 Research article
 Open Access
 Published:
Identification of hydrogeological evolution using hydrogeologyseismology analysis of groundwater head fluctuation related to the 1999 M_{W} = 7.5 ChiChi earthquake
Progress in Earth and Planetary Science volume 7, Article number: 51 (2020)
Abstract
At 1:47 a.m. on September 21, 1999, the M_{w} 7.5 ChiChi earthquake struck Taiwan. The purpose of this study is to (1) apply multiple spatiotemporalfrequency analysis to filter the postseismic change in groundwater head to explore the implicit drawdown associated with excess porewater pressure release and effective stress relief during postseismic evolution and (2) establish a stochastic experimental optimization model for identifying the hydrogeological evolution. The approaches used for postseismic drawdown filtration include multirank principal components decomposition, multifrequent wavelet transforms decomposition, and multilevel wavelet denoising. This study especially evaluates the following advanced postseismic evolving parameters: (1) harmonic average leaking/injecting rate, (2) distance between the acting position and monitoring well, (3) storage coefficient under effective stress relief and formation compression, and (4) transmissivity for excessporewater pressure release. This study applies the integrated methodology on 179 monitoring wells in the ChouShui River alluvial fan. Results show that the overlying principal components PCs and lowlevel wavelet denoising can filter additional sources/sinks, in which the extracted drawdown from PC1+PC3 was related to the excess porewater pressure relaxation process, that from PC2+PC5+PC6+PC7 and highfrequency wavelet denoised detail cD_{2} related to the earth tidal fluctuation effect, and that from PC4+PC9 and cD_{3} related to the barometric effect. According to the Riemann integral and an objective function value duration curve, calculated occurrence probability from the stochastic optimization for SC2, the storage coefficient was reduced from preseismic pumping test value 0.00107, postseismic 27th hour evolving value 0.000826 to postseismic pumping test value 0.000578 in 2004, and the transmissivity increased from preseismic test value 92.4 m^{2}/h, postseismic 27th hour evolving value 98.6 m^{2}/h to postseismic test value 147.6 m^{2}/h. The results demonstrate that the SC2 and GH3 zones suffer from crustal compression and the permeability was increased to dissipate excess porewater pressure and effective stress.
Introduction
The crustal stress (strain) caused by earthquakes is a natural driving force, causing the groundwater head to change in a short period in a large area subject to a consistent stressor (Wakita 1975). Hence, it provides a medium for understanding the hydrogeological evolution of the formation through earthquakeinduced changes in groundwater head. The changes in groundwater head caused by earthquakes can be divided into coseismic oscillation and steplike changes (Wang et al. 2004a). The coseismic oscillation is mainly subject to the elastic change caused by seismic wave transmission, in which the groundwater head returns to the original head after the seismic wave dissipates. The steplike changes in groundwater head are mainly affected by the inelastic change of the crustal stress (or strain), and the groundwater head does not return to the original head immediately after the earthquake, resulting in possible permanent head changes after a period of time.
The instantaneous rise and fall of groundwater head induced by the earthquake are mainly produced by the fact that the stratum is squeezed by the seismic stress, causing the water pressure of the large area to rise, so that the excess porewater pressure may be dissipated to the surface. The coseismic steplike change in groundwater head is usually accompanied by a postseismic recovery phenomenon, which represents the porewater pressure in the stratum stabilizing. This behavior is mainly controlled by the hydrogeological characteristics of the porous medium that can be identified from the dissipation or replenishment of porewater pressure (Roeloffs 1996).
Three categories of previous studies regarding abnormal changes in groundwater head caused by earthquakes can be classified according to timed changes in groundwater head. In research focusing on preseismic hydrogeological abnormalities, Roeloffs and Quilty (1997), Koizumi et al. (2004), and Hartmann and Levy (2006) found that some monitoring wells recorded abnormal groundwater head and hydrochemical changes before earthquakes. Regarding research focusing on abnormalities in coseismic groundwater head change, Roeloffs (1996), Wakita (1975), and Kagabu et al. (2020) found that the regional distribution of coseismic groundwater head rise and fall closely coincided with areas of contraction and dilatation expected from faulting. In research focusing on postseismic hydrogeological changes, Rojstaczer and Wolf (1992), Montgomery and Manga (2003), and Petitta et al. (2018) found that earthquakes result in persistent or permanent changes in aquifer characteristics, hydrogeological parameters, and groundwater hydrology.
The 1999 ChiChi (M_{w} 7.5) earthquake was the largest to hit Taiwan in the past century. A dense network of hydrological stations on the ChouShui River alluvial fan near the epicenter of the earthquake captured the groundwater head changes before, during, and after the earthquake. The proximity to a large earthquake and the dense network of hydrological stations provided a rare opportunity to examine the effect of an earthquake on groundwater. Several studies have discussed the abnormal groundwater head changes in response to this earthquake (Chia et al. 2008; Wang et al. 2016; Wang et al. 2004b). However, large earthquakes may cause partial aquifer/aquitard rupture and an increase in the permeability of aquifer/aquitard (Elkhoury et al. 2006; Hasegawa et al. 2019; Katayama et al. 2015; Ujiie and Kimura 2014), but few previous studies considered the earthquakeproduced leaking/injecting effect on the postseismic changes in groundwater head drawdown. Moreover, the postseismic poreelastic mechanical parameters and hydrogeological parameters associated with excess porewater pressure release and effective stress relief should be in evolution, so many parameters may increase the uncertainties and error of simulating the drawdown in groundwater head. Furthermore, previous studies have seldom used a stochastic approach to simulate all possible natural hydrogeological scenarios, in which the measured groundwater head may involve additional factors such as sources/sinks and boundary conditions, where the corresponding magnitude may be filtered by proper spatiotemporalfrequency analysis.
Accordingly, the purpose of this study is to (1) apply multiple spatiotemporalfrequency analysis approaches to filter the postseismic change in groundwater head to explore the implicit drawdown associated with excess porewater pressure release and effective stress relief during an evolving period and (2) establish the stochastic experimental optimization model for identifying the hydrogeological evolution under the leaking/injecting effect. To identify the hot zones suffering from intense crustal strain caused by an earthquake, this study investigates/detects the threedimensional concussive patterns of spatiotemporal change in copostseismic groundwater head and obvious permeability increase. The approaches used for the identification of drawdown related to excess porewater pressure release include multirank principal components decomposition, multifrequent wavelet transform decomposition, and multilevel wavelet denoising. Moreover, this study establishes an indicator for evaluating the postseismic hydrogeological harmonic average evolving parameters according to the Riemann integral and an objective function value duration curve.
Methodology
Procedures
The research procedures of this study consist of four parts: (1) explicit global hydrogeologyseismology analysis across multiple zones, aquifers, and aquitards during excess porewater pressure release; (2) implicit regional postseismic spatiotemporalfrequency analysis for nonporewater pressure releaserelated groundwater head filtration; (3) establishment of the stochastic experimental optimization model; and (4) solution of the stochastic optimization model for identifying the evolving process. The flowchart of the methodology is shown in Fig. 1 and the steps are described as follows:
Step 1. According to the measured coseismic uplift and postseismic drawdown in groundwater heads, this study firstly investigates three natural phenomena in hydrogeologyseismology: (1) vertical concussion hot zones and spatial patterns of changes in copostseismic groundwater head, (2) spatiotemporal pattern of postseismic uplift/downward partial derivatives, and (3) detect the hot zones of vertical permeability increase. Accordingly, the threedimensional hot zones suffering earthquakeinduced intense crustal strain are identified and selected.
Step 2. Filter the measured groundwater head associated with the seismically induced crustal strain followed by relaxation from excess porewater pressure. Three different signal decomposition approaches are used to evaluate the strain relaxation process, highlowrank singular value decomposition (SVD), highlowfrequency wavelet transform decomposition, and highlowlevel softthresholding denoising. These approaches are used in conjunction with the measured earthoceanic tidal head, barometric pressure, and other geophysical factors.
Step 31. Embed a postseismic groundwater head drawdown simulation model into an established stochastic experimental optimization loop: (1) set an initial solution for condition ID number i, boundary condition b, and drawdown analysis scheme α; (2) set the parameterized decision variables (i.e., harmonic average leaking/injecting rate \( \overline{Q}\left(F,r,z\right) \), the distance between harmonic acting position and monitoring well \( \overline{r}\left(F,r,z\right) \), storage coefficient under effective stress compression \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \), and transmissivity for excess porewater pressure release \( \overline{T}\left(F,r,z,T\right) \); and (3) establish the postseismic groundwater head drawdown simulation model by transforming the force point source analytical solution of pore elasticity theory (Roeloffs 1996) into poreelastic homologous Theis equation to reduce the number of parameters and uncertainties under crustal straininduced leaking/injecting and excess porewater pressure release through the evolving parameters.
Step 32. Establish the stochastic optimization model, which is composed of a stochastic objective function \( {g}_{\alpha, b,i}\left(\overline{Q},\overline{r},\overline{S},\overline{T}\right) \), the physical constraints of harmonic variables \( \overline{Q}\left(F,r,z\right) \), \( \overline{r}\left(F,r,z\right) \), \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \), and \( \overline{T}\left(F,r,z,T\right) \), in which the objective duration curve (ODC) of exceedance probability vs. objective function value is computed from a series of stochastic experimental calculated g_{α, b, i}.
Step 4. Apply the simulated annealing algorithm to solve the stochastic optimization model, and output all possible excess porewater pressure release scenarios. Evaluate the postseismic hydrogeological evolving parameters according to Riemann integral and the computed ODC, and finally discuss the preseismic and postseismic evolution.
Global spatiotemporal investigation of hot zones pattern of change in copostseismic groundwater head and detection of permeability increase
This study first analyzes the explicit information embedded in the steplike sudden rise/drop in coseismic groundwater head followed by a postseismic release/recovery behavior. The designed research topics include:

(1)
Investigate the hot zone of intense changes in copostseismic groundwater head h across multiple zones κ and aquifers f using quartile analysis, in which the partial derivatives of groundwater head vs. time should be larger than 5 cm, i.e., \( \mid \frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\mid \ge 5\kern0.20em \mathrm{cm} \).

(2)
Investigate the concussion change in copostseismic groundwater head across multiple zones κ and aquifers f (\( \frac{\partial {h}_f\left(t,\kappa \right)}{\partial t} \)) by using quartile analysis and the spatial patterns of change in coseismic groundwater head across multiple aquifers (i.e., \( \frac{\partial {h}_f\left(t,\kappa \right)}{\partial {t}_0} \)).

(3)
Investigate the spatiotemporal pattern of postseismic uplift/downward partial derivatives of groundwater head vs. time during excess porewater pressure release periods t_{1} and t_{2} across multiple zones and aquifers (\( \underset{t}{\mathrm{E}}\left[\frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\left\begin{array}{l}\\ {}t={t}_1\sim {t}_2\end{array}\right.\right] \)).

(4)
Detect the hot zone patterns of obvious vertical permeability increase during excess porewater pressure release evolving periods t_{1} and t_{2} across multiple zones and aquitards T (\( \underset{t}{\mathrm{E}}\left[\frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\cdot \frac{\partial {h}_{f+1}\left(t,\kappa \right)}{\partial t}\left\begin{array}{l}\\ {}t={t}_1\sim {t}_2\end{array}\right.\right] \)), in which the aquitard T is coated by aquifer f and f + 1. The condition \( \underset{t}{\mathrm{E}}\left[\frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\cdot \frac{\partial {h}_{f+1}\left(t,\kappa \right)}{\partial t}\left\begin{array}{l}\\ {}t={t}_1\sim {t}_2\end{array}\right.\right]<0 \) for the zone κ means that the groundwater head in one of the aquifers among f and f + 1 rises and that of the adjacent aquifer falls, indicating a leak to another aquifer through aquitard T.
Hydrogeologyseismology spatiotemporalfrequency analysis of groundwater head fluctuation to filter factors related to the strain and relaxation process
According to Roeloffs (1996), the phenomenon of the steplike sudden rise/drop in coseismic groundwater head followed by a release/recovery behavior is mainly caused by crustal strain followed by relaxation. However, the postseismicmeasured groundwater heads might be affected by factors (source/sink) such as rainfall R_{rain}, artificial pumping P_{human}, earth tidal T_{earth}, oceanic tidal T_{ocean}, barometric effect B_{baro}, and other factors O (which include well loss, wellbore storage effects, and noise from automatic recording). These existing sources/sinks, which are formulated as W in Eq. (1), will affect groundwater flux. The governing equation of the threedimensional groundwater flow can be expressed as (McDonald and Harbaugh 1988):
where K_{xx}, K_{yy}, and K_{zz} are the hydraulic conductivities along the x, y, and z directions, respectively [LT^{−1}]; S_{s} is the specific storage of porous media [L^{−1}]; h is the groundwater head [L]; and W is the volumetric flux per unit volume representing sources (recharge) and/or sinks (pumpage) of the groundwater system.
To filter the postseismic drawdowns that are associated with the crustal strain followed by the relaxation process of excess porewater pressure, this study applies multirank principal component analysis, multifrequency wavelet transform, and multilevel wavelet denoising to decompose the groundwater head fluctuation into a series of intrinsic mode functions (IMFs). The detailed calculation methods are described in the following sections.
Principal component analysis (PCA)
Longuevergne et al. (2007) firstly applied PCA to decompose the groundwater head hydrograph into a series of intrinsic mode functions from lowrank to highrank components that were assigned to different root causes. This study uses single or highlow rank principal component combinations to identify the factors/causes associated with the postseismic pressure relaxation process. Singular value decomposition (SVD) used in PCA (Pearson 1901) can produce a series of noncorrelated new variables (partially measured groundwater head by a geophysical drawdown factor) for the analyzed spatiotemporal variables (originally measured groundwater head), and the deviation matrix X can be expressed by the factor score \( \tilde{\mathbf{C}} \) and the factor loading F as follows (Longuevergne et al. 2007):
where the row vector of the n × κ_{f}order matrix U is an orthonormal left singular vector {u_{1, · , f}, …, u_{n, · , f}}; \( \mathbf{D}=\operatorname{diag}\left({\varepsilon}_{1,f},\dots, {\varepsilon}_{\upkappa_f,f}\right) \); \( \boldsymbol{\Sigma} =\operatorname{diag}\left({\sigma}_{1,f},\dots, {\sigma}_{\upkappa_f,f}\right) \), and \( {\sigma}_{1,f}\ge \dots \ge {\sigma}_{\upkappa_f,f}\ge 0 \) is a singular value. The row vector of the κ_{f} × κ_{f}order matrix V is an orthonormal right singular vector \( \left\{{\mathbf{v}}_{1,\cdotp, f},\dots, {\mathbf{v}}_{\upkappa_f,\cdotp, f}\right\} \). The singular value σ_{j, f} determines the eigenvalue \( {\lambda}_{j,f}=\frac{1}{n\hbox{} 1}{\sigma}_{j,f}^2 \). Next, let \( \mathbf{F}=\left[{\phi}_{j,{\kappa}_f,f}\right] \) be the κ_{f} × κ_{f}order factorloading matrix, which \( {\phi}_{j,{\kappa}_f,f} \) is the correlation coefficient between the \( {\kappa}_f^{\mathrm{th}} \) monitoring well’s groundwater head and the j^{th} principal component coefficient.
The mathematical meaning in Eq. (2) can be interpreted as there being L independent factors causing the drawdown in groundwater head. The factor score (transpose) matrix \( {\tilde{\mathbf{C}}}^{\mathrm{T}}=\left[\tilde{{\mathbf{c}}_{1,\cdotp, f}}\kern0.5em \cdots \kern0.5em \tilde{{\mathbf{c}}_{n,\cdotp, f}}\right] \) stores n time serial observations for L = κ_{f} factors in aquifer f, where \( \tilde{{\mathbf{c}}_{t,\cdotp, f}}={\left(\tilde{c_{t,1,f}},\dots, \tilde{c_{t,j,f}},\dots, \tilde{c_{t,L,f}}\right)}^{\mathrm{T}} \). Hence, Eq. (2) can be rewritten as:
To filter out the factors that are not associated with postseismic pressure relaxation among the L factors, this study uses the partial component j of the entrywise factor score \( \tilde{c_{t,j,f}} \), factor loading \( {\phi}_{j,{\kappa}_f,f} \), and sample standard deviation for scale \( 1/{\varepsilon}_{\kappa_f,f} \) to reduce the nonrelated factors. The postseismic relaxation processrelated groundwater head \( {h}_{t,{\kappa}_f,f}^{\mathrm{post}\hbox{} \mathrm{sei}\hbox{} \mathrm{relax}} \) and the corresponding drawdown \( {h}_{t,{\kappa}_f,f}^{\mathrm{post}\hbox{} \mathrm{sei}\hbox{} \mathrm{relax}}{h}_{0,{\kappa}_f,f}^{\mathrm{post}\hbox{} \mathrm{sei}\hbox{} \mathrm{relax}} \) can be generated as follows:
where θ_{j} is the binary variable to determine whether component j is related to the postseismic relaxation process. If component j is related, then θ_{j} = 1; otherwise, θ_{j} = 0.
Wavelet transform
The application of wavelet transform to groundwater, hydrology, and water resources are still very new. Smith et al. (1998), Cannas et al. (2006), and Adamowski and Chan (2011) found that wavelet transform is a feasible data preprocessing method to isolate or filter relevant factors. Meyer (1990) and Mallat (1989) used multiscale analysis to decompose complex signals through a wave filter into approximation space/coefficients (lowfrequency major component) and detail space/coefficients (highfrequency detailed component), as shown in Fig. 2a.
This study applies the scaling filter associated with the approximation coefficient and reduces partially leveled detail coefficient to identify the groundwater head drawdown factors which are associated with the postseismic pressure relaxation process, where uses the fast discrete Daubechies wavelet transform (Daubechies 1988) to calculate the above filters and coefficients. This gives a representation in basis functions of the corresponding subspaces using a linear combination of the scaling function φ(x) and the wavelet function ψ(x), as expressed below:
with the approximation coefficient \( {a}_k^{\left(\nu \gamma \right)} \) and detail coefficient \( {d}_k^{\left(\nu l\right)} \) at k^{th} level:
where γ = 1, 2, …, ν + 1 is the γth recursion step resulting in the linear combination between the multilevel approximation and detail space, and the raw groundwater head signal x = X(t) underwent three levels of the wavelet transform, as shown in Fig. 2b. For N ∈ ℕ, a Daubechies wavelet ψ(x) and scaling function φ(x) of class D2N is defined as follows:
where N is the number of Daubechies coefficients η_{k}, and η_{0}, …, η_{2N − 1} ∈ ℝ are the constant filter coefficients satisfying the conditions:
This study uses the partial Daubechies scaling father wavelet associated with the corresponding approximation coefficients in the (ν − γ)^{th}leveled component calculated drawdown in the stochastic experiment optimization to evaluate the possible postseismic hydrogeological evolution.
Wavelet denoising
The principle of wavelet denoising is to retain the wavelet coefficient to define the factors/causes associated with the postseismic relaxation process. Suppose that there are n noisy samples of a function f^{de}(t) and we observe data \( \mathbf{x}=\left({h}_{1,{\kappa}_f,f}{h}_{0,{\kappa}_f,f},\dots, {h}_{t,{\kappa}_f,f}{h}_{0,{\kappa}_f,f},\dots, {h}_{n,{\kappa}_f,f}{h}_{0,{\kappa}_f,f}\right) \). The most general denoising problem for the noisy groundwater head signal is expressed as follows Donoho (1995):
where f^{de}(t) is a denoised function sampled at equally spaced points and {e(t)} are independent identical distributed (i.i.d.) in N(0, 1). The denoising objective is to suppress the signal noise X(t) and recover f^{de}(t). Our goal is to estimate \( {\hat{\mathbf{f}}}^{\mathrm{de}}=\left[{\hat{f}}^{\mathrm{de}}(1),\dots, {\hat{f}}^{\mathrm{de}}(t),\dots, {\hat{f}}^{\mathrm{de}}(n)\right] \) with small mean square error, as expressed below:
where \( {\hat{d}}_k^{\left(\nu l\right)} \) is the denoised detail coefficient in level ν − l. The noise is concentrated in the highfrequency IMFs, so those are removed by using the fast discrete Daubechies wavelet transform with a soft threshold. The softthresholding rule is usually referred to as wavelet shrinkage since it reduces the high amplitude detail coefficients \( {d}_k^{\left(\nu l\right)} \) so that their absolute value is below a certain threshold level, denoted by λ(l, k), which can generally be a function of the decomposition/resolution level l and the index k. This study calculates the denoised wavelet coefficients \( {\hat{d}}_k^{\left(\nu l\right)} \) using softthresholding rule \( {\delta}_{\lambda}^{\mathrm{soft}} \), which the associated threshold λ(l, k) is calculated according to Donoho (1995), as expressed below:
where σ(l, k) is the standard deviation of the noise signal in level l. The last step is to compute wavelet reconstructed denoised signal \( {\hat{f}}^{\mathrm{de}}(t) \) based on the original approximation coefficients \( {a}_k^{\left(\nu \gamma \right)} \) of level ν − γ and the modified detail coefficients \( {\hat{d}}_k^{\left(\nu l\right)} \) of levels 1 to γ.
Formulations of copostseismic groundwater head recovery equation
Force point source analytical solution
According to Roeloffs (1996), considering a saturated porous medium in an infinite domain, assuming that the medium is homogeneous isotropic soil and is subjected to three concentrated forces produced by the earthquakeinduced crustal strain along the coordinate axis (F_{r}, F_{θ}, and F_{z}) and the Qvolume porewater is injected at t = 0. When t > 0, the concentrated force continues to act on the medium. If the media displacement u_{r}, u_{θ}, and u_{z} and the excess porewater pressure P are the variables and the coordinate origin is placed on the force point source, the governing equation of the pore elasticity theory can be expressed in cylindrical coordinates:
where r is the horizontal distance from the origin, z is the depth of the force point source, ε is the volume strain (\( \varepsilon =\frac{\partial {u}_r}{\partial r}+\frac{u_r}{r}+\frac{1}{r}\frac{\partial {u}_{\theta }}{\partial \theta }+\frac{\partial {u}_z}{\partial z} \)), G is the shear modulus, η = (1 ‐ ν)/(1 ‐ 2ν), ν is the Poisson’s ratio, γ_{w} is the specific weight of water, n_{p} is porosity, β_{f} is water compressibility, δ is the Dirac delta function, and H is the Heaviside step function. If the pumping/injecting water effect (Q = 0) and the tangential force F_{θ} are ignored, assuming the forces along horizontal and vertical directions are uniformly applied (F_{r} = F_{Z} = F), letting D_{t} ' = R^{2}/4Dt and C_{0} = 1/8πGηS_{s}, where S_{s} = K/D and D = T/S is the hydraulic diffusivity which can be estimated from the transmissivity (T) and storage coefficient (S). After simplification, the point source force (F) can determine the change in groundwater head (Δh_{t}) at different positions:
where R is the distance between the force point source and monitoring point (\( R=\sqrt{r^2+{z}^2} \)), erf(·) is the error function, and s ' (t) is the drawdown of the monitoring wells. The parentheses in Eq. (19) represent the excess porewater pressure corresponding to the change in groundwater head was gradually removed as time increases.
However, during copostseismic pressure release, the hydrogeological properties in the porous media are evolving, so accurate estimations of G, η, S_{s}, F, T, S, and coefficients of the error function in Eq. (19) for different periods are difficult and should be timevariant, and should not equal the preseismic values.
Reduction of parameter and uncertainty by using transformed pore elasticitybased equation
In reality, when an earthquake or swarm earthquake occurs, the forces along the horizontal and vertical directions are usually nonuniform (F_{r} ≠ F_{Z} ≠ F) and the tangential force F_{θ} cannot be ignored, because the source, depth, and scale of each earthquake are different. Besides, the seismic point source forces often cause formation fracture, which produces the postseismic leaking/injecting Q ≠ 0. Let the harmonic average leaking/injecting rate \( \overline{Q} \) act on a position with a harmonic average distance of \( \overline{r} \) between monitoring wells. To reduce the number of variables/parameters in Eq. (19) along with uncertainties, this study uses the transformed Theis equation to simulate the postseismic groundwater head drawdown that finishes with a relaxation process. For the postseismic groundwater flow, the approximate analytical solution of the transformed Theis Equation (Theis 1935) can be expressed as follows:
where H^{0} is the initial groundwater head, \( \overline{Q}\left(F,r,z\right) \) is the harmonic average leaking/injecting rate during a postseismic poreelastic release/recovery period, \( \overline{T}\left(F,r,z,T\right) \) is the postseismic transmissivity of the aquifer for excess porewater pressure release with the leaking/injecting effect, \( \overline{r}\left(F,r,z\right) \) is the harmonic average distance between the leaking/injecting position and monitoring well, \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \) is the postseismic aquifer storage coefficient with effective stress relief and formation compression, and \( W\left(\overline{u}\right) \) is the harmonic average well function.
From Roeloffs (1996) to this section, although the transformed Theis equation cannot estimate the groundwater head at t = 0, the numbers of variables/parameters for estimating postseismic recovery groundwater head/drawdown were reduced from over 8 to 4. However, the uncertainty still exists that every possible solution along with reliable occurrence probability should be thoroughly considered.
Establishment of the stochastic experimental optimization model
This study thoroughly considers all of the possible excess porewater pressure releaserelated parameter (\( \overline{Q},\overline{r},\overline{S},\overline{T} \)) solutions in a specific postseismic period under different kinds of initial conditions i, and boundary conditions b that best fit the filtered drawdown associated with the relaxation process using ensemble analysis scheme ID number α. This study designs the uncertainty/reliability (occurrence probability) of each generated solution to be quantified by the exceedance probability of the objective function value duration curve (ODC). The lower objective function value (lower uncertainty) in the ODC would result in a higher exceedance probability. The filtered measured drawdown that is more like the actual relaxation process and better fit between the filtered drawdown and calculated drawdown would result in a lower objective function value. The optimized solutions corresponding to the objective function value is within a set threshold g^{thresh} surrounded minimum that is sent to evaluate the postseismic release scenario through hydrogeology. This study uses the root mean square error (RMSE) as an objective function g_{α, b, i} to calculate the fitting accuracy between the measured/filtered drawdown \( s{\hbox{'}}_{\alpha}^{\mathrm{mea}}(t) \) and the calculated drawdown \( s{\hbox{'}}_{b,i}^{\mathrm{cal}}\left(t,\overline{Q},\overline{r},\overline{S},\overline{T}\right) \), as expressed below:
where \( s{\hbox{'}}_{\alpha}^{\mathrm{mea}}(t) \) is the drawdown associated with the pressure relaxation filtered using analysis scheme ID number α; \( s{\hbox{'}}_{b,i}^{\mathrm{cal}}\left(t,\overline{Q},\overline{r},\overline{S},\overline{T}\right) \) is the calculated drawdown from Eq. (22) under boundary conditions b and initial condition i; Α, Β_{,} and Ι are the total number of signal analysis schemes, boundary conditions, and initial conditions, respectively; and \( {\rho}_{\cdotp, \cdotp, \cdotp}^{\mathrm{thresh}} \) is the exceedance probability threshold corresponding to the objective function value threshold g^{thresh}, in which the transformation between \( {\rho}_{\cdotp, \cdotp, \cdotp}^{\mathrm{thresh}} \) and g^{thresh} is made by the ODC.
From Eqs. (15) to (18), one can find that F is positively proportional to \( \frac{Q}{r} \) (\( F\propto \frac{Q}{r} \)). Hence, the upper and lower bounds of variables \( \overline{Q}\left(F,r,z\right) \) and \( \overline{r}\left(F,r,z\right) \) of an analyzed monitoring wellcontrolled zone ν can be evaluated by those of a reference well ρ with postseismic pumping test according to the distance r^{∗} between the crustal straininduced fault (leaking/injecting source) and monitoring well, as expressed below:
where \( {\overline{Q}}_{\rho}\left(F,r,z\right) \) and \( {\overline{Q}}_{\nu}\left(F,r,z\right) \) are the harmonic average leaking/injecting rates of the referenced and nonreferenced monitoring zone, respectively; \( {\overline{Q}}_{\rho}^{\mathrm{LB}}\left(F,r,z\right) \) and \( {\overline{Q}}_{\rho}^{\mathrm{UB}}\left(F,r,z\right) \) are the lower and upper bounds of the harmonic leaking/injecting rate of the referenced monitoring zone, respectively; \( {r}_{\rho}^{\ast } \) and \( {r}_{\nu}^{\ast } \) are the distances between the straininduced fault and monitoring well of the referenced and analyzed monitoring zone, respectively; \( {\overline{r}}_{\rho}\left(F,r,z\right) \) and \( {\overline{r}}_{\nu}\left(F,r,z\right) \) are the harmonic distances between the leaking/injecting position and monitoring well of the referenced and analyzed monitoring zone, respectively; and \( {\overline{r}}_{\rho}^{\mathrm{LB}}\left(F,r,z\right) \) and \( {\overline{r}}_{\kappa}^{\mathrm{UB}}\left(F,r,z\right) \) are the lower and upper bounds of the harmonic distance between the leaking/injecting position and monitoring well of the referenced zone ρ and all controlled zone κ, respectively.
To guarantee the reality, reliability, and accuracy of the optimized solution from a series of stochastic experimental optimization, this study sets the upper and lower bounds for decision variables according to the in situ soil formation and material, as expressed in Eq. (29). Accordingly, the governing equation of the groundwater head drawdown calculation, expressed in Eqs. (22)–(24), should also be included as constraints.
Moreover, in this study, based on the Riemann integral and mean value theorem, the optimized stochastic situational solutions of each exceedance probability and probability density were used to calculate the harmonic hydrogeological evolving parameters \( {\overline{S}}_{\kappa}\left(G,\eta, {S}_s,F,r,z,S\right) \) and \( {\overline{T}}_{\kappa}\left(F,r,z,T\right) \). First, the expected postseismic evolving parameter of each exceedance probability (EP) in the ODC is imported for the derivation to consider all scenarios. When calculating the expected parameters of two adjacent EPs such as \( {\overline{S}}_{\kappa}^{\mathrm{E}}\left({P}_{\kappa, \rho}\sim {P}_{\kappa, \rho +1},G,\eta, {S}_s,F,r,z,S\right) \) and \( {\overline{T}}_{\kappa}^{\mathrm{E}}\left({P}_{\kappa, \rho}\sim {P}_{\kappa, \rho +1},F,r,z,T\right) \), the occurrence probability is the difference between the two EPs P_{κ, ρ + 1} − P_{κ, ρ}, while the evolving parameter value is their average, namely \( \left(\frac{{\overline{S}}_{\kappa}\left({P}_{\kappa, \rho },\cdotp \right)+{\overline{S}}_{\kappa}\left({P}_{\kappa, \rho +1},\cdotp \right)}{2}\right) \) and \( \left(\frac{{\overline{T}}_{\kappa}\left({P}_{\kappa, \rho },\cdotp \right)+{\overline{T}}_{\kappa}\left({P}_{\kappa, \rho +1},\cdotp \right)}{2}\right) \). Then, to evaluate the mixed influence of porewater pressure release, effective stress relief, formation compression, and leaking/injecting effect, the harmonic evolving parameters are calculated from the summation of the expected parameters of all different EPs, as derived below:
where \( {\overline{S}}_{\kappa}^{\mathrm{HAE}}\left(\cdotp \right) \) and \( {\overline{T}}_{\kappa}^{\mathrm{HAE}}\left(\cdotp \right) \) are the postseismic harmonic average storage coefficients under effective stress compression and the transmissivity for pressure release of the zone κ, respectively; P_{κ, ρ} is the ρ^{th} EP of the evolution parameter; and ρ^{max} is the maximum serial number of EP. The aboveformulated optimization problem is nonlinear. This study solved it by using the simulated annealing algorithm in the MATLAB optimization toolbox.
Solving stochastic optimization model using simulated annealing algorithm
This study used a simulated annealing algorithm proposed by Kirkpatrick et al. (1983) to optimize/solve the minimum value of the objective function for different initial solutions i, boundary conditions b, and analytical schemes α. The decision variables are four excess porewater pressure releaserelated parameter solutions considering the leaking/injecting rate \( \overline{Q}\left(F,r,z\right) \), acting distance \( \overline{r}\left(F,r,z\right) \) between the position of \( \overline{Q} \) and a monitoring well, storage coefficient under effective stress reliefcompression \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \), and transmissivity for excess porewater pressure release \( \overline{T}\left(F,r,z,T\right) \). The optimization flowchart using simulated annealing is shown in Fig. 3, and the optimization steps are described below:
Step 1. Set initial solution i, boundary condition b, and analysis scheme α, and then set the initial temperature \( {T}_{\mathrm{c}}^0=100 \) and initial decision variables \( {\boldsymbol{\upxi}}^0={\left[\overline{Q},\overline{r},\overline{S},\overline{T}\right]}^0 \). Accordingly, input the settings into the postseismic poreelastic groundwater head drawdown simulation model to acquire the analytical solution and calculate the initial objective function value g^{0}.
Step 21. Randomly generate a neighborhood decision variable \( {\left({\boldsymbol{\upxi}}^{\hbox{'}}\right)}^{\tau }={\left[\overline{Q}\hbox{'},\overline{r}\hbox{'},\overline{S}\hbox{'},\overline{T}\hbox{'}\right]}^{\tau } \) and input it into the postseismic groundwater head drawdown simulation model to calculate the analytical solution, then calculate the corresponding objective function value (g')^{τ} for a searching iteration ID number τ.
Step 22. If (g')^{τ} ≤ g^{τ}, then accept the neighborhood solution, ξ^{τ + 1} = (ξ^{'})^{τ}. If (g')^{τ} > g^{τ}, then a number ϑ^{τ} that lies between 0 and 1, 0 < ϑ^{τ} < 1, is randomly generated and the Metropolis acceptance criterion (Metropolis et al. 1953) is used to calculate the acceptance probability, i.e., \( {P}^{\tau }=1/\exp \left(\frac{{\left(g\hbox{'}\right)}^{\tau }{g}^{\tau }}{T_{\mathrm{c}}^{\tau }}\right) \), where \( {T}_{\mathrm{c}}^{\tau } \) is the current temperature at iteration τ associated with the cooling rate (i.e., \( {T}_{\mathrm{c}}^{\tau }={T}_{\mathrm{c}}^0\times {0.95}^{\tau } \)). The annealing parameter τ is the same as the iteration number until reannealing. When ϑ^{τ} < P^{τ}, the new variable (ξ^{'})^{τ} is accepted, ξ^{τ + 1} = (ξ^{'})^{τ}; otherwise, it is not accepted, and the process returns to step 21.
Step 3. Examine whether ξ^{τ + 1} meets the stopping condition (i.e., the gradient of objective function g^{τ} vs. iteration times \( \frac{\partial g}{\partial \tau } \) is lower than 1 × 10^{−10} after 2000 iterations). If the stopping condition is satisfied, then output the optimized g value and the corresponding decision variables. If the condition is not satisfied, a reannealing scheme is added to strengthen the ability for identifying the global minimum. If 100 more new variables during the past iterations are accepted, the annealing parameter τ is replaced by \( \tau =\tau \hbox{'}=\log \left(\frac{T_0^{\tau }}{T_{\mathrm{c}}^{\tau }}\frac{\max \left(\frac{\partial g}{\partial \tau}\left\begin{array}{l}\\ {}\tau =1\sim 100\end{array}\right.\right)}{\frac{\partial g}{\partial \tau}\left\begin{array}{l}\\ {}\tau =100\end{array}\right.}\right) \) to increase \( {T}_{\mathrm{c}}^{\tau \hbox{'}} \) for jumping out of local minima; otherwise, if it has not accepted, then set τ = τ + 1 and return to step 21.
Application
Study area and preseismic hydrogeology
The study area of this research was the groundwater system of the ChouShui River alluvial fan, located in middlewest Taiwan. This alluvial fan is mainly recharged by the ChouShui River, which extends from the Baguah Mountain and Dulliu Hill in the east to the Taiwan Straits in the west and from Wu River in the north to Peikang River in the south. The radius of the alluvial fan is approximately 40 km and the area is approximately 2079 km^{2}. The ChouShui River alluvial fan can be divided into proximal, mid, and distal fan regions according to topography, geology, and strata, as shown in Fig. 4a.
The groundwater strata in the ChouShui River alluvial fan is approximately 330 m thick, which can be divided into four aquifers and four aquitards, as shown in Fig. 4b. Aquifer 1 covers the whole region and varies in thickness from 19 to 103 m. The proximal fan is mainly composed of gravel and coarse sand, which grades into fine sand and clay in the middle fan and distal fan. The depth of aquifer 2 is 35 to 217 m from the ground surface, and preseismic S is about 1.15 × 10^{−4} to 2.98 × 10^{−3} (Jang et al. 2008). Aquifer 3 covers the region at a depth of 140 to 275 m, and S is about 5.66 × 10^{−5} to 1.07 × 10^{−3}. The preseismic average transmissivity (T) of confined aquifers 2 and 3 is 50.63 and 29.59 m^{2}/h, respectively. The comparison of preseismic and postseismic groundwater head along profile AB is shown in Fig. 4f, in which only well Tanaka2(TN2) showed dropping in coseismic groundwater head and the others showed rising.
Measured hydrogeologyseismology nature resulting from the strain and relaxation process
At 1:47 a.m. on September 21, 1999, the M_{w} 7.5 ChiChi earthquake struck Taiwan. The epicenter of this earthquake was located at 23.85 °N, 120.82 °E, which is in ChiChi Town in Nantou County, at a depth of approximately 8.0 km. 157 monitoring wells on the west side of the Changhua Fault showed oscillatory and steplike coseismic changes in groundwater head, followed by stable drawdownlike decreases/increases in heads after suddenly rising/dropping, as shown in Fig. 5a.
Jang et al. (2008) had previously conducted onsite pumping tests for the SC2, Shihu2(SH2), and Yanlin2(YL2) monitoring wells in 2004, where only SC2 in aquifer 2 experience the aforementioned postseismic response. To evaluate the hydrogeological scenario in different aquifers, the GH3 monitoring well in aquifer 3 was also selected as a study subject. The monitoring wells in the middle fan that experience the seismic response were Yanlin1 (YL1), Huatan2 (HT2), and Ganghou1 (GH1) wells in aquifer 2 and Shihu3 (SH3), Hesing2 (HS2), Jiulong3 (JL3), and Dongguang4 (DG4) in aquifer 3. These seven monitoring wells were combined with the SC2 and GH3 for a total of nine wells, which the datasets for comparison and subsequent signal analyses of the precopostseismic change in groundwater head is illustrated in Fig. 5b.
Results and discussions
Investigated hot zones and pattern of change in copostseismic groundwater head across multiple subfans and aquifers
Initially, this study investigated the hot zones of intense change in copostseismic hourly groundwater head across multiple subfans through quartile analysis, in which the partial derivatives of groundwater head vs. time (Δt = 1 h) followed \( \mid \frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\mid \ge 5\kern0.20em \mathrm{cm} \), as shown in Fig. 6. From Fig. 6a, this study discovers that the concussive ranges of postseismic groundwater head of the southern upstream and downstream middle fan are larger than those of the northern middle fan, respectively. This is because the northern upstream boundary comprises the Changhua Fault, and the Chiuchiungkeng Fault is only minorly connected with the southern upstream boundary. Hence, the earthquakeinduced strain force from the ChiChi earthquake epicenter was not completely blocked by a fault in the southern alluvial fan.
The proximal fan suffers huge crustal strain force, so some extreme values of coseismic groundwater head suddenly rise over 60 m. Moreover, Fig. 6b exhibits plenty of extreme coseismic groundwater heads in the middle distal fan, indicating that this area experiences large stratumheighteninginduced excess porewater pressure. Figure 5b verifies that the stratum was heightened 0.81 to 4.52 m calculated based on the preseismic head (t = 0 h) and postseismic approaching steady head (t = 55 h). The middlenorth distal fan only sees an instantaneous drop in head, meaning that this area is in the crustal extension zone, and the middlesouth distal fan and proximal fan shows both increase and decrease in coseismic groundwater head.
This study investigated the quartile change patterns in copostseismic hourly groundwater head (\( \frac{\partial {h}_f\left(t,\kappa \right)}{\partial t} \)) across multiple subfans. Figure 7 indicates that the extreme values occur substantially across the proximal, middle, and distal fans, in which the rapid rise and drop in coseismic groundwater head range from 0.34 to 6.55 m and − 0.31 to − 2.47 m, respectively. Only the most southern distal fan did not experience the drop in head. Next, we investigated the spatial patterns in the coseismic head changes. Figure 8 shows that the increase is mainly concentrated in the middle fan, and the decrease dominates at the southern boundary and the northern proximal fan across multiple aquifers.
Investigated spatiotemporal pattern of postseismic uplift/downward partial derivatives of groundwater head vs. time
This study investigated the spatiotemporal pattern of postseismic uplift/downward partial derivatives of groundwater head vs. time (\( \underset{t}{\mathrm{E}}\left[\frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\left\begin{array}{l}\\ {}t={t}_1\sim {t}_2\end{array}\right.\right] \)) during t = 1 and t = 27 across multiple aquifers f. From Fig. 9, this study discovers that in shallow aquifers 1 and 2, the excess porewater pressure release rate is faster than that in deep aquifers 3 and 4, in which the middle fan in aquifer 1 and the proximal fan in aquifer 2 already exhibit drawdownlike drops in groundwater head during t = 1 and t = 27. Otherwise, the crustal straininduced excessporewater pressure is continuously pressurized at aquifers 3 and 4, except for the southern boundary.
Additionally, from Fig. 10, this study demonstrates that the partial derivative groundwater head in the continuously pressurized zones in Fig. 9 during postseismic t = 28 and t = 55 are transformed into decreasing zones starting to release the excess pressure. Only some portions of the southern boundary in aquifers 1–3 and northern boundary in aquifer 4 continuously suffer crustal strain pressure.
Detected spatiotemporal hot zones of obvious vertical permeability increase
Hot zones of obvious vertical permeability increase across multiple aquitards and zones κ (\( \underset{t}{\mathrm{E}}\left[\frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\cdot \frac{\partial {h}_{f+1}\left(t,\kappa \right)}{\partial t}\left\begin{array}{l}\\ {}t={t}_1\sim {t}_2\end{array}\right.\right] \)) are investigated by using the Kriging gridding method. Figure 11 shows that the condition \( \underset{t}{\mathrm{E}}\left[\frac{\partial {h}_f\left(t,\kappa \right)}{\partial t}\cdot \frac{\partial {h}_{f+1}\left(t,\kappa \right)}{\partial t}\left\begin{array}{l}\\ {}t={t}_1\sim {t}_2\end{array}\right.\right]<0 \) is met in part of the middle and proximal fans in aquitards 1 and 2 and at the distal fan in aquitard 3 during postseismic t = 1 and t = 27. During the evolving period t = 28 to t = 55, the temporal rise/drop in groundwater head between the adjacent aquifers which cladding aquitards 1 and 3 in the middle fan and proximal fan all turn into the opposite, so it means that the permeability became increasing after t = 28. Although aquitard 3 did not show permeability increase especially in the proximal fan during t = 1 and t = 27 hours, the permeability became increasing in proximal fan, middle fan, and most distal fan during t = 28 and t = 55.
Decomposed pattern of measured copostseismic groundwater heads fluctuation using PCA
From the above four explicit analysis for the steplike sudden rise/drop in coseismic groundwater head followed by a postseismic release/recovery behavior, it is clear that the middle fan of the ChouShui River alluvial fan is the complex critical evolving porewater pressure area, which is consistent with the selected groundwater monitoring wells for hydrogeological implicit evolving scenarios analysis (YL1, HT2, GH1, SH3, HS2, JL3, DG4, SC2, and GH3). Furthermore, there was no rainfall, R_{rain}, at the ChouShui River alluvial fan before or after the ChiChi earthquake, and the earthquake resulted in a large blackout, which eliminated possible impacts from artificial pumping, P_{human}.
SVD in PCA was performed on the postseismic groundwater head recovery hydrograph for nine monitoring wells. Accordingly, there were nine principal components (nine factors) produced for the analyzed zones SC2 and GH3, as shown in Figs. 12 and 13a. Results found that principal component 1 (PC1) and the first three PCs could explain 97.41% and 99.36% of the originally measured groundwater head, respectively, according to the eigenvalue of each component. From Figs. 12 and 13a, we show that the decomposed drawdown in groundwater head in PC1, PC2, and PC3 appears to be the refined drawdown and excess porewater pressure release/recovery behavior, and the other highrank decomposed drawdown may be related to other geophysical sources/sinks (earth tidal, barometric pressure, recharge) and noise. Therefore, this study used four combinations of PC1, PC1+PC2, PC1+PC3, and PC1+PC2+PC3 to decompose and filter the drawdown hydrograph from the originally measured groundwater heads, respectively, to extract the factors associated with the excess porewater pressure relaxation process, as shown in Fig. 13a. The filtered drawdown is set as \( s{\hbox{'}}_{\alpha}^{\mathrm{mea}} \) in Eq. (25) for optimization.
Wavelet transformed pattern of copostseismic groundwater heads fluctuation
One to three levels of wavelet transform were executed on the postseismicmeasured groundwater heads of SC2 and GH3 monitoring wells to calculate the approximation coefficients cA_{1}, cA_{2}, and cA_{3} (lowfrequency). These coefficients, as shown in Fig. 14a, were used to refine the drawdown identifying the composition factors associated with the relaxation process. The decomposed drawdown hydrographs using multilevel wavelet transforms are shown in Fig. 13b, in which the decomposed drawdown is set as \( s{\hbox{'}}_{\alpha}^{\mathrm{mea}} \) in Eq. (25) for optimization. The abovementioned figures demonstrate that the decomposed lowlevel drawdown hydrograph shows more refined drawdown than the highlevel drawdown hydrograph, because the Daubechies wavelet filter causes the highlevel drawdown hydrograph at log(t) > 5 h to exhibit wave shapes.
Wavelet denoising pattern of copostseismic groundwater head fluctuation
One to three levels of wavelet denoising were performed on the postseismicmeasured groundwater heads of SC2 and GH3 monitoring wells to eliminate noise and retain the factors associated with excess porewater pressure relaxation, as shown in Fig. 14b, and the corresponding drawdown was calculated as in Fig. 13c. The abovementioned figures show that the denoised lowlevel drawdown hydrograph represents the more refined drawdown and excess porewater pressure release/recovery behavior than the highlevel denoised hydrograph, because the applied softthresholding proposed by Donoho (1995) allows some detail factors in the highlevel denoised drawdown to be removed. In particular, the multilevel denoised drawdown hydrographs of monitoring well GH3 are almost the same as the originally measured drawdown, suggesting that noise in the deep aquifer is rare.
Identified hydrogeological evolving status at 27 h of postseismic strain and relaxation
Parameter setting of the stochastic experimental optimization
Before the stochastic optimization, the boundary condition and initial condition should be set properly. The monitoring wellcontrolled zone SC2 has a postseismic pumping test for the hydrogeological parameter in 2004; thus, SC2 was set as the reference well to estimate the boundary condition of GH3 which did not have a postseismic pumping test. From Fig. 4, we see the connection between the epicenter and SC2 monitoring well and between the epicenter and GH3 well all pass through the Chelungpu Fault outside of the ChouShui River alluvial fan and Changhua Fault. This study uses the Changhua Fault as a reference to calculate the distance ratio \( {r}_{\mathrm{GH}3}^{\ast }/{r}_{\mathrm{SC}2}^{\ast }=24947\ \mathrm{km}/8182\mathrm{km}=3.049 \), and then the boundary condition of GH3 well under the decreasing crustal strain transmissive force according to the distance between the well and fault, as expressed below:
Moreover, according to the in situ hydrogeological soil composition of SC2 and GH3 in a confined aquifer, the corresponding boundary condition is set as expressed below:
Furthermore, this study designs a series of experiments for optimization to ensure identifying the global minimum along with an occurrence probability calculated by the objective duration curve for the objective function value g_{α, b, i} from different kinds of initial conditions i, boundary conditions b, and analysis schemes α in Eq. (25). In the solution space (boundary conditions) for the decision variables \( {\boldsymbol{\upxi}}^{\tau }={\left[\overline{Q},\overline{r},\overline{S},\overline{T}\right]}^{\tau } \), Eqs. (28) and (29), this study uniformly generated 4800 combinations of initial decision variables (initial conditions), \( {\boldsymbol{\upxi}}_{\alpha, b,i}^{\tau } \) (i = 1,2,…,4800), for the experimental optimization.
Ensemble optimized pattern of porewater pressure and effective stress delivered through hydrogeology in SiChou(2)
After the stochastic experimental optimization, the evaluated evolving parameters (leaking/injecting rate \( \overline{Q}\left(F,r,z\right) \), harmonic average distance between the acting position and monitoring well \( \overline{r}\left(F,r,z\right) \), storage coefficient under effective stress compression \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \), and transmissivity for excess porewater pressure release \( \overline{T}\left(F,r,z,T\right) \)) for different initial conditions i, boundary conditions b, and analysis schemes α were collected. The stochastic optimized solutions for \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \) and \( \overline{T}\left(F,r,z,T\right) \) associated with the corresponding objective function value g_{α, b, i} for SC2 is shown in Fig. 15a, in which few local minima of the objective function corresponding to less uncertainly (larger reliability) and higher occurrence probability are revealed. According to the Riemann integral and an ODC calculated occurrence probability for SC2, as shown in Fig. 15c, the evolving storage coefficient was reduced from preseismic onsite pumping test value 0.00107, and postseismic 27th hour value 0.000826 to postseismic pumping test value 0.000578 in 2004 (Jang et al. 2008), and the evolving transmissivity increased from a preseismic value of 92.4 m^{2}/h, postseismic 27th hour value 98.6 m^{2}/h to postseismic value of 147.6 m^{2}/h in 2004, indicating that the SC2 zone experienced large crustal compression, and permeability was increased to dissipate excess porewater pressure and effective stress. The magnitude of change between the postseismic 27th hour evolving value and postseismic pumping test value in 2004 shows that S decreased by 49.6%, whereas T increased by 11.2%. Among all of the stochastic optimized solutions, when the decomposed combination of PC1+PC3 was used to calculate the drawdown associated with the excess porewater pressure relaxation process, as shown in Fig. 16a, the optimized objective function value (RMSE 0.06327 m) was the lowest among the three filtration analysis methods.
Ensemble optimized pattern of porewater pressure and effective stress delivered through hydrogeology in GangHou(3)
The stochastic experimental optimized solutions for \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \) and \( \overline{T}\left(F,r,z,T\right) \) associated with the corresponding objective function value g_{α, b, i} for zone GH3 is shown in Fig. 15b, in which the entire local minimum shows that the evolving transmissivity increased to release excess porewater pressure. According to Riemann integral and an ODC calculated occurrence probability for GH3, as shown in Fig. 15c, the evolving storage coefficient was reduced from a preseismic pumping test value of 0.000149 to postseismic 27th hour value 0.000112, which is a decrease of 24.8%; and the evolving transmissivity increased from a preseismic value of 28.8 m^{2}/h to postseismic 27th hour value 120.7 m^{2}/h, meaning that the GH3 zone also experienced large crustal compression and permeability was increased to dissipate excess porewater pressure and effective stress in the deep confined aquifer. Among all of the stochastic experimental optimized solutions, when postseismicmeasured groundwater heads were used to calculate drawdown and optimized parameters, the objective function value (RMSE 0.01221 m) was smaller than that from other filtration analysis methods. Furthermore, Fig. 14b2 could found that after level 1 to 3 wavelet denoising was used to refine the postseismic measured groundwater head of GH3, the results were identical to those of the original postseismic measured groundwater head. Therefore, it could be deduced that the postseismic measured GH3’s drawdown in deep aquifer is only caused by the excess porewaterpressure relaxation process.
Comprehensive discussion
Freeze and Cherry (1979) found that storage coefficient S is related to the compressibility of water and soil, so that the S can be expressed as Eq. (30). Accordingly, it can be seen that when the earthquake compresses the aquifer, the soil porosity η and the coefficient of compressibility of the soil α decrease, so that S also decreases.
