城市案例研究，证实其可有效复演洪水淹没与面源污染物迁移过程。进一步分析表明，在复杂城市场景中开展可靠模拟，需采用优于5米的空间分辨率。颗粒物追踪功能可实现污染物来源的时空定位识别。本框架为科研人员、政策制定者与环境从业者推进城市水质治理工作提供了极具价值的工具。

© 作者，CC BY 4.0