WATER PEACE AT HOME WATER PEACE IN THE WORLD

General

Aquifer permeability change caused by a near‐field earthquake, Canterbury, New Zealand

 

H. K. Rutter 

S. C. Cox 

N. F. Dudley Ward 

J. J. Weir

First published:01 November 2016

https://doi.org/10.1002/2015WR018524

Citations: 7

Abstract

The MW 7.1 Darfield (Canterbury) earthquake, 4 September 2010, generated widespread hydrological effects in New Zealand ranging from instantaneous changes of piezometric levels, to more sustained postseismic changes in spring flow, river discharge and groundwater levels, and increased turbidity and declined yields of water abstracted from wells. Four years later, piezometric levels remained elevated in deeper (>40 m) aquifers along the north‐western (upper) side of the Canterbury Plains near the Greendale Fault, with changes in mean piezometric level reaching +13 m. Linear reservoir modeling (eigen modeling) suggests that sustained high groundwater was not the result of changes in abstraction or land surface recharge. Step‐drawdown tests at six wells within 15 km of Greendale Fault were carried out prior to the earthquake and were retested following fault rupture. Eden‐Hazel analysis of discharge/drawdown relationships discriminates potential sources of head losses, and how these changed (or otherwise) as a result of the earthquake. Objective application of Eden‐Hazel analysis provided confidence levels for the interpretation, including when step tests provide reliable/unreliable estimates of transmissivity change. Increases in both aquifer losses and well losses were observed in four wells, reflecting both a change in sediment transmissivity and decrease in well efficiency. At two locations, the data were unable to provide results that can be interpreted with confidence. As the majority of local groundwater flow occurs through high‐permeability open framework gravel lenses, we suggest that reduction in the permeability of these gravels, due to fine‐sediment incursion, is the cause of the reduction in transmissivity and increase in well losses.

1 Introduction

Earthquakes cause both temporary and permanent strains in the Earth's subsurface, so both immediate response of groundwater levels and/or some form of transient through to permanent change in hydrologic properties and groundwater flow can be expected. Ground close to faults becomes stressed or destressed as a result of the offset caused by the fault (static stress). The passage of seismic waves also generates dynamic (transient) stresses in bed rock and aquifers. Both stresses increase with the magnitude of the earthquake but decay quite differently with distance from the epicenter [Manga and Wang2007]. Static stresses decrease at approximately 1/r3 (where r is distance from an earthquake source), whereas dynamic stresses are proportional to the seismic wave amplitude, for the most part decreasing more gradually (∼1/r2). As a result, both dynamic and static stress may be significant close to an earthquake epicenter (at near‐field and intermediate‐field distances of up to a few fault lengths), but only dynamic stresses are likely to be significant at greater (far‐field) distances.

There is a desire to elucidate the mechanism(s) that cause changes in groundwater levels and stream flows following earthquakes from both an academic view point and practical standpoint. Static strain is generally thought to result in fluid flow toward dilational sites and away from compressional sites which is widely adopted as a model explaining mineral deposit formation [e.g., Sibson1994]. Surface observations of sustained changes in groundwater level in the near‐field to intermediate‐field have also been attributed to coseismic static strain changes, with an increase in piezometric levels in zones of contraction and a fall in zones of dilation [Jonsson et al., 2003]. However, hydraulic responses to earthquakes are complex. Other workers have found responses are commonly inconsistent with the patterns of static strain [e.g., Cox et al., 2012Shi et al., 2015], and there is an extensive collection of literature relating observed water level changes to dynamic stresses, particularly at far‐field distances [see Wang and Manga2010a2010b, and references therein]. In an investigation of the response of 30 m deep wells to three local and five distant earthquakes, Roeloffs [1998] considered that static strain could not explain the water level rises partly due to the fact that the water level movement was in the wrong direction. Kitagawa et al. [2006] found that only about half of the wells recorded a water level change that was consistent with the coseismic strain, implying that dynamic strains may also be important.

In contrast to static deformation, the dynamic deformation of sediments is affected by inertial forces, which depend on loading rates and number of loading cycles [Manga and Wang2007]. Permanent deformation occurs when the magnitude of shear strain reaches threshold values, which vary for different rocks and geological conditions. A notion that earthquakes could change permeability was first proposed by Waller [1966], who suggested that a rearrangement of aquifer grains could cause compaction and reduce permeability. However, following the Alaskan 1964 earthquake, groundwater levels fell, rather than increased (as might be expected through aquifer compaction), so Waller could only visualize a strain that could rearrange particles such that there was increase in permeability. He suggested that this might be due to the long time period of the earthquake and aftershock sequence, and the refraction of seismic waves by different geologies.

Potential permeability changes as a result of earthquakes have now been documented by many workers, with various reasons postulated as to the mechanisms involved. At far‐field distances, causative stresses are likely to be transient and not expected to induce a permanent strain in the rock mass, so proposed mechanisms involve transient phenomenon, such as mobilization of particles (including colloids) and/or mobilization of bubbles and droplets trapped in pores by capillary forces [e.g., Brodsky et al., 2003Montgomery and Manga2003Manga et al., 2012]. Such transient stresses have also generated both permeability increases and decreases in small‐scale laboratory experiments [e.g., Liu and Manga2009Elkhoury et al., 2011].

Although permeability changes have been inferred through studies of stream responses, groundwater levels, and temperature [e.g., Rojstaczer et al., 1995Wang et al., 2004a2004bManga and Rowland2009Wang et al., 2012, 2013], or quantified through injection experiments and strain measurement [e.g., Kitigawa et al., 2007Mukai and Fujimori2007] or tidal analysis of boreholes [e.g., Elkhoury et al., 2006], surprisingly few studies have attempted to define earthquake‐induced changes using standard pump testing methods. To our knowledge, only Jang et al. [2008], who described changes in an alluvial fan aquifer as a result of the 1999 M7.3 Chi Chi earthquake in Taiwan, have used pump testing to define aquifer property changes due to seismic shaking (but see also Brodsky et al. [2003] and Doan et al. [2006]). Aquifer parameters were explored using constant rate aquifer tests in three wells before and after the earthquake by Jang et al. [2008]. There were decreases in the postearthquake storativity at two of the wells, where analysis indicated that aquifer sediments were compressed and the aquifer storage capacity had decreased as a result of the earthquake. Changes in the postearthquake transmissivity ranged from a 61% increase to 0.8% decrease. Their conclusion was that, after the earthquake, a decrease in storativity and an increase in transmissivity decreased the pumping drawdown, whereas a decrease in storativity and no transmissivity changes caused an increased drawdown.

