Seawater intrusion in coastal aquifers is a significant problem that can be addressed through the construction of subsurface dams or physical cut-off barriers. An alternative method is the use of microbially induced carbonate precipitation (MICP) to reduce the hydraulic conductivity of the porous medium and create a physical barrier. However, the effectiveness of this method depends on various factors, and the scientific literature presents conflicting results, making it challenging to generalise the findings. To overcome this challenge, a statistical and machine learning (ML) approach is employed to infer the causes for the reduction in hydraulic conductivity and identify the optimum MICP parameters for preventing seawater intrusion. The study involves data curation, exploratory analysis, and the development of various models to fit the input data (k-Nearest Neighbours – kNN, Support Vector Regression – SVR, Random Forests – RF, Gradient Boosting – XgBoost, Linear model with interaction terms, Ensemble learning algorithms with weighted averages – EnL-WA and stacking – EnL-Stack). The models performed reasonably well in the region where permeability reduction is sensitive to carbonate increase capturing the permeability reduction profile with respect to cementation level while demonstrating that they can be used in initial assessments of the specific conditions (e.g., soil properties). The best performing algorithms were the EnL-Stack and RF followed by XgBoost and SVR. The MICP method is effective in reducing hydraulic conductivity provided that the various biochemical parameters are optimised. Critical biochemical parameters for successful MICP formulations are the bacterial optical density, the urease activity, calcium chloride concentration and flow rate as well as the interaction terms across the properties of the porous media and the biochemical parameters. The models were used to identify the optimum MICP formulation for various porous media properties and the maximum permeability reduction profiles across cementation levels have been derived.

中文翻译：

---

应用微生物诱导碳酸盐沉淀作为控制海水入侵的物理屏障的统计和机器学习分析

沿海含水层的海水入侵是一个重大问题，可以通过建造地下水坝或物理隔离屏障来解决。另一种方法是使用微生物诱导碳酸盐沉淀（MICP）来降低多孔介质的水力传导率并形成物理屏障。然而，这种方法的有效性取决于多种因素，并且科学文献提出了相互矛盾的结果，因此很难概括这些发现。为了克服这一挑战，采用统计和机器学习 (ML) 方法来推断导水率降低的原因，并确定防止海水入侵的最佳 MICP 参数。该研究涉及数据管理、探索性分析以及开发各种模型来拟合输入数据（k 最近邻 - kNN、支持向量回归 - SVR、随机森林 - RF、梯度提升 - XgBoost、具有交互项的线性模型、具有加权平均值的集成学习算法 – EnL-WA 和堆叠 – EnL-Stack）。这些模型在渗透率降低对碳酸盐增加敏感的区域表现相当好，捕获了相对于胶结水平的渗透率降低曲线，同时证明它们可以用于特定条件（例如土壤特性）的初始评估。性能最好的算法是 EnL-Stack 和 RF，其次是 XgBoost 和 SVR。只要优化各种生化参数，MICP 方法就能有效降低水力传导率。成功的 MICP 配方的关键生化参数是细菌光密度、脲酶活性、氯化钙浓度和流速以及多孔介质特性和生化参数的相互作用项。该模型用于确定各种多孔介质特性的最佳 MICP 配方，并得出了不同胶结层的最大渗透率降低曲线。