Study Of Data Assimilation For Saturated-unsaturated Flow And Solute Transport

The behavior of water flow and solute transport and transformation in saturated and unsaturated system has always been the hotspot in the field of groundwater resources and environmental protection.Many sophisticated programs have been developed to simulate flow and the coupling process of solute transport and multicomponent chemical reactions.However,the saturated and unsaturated zone under real circumstance is a complex,open system,any numerical method,governing equations or boundary conditions are only the abstract of the complex real problems.Besides,these numerical models are typically characterized by flow,transport,and reaction parameters,which can not be measured directly and need to be estimated through inverse modeling.Data assimilation has been widely used as an efficient inverse method for model parameter and state estimation.In particular,Ensemble Kalman filter(En KF)has become popular for its simplicity and affordable computational cost in saturated and unsaturated flow and solute modeling.This work firstly reviews the progress in saturated-unsaturated water flow and solute transport/transformation theories and data assimilation applications.It is supposed that there are two major difficulties in exsisting literature: the first problem is.how to characterize and quantify model structural error;the second problem is how to resolve mismatch between the spatial scale at which model simulation is run and scale at which observation is measured.It is necessary to explore the applicability of data assimilation techniques in these two complex problems,and disscuss the effects of different uncertainty sources,such as imperfectly parameter,poor initial condition,error meteorological forcing data as well as inadequate model structure,and observation scales on data assimilation performance.The detailed contents and conclusions are presented as follow:1)A Richards-based soil moisture data assimilation framework is built based on En KF algorithm and SWAP(soil-water-atmosphere-plant)model,a one –dimensional field-scale experiment is designed to examine its capability and feasibility.The parameter uncencerty is firstly considered.Results show that joint assimilating states and parameters can effectively improve the assimilation results and parameter estimation compared with open-loop results and state-only assimilation strategy.2)The assimilation considering the initial condition and meteorological upper boundary uncertainty leads to an improvement in soil moisture states and parameter estimations than that without considering these potential uncertainties.Unlike the other error sources,a poor description to the meteorological forcing has negative influence on surface soil moisture estimation,which might be attributed to the persistent disturbances of evaporation uncertainty and the lack of observations at shallow soil depth.3)Due to the strong spatial and temporal variability in field soil water,a key problem in the application of soil moisture data assimilation methods is the difference in spatiotemporal scale between the model and measurements.To this end,soil water temporal stability analysis and a random combination method from available measuring sites were proposed in reducing upscaling uncertainty for assimilating procedure.Results show that the work on soil moisture spatiotemporal characteristics has significant implications for the further exploration of optimal data assimilation strategy.This attempt which helps to decrease the efforts and guarantee the accuracy for updating through data assimilation should have high research priority.4)The transport of reactive contaminants within saturated and unsaturated system is typically affected by a large number of nonlinear and coexsisting physical,chemical,and biological processes.A coupled geochemical model based on HYDRUS-1D and PHREEQC software package is developed in this paper to simulate solute transport and nitrate reactions within one-dimension saturated flow system,but with model structural inadequacy.En KF assimilation method is integrated into the imperfect coupled model to simultaneously estimate the nitrate concentration and chemical parameters.Results show that without the explicit treatment of model structural error,parameter compensation in traditional En KF method leads to unreasonable parameter estimates and biased model predictions.The unresolved model structural error raises questions regarding suitability of traditional data assimilation algorithm5)A new hybrid framework is presented to statistically deal with model structural error during DA-based parameter estimation procedure in this paper.A dynamic datadriven error model based on Gaussian process(GP)regression is sequentially integrated into En KF data assimilation,hereby named En KF-GP method.The hybrid En KF-GP method jointly infers model parameters and structural error,and thereby reduces the impacts of parameter compensation on model prediction.For an imperfect reactive transport model with different model structural errors,results show that En KF-GP method significantly alleviates parameter compensation and provides improved model predictions.This work is expected to be helpful for better understanding and modeling of flow and solute transport and transformation since model inadequacy is inevitably encountered in saturated and unsaturated system.Given the ubiquity of such structural error,as well as the critical challenge of resolving it,it is hoped that the En KF-GP method proposed herein continue to be tested and applied to real-world practice.