In this contribution, we examined sites in unconsolidated Quaternary gravel aquifers of the Canterbury Plains in New Zealand, where hydrological changes were observed as a result of the Mw 7.1 Darfield earthquake on 4 September 2010. Our aim was to explore the potential for hydrological changes to have been caused by either changes in recharge to/discharge from the aquifer system or a change in aquifer properties. First, we identified locations where significant long‐term hydrological effects have occurred, lasting at least 3 years after the earthquake. The observed pattern of long‐term hydrological effects is inconsistent with zones of contraction and dilation expected about a strike‐slip fault. Eigen modeling suggests that the long‐term changes cannot be attributed to changes in rainfall or abstraction and so potentially reflect changes in aquifer properties. We then tested this hypothesis using repeat step‐drawdown tests, comparing preearthquake and postearthquake results in order to identify changes caused by the earthquake. Step‐drawdown testing is commonly used to determine aquifer transmissivity, evaluate the efficiency of a new well, and determine the characteristics of an aquifer. It can also provide information on the condition of an existing well, such as whether well losses had changed and might be improved by some form of rehabilitation (known as well development). From step‐drawdown tests on four of the six wells tested, we show clear earthquake‐related changes in well performance and aquifer permeability, which we infer to be widespread in the near‐field region of the earthquake. The results are important, not only because of user concerns and the economic importance of irrigation but also because there are so few international examples of testing and aquifer property changes following large earthquakes.

2 Setting

2.1 Background to the Darfield Earthquake

On 4 September 2010 at 0436 (NZST), a moment magnitude (Mw) 7.1 earthquake occurred near Darfield, Canterbury in the South Island of New Zealand (Figure 1) [Gledhill et al., 2011]. A concealed fault beneath the Canterbury alluvial outwash plains ruptured to the surface to produce the 30 km long E–W striking Greendale Fault, that had predominantly right‐lateral strike‐slip motion with maximum displacement of 5.3 m [Quigley et al., 2010, 2012]. A rich aftershock sequence ensued, the largest being a Mw 6.3 earthquake on 22 February 2011 [Kaiser et al., 2012] beneath the already damaged city of Christchurch. The previously unknown Greendale Fault ruptured through an inland area of unconfined and semiconfined alluvial aquifers, where piezometric levels are typically below surface. There were coseismic and postseismic changes in groundwater levels and stream flow as a result of the earthquake, with responses varying both spatially and with well depth [Cox et al., 2012Gulley et al., 2013]. Although not amongst those monitored with transducers, there were at least four wells immediately adjacent to Greendale Fault that farmers reported had developed artesian flows in the first 24 h after the earthquake, bringing turbid water to the surface where piezometric levels are normally around 20–50 m below ground (Figure 1) [Cox et al., 2012]. Artesian flows subsided in the following days to weeks, but piezometric levels remained high or rose in many deeper wells, particularly along the north‐western (upper) side of Canterbury Plains.

 

image Figure 1

Open in figure viewerPowerPoint

Geological map and cross section through the Canterbury aquifer system (section adapted after Environment Canterbury). Blue contours show the symmetric distribution of coseismic (12 h) piezometric groundwater level changes mapped in deep aquifers by Cox et al. [2012].

Uplift and subsidence reached 1.5 m along the fault trace [Quigley et al., 2012] but at distances of five kilometers or more from the fault they were no greater than 200 mm [Beavan et al., 2010]. As the changes in ground elevation were between 1 and 2 orders of magnitude smaller that the changes in mean groundwater level and were centered about the Greendale Fault, the vertical movement on the fault cannot account alone for the observed spatial pattern in groundwater change. While of great interest, specific relationships between hydrological changes in groundwater with degrees of shaking and deformation recorded by GeoNet [2015] are the subject of a future paper and are not further discussed here.

2.2 Local Hydrogeology

The Canterbury Plains lie along the eastern South Island of New Zealand. They are composed of gravel alluvium and glacial outwash deposits that are around 300–600 m thick, and sit on basement rocks of Permian‐Jurassic Torlesse greywacke, or locally on a 0–1.5 km‐thick sequence of late Cretaceous‐Tertiary sedimentary rock and minor volcanics (Figure 1) [Brown2001Forsyth et al., 2008]. The alluvium is a complex of coalesced floodplain gravel and was deposited by braided glacial meltwater rivers that carried detritus eroded from the uplifting Southern Alps during the Pleistocene period. Close to the coast, changes in sea level of around 200 m, and westward transgression of the sea during interglacial periods, resulted in a coastal zone where estuarine and shallow marine sediments are interbedded with the alluvial gravels. These fine‐grained marine/estuarine sediments are now found up to 15 km inland of the present‐day shoreline. Fan and alluvial gravel sequences form aquifers, whereas the marine/estuarine sediments of silt, clay, peat, and shelly sand act as aquitards. An artesian system occurs in the coastal area due to the upward hydraulic gradient and the confining nature of the system. Further inland, the layered structure of the coastal aquifers is less obvious and groundwater is semiconfined, occurring within a heterogeneous mix of alluvial outwash gravels and finer over bank deposits. The system is schematically represented in Figure 1b. There is a prominent downstream fining and sorting of gravel clasts with transport distance in present river channels [Browne2004] that potentially also occurs within the entire aquifer system across the Canterbury Plains.

Groundwater monitoring is carried out by the Canterbury Regional Council (Environment Canterbury), who maintain over 125 local monitoring wells recording at 15 min intervals at depths ranging from 5 to 405 m. The six wells used in this study are not part of this monitoring system, although there are some water level data from these wells on the Environment Canterbury website [ECan2015]. Around 500 additional wells, including both dedicated monitoring sites and production wells, have had data collected by manual measurement at weekly, monthly, or less‐regular intervals. Regional piezometric surveys (Figures 1a and 2) indicate groundwater beneath the Canterbury Plains flows in a south‐eastward direction from the foothills toward the coast, down the topographic gradient. Groundwater flow is dominated by flow though better‐sorted, more‐permeable gravels that have been reworked and sorted by alluvial processes, known as open framework gravels (OFGs). These are separated by poorly sorted, relatively impermeable sandy and silty gravels. Of significance, are observations that groundwater flow occurs predominantly through the high‐permeability gravel lenses. Previous work suggested that 98% of the flow occurs through approximately 1% of the sediment thickness, through these relatively thin lenses of open framework gravels [Dann et al., 2008]. High well losses that occur in many Canterbury wells are a result of turbulent flow in these lenses as water is drawn into the well (V. Brunetaud, Update of the Bal equation for Canterbury Plains wells, unpublished data, Université Montpellier, France, 2008).

Throughout much of the area, there is a shallow unconfined aquifer, with a water table in hydraulic connection with adjacent surface water courses. This overlies intermediate and deeper aquifers. Groundwater yields tend to vary laterally over short distances [Bal1996], also a reflection that localized channels of more‐permeable gravel are a significant feature of the flow regime [see also White et al., 2007Lough and Williams2009]. The aquifer system is regionally very important, providing 80% of the region's drinking supply and 50% of water for agriculture [Brown and Weeber1992Brown2001].

