College of New Energy and Environment, Jilin University, Changchun, 130021, China.
Key Laboratory of Groundwater Resources and Environment (Jilin University), Ministry of Education, Changchun, China.
Environ Sci Pollut Res Int. 2020 Aug;27(22):28077-28090. doi: 10.1007/s11356-020-09177-2. Epub 2020 May 14.
Effective coastal aquifer management typically relies on numerical models to analyze the seawater intrusion (SI) process. Before using groundwater simulation models to predict the extent of SI in the future, preparing input data is an extremely necessary and important step. For precipitation and sea-level rise (SLR), which are two of the most influential factors for SI, it is difficult to precisely forecast their variations. Current studies of using numerical models to predict future SI often overlook the uncertainty of these two factors. This can result in compromised predictions of SI. In this study, a three-dimensional variable-density groundwater simulation model was established for a coastal area in Longkou, China. Then, the Monte Carlo method was applied to perform uncertainty analysis for the input data of precipitation and SLR of the SI model. In order to reduce the huge computational load brought by repeated invocation of the SI model during the process of Monte Carlo simulation, a surrogate model based on a multi-gene genetic programming (MGGP) method was developed to replace the SI simulation model for calculation. A comparison between the MGGP surrogate model and the Kriging surrogate model was carried out, and the results show that the MGGP surrogate model has a distinct advantage over the Kriging surrogate model in approximating the excitation-response relationship of the variable-density groundwater simulation model. Through statistical analysis of Monte Carlo simulation results, an object and reasonable risk assessment of SI for the study area was obtained. This study suggests that it is essential to take the uncertainty of precipitation and SLR into account when modeling and predicting the extent of SI.
有效的沿海含水层管理通常依赖于数值模型来分析海水入侵（SI）过程。在使用地下水模拟模型预测未来的 SI 范围之前，准备输入数据是一个极其必要和重要的步骤。对于降水和海平面上升（SLR）这两个对 SI 影响最大的因素之一，很难精确预测它们的变化。目前使用数值模型预测未来 SI 的研究往往忽略了这两个因素的不确定性。这可能导致 SI 的预测不准确。在这项研究中，为中国龙口的一个沿海地区建立了一个三维变密度地下水模拟模型。然后，应用蒙特卡罗方法对 SI 模型的降水和 SLR 输入数据进行不确定性分析。为了减少蒙特卡罗模拟过程中反复调用 SI 模型带来的巨大计算负荷，开发了一种基于多基因遗传编程（MGGP）方法的代理模型来替代 SI 模拟模型进行计算。对 MGGP 代理模型和克里金代理模型进行了比较，结果表明，MGGP 代理模型在近似变密度地下水模拟模型的激励-响应关系方面具有明显优势。通过对蒙特卡罗模拟结果的统计分析，对研究区域的 SI 进行了客观合理的风险评估。本研究表明，在建模和预测 SI 范围时，必须考虑降水和 SLR 的不确定性。