where ρ is the density of water, g is the gravitational acceleration, b is the thickness of the aquifer, and α and β are the coefficients of soil and water compressibility, respectively.
According to Roeloffs (1996), the earthquakeinduced instantaneous rise or fall in groundwater heads is usually reflected by volume changes caused by the sliding (or dislocation) of aquifers composed of poroelastic media at seismic faults. If an aquifer is located at a crustal compression zone, its groundwater heads will rise; conversely, if the aquifer is located at a crustal expansion zone, its groundwater heads will fall. Lee et al. (2002) found that during the ChiChi earthquake, the westward thrusting of the Chelungpu Fault resulted in the westward displacement of the western Changhua Fault. Therefore, the region west of Changhua Fault is a crustal compression zone, and thus its groundwater head rises. Because the SC2 and GH3 monitoring wells both showed a rise in coseismic groundwater head and a fall in the postseismic identified evolving storage coefficient \( \overline{S}\left(G,\eta, {S}_s,F,r,z,S\right) \), it is proven that the earthquake resulted in crustal contraction in this region. Moreover, according to Wang et al. (2001), because of the asymmetrical shear stress effects on the soil after an earthquake, when the shear stress is sufficiently large to damage soil, it will expand and crack and resulting in a postseismic transmissivity \( \overline{T}\left(F,r,z,T\right) \) increase. Meanwhile, Montgomery and Manga (2003) noted that the occurrence of large earthquakes will cause irreversible changes in hydrogeological parameters such as storage coefficient S, hydraulic conductivity K, transmissivity T, and porosity η. Examination of the results of this study shows that the trends of the identified harmonic averages for the evolving hydrogeological parameters of zones SC2 and GH3 conform to the change mechanisms in postseismic S and T proposed by the aforementioned researchers. Therefore, the analyzed postseismic filtered drawdown hydrograph associated with the applied stochastic experimental optimization model is a useful methodology.
By comparing the frequency significance and variation pattern of the measured geophysical factors (earthoceanic tidal head, barometric pressure) with the decomposed drawdown, this study finds that the extracted drawdown from PC1+PC3 shown in Fig. 16a could be understood as the effects of the excess porewater pressure relaxation process. According to Bredehoeft (1967) and Roeloffs (1988), the M2 (12.4 h cycle) tidal constituent was used as the tidal force to calculate its effects on groundwater heads, and the results showed that a variation of 1–2 cm and 0.4–8 cm in amplitude could be produced. Similarly, the tide along the seashore near the ChouShui River alluvial fan is also mainly governed by the M2 (12.4 h cycle) tidal constituent so that there are two high tides and two low tides each day, in which the tide data from the Mailiao (ML) station measured by the Central Meteorological Bureau during the ChiChi earthquake is shown in Fig. 17c. The planimetric positions of the ML tide station and TC barometer station are shown in Fig. 4. Accordingly, the drawdown extracted from PC2+PC5+PC6+PC7 in Fig. 17a and cD_{2} in Fig. 17b show the earth tidal fluctuation effect, T_{earth}. Moreover, Jacob (1940) found that the changes in groundwater head and barometric pressure show opposite trends. The barometric data from Taichung (TC) station operated by the Central Meteorological Bureau is shown in Fig. 17d. Accordingly, the drawdown extracted from PC4+PC9 in Fig. 17a and cD_{3} in Fig. 17b could be approximately regarded as the barometric effect, B. With regard to the drawdown extracted from PC8 in Fig. 17a and cD_{1} in Fig. 17b, because its magnitude is a cyclic small hydrograph with periodicity, it could be regarded as white noise and other interference factors, O.
Conclusions
The purpose of this study was to (1) apply multiple spatiotemporalfrequency analysis approaches to filter the postseismic change in groundwater head to explore the implicit drawdown associated with excess porewater pressure release and effective stress relief during postseismic evolution and to: (2) establish the stochastic experimental optimization model for identifying the hydrogeological evolution. To identify the hot zones experiencing intense crustal strain caused by an earthquake, this study investigates/detects the threedimensional concussive patterns of spatiotemporal change in copostseismic groundwater head and vertical permeability increase. The signal analysis used to describe pore pressure release during drawdown includes multirank principal components analysis using SVD, multifrequent wavelet transform decomposition and multilevel wavelet denoising. This study especially identifies the following advanced postseismic evolving parameters: (1) leaking/injecting rate, (2) harmonic average distance between the acting position and monitoring well, (3) storage coefficient under effective stress relief and formation compression, and (4) transmissivity for excess porewater pressure release related to leaking/injecting effect. A transformed postseismic drawdown simulation model based on pore elasticity theory for parameter and uncertainty reduction is embedded into the stochastic optimization model. Moreover, this study establishes an indicator for evaluating the postseismic hydrogeological harmonic average evolving parameters according to the Riemann integral and an objective duration curve.
At 1:47 a.m. on September 21, 1999, the M_{w} 7.5 ChiChi earthquake struck Taiwan. This study applied the integrated methodology to 179 monitoring wells in the ChouShui River alluvial fan. The explicit hydrogeologyseismology analysis results show that (1) the rapid increase in copostseismic groundwater head ranged from 0.34 m to 6.55 m and the rapid decrease from − 0.31 to − 2.47 m; where the rising change in coseismic groundwater head was mainly concentrated in the middle fan, and the decreases were at the south boundary and northern proximal fan in aquifers 1, 2 and 3; (2) compared to the middle fan in aquifer 1 and the proximal fan, the groundwater head in the continuously pressurized zones at aquifers 3 and 4 except for the southern boundary during postseismic t = 28 and t = 55 are decreases and starts to release excess pressure with a lagtime of 27 h; and (3) the permeability increased in part of the middle and proximal fans in aquitards 1 and 2 and in the distal fan in aquitard 3 during postseismic t = 1 and t = 27 h. During t = 28 and t = 55, the middle and proximal fans in aquitards 1 and 3 all show increasing permeability.
Furthermore, the overlying PCs and lowlevel wavelet denoising can filter the additional source/sink factors, which affects the fluctuation of the postseismic drawdown hydrograph. By comparing the frequency significance and variation pattern of the measured geophysical factors with the decomposed drawdown, this study shows that the extracted drawdown from PC1+PC3 could be approximately regarded as the effects of the excess porewater pressure relaxation process. Accordingly, the drawdown extracted from PC2+PC5+PC6+PC7 and highfrequency wavelet denoised detail cD_{2} could be approximately regarded as the earth tidal fluctuation effect. Moreover, the drawdown from PC4+PC9 and cD_{3} could be understood as the barometric effect, and the drawdown from PC8 and cD_{1} as white noise and other interference factors.
After the stochastic experimental optimization process, according to the Riemann integral and an ODC calculated occurrence probability for SC2, the storage coefficient was reduced from the preseismic onsite pumping test value 0.00107, postseismic 27th hour evolving value 0.000826, to the postseismic pumping test value 0.000578 in 2004, and transmissivity was increased from the preseismic value 92.4 m^{2}/h, postseismic 27th hour value 98.6 m^{2}/h, to postseismic value 147.6 m^{2}/h. For GH3, the storage coefficient was reduced from the preseismic value 0.000149 to postseismic 27th hour evolving value 0.000112, which is a decrease of 24.8%, and the transmissivity increased from preseismic value of 28.8 m^{2}/h to postseismic 27th hour value of 120.7 m^{2}/h. These results demonstrate that zones SC2 and GH3 suffered large crustal compression resulting in permeability increase to dissipate excess porewater pressure and effective stress in the deep confined aquifer.
Availability of data and materials
The analyzed data was measured by the Water Resources Agency, Taiwan, and acquired from the website https://gweb.wra.gov.tw/Hydroinfo/. Due to the copyright and intellectual property rights of the data used in this study, the readers can acquire the data through the above website or by contacting the author for data requests.
Abbreviations
 CWT:

Continuous wavelet transform
 D:

Hydraulic diffusivity
 DWT:

Discrete wavelet transform
 EP:

Exceedance probability
 F:

Point source force
 IMF:

Intrinsic mode function
 ODC:

Objective function value duration curve
 PC:

Principal component
 PCA:

Principal component analysis
 RMSE:

Root mean square error
 S:

Storage coefficient
 SVD:

Singular value decomposition
 T:

Transmissivity
References
Adamowski J, Chan HF (2011) A wavelet neural network conjunction model for groundwater level forecasting. J Hydrol 407:28–40
Bredehoeft JD (1967) Response of wellaquifer systems to earth tides. J Geophys Res 72:3075
Cannas B, Fanni A, See L, Sias G (2006) Data preprocessing for river flow forecasting using neural networks: wavelet transforms and data partitioning. Phys Chem Earth 31:1164–1171
Chia Y, Chiu JJ, Chiang YH, Lee TP, Wu YM, Horng MJ (2008) Implications of coseismic groundwater level changes observed at multiplewell monitoring stations. Geophys J Int 172:293–301
Daubechies I (1988) Orthonormal bases of compactly supported wavelets. Commun Pure Appl Math 41:909–996
Donoho DL (1995) Denoising by softthresholding. Ieee T Inform Theory 41:613–627
Elkhoury JE, Brodsky EE, Agnew DC (2006) Seismic waves increase permeability. Nature 441:1135–1138
Freeze RA, Cherry JA (1979) Groundwater. PrenticeHall, Englewood Cliffs
Hartmann J, Levy JK (2006) The influence of seismotectonics on precursory changes in groundwater composition for the 1995 Kobe earthquake, Japan. Hydrogeol J 14:1307–1318
Hasegawa R, Yamaguchi A, Fukuchi R, Hamada Y, Ogawa N, Kitamura Y, Kimura G, Ashi J, Ishikawa T (2019) Postseismic fluid discharge chemically recorded in altered pseudotachylyte discovered from an ancient megasplay fault: an example from the Nobeoka Thrust in the Shimanto accretionary complex, SW Japan. Prog Earth Planetary Sci 6
Jacob CE (1940) On the flow of water in an elastic artesian aquifer. Eos T Am Geophys Un 21:574–586
Jang CS, Liu CW, Chia Y, Cheng LH, Chen YC (2008) Changes in hydrogeological properties of the River Choushui alluvial fan aquifer due to the 1999 ChiChi earthquake, Taiwan. Hydrogeol J 16:389–397
Kagabu M, Ide K, Hosono T, Nakagawa K, Shimada J (2020) Describing coseismic groundwater level rise using tank model in volcanic aquifers, Kumamoto, southern Japan. J Hydrol 582
Katayama I, Kubo T, Sakuma H, Kawai K (2015) Can clay minerals account for the behavior of nonasperity on the subducting plate interface? Prog Earth Planetary Sci 2:30
Kirkpatrick S, Gelatt CD, Vecchi MP (1983) Optimization by simulated annealing. Science 220:671–680
Koizumi N, Kitagawa Y, Matsumoto N, Takahashi M, Sato T, Kamigaichi O, Nakamura K (2004) Preseismic groundwater level changes induced by crustal deformations related to earthquake swarms off the east coast of Izu Peninsula, Japan. Geophys Res Lett 31
Lee M, Liu TK, Ma KF, Chang YM (2002) Coseismic hydrological changes associated with dislocation of the September 21, 1999 Chichi earthquake, Taiwan. Geophys Res Lett 29
Longuevergne L, Florsch N, Elsass P (2007) Extracting coherent regional information from local measurements with KarhunenLoeve transform: case study of an alluvial aquifer (Rhine valley, France and Germany). Water Resour Res 43
Mallat SG (1989) A theory for multiresolution signal decomposition  the wavelet representation. Ieee T Pattern Anal 11:674–693
McDonald MG, Harbaugh AW (1988) A modular threedimensional finitedifference groundwater flow model. USGPO
Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH, Teller E (1953) Equation of state calculations by fast computing machines. J Chem Phys 21:1087–1092
Meyer Y (1990) Wavelets, quadrature mirror filters and numerical imageprocessing. Lect Notes Math 1438:14–25
Montgomery DR, Manga M (2003) Streamflow and water well responses to earthquakes. Science 300:2047–2049
Pearson K (1901) On lines and planes of closest fit to systems of points in space. Philos Mag 2:559–572
Petitta M, Mastrorillo L, Preziosi E, Banzato F, Barberio MD, Billi A, Cambi C, De Luca G, Di Carlo G, Di Curzio D, Di Salvo C, Nanni T, Palpacelli S, Rusi S, Saroli M, Tallini M, Tazioli A, Valigi D, Vivalda P, Doglioni C (2018) Watertable and discharge changes associated with the 20162017 seismic sequence in central Italy: hydrogeological data and a conceptual model for fractured carbonate aquifers. Hydrogeol J 26:1009–1026
Roeloffs E (1996) Poroelastic techniques in the study of earthquakerelated hydrologic phenomena. Adv Geophys 37(37):135–195
Roeloffs E, Quilty E (1997) Case 21: Water level and strain changes preceding and following the August 4, 1985 Kettleman Hills, California, earthquake. Pure Appl Geophys 149:21–60
Roeloffs EA (1988) Hydrologic precursors to earthquakes  a review. Pure Appl Geophys 126:177–209
Rojstaczer S, Wolf S (1992) Permeability changes associated with large earthquakes  an example from LomaPrieta, California. Geology 20:211–214
Smith LC, Turcotte DL, Isacks BL (1998) Stream flow characterization and feature detection using a discrete wavelet transform. Hydrol Process 12:233–249
Theis CV (1935) The relation between the lowering of the piezometric surface and the rate and duration of discharge of a well using ground water storage. Eos T Am Geophys Un 16:519–524
Ujiie K, Kimura G (2014) Earthquake faulting in subduction zones: insights from fault rocks in accretionary prisms. Prog Earth Planetary Sci 1:7
Wakita H (1975) Water wells as possible indicators of tectonic strain. Science 189:553–555
Wang CY, Cheng LH, Chin CV, Yu SB (2001) Coseismic hydrologic response of an alluvial fan to the 1999 ChiChi earthquake, Taiwan. Geology 29:831–834
Wang CY, Liao X, Wang LP, Wang CH, Manga M (2016) Large earthquakes create vertical permeability by breaching aquitards. Water Resour Res 52:5923–5937
Wang CY, Wang CH, Kuo CH (2004a) Temporal change in groundwater level following the 1999 (Mw = 7.5) ChiChi earthquake, Taiwan. Geofluids 4:210–220
Wang CY, Wang CH, Manga M (2004b) Coseismic release of water from mountains: Evidence from the 1999 (Mw = 7.5) ChiChi, Taiwan, earthquake. Geology 32:769–772
Acknowledgments
We thank Mary who works for the company Editage for the English language editing services, and we thank the Water Resources Agency, MOEA, for providing valuable hydrogeologyseismology data. Besides, the authors are indebted to the reviewers for their valuable comments and suggestions.
Funding
This research was partially supported by the Ministry of Science and Technology, Taiwan (Grant Nos. MOST 1072625M002005 and 1082625M002007).
Author information
Affiliations
Contributions
Dr. ChienLin Huang proposed the topic and conceived and designed the study. Ph.D. student ChunHao Yao carried out the experimental study. Dr. ChienLin Huang and Professor NienSheng Hsu analyzed the data and helped in their interpretation. Professor NienSheng Hsu and Associate Professor Gene J.Y. You collaborated with the corresponding author in the construction of manuscript. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Huang, CL., Hsu, NS., Yao, CH. et al. Identification of hydrogeological evolution using hydrogeologyseismology analysis of groundwater head fluctuation related to the 1999 M_{W} = 7.5 ChiChi earthquake. Prog Earth Planet Sci 7, 51 (2020). https://doi.org/10.1186/s4064502000346y
Received:
Accepted:
Published:
Keywords
 Hydrogeological evolution
 Hydrogeologyseismology analysis
 Copostseismic groundwater head fluctuation
 Earthquake
 Excess porewater pressure
 Crustal compaction