2.3 Initial Hydrological Responses to the Darfield Earthquake

Similar to many other studies, a series of short‐term to medium‐term hydrological responses were observed within the first hours, days, and up to 1 year following the Mw 7.1 Darfield earthquake [Cox et al., 2012]. These occurred throughout New Zealand, from near‐field to far‐field distances, and can be summarized as follows:

  1. A “spike” pressure change.
  2. A positive step offset in piezometric level that is an instantaneous positive change.
  3. A negative step offset in piezometric level.
  4. Combinations of spike and offset changes.
  5. Changes in the rate of recharge or recession.
  6. Lack of a recession.

 

In addition to observed changes in water level, farmers and domestic users reported increased turbidity of water abstracted from wells in Canterbury and increased drawdown in pumping wells, indicating increased resistance to flow [ECan2011]. These short‐term to medium‐term effects had widespread impact on wells, pumps, and use of aquifers and became an immediate concern for groundwater users. Questions were subsequently asked as to whether aquifer damage had occurred, either locally near the fault or regionally across Canterbury. Preliminary assessment by Cox et al. [2012] suggested some wells continued to show effects a year after the earthquake, either in terms of a sustained change in piezometric level or in terms of well performance.

3 Observations of Long‐Term Effects on Groundwater

In the years after the earthquake, we observed that a number of the wells across mid‐Canterbury still showed a sustained rise in water levels as a result of the Mw 7.1 Darfield earthquake on 4 September 2010, some exceeding previous record high levels. To provide a spatial overview of these changes, we derived mean piezometric levels from all available groundwater level data over the period 4 years before (pre) versus the 3 years after (post) the earthquake, calculated the change in the mean levels, then carried out a statistical assessment of these data to assess whether the difference in means was significant (p < 0.05). Four years preearthquake was chosen to allow for seasonal and annual fluctuations in groundwater level, while minimizing external effects such as land use intensification and increased irrigation, which has also had an impact on groundwater levels over longer time periods. We tested different postearthquake periods, including 1, 2, and 3 years postearthquake. Within the different time periods, the majority of wells that had significant changes in piezometric levels after 1 year still had significantly different levels after 3 years, and 3 years was chosen as it made maximum use of the observations. Some of the sites and equipment suffered damage as a result of the 4 September 2010 earthquake and ongoing aftershock sequence (specifically the Mw 6.3 22 February 2011 aftershock). Consequently, not all wells could be used: any with less than 10 postearthquake data points were removed from the assessment as, due to the fact that piezometric levels can be affected by seasonal, barometric, tidal, and pumping effects, we found that using wells with less than 10 data points reduced our certainty in the results. The majority of wells across the plains suffered damage as a result of the Darfield earthquake, and subsequent aftershocks caused little damage to these, with most damage occurring to wells in the Christchurch area. The magnitude of the difference in mean groundwater level from preearthquake to postearthquake was plotted (Figure 2) and showed a systematic spatial pattern. Although sustained rises occurred throughout the Canterbury region, changes were widespread and consistent along the north‐western (upper) side of the Canterbury Plains toward the western end of the Greendale Fault. Sustained rises reached 13 m and were greater than 10 m within about one fault length (35 km) of the earthquake epicenter. By way of contrast, there appeared to be a decrease in mean groundwater level in the coastal region and/or further afield from the fault rupture. Although there are a large number of wells without any significant change, the lack of significance (in wells greater than 40 m depth) was, for the most part, found to be caused by insufficient or poor quality data, rather than an actual lack of a response to the earthquake.

 

image Figure 2

Open in figure viewerPowerPoint

Map showing wells where sustained changes in 3 year water level preearthquake to postearthquake that are statistically significant (colored by change), wells where changes were not significant (light grey, p > 0.05), and wells where repeat pump testing has been completed (black circles, labeled by well number). Grey contours at 5 m intervals show the overall shape of the potentiometric surface.

The changes observed in wells less than 40 m depth were not as consistent as those in deeper wells. When changes were examined with respect to well depth, within 10 km of the fault line, only 26% (9/35) of the shallow (<40 m depth) wells had a statistically significant difference in mean groundwater levels after the earthquake compared with before, with the typical rise being 1.0 m. A greater proportion of intermediate wells (40–80 m depth) and deep wells (>80 m depth) showed significant changes (74% or 26/35), with a typical rise in intermediate to deep wells' water levels of 6.2 m. By removing the shallow wells, it was possible to obtain a more consistent pattern in terms of the change in piezometric level. Figure 3 shows the distribution of sustained changes for wells greater than 40 m, overlain on a colored grid of this change (derived using a natural neighbors interpolation in ArcGIS). The figure illustrates the north‐east to south‐west orientation of sustained high groundwater levels along the upper, north‐western side of the Canterbury Plains, from approximately 20 km north east of the Greendale Fault to 30 km south west. The sustained increases at the western end of Greendale Fault occur both to the north and south of the fault trace, where static stress calculated from right‐lateral fault slip predict opposing zones of coulomb stress at the tips of the fault with decrease (north) and increase (south), corresponding to volumetric dilation and compression, respectively [e.g., King et al., 1994, Figure 1Sibson1994, Figure 5Zhan et al., 2011, Figure 9]. There is therefore no spatial concurrence between the pattern of static stress and piezometric level rise. Short‐term transient groundwater rises also occurred near‐symmetrically around Greendale Fault (Figure 1) [Cox et al., 2012, Figure 4]. The lack of correspondence between first‐order static stress predictions of groundwater rise/fall and observed long‐term groundwater rise suggests the mechanism responsible for the groundwater change is unlikely to be associated with static stress‐related volumetric change.

 

image Figure 3

Open in figure viewerPowerPoint

Map showing intermediate wells and deep wells where there are statistically significant sustained changes in 3 year water level preearthquake to postearthquake (colored by change). Intermediate 40–80 m depth wells colored by rectangles, deep >80 m wells (circles). Wells for which eigen models have been developed are distinguished with crosses and annotated well numbers. A region of anomalous seismic velocity at 3 km depth is shown by the purple Vp/Vs = 1.6 contour [Reyners et al., 2014]. The green contour shows the approximate position of 0.5 g peak ground acceleration experienced around the Greendale Fault during the Mw 7.1 Darfield earthquake [using Gledhill et al., 2011].

 

image Figure 4

Open in figure viewerPowerPoint

Eigen model for well L36/1226. Note the lack of a recession in 2010, which contributed to the sustained rise. Groundwater levels are shown in meters above mean sea level (m amsl). The location of this well and other wells for which eigen models were completed is shown in Figure 3.

