非饱和土壤水流和溶质运移的多源多保真度数据同化方法研究

Establishing effective practices in agricultural water management is based on the accurate prediction of water flow and solute transport in the vadose zone, where one key prerequisite is the accurate acquisition of related model parameters. Data assimilation (DA) is an important method as it can reduce the uncertainty of model parameters and increase the prediction accuracy by fusing the information contained in the measurement data. However, it poses a huge challenge for the state-of-the-art DA algorithms when dealing with water flow and solute transport in complex and heterogeneous soils. In this project, based on both experimental and numerical studies, we address this issue by proposing a multi-fidelity data assimilation method for water flow and solute transport in unsaturated soils with multiple sources of measurement data. To maximize the reduction of system uncertainty, we also conduct Bayesian experimental design in this project to find the optimal measurement scheme and reduce the cost of monitoring. This project will improve the quantification of water flow and solute transport in soils, and provide both theoretical and technical support for agricultural water management.

制定有效的农业水资源管理措施需要建立在对包气带中水流和溶质运移准确预测的基础上，而准确获取相关参数是其关键前提。数据同化可以融合观测数据信息，降低模型参数的不确定性，从而提高预测精度，因而具有重要意义。然而，现有的数据同化算法在处理复杂非均质土壤中水流和溶质运移时遇到了较大的挑战。为此，本项目将结合实验研究和数值模拟，为非饱和土壤水流和溶质运移发展出高效的多源多保真度数据同化方法。为了最大程度地降低土壤系统的不确定性，本项目还将开展贝叶斯试验设计，寻找最优观测方案，降低监测成本。本项目的实施有助于提高非均质土壤中水流和溶质运移的模拟精度，为农业水资源管理提供有力的理论和技术指导。

在农业、环境和生态等领域，认识和模拟土壤和地下水中水流和溶质运移过程具有十分重要的意义，而准确识别相关模型参数（如水力参数和污染源信息）是其关键前提。数据同化可以融合多源观测数据（如水头、含水量、浓度和温度）信息，降低模型不确定性，从而提高模型预测效果。由于地下多孔介质的非均质性与观测数据的相对稀缺性，准确高效地模拟土壤和地下水中水流与溶质运移过程较为困难。本项目充分利用多源监测数据，发展了适用于非均质多孔介质水力参数和污染源参数反演的新型数据同化框架：（1）利用多保真度数据和有监督降维方法构建了土壤和地下水系统的数据驱动模型，提高了数据同化效率；（2）提出了基于卡尔曼更新的提议分布，提升了马尔科夫链蒙特卡洛方法反演参数后验的效果；（3）发展了基于深度学习的新型参数更新方法，改善了对非高斯分布参数的识别能力。上述数据同化框架的有效性通过数值模拟与室内试验研究得到验证。本项目的实施有助于提高对非均质地下介质中水流和溶质运移过程的理解和准确模拟，为地下水和农业水资源的管理提供有效的理论和技术支撑。