Study On Estimating Macro-scale Coefficients Of Reactive Solute Transport

Describing contaminant transport in aquifers is critical for the remediation of groundwater pollution.Advection and dispersion model is the most widely used model.But,properties of subsurface porous media(hydraulic conductivity,porosity,dispersivity,decay rate,activity microbial biomass)are heterogeneous and directly observing or measuring of contaminant transport in subsurface is usually difficult.So the measured data is always limited.These situations make it difficult to estimate dispersivity or degradation rate for advection and dispersion model in heterogeneous aquifers.The leading question behind this work is hence,how we can properly estimate the coefficient of advection and dispersion model under various heterogeneities or limited data at macro-scale.In the early part of this thesis,stochastic method is utilized to derive field-scale reactive contaminant transport equation and corresponding macro-scale coefficients for first order decay and Mono degradation in stratified aquifers.The effects of various heterogeneities of subsurface porous media on field-scale solute transport are analyzed.Emphasis is placed on effects of local transverse dispersion.However,the application of these analytical works is limited due to the assumptions of stationarity and ergodicity.In recent years,the deep learning method has achieved a groundbreaking development in image identification and has began to be used as a substitution method to solve problems in different research fields.Deep learning method can process data with different layers and automatically capture and transform features.So,the burgeoning deep learning method is used to build surrogate models to estimate transport and transformation coefficients in the latter part.The main outcomes of this work are as follows:(i)The mean solute transport equation and corresponding macro-scale coefficients for contaminant under first order decay is derived in a stratified aquifer.The effective decay rate is always less than the mean value and local transverse dispersion can enlarge the effective decay rate.Increasing in Damkohler number and coefficient of variation of decay rate may decrease the macrodispersivity and effective decay rate.(ii)The mean solute transport equations and corresponding macro-scale coefficients for dissolved oxygen and contaminant for an oxygen limited contaminant biodegradation are derived.The effective biodegradation rate for the dissolved oxygen and contaminant and the effective retardation factor for the contaminant are always less than the mean values,due to the heterogeneities in biodegradation and adsorption processes.There is a nonlinear relationship between the magnitudes of macroscale coefficients and the mean concentration of the contaminant and the dissolved oxygen.Increasing in local transverse dispersion can relieve the effects of heterogeneities in biodegradation and adsorption processes and hence enlarge the effective biodegradation rate and the effective retardation factor.The macrodispersivity for the contaminant and the dissolved oxygen have the same rationale.(iii)Directly mapping relationship between hydraulic conductivity field and longitudinal macrodispersivity is built with convolutional neural network(CNN).From several numerical experiments,it is found that a CNN trained by a group of conductivity field with a specific variance can be used to predict macrodispersivity of conductivity field with different variance.Under a specific size of training dataset,the performance of the trained CNN decrease with increasing heterogeneity in conductivity field.(iv)A physical informed neural network for transport of contaminant under first order decay is built to predict velocity,dispersion coefficient and decay rate and identify corresponding transport and transformation processes.The deep neural network can give accurate estimates of velocity,dispersion coefficient and decay rate and identify the corresponding physical process from a small number of measured data(less than 1% of potential measured data)with 10% error.The entire spatial and temporal distribution of contaminant concentration can also be provided.The impacts of data error on estimation of coefficients increase with the number of physical processes involved in solute transport.The estimation of concentration distribution seems to be independent of solute transport scenearies.In general,this work derives new macro-scale analytical models for solute transport in stratified aquifers based on stochastic theory and the effects of local transverse dispersion on macro-scale coefficients is analyzed.Then,surrogate models or supplementary methods for estimating solute transport coefficients is build based on deep learning techniques for proof-of-concept.