基于数据同化的土壤水力参数反演方法:研究进展与展望  

Data Assimilation for Soil Hydraulic Parameter Estimation:Progress and Perspectives

作　　者：满俊 张江江 郑强 尧一骏 曾令藻[4] MAN Jun;ZHANG Jiangjiang;ZHENG Qiang;YAO Yijun;ZENG Lingzao(Key Laboratory of Soil Environment and Pollution Remediation,Institute of Soil Science,Chinese Academy of Sciences,Nanjing 210008,China;Yangtze Institute for Conservation and Development,Hohai University,Nanjing 210024,China;Department of Mathematics and Theory,Peng Cheng Laboratory,Shenzhen 518055,China;Zhejiang Provincial Key Laboratory of Agricultural Resources and Environment,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]中国科学院土壤环境与污染修复重点实验室(南京土壤研究所),南京210008 [2]长江保护与绿色发展研究院(河海大学),南京210024 [3]数学与理论部(鹏城实验室),深圳518055 [4]浙江省农业资源与环境重点实验室(浙江大学),杭州310058

出　　处：《土壤学报》2023年第6期1543-1554,共12页Acta Pedologica Sinica

基　　金：国家重点研发计划项目(2021YFC1808904);国家自然科学基金项目(42107066);江苏省自然科学基金项目(BK20201105)资助。

摘　　要：土壤水力参数及其非均质性刻画关乎到诸多土壤与地下水等领域的量化模拟研究问题。受限于时间和采样成本,传统的直接测定方法并不能很好地解决这个问题。随着物联网技术的发展,与土壤水运动有关的一些状态表征量(如含水量和水头)已经能够通过传感器实时获得。如何充分融合这些观测数据信息,反演出土壤水力参数是当前的一个研究热点。数据同化方法能够通过融合观测数据与模型预测值信息,实现对模型参数的反演估计。本文系统分析了土壤水力参数不确定性的来源及测定方法,阐述了常用数据同化方法的基础理论及其在土壤水力参数反演方面的应用,并从计算效率和反演精度两方面着重论述了数据同化方法的最新前沿进展,最后探讨了数据同化方法未来的发展方向。研究表明:数据同化方法能够突破传统测定方法的限制,用于土壤水力参数及其非均质性刻画。尽管如此,由于土壤非饱和流模型的强非线性以及原位观测数据的相对稀缺性等问题的存在,当前数据同化方法的计算效率和反演精度还有待进一步提升。未来可从发展监督式降维方法、多源多尺度数据融合以及耦合物理机制的机器学习等方面深化土壤水力参数反演方法研究,这有利于农业土水管理、污染防治和修复等工作的合理开展。The characterization of soil hydraulic parameters and their heterogeneity is related to many scientific problems in soil and groundwater fields.Due to the limitation of time and sampling cost,the traditional experimental approaches cannot address this issue adequately.With the development of Internet of Things technology,the state variables related to soil water movement(such as water content and pressure head)can be acquired in real time through sensors.This has sparked some debates about how to estimate the soil hydraulic parameters using these measurements.Data assimilation methods can estimate the soil hydraulic parameters by integrating the measurements into numerical models.This paper systematically analyzes the uncertainty sources and measurement approaches of soil hydraulic parameters,expounds on the basic principles of several common data assimilation methods and their applications in soil hydraulic parameter inversion,and discusses the latest advances in data assimilation methods from aspects of computational efficiency and accuracy.Finally,the development direction of data assimilation methods is provided.The results show that the data assimilation methods can break through the limitation of the traditional experimental approach,and thus are suitable for the characterization of soil hydraulic parameters and their heterogeneity.However,limitations such as the strong nonlinearity of the unsaturated flow model,spatial heterogeneity of soil and sparsity of in-situ measurements do exist.It is,therefore,essential for us to unfold in-depth research on soil hydraulic parameter inversion from the aspects of supervised dimension reduction method,multi-source and multi-scale data fusion,and coupling of machine learning with physical mechanisms,thereby assisting agricultural soil and water management as well as the prevention,control,and remediation of pollution in agroecosystems.

关 键 词：数值模拟 土壤水力参数 参数反演 数据同化 优化试验设计 机器学习 

分 类 号：S152[农业科学—土壤学]