There were many short‐term and medium‐term changes in groundwater level that were clearly coincident with the time of the earthquake that had potential to reflect the start of a long‐term response/change [e.g., Cox et al., 2012, Figure 3A]. However, we also recognized the potential for long‐term changes to reflect influences other than the earthquake, such as differences in climatic conditions and/or abstraction. To test for this possibility, we adopted an eigen modeling approach to predict expected groundwater fluctuations (caused by either natural or abstractive drivers) [Bidwell and Morgan2002Bidwell2003Pulido‐Velazquez et al., 2005Williams et al., 2008Cox et al., 2012]. The eigen model is a one‐dimensional representation of the plains that assumes that the aquifer system is one large homogeneous aquifer with all the wells responding similarly, depending on their position in the catchment. The model is a simple single‐input single‐output linear transfer function model, relating monthly water level measurements to monthly totals of recharge and pumping (estimated from a daily water balance model based on measured climate variables), and abstraction estimated from the Canterbury Regional Council's consents database. The model characterizes the dynamics of a quasi‐infinite number of linear storage reservoirs, each of which drains at an exponential rate (the eigen‐values). For this modeling, multiple zones were set up, and for each, land surface recharge and pumping were calculated. A constant river recharge was assumed. The approach involves calibration of the modeled groundwater levels to measured preearthquake groundwater levels, then prediction of postearthquake groundwater levels in the catchment, based on the postearthquake recharge and abstraction.

Eigen models were developed for 12 wells across the Canterbury Plains (Figure 3) where there was an apparent sustained rise in piezometric levels, to assess whether levels postearthquake were actually different to those preearthquake, or whether they were a reflection of a change in climate or abstraction. An example, with predicted groundwater levels is provided in Figure 4. The blue line represents the calibration period and the green line shows the predicted groundwater levels postearthquake. This model highlights a sustained offset in piezometric level of L36/1226 that has occurred during the 4 years since the earthquake. Models for nine other wells (L35/0163, L36/0064, L36/0058, L36/0092, L36/0282, L36/0663, L36/0205, M35/1000, and K36/0439) also showed that the change in piezometric level was unlikely to be attributed to changes in climate or groundwater abstraction. Results from the other two models (M36/1926 and K36/0495) also suggested this but were less convincing. Thus, the sustained rise in piezometric level is suggested to result from a change in aquifer properties and the aim of further modeling is to assess whether this is the case. The eigen models and results of numerical modeling will be presented in a future publication.

4 Method

Step‐drawdown testing provides a simple approach to estimating aquifer transmissivity (hydraulic conductivity multiplied by aquifer thickness) and accounting for losses in a pumped well [Jacob1947Eden and Hazel1973]. In a step‐drawdown test, the discharge rate in the pumping well is increased from an initially low constant rate through a sequence of pumping intervals (steps) of progressively higher (or lower) constant rates. For this study, we reviewed the area of sustained groundwater changes in the Canterbury Plains near the Greendale Fault (Figure 2), collating results of step‐drawdown testing prior to the Mw 7.1 Darfield earthquake of 4 September 2010. This study carried out new, repeat, postearthquake tests at a selection of the sites where there step‐drawdown testing before the Mw 7.1 earthquake had been reliable, in order to compare preearthquake and postearthquake estimates of transmissivity and changes induced by the earthquake.

Step testing provides a yield/drawdown relationship, which enables a quick assessment of whether or not changes might have occurred. The drawdown incorporates both well losses and drawdown due to head losses in the aquifer, and any change from preearthquake to postearthquake drawdown, is due to a change in one or both of these. A detailed analysis of step test analysis has traditionally involved a subjective, curve‐matching approach [Kruseman and de Ridder1994]. In this paper, we carry out a least squares estimation of preearthquake and postearthquake transmissivity from step test data. This was done by fitting the Eden‐Hazel model to each data set, as discussed below, which enables us to quantify the uncertainty in parameter estimates and decide whether there is a statistically significant difference between preearthquake and postearthquake estimates.

The Eden‐Hazel method [Eden and Hazel1973] is based on the following well‐known approximation to the Theis formula for drawdown in an infinite homogenous and isotropic confined aquifer:

urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0001(1)

where s = drawdown (m), Q = pumping rate (m3/d), t = time (days), and T = transmissivity (m2/d).

 

In the Eden‐Hazel method, the approximation (1) is applied to model drawdown over each constant rate interval in the pumped well together with a nonlinear turbulent loss term cQ2. If the pumped discharge Q varies as a sequence of values Qi beginning at times ti, then by the principle of superposition the drawdown at time t during the jth step, urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0002 is given by

urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0003(2)

where a, b, and c are model parameters and urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0004 is the jump in pump rate at the start of the ith step.

 

Drawdown in the pumped well is a balance of the different components of well and aquifer losses, which determine both the amount of drawdown and the shape of the drawdown curve. Thus, although we can have the same yield/drawdown relationship in different wells (or in the same well preearthquake and postearthquake), closer examination of the shape of the curves can suggest that this is due to a different combination of losses. For example, a flatter curve results from higher transmissivity (lower “b”). Although described in much more detail in Eden and Hazel [1973], it is useful to review the meaning of the coefficients here:

  1. The “a” term, is a constant related to the bore construction and includes the effective radius of the bore. Development of a well increases the effective radius, reducing head loss within this zone, and increases the efficiency of the well (decreases well losses, and hence, drawdown). An increase in this term indicates that the material outside the bore has been rearranged to be less permeable to water entering the well, whereas a decrease means that any rearrangement has resulted in less impedance to flow.
  2. The “b” term relates solely to the transmissivity and not to any losses associated with the well.
  3. The nonlinear head loss term “c” represents the turbulent entry losses to the well. In the Canterbury gravel aquifer system, with around 98% of flow being through 1% of the aquifer thickness (through open framework gravels) [Dann et al., 2008], turbulent losses are expected (and are observed) to be high due to turbulent flow through these lenses in the vicinity of the well (Brunetaud, unpublished data, 2008).

 

Note that T is inversely proportional to b through

urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0005(3)

 

The Eden‐Hazel model was fitted to preearthquake and postearthquake test data, and T was estimated from equation 3. We then carried out a hypothesis test on T, with the null hypothesis that there is no significant difference between Tpre and Tpost.

Assuming that b is normally distributed about the least squares estimate urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0006 with standard deviation δ = RMSE (root‐mean‐square error) x SD( urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0007), where RMSE is the root‐mean‐square error of the model fit, SD( urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0008) is the standrard deviation of urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0009 then T has a reciprocal normal distribution. The easiest way to carry out the hypothesis test is to draw random samples from N ( urn:x-wiley:00431397:media:wrcr22355:wrcr22355-math-0010 and obtain an approximate distribution of Tpre–Tpost. Table 2 shows the preearthquake and postearthquake estimates of transmissivity together with the p values for each well.

