CHRISTCHURCH CITY GROUNDWATER MODEL

Transcription

1 CHRISTCHURCH CITY GROUNDWATER MODEL Mike Thorley 1, Peter Callander 1, Howard Williams 1, Hilary Lough 1, Mike Kininmonth 2, Bruce Henderson 2 1 Pattle Delamore Partners Ltd Level 2, Radio New Zealand House, 51 Chester Street West, P.O. BOX 389, Christchurch, Phone , Fax , mike.thorley@pdp.co.nz Web: 2 Christchurch City Council PO BOX 237, Christchurch Abstract: One tool used in the assessment of the well security for Christchurch City Council (CCC) public supply wells is a numerical groundwater model. The groundwater model enables an integrated analysis of independently determined hydrogeologic parameters controlling groundwater flow paths to the entire network of public supply wells. The numerical model has been developed using the USGS model code MODFLOW and the Waterloo Hydrogeologic Inc pre and post processor Visual MODFLOW 4.1 which is a three-dimensional, finite difference code. The model characterization involved taking field data describing the aquifer system and translating this information into input variables that the model code uses to solve governing equations of flow. The input variables for the model were supplied by CCC and Environment Canterbury, and included topographic data, climate data, surface water flows, hydrogeological data, and groundwater abstraction data. A component of the modelling software (MODPATH) provides the ability to trace water flows backwards in time and space. Backwards-tracking particles were released from model cells surrounding a selection of 31 separate public drinking water supply wells or pumping stations. Steady state simulations were predominantly used to allow the particle tracking module MODPATH to produce the full range of path lines and travel times of particles arriving at public supply wells. The model also provides excellent visual representations of both the supply network and subsurface flow. This presents resource managers and decision makers with a clear understanding of the nature of the types of pathways through which groundwater reaches the wells. Keywords: Groundwater modelling, hydrogeology, MODFLOW, Christchurch, drinking water protection. INTRODUCTION A methodology for assessing the secure status of a large network of multiple wells has been developed by the Christchurch City Council (CCC) and Pattle Delamore Partners Ltd (PDP) to achieve the objectives of the proposed Drinking Water Standards New Zealand (MoH, 2005). One tool in this assessment methodology is the development of a numerical groundwater model. The groundwater model enables an integrated analysis of independently determined hydrogeologic parameters controlling groundwater flow paths to the entire network of public supply wells. Additionally, the model provides excellent visual representations of the subsurface flow system presenting resource managers and decision makers with a clearer understanding of how the groundwater reaches the water supply wells. This paper outlines the development of the numerical groundwater model and presents some of the outcomes determined using the model to improve the understanding of groundwater flow to the public supply wells in Christchurch City. MODEL SETUP The numerical model has been developed using the USGS model code MODFLOW and the Waterloo Hydrogeologic Inