ASI-GCN 系列

受污染场地的钻孔采样有限，导致土壤污染物数据稀疏且分布不均匀。传统的插值方法，例如普通克里金法 （OK） 和反距离加权法 （IDW），在应用于此类稀疏数据时可能会掩盖土壤污染的局部变化，从而降低插值精度。为了克服这一挑战，我们提出了一种自适应空间插值图卷积网络 （ASI-GCN） 模型。该模型将空间自相关的原理与图卷积网络 （GCN） 的功能协同集成在一起。通过合并这些元素，ASI-GCN 模型有效地限制了污染物浓度的转移。它巧妙地捕捉空间结构中的细微变化，从而提高土壤污染表征的精度。我们基于从 15 个钻孔中收集的 215 个土壤样本，将模型应用于北京的一家焦化厂。为了评估模型的稳健性，采用了三种具有不同挥发特性的污染物，包括砷（As，非挥发性）、苯并（a）芘（BaP，半挥发性）和苯（Ben，挥发性）。当 ASI-GCN 及其衍生物 ASI-GCN_RC_G 和 ASI-GCN_RC_K 用于在未采样站点进行预测时，使用留一法交叉验证程序来估计它们的性能。结果表明，ASI-GCN 模型表现出更好的性能，As、BaP 和 Ben 的 R² 值分别为 0.52、0.67 和 0.57，RMSE 值分别为 5.041 mg/kg、1.699 mg/kg 和 167.710 mg/kg，优于 OK 和 IDW 等传统模型。进一步分析揭示了污染物波动性对垂直迁移模式的显著影响。非挥发性 As 主要分布在填充砂层和粉砂层中，半挥发性 BaP 集中在粉砂层中，而挥发性 Ben 主要分布在粘土层和细砂层中。ASI-GCN 模型通过利用有限的钻孔数据捕获复杂的污染物分布，为风险评估和修复策略提供关键见解，展示了其在各种污染场地中的应用潜力。Limited borehole sampling at contaminated sites results in sparse and unevenly distributed soil pollutant data. Traditional interpolation methods, such as Ordinary Kriging (OK) and Inverse Distance Weighting (IDW), may obscure local variations in soil contamination when applied to such sparse data, thereby reducing interpolation accuracy. To address this challenge, we propose an Adaptive Spatial Interpolation Graph Convolutional Network (ASI-GCN) model. This model synergistically integrates the principle of spatial autocorrelation with the functionality of Graph Convolutional Networks (GCN). By combining these elements, the ASI-GCN model effectively constrains the propagation of pollutant concentrations and adeptly captures subtle changes in spatial structures, thereby improving the accuracy of soil contamination characterization. We applied the model to a coking plant in Beijing using 215 soil samples collected from 15 boreholes. To evaluate the robustness of the model, three pollutants with different volatilization characteristics were selected, including arsenic (As, non-volatile), benzo(a)pyrene (BaP, semi-volatile), and benzene (Ben, volatile). When ASI-GCN and its variants ASI-GCN_RC_G and ASI-GCN_RC_K were used for prediction at unsampled sites, a leave-one-out cross-validation procedure was adopted to estimate their performance. The results demonstrate that the ASI-GCN model exhibits superior performance, with R² values of 0.52, 0.67, and 0.57 for As, BaP, and Ben, respectively, and RMSE values of 5.041 mg/kg, 1.699 mg/kg, and 167.710 mg/kg, outperforming traditional models such as OK and IDW. Further analysis reveals the significant impact of pollutant volatility on vertical migration patterns. Non-volatile As is mainly distributed in filled sand layers and silty layers, semi-volatile BaP is concentrated in silty layers, while volatile Ben is mainly distributed in clay layers and fine sand layers. By leveraging limited borehole data to capture complex pollutant distributions, the ASI-GCN model provides critical insights for risk assessment and remediation strategies, demonstrating its application potential across various contaminated sites.