Table 1. Well Details With Local Estimates of Ground Surface Change (Uplift/Down Throw) After Beavan et al. [2010] at Each Well
  M36/6998 M36/7466 L36/1700 L36/0593 L36/1776 L36/2159
Latitude_WGS84 −43.5873 −43.642 −43.5637 −43.7029 −43.561 −43.6170
Longitude_WGS84 172.2690 172.413 172.0270 172.2300 172.102 172.0920
Depth (m) 87 118 113 88 208 81
Distance (m) to Greendale Fault 70 8200 1200 12,000 3900 2400
Uplift/down throw (mm) 460 −5 −448 67 −516 222
Table 2. Step Test Results With Confidence Intervals on the Coefficient Estimatesa
  M36/6998 L36/1700 L36/0593 L36/1776

Pre‐EQ

Mar 2010

Post‐EQ

Sep 2010

Pre‐EQ

Oct 2003

Post‐EQ

Oct 2014

Pre‐EQ

Sep 2009

Post‐EQ

Apr 2014

Pre‐EQ

May 2005

Post‐EQ

May 2015

a (95% CI) 0.289 (0.278–0.300) 0.300 (0.255–0.346) 0.344 (0.333–0.355) 1.795 (1.725–1.866) 8.731 (8.546–8.917) 5.578 (5.318–5.838) 4.587 (4.509–4.665) 3.419 (3.108–3.731)
b (95% CI) 0.045 (0.036–0.055) 0.117 (0.077–0.157) 0.111 (0.103–0.119) 0.255 (0.201–0.310) 0.532 (0.418–0.646) 1.862 (1.659–2.064) 0.512 (0.441–0.583) 0.640 (0.567–0.713)
c (95% CI) 0.074 (0.072–0.076) 0.187 (0.177–0.196) 0.204 (0.201–0.207) 0.762 (0.739–0.785) 8.914 (8.642–9.185) 9.137 (8.879–9.394) 0.237 (0.197–0.278) 0.542 (0.398–0.686)
RMSE 1.766 3.637 0.080 0.277 0.300 0.336 0.201 0.325
Min well losses (%) 30.0 30.0 45.4 52.3 85.0 71.0 31.5 25.1
Max well losses (%) 62.0 86.0 63.8 64.4 89.0 82.0 39.0 31.4
T (m2/d) 5830 2250 2370 1030 500 140 510 410
p value ≪0.01 ≪0.01 ≪0.01 0.007
Comments Tpre is significantly different to Tpost at the 95% CI. Analysis without recovery data Tpre is significantly different to Tpost at the 95% CI. Analysis without recovery data Tpre is significantly different to Tpost at the 95% CI. Analysis without recovery data Tpre is significantly different to Tpost at the 95% CI. Analysis without recovery data
  • a Well losses are calculated based on a and c and represent the percentage of total drawdown that is contributed by linear and turbulent well losses. These vary with pumping rate, with higher losses at higher rates. Canterbury wells are known to have high well losses, as discussed in Brunetaud (unpublished data, 2008).

An important consideration in fitting step test data (and pump test data in general) is whether the recovery data should be included in the fit. The conventional wisdom is that “more is better” and that, as a rule, you should use the recovery data. However, from a statistical point of view, one needs to ask whether the recovery data adds to the quality of the fit, and in our experience it did not for these tests. That is, inclusion of recovery data increased the error in terms of the model fit to the observed data, and can result in a less stable fit. To be consistent, for each test, we considered both drawdown and drawdown plus recovery data, and took the fit with the smaller RMSE. We also found that exclusion of one or more data points can result in quite a different model fit. This is not too surprising, particularly at the start of the steps, since it is actually quite hard, in practice, to achieve an instantaneous jump. Some of the tests had obvious operational problems, where flow at the start of a step was higher than desired, and was then adjusted down. As a result, we manually excluded spurious data at the start of each step.

5 Aquifer Test Results

Aquifer test results are examined for each well in turn. Figure 5 shows the yield/drawdown data for each pair of results for each well. The locations of wells tested (Figure 2) are shown in detail in Figure 6, together with the results. Well details are shown in Table 1 and the results summarized in Table 2. Wells are referred to by the Environment Canterbury well number. Well logs, step test results, and other information can be found in the Environment Canterbury database [ECan2015] and the results presented in this paper are provided as supporting information.

 

image Figure 5

Open in figure viewerPowerPoint

Yield‐drawdown relationships for the wells tested.

 

image Figure 6

Open in figure viewerPowerPoint

Transmissivity values for wells where both (a) preearthquake and (b) postearthquake step testing has been carried out, relative to Greendale Fault (black line) (epicenter = yellow star, initiated at 10 km depth) and major rivers (blue lines).

M36/6998. This is a 87 m deep well located in very close to the Greendale Fault trace. It is highly productive, with abstraction rates of over 100 L/s. Prior to the September earthquake, it was tested at rates of up to 103 L/s, resulting in a maximum of 5 m drawdown. After the earthquake, drawdown approximately doubled, as illustrated by the flow/drawdown curve (Figure 6a). A step test had been carried out on 17 March 2010. A second step test was carried out on 24 September 2010, after the increase in drawdown had been observed. The flow‐drawdown curves show a significant change between the preearthquake and postearthquake tests. A further test was carried out in May 2013, showing similar results to the test carried out in September 2010. The similarity of the test results for both postearthquake tests provided us with confidence about the repeatability of the step testing, and that observed changes were persistent over the nearly 3 year time span, in spite of being pumped for irrigation water throughout this time. The fact that there was little change in the step‐drawdown results over the 3 year time span also provides evidence that the drawdown testing itself did not “reset” the system and return the aquifer to its preearthquake state. Assessment of the head loss coefficients showed the observed increase in drawdown is partly due to the increase in aquifer losses (parameter b, equation 2), and also partly due to an increase in turbulent well losses (parameter c). Well losses vary with pumping rate, being lower at lower rates and increasing as flow increases. Table 2 provides the range of well losses preearthquake and postearthquake, together with the transmissivity changes. The results suggest that there is a decrease in transmissivity, but that there is also an increase in turbulent flow in, and adjacent to, the well. The implications of this, in terms of what might have changed within the aquifer, are reviewed in the Discussion. The decrease in transmissivity was significant at the 99% level.

L36/1700. This is a 113 m deep well located toward the western end of the Greendale Fault on its northern side. Although water level monitoring ceased in May 2011, the levels showed a statistically significant increase between September 2010 and May 2011, with a difference of over 10 m preearthquake to postearthquake. A step test had been carried out on 28 October 2003, prior to the earthquake. A second step test was completed on 28 October 2014 after the earthquake. Although there was a considerable time between the two tests, there were no appreciable earthquakes during this period that would have affected the well. Deterioration of the well screen occurs occasionally over extended time periods, but this would be unlikely to have significantly affected the well over 10 years (I. Haycock, McMillan Drilling, personal communication, 2016). The flow‐drawdown curves (Figure 5b) show a significant change between the preearthquake and postearthquake tests. Eden‐Hazel analysis results are summarized in Table 2. Transmissivity decreased from 2370 to 1030 m2/d, the difference being significant at the 99% level.

Assessment of the head loss coefficients (Table 2) suggests that the observed increase in drawdown is due to increases in all components of head loss, not just the laminar aquifer losses. While there is some decrease in transmissivity, there are also increases in turbulent and linear losses adjacent to, and through, the well screen.

L36/0593. This is an 89 m deep well located over 10 km south of the Greendale Fault. Prior to the September earthquake (September 2009), it was tested at rates of up to 21 L/s, resulting in a maximum of over 27 m drawdown. After the earthquake (April 2014), there was little change in drawdown (Figure 5c). While this might be taken to mean there was no change in aquifer properties, examination of the test results shows that this is a consequence of the balance between components of well and aquifer losses. The “b” term (inverse of transmissivity) has increased significantly showing a decrease in transmissivity, but the “a” term has reduced. The physical reason for a reduction in “a” might be that the material around the well screen has been rearranged in a way that reduces impedance to flow through the screen. This and the physical reasons for changes in other coefficients are covered in more detail in the discussion. Eden‐Hazel analysis results are summarized in Table 2. In spite of an appearance of little change in the yield‐drawdown relationship (Figure 7), the analysis suggested transmissivity decreased from 500 to 140 m2/d, which was significant at the 99% level (Table 2).

M36/7466. This is a 118 m deep well located about 9 km south east of the eastern end of the Greendale Fault, in an area where there are few monitoring wells with long‐term data and this area may be outside the area with a sustained rise in piezometric levels. Prior to the September earthquake (January 2004), it was tested at rates of up to 80 L/s, resulting in a maximum drawdown of nearly 7 m, and postearthquake (December 2014) at rates of up to 50 L/s, with a drawdown of 2.4 m (Figure 5d). This well showed an improvement in the yield‐drawdown relationship, and an apparent increase in transmissivity. Analysis of the results was hindered by the fact that the original test had very noisy data, the second test had a spike at the start of each step, and that the steps were flat, which means that they could not be meaningfully interpreted. Without including the recovery data, parameter b was not able to be determined from the data, and defaulted to zero. With recovery data, there was a very wide range (and hence uncertainty) in the postearthquake results. As a consequence, we were unable to interpret the data, although the reduction in drawdown does suggest an increase in transmissivity or a decrease in turbulent losses, that is, the opposite to what we observe in other wells.

L36/1776. This is a 208 m deep well located about 3.9 km to the north of the western end of the Greendale Fault. Prior to the September earthquake (May 2005), it was tested at rates of up to 47.2 L/s, resulting in a maximum drawdown of around 17.7 m, and postearthquake at rates of up to 51.3 L/s, with a drawdown of 20.3 m (Figure 5e). This well showed little change in the yield‐drawdown relationship as a result of the earthquake. However, the analysis showed that both aquifer losses and turbulent well losses had increased, whereas the “a” coefficient had decreased, suggesting less impedance to linear flow into the well. The results of the analysis showed that transmissivity decreased from 520 to 410 m2/d, and that this decrease was statistically significant (Table 2).

L36/2159. This is an 81 m deep well, located approximately 2 km south of the Greendale Fault trace, toward the western end of the fault where around 100 mm ground uplift occurred during the earthquake [Beavan et al., 2010]. Step testing had previously been carried out in June 2008 and was repeated after the earthquake in October 2013 (Figure 5f). The owner had noticed the well was not performing to the standard it had been prior to the earthquake, and this was confirmed through comparing the preearthquake and postearthquake flow‐drawdown relationships. It was not possible to reliably analyze the data. In particular, parameter “c” was not physically constrained by the drawdown data and defaulted to zero, and no reliance could be placed on the results. This underlines the difficulties in interpreting step test data for wells with high transmissivity. However, the increase in drawdown suggests that there was an increase in well losses and/or a decrease in transmissivity.

6 Discussion

Long‐term offsets in piezometric level caused by a change in aquifer properties, may be expected in the “near field” of an earthquake where the earthquake subjects the Earth's crust (including the groundwater system) to static stresses and permanent strain (deformation) In many documented examples, earthquake‐induced changes do not seem to be permanent; rather, permeability seems to return to some sort of baseline value over months to years [e.g., Elkhoury et al., 2006Xue et al., 2013]. Such coseismic changes commonly involve permeability increases in fractured rock and have been attributed to mobilization of fine particles, with permeability recovering to the preearthquake state occurring as fractures clog up again. In this study, we have observed the opposite: permeability appears to decrease, and the effects appear to last over multiple years and are potentially permanent. We hypothesize that the difference may be that, in Canterbury, permeability is associated with flow through high‐permeability gravel lenses, within unconsolidated aquifers, and that these may be clogged, rather than a fracture being unclogged, as a result of seismic shaking. The fact that there was little change in the step‐drawdown results in M36/6998 over the 3 year time span between postearthquake step tests provides evidence that abstracting water for irrigation, and the drawdown testing itself, did not “reset” the system and return the aquifer to its preearthquake state.

Step tests provide a yield/drawdown relationship for a particular well, which can be informative in terms of showing how this has changed with time. However, the details of the drawdown, in terms of both the magnitude of drawdown and the shape/slope of each step, allow us to estimate the different losses causing the drawdown. We recognized the fact that step test interpretation can be subjective, with numerous possible combinations of parameters for a single test, and hence took a more consistent approach to test analysis.

The results showed that, in all cases, “b” and “c” both increased. This suggests a decrease in transmissivity and an increase in turbulent flow in and around the well. Brunetaud (unpublished data, 2008) examined the reasons for high turbulent losses in Canterbury wells, and concluded that these were due to turbulent flow in the open framework gravels that were responsible for flow through the aquifer system. An increase in both these losses could be explained by some reduction in ease of flow through the high‐permeability gravel lenses. The change in “a” reflects a change in the “skin” effect and/or possibly storativity and is considered to reflect a rearrangement of the material close to the well (C. Hazel, personal communication, 2016). In some cases “a” increases, and in others it decreases. Both are theoretically possible as a result of the effects of the earthquake. Development of a well when it is drilled, rearranges the material around the well, increasing the effective radius and reducing the head loss (“a” is reduced). During the earthquake, water would have been surging in and out of the well, and it is possible that this surging could have further developed the well. It is also possible that, in some cases, the movement of water in and out of the well and the effects of shaking could have resulted in additional fines adjacent to the well. The transmissivity of the aquifer cannot be returned to its former state, but it is quite likely that further development of affected wells could increase the effective radius, increase the efficiency of the well and hence result in a smaller drawdown to achieve a particular discharge rate. Such development may also reduce the turbulent head loss component.

While permeability, and porosity, within the open framework gravels are likely to decrease with increased stress and positive compressive strain, dynamic stresses may also induce permanent changes, potentially through affecting void connectivity or flow paths. Such changes can occur at very small strains (e.g., <10−5) [Manga et al., 2012]. We are proposing that shaking‐induced changes in the grain‐size distribution and “unsorting,” with incursion of fine‐grained silt and sand from less‐permeable layers into the open framework gravel lenses previously sorted by alluvial processes, can conceivably change small‐scale flow paths, but have a bulk effect on transmissivity over the scale of an aquifer.

Step testing measures bulk properties at, and in the near vicinity, of the well being tested. If the flow is accepted as being predominantly within high‐permeability lenses, such as the open framework gravels (OFGs) of the Canterbury Plains [Dann et al., 2008], the transmissivity measured will be the sum of the permeability and thickness of all the lenses within the screened interval. We suggest the Mw 7.1 Darfield earthquake caused some reduction in permeability of the OFGs in the area where we observe a piezometric level rise. The changes appear to have occurred throughout the near‐field (i.e., one fault length distance) reaching beyond the area where vertical ground motion about Greendale Fault, due to combined fault offset and/or sediment settlement (compaction), has been observed by combined GPS and InSAR survey [Beavan et al., 2010]. For the OFGs, this decrease in transmissivity could be due to a number of possible mechanisms, including (i) re‐sorting of the aquifer sediments resulting in partial clogging of the open framework gravels, (ii) clasts within the gravel lenses that suffered some “repacking” during shaking such that the flow pathways have been modified (such as clast rotation suggested by Friedrich et al. [2015]) or locally (iii) truncation of the gravel lenses across the fault.

These mechanisms are likely to result in a reduction in permeability and storativity within the OFGs but do not necessarily require a major compaction of the entire sediment sequence. Given that strains of 10−6 or 10−5 are all that is required to stimulate permeability changes in wells [Manga et al., 2012], which corresponds only to a 0.1–1 mm ground subsidence over a 100 m depth, the strains responsible for the observed changes in transmissivity in Canterbury could be very small. A change in storativity during earthquakes has been used to model and explain short‐term hydrologic responses and vertical flow between aquifers [Dudley Ward2015]. Unfortunately step testing cannot provide an estimate of storativity, so further work will be required to quantify the extent to which our observed changes might involve a measurable strain and storativity change.

Our results confirm that there were changes in both aquifer and well properties after the Mw 7.1 Darfield earthquake. A decrease in aquifer transmissivity would explain the observed rises in piezometric levels. If re‐sorting of the sediments occurred, perhaps through fine‐grained material being loosened and redeposited within the gravel lenses, then it is reasonable to assume that the concomitant reduction in pore space of the individual gravel lenses would result in both a decrease in permeability and an increase in turbulent flow close to the well, due to the increased pore water velocities. In contrast, truncation of the gravel lenses across the fault might result in decreased transmissivity but would not explain any increase in turbulent losses. The pattern of sustained increase in piezometric level also does not show clear changes across the fault, as would be expected if the fault was acting as a barrier. Instead, we observe variations in the sustained changes from the north‐western (upper) side of the Canterbury Plains toward the coast. This may be a function of regional‐scale differences in degree of sorting and downstream fining that are presently observed in the present‐day rivers [Browne2004], with less‐sorted gravels nearer to the mountain source having greater fine content and preexisting potential for mobilization of fine particles during shaking.

The magnitude of the change in transmissivity does not show a direct correlation with the magnitude of piezometric level rise, but the lack of a close relationship is not unexpected. The tests are site specific, and are a measure of the properties in the immediate vicinity of the well. Transmissivity values for wells across the plains are spatially variable (Brunetaud, unpublished data, 2008): there is no regional‐scale pattern across the plains, as expected for the type of heterogeneous, alluvial and glacial outwash gravel system that exists. Similarly, there is no pattern to the decrease in transmissivity values that we obtained. The regional piezometric rise that has been mapped reflects the amalgamation of effects across the area. The step tests reflect the impacts at specific localities. In spite of the lack of a spatial correlation with the increase in piezometric levels, this work does confirm that, in the area where we see a sustained rise in piezometric level, the transmissivity has decreased. The degree to which the piezometric level rise, and also the decrease in transmissivity, is related to static stress change versus dynamic stress damage is a goal for our ongoing research.

Through the eigen models, we were able to rule out a change in climate or abstraction causing a change in piezometric levels. However, our study has not completely been able to rule out the possibility that the observed rise in piezometric level might be related to an increase in bedrock permeability and flow into aquifers driven by mountains, as hypothesized by Wang et al. [2004b2013] in Taiwan. A strong anomaly in seismic P to S wave velocity ratio < 1.6 was observed at 3 km depth in the vicinity of Greendale Fault by Reyners et al. [2014], who attributed the anomaly to widespread open fractures in the greywacke bedrock. Located in the center of the Canterbury Plains, the anomaly does not coincide with the area of sustained groundwater rise along the north‐western edge of the plains. However, we cannot rule out the possibility that there is increased recharge from the Southern Alps to the west of the fault.

Our favored interpretation is that there was incursion of fine‐grained sediment into open framework gravels, sourced from surrounding less‐sorted, less‐permeable sediments during intense shaking, which produced local and variable changes in transmissivity that were widespread throughout the near‐field region. In turn, the decrease in transmissivity has affected the flow regime, and potentially storativity, resulting in a sustained change in mean groundwater levels. That fine‐grained sediment can be mobilized in during earthquakes is most clearly evidenced in Christchurch, where the Mw 7.1 and subsequent aftershocks caused widespread surface ejection of sand and silt from liquefied layers at depth that devastated the city [Cubrinovski et al., 2012] in what was one of the most pervasive and severe liquefaction events on record. In the vicinity of Greendale Fault, where the intensity of shaking was greater, but depth to groundwater was far greater, and in gravel‐dominated sediments, little liquefaction ejecta were observed at the surface (though small blows were observed in Selwyn riverbed) but could potentially have occurred, and remained unnoticed, in any saturated, weak, fine‐grained sediment layers at depth. Supporting evidence for at least some increase in fine‐grained material within the open framework gravels is the fact that a large number of wells pumped turbid water after the earthquake [ECan2011]. Anecdotally, while the turbidity improved with time in most cases, there was not always a concomitant improvement in performance. Drillers have noted that, postearthquake, wells have been much more difficult to develop after drilling, with sand and silt being pumped for extended periods of time, and in some cases never being developed to the point that clear water was obtained (Haycock, personal communication).

If this hypothesis is correct, it raises questions as to how many times the sediments have been shaken and affected in a similar way since they were deposited, and the extent to which we the observed hydrological process may be important over geological timescales. Records from nearby GeoNet seismometers [Gledhill et al., 2011] suggest surface shaking experienced at wells that developed sustained >5 m changes in piezometric level experienced peak ground accelerations of at least 0.1 g, but most with >8 m changes experienced >0.5 g (Figure 3). The Greendale Fault is thought to have a rupture recurrence interval of between 20,000 and 30,000 years [Hornblow et al., 2014], but probabilistic shaking models for New Zealand [Stirling et al., 2012] suggest 0.1–0.5 g accelerations can be expected from other fault sources [Litchfield et al., 2014] at recurrence frequencies between 100 and 10,000 years, respectively (but prior to the Mw 7.1 Darfield earthquake, none were recorded since strong‐motion instrumentation began in 1966). Most of the shallow aquifers within the late Last Glacial outwash gravel, are only thought to be between 24,000 and 12,000 years old [Forsyth et al., 2008], but some wells may have screens in slightly older gravel pertaining to previous Late Pleistocene glacial episodes. The older aquifers could potentially have experienced 200 or more events of shaking with sufficient intensity to affect their permeability, whereas most aquifers have probably experienced less than 10 such events. During the same time period, however, there were substantial changes in climate and sea level [Clement et al., 2016] that affected base levels, drainage and groundwater flow within the Canterbury Plains. It is quite conceivable, therefore, the Darfield Mw 7.1 earthquake was potentially the first, or one of just a few, episodes to have affected flow within the open framework gravels.

We carried out an assessment as to whether transmissivity decreased with depth but obtained inconclusive results, most likely due to the high variability in transmissivity values over short spatial distances. In conclusion, over the relatively short depths of our wells, we do not see any systematic change that reflects the rate at which the aquifer is “unmixing” and changing from a well‐sorted gravel aquifer with high transmissivity into a (lower entropy state) weathered, mixed aquifer with lower transmissivity. Accounting for the natural heterogeneity and anisotropy of the aquifer system, it is difficult to say anything conclusive about how important the hypothesized process is over the geological timescale.

7 Summary and Conclusions

The Mw 7.1 Darfield (Canterbury) earthquake on 4 September 2010 generated widespread hydrological effects such that most monitored wells in Canterbury, and many far‐field wells in other regions, showed some observable response to the earthquake. Three years on from the earthquake, groundwater levels in intermediate aquifers (40–80 m) and deep aquifers (80 m or more depth) in the vicinity of the newly ruptured Greendale Fault were still elevated, with changes in mean piezometric levels reaching as much as +13 m above the mean level during 4 years prior to the earthquake. Eigen modeling indicated the sustained high water levels are not due to changes in abstraction or land surface recharge. Assuming there is no (unseen) increase in recharge to the aquifers from below or from the Southern Alps, such as might be caused by an increase in basement‐rock permeability, we adopt a simple hypothesis that the elevated groundwater levels were due to a decrease in sediment permeability. This was tested by carrying out step tests at wells and comparing the preearthquake and postearthquake well performance.

Although the interpretation of the step test data can produce ambiguous results, our results indicate that reduced well performance in four deep wells in the Canterbury Plains near Greendale Fault can be attributed to both a decrease in transmissivity of the sediments and an increase in well losses. The step testing shows that there is some reduction in transmissivity at four of the locations tested, and that this was statistically significant at the 99% level. Although a satisfactory result could not be obtained for L36/2159, the yield/drawdown relationship declined, suggesting there was a decrease in transmissivity or an increase in well losses at that location. The only well that showed the opposite result, that is, an improvement in the yield/drawdown relationship, was M36/7466, which is outside the area that we could define as having a sustained rise, and may not have been affected by the earthquake in the same way as the other wells. For wells L36/2159 and M36/7466, the data were poor and/or the model coefficients could not be determined by the data, and we were not able to derive transmissivity values in which we have confidence. A decrease in transmissivity can explain the observed sustained high groundwater levels, for which it is difficult to provide an alternative explanation.

Understanding the behavior of groundwater systems during earthquakes is important in a seismically active country that critically relies on groundwater for irrigation and drinking water. The apparent reduction in aquifer transmissivity from the test results is in agreement with these being located within the area of observed sustained groundwater level rise. An increase in piezometric level, in the absence of a change in recharge or discharge, can be sustained if there is a reduction in the permeability or thickness. As 98% of the flow occurs through approximately 1% of the sediment thickness, through relatively thin lenses of open framework gravels, any reduction in the permeability of these gravels would have an effect on the overall transmissivity of the sediments. It could also result in a reduction in porosity/storativity, as suggested by Dudley Ward [2015]. We suggest that a reduction in the permeability of open framework gravels best accounts for the decrease in transmissivity and increase in piezometric levels that has been observed.

Acknowledgments

This study was funded by the Royal Society of New Zealand Marsden Fund (2012‐GNS‐003). Many thanks for the initial discussions about postearthquake impacts with Tim Ezzy (ex‐Environment Canterbury) and the Environment Canterbury groundwater section as a whole, and for the “heads up” about changes in pumping water levels at Brian Fechney's well. Also to Bowden Consulting for allowing us to have access to the original data. Additionally thanks to Aqualinc directors (especially John Bright) who allowed us to pursue the initial investigations with no external funding. Without the support and information from both Tim Ezzy and John Bright, this project may never have been started. Konrad Weaver, David Painter, Andrew Dark, and Mark Flintoft are thanked for their contributions, discussions, ideas, and enthusiasm. Many thanks to Caroline Fraser for her expertise in R and statistical methods, allowing rapid analysis of data sets. Many thanks also to David Fletcher for helpful discussions on the statistical approach to step test analysis. Particular thanks to Colin Hazel, who assisted in interpretation of data, and was a mine of useful information with regards to step test analysis. With regards to data, the research has utilized groundwater monitoring by Environment Canterbury. Brain Fechney, Brindley Riches, Alan Garrett, Graham Wells, Jack Mackie, and John Grigg have provided access to data, access to wells, and allowed us to carry out repeat tests for no immediate benefit to themselves. Pump testing was carried out by Dan Farrow and Julian Weir (Aqualinc). The data used are listed in the manuscript and are submitted as supporting information.

Source :https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2015WR018524

Share Your Comments

Only members can comment, Click here to sign up for free right now

(Your e-mail address will not be published)
Submit Review
No Comments Yet