基于不确定性估计的土壤重金属污染监测采样设计研究

Soil heavy metal pollution poses a serious threat to ecologically environmental health and agricultural product safety. There exist many questions in the current sampling design for soil heavy metal, such as the difficulty to determine the sampling locations and difficulty to meet the needs of soil remediation decision because low sampling efficiency and low investigation accuracy. In this project, the rice-growing area and a brownfield of Fuyang district in Zhejiang province were chosen as the study area, based on the historical data obtained by the investigation on soil heavy metal pollution, the spatial and temporal variation of soil heavy metal will be modeled using the temporal-spatial analysis models and the drivers will be explored，the hot spot, abnormal area and high risk area of heavy metal pollution will be identified. For the single variable of heavy metal, the spatial distribution of pollution uncertainty will be analyzed and the additional sampling methods will be developed based on the uncertainty originated from spatial prediction and the uncertainty originated from polluted area delineation, respectively. The remediation volumes and cost, and their uncertainties in a selected brownfield will be calculated and the multi-stage additional method will be developed based on the uncertainty estimation and accuracy constraints. For the multi variables of heavy metal, an optimal sampling method will be developed for the monitoring of heavy metal pollution based on the analysis of the multivariate distribution and spatial structure of the multi historical heavy metal elements. The study will be a good extension to the soil sampling theory, can improve the existing monitoring network of heavy metal pollution and solve the basic problems of regional pollution investigation and risk assessment, so as to provide technical support for soil environment quality monitoring, soil heavy metal pollution control and remediation.

土壤重金属污染威胁到生态健康与农产品安全。项目针对目前我国土壤重金属污染监测采样布点难、效率低、监测精度不能满足修复决策的需求等问题，充分利用浙江省富阳研究区十年来的土壤重金属调查\普查先期数据，建立土壤时空变异性模型，探究重金属污染时空变异规律和影响因素，识别重金属污染热点区、异常区和高风险区域；针对单一重金属变量，通过分析污染不确定性空间分布，分别发展基于空间变化推测不确定性和污染范围界定不确定性的耕地土壤补样方法；通过估算污染修复体和修复费用及其不确定性，发展不确定性估计和精度约束的污染场地多阶段补样方法；针对多个重金属变量，通过分析先期重金属元素的多元分布和空间结构，发展面向多目标变量土壤重金属污染监测的优化采样方法。该研究可丰富土壤抽样的理论和方法，完善现有重金属污染监测布点网络，解决区域土壤环境污染调查和风险评价的基础问题，为土壤质量监测和重金属污染综合防治与修复提供技术支撑。

采集能够反映土壤重金属空间变化规律的典型样点，并保证土壤重金属污染监测和环境质量评价的精度是土壤重金属污染修复和控制的基础和前提。本项目充分利用土壤重金属调查/普查先期数据，探究重金属污染时空变异规律和影响因素，解析重金属污染源，识别重金属污染热点区、异常区和高风险区域；基于空间域和属性域上土壤重金属推测的不确定性，构建针对单一重金属元素的土壤采样（补样）方法；基于多源环境变量的深度学习和数据挖掘，构建针对多个重金属元素的优化采样方法。最后基于质量平衡模型、累计模型和时间模型等，模拟不同情景下重金属元素含量变化和污染风险，提出完善土壤重金属现有监测布点网络的方案。研究发现，我国绝大部分省份都存在不同重金属元素的富集和污染，省级层面重金属污染热点主要由区域性采矿、工业活动、污水灌溉和大气沉降造成，部分省份富集率较高则与低背景值和人类活动的贡献有关；重点典型研究区10年间的数据分析表明，经过10年工农业生产排放，研究区土壤重金属累积严重，形成了以Hg、Ni、Cr为主的污染格局，环境安全风险大。采样研究表明：空间加密采样方法针对单变量土壤重金属元素，而条件拉丁超立方采样方法（cLHS）可以兼顾多变量土壤重金属元素，而后者比前者的精度高，且总样本点越多，采样精度越高。不同重金属的样本数量对采样精度有不同程度的影响。Pb等重金属的样本数量对采样精度影响较大，而Zn等重金属的样本数量对采样精度影响较小。当样本达到一定数量时，增加更多的样本数量对提高采样精度的贡献较小。我们提出了基于深度学习算法的、可实现对多个重金属元素同时采样的优化采样方案框架，该方案能保证采样精度不降低而样本量减少30%。