Characterizing The Heterogeneity Of Aquifers Using Jointly Multi-source Data Based On Hydraulic Tomography

Numerical simulation of groundwater flow is the most important method to predict and deal with environmental geologic problems associated with groundwater(Such as groundwater pollution,mine water inrush,land subsidence).However,the complexity and diversity of geological units which constitute aquifers and cause hydraulic parameters(such as hydraulic conductivity K,specific storage S_s)to be heterogeneous in space.K and S_s play important roles in the numerical simulation of groundwater flow.On one hand,due to the difficulty of estimating S_s,the current research degree of K is higher than S_s.Some scholars may only consider the heterogeneity of K and take S_s to be homogeneous when conducting simulation studies.On the other hand,in practice,insufficient measured data with the heterogeneous information resulting in limited knowledge of the aquifer heterogeneity seriously restricts the prediction accuracy of groundwater model,which limits the forecasting and governance effect of environmental geologic problems management such as groundwater pollution.To better have a good knowledge of the heterogeneous characteristic of aquifers,this thesis took the laboratory sandbox heterogeneous aquifer as the research object,and first conducted aquifer hydraulic experiments with different stimuli(hydraulic tomography(HT)surveys,natural gradient(NG),precipitation/infiltration(PI)events and permeability tests of rock cores to collect multi-source data,including kinds of direct(such as K values of in-situ core sampling points)and indirect measurement data(such as observed heads from experiments of different stimuli,electric conductivity from electrical resistivity tomography)of aquifers.Then,the high-precision hydraulic parameter estimates were obtained by data fusion based on hydraulic tomography(HT),and deeply understand the spatial variability law of K and S_s,and when a project cannot obtain accurate K and S_s estimates due to insufficient measured data,which provides a scientific basis for the groundwater model assignment.The results lead to the following conclusions:(1)Utilizing the simultaneous successive linear estimator to interpret heads of different stimuli experiments to obtain K fields,which revealed the effectiveness of characterizing heterogeneous characteristics of different experiments and the uncertainty of predicting the drawdown based on different estimated K fields.HT surveys yielded the highest resolution of the estimated conditional effective K field,which can accurately realize the fine characterization of strata.The estimated K field from NG experiment followed,which could reflect the the heterogeneous characteristics roughly and met the project of imprecise requirements of estimated parameters.The head measurements from the PI experiment were least effective,which can only capture the location of low-K zones,incorrect boundary flux and inadequate flow modeling in the unsaturated zone might taint the conditional effective K field from PI events.This conditional Monte Carlo simulation based on the successive linear estimator algorithm and the Karhunen-Loeve expansion method was accomplished to address the uncertainty.The uncertainty of drawdown predictions decreases gradually with the increase of the estimate accuracy.HT yielded the least uncertainty of drawdown predictions,and those of PI events were the largest.(2)The cross-correlation analysis between K and observed heads could be used to explain the inner mechanism of the difference in estimates.The cross-correlation spatial patterns on the same observation well of different stimuli are different under the same groundwater monitoring network,which means that the head measurements collected at the same observation locations of different stimuli carry nonredundant information about heterogeneity of hydraulic parameters.Therefore,data fusion is helpful to improve the accuracy of hydraulic parameters estimates.Successive pumping tests in HT surveys at different locations create different flow fields in order to product non-fully-redundant information about aquifer heterogeneity,the regions of high cross-correlation values cover almost the entire area of the aquifer.Thus,HT surveys can obtain the highest precision hydraulic parameter estimates.The flow direction of NG and PI events is monotonous,and the regions of high cross-correlation values only cover part area of the aquifer.Thus,the information of aquifer heterogeneity is less than HT surveys.(3)Based on HT surveys and the simultaneous successive linear estimator,a method system was established to obtain the high-precision K estimates by integrating the multiple kinds of data of aquifers.Taking the correct K data of in-situ core sampling points as an additional constraint of the calibration model of head data from aquifer hydraulic tests can improve the K estimates and correct the stratigraphic boundary,because the in-situ sample parameters contain in-situ geological information.In particular,the improved K estimates of NG events not only can ensure the accuracy,but also meet the accuracy requirements of common projects without pumping tests.It is not satisfactory to use one inverse model to jointly integrate all heads of different experiments to obtain K estimates.The result of one experiment containing more correct information can be used as the prior information of HT inverse model to improve the accuracy of estimate and reduce the uncertainty of predicted drawdown at the same time.Cluster analysis deals with the physical property data(such as K and electrical conductivity)of the aquifer medium,which can generally obtain the interlayer zoning of the aquifer(manifested as interlayer heterogeneity).The K value of each zone can be obtained from the specified permeability value of the in-situ core permeability measurement,and then we obtained useful heterogeneous information between layers.The estimated K field from HT after interlayer zoning by clustering analysis can be used as the initial K field of the HT inverse model.The accuracy of K estimates can be greatly improved without adding pumping tests and observation wells.There is no general petrophysical model between K and electrical conductivity,which varies with the formation.In this thesis,clustering analysis is used to convert the electrical conductivity obtained by low-cost electrical resistivity tomography(ERT)into the K field which can be served as the prior information of HT inverse model.After corrected by heads,high-precision K estimates can be obtained even in the areas lacking observed head data.It is possible that more non-redundant data collected by geophysical exploration are used in the Sim SLE inversion process.This method can be general method to fuse the data of electrical conductivity,head and parameters of in-situ core sampling points to accurately characterize aquifer heterogeneity.(4)Based on the comprehensive analysis of results of sandbox experiments and field examples in literatures,and combined with the mathematical empirical formula of K and S_s,we systematically analyzed their spatial distribution regularities,and when their estimates cannot be accurately obtained due to the insufficient measured data,our results can provide a scientific basis for assigning parameters in groundwater modeling.K and S_s are closely related to the physical properties of the aquifer medium.However,due to the correlation scale and spatial variability,their spatial variation rules are different.The K estimates is positively correlated with the particle size of sands,and the correlation scales of the estimated ln K values agree with the average dimension of the real sand layers of different grain sizes.The correlation scales for ln S_s are longer than ln K,and There is no clear relationship between S_s and the particle size of the sands,indicating that S_s is less dependent on the grain size.The results of sandbox experiments show that S_s generally show a decreasing trend with an increasing depth,but K does not show a similar trend due to limit of the vertical scale of the sandbox aquifer.The changes in the void ratio of the aquifer matrix over the depth due to overburden stress determine the pore-volume compressibility and matrix permeability at specific depths within aquifers,which in turn supports the depth-dependent profiles of S_s,and S_s had a high sensitivity to overlying materials stresses than K.In porous media,there is no clear spatial correlation exists between K and S_s,and there is not a viable empirical relational expression to estimate S_s from K.Although it is difficult to estimate the S_s filed and its variability is smaller than that of K,ignoring its heterogeneity will reduce the prediction accuracy of groundwater flow.Therefore,in the analysis of unsteady-state groundwater flow,the heterogeneity of K should not be considered only,let alone the equivalent homogenous model,and the heterogeneity of S_sshould be considered.When hydraulic test data are lacking,accurate knowledge of a decreasing S_s with aquifer depth could improve groundwater flow analyses and groundwater storage assessment in aquifers for groundwater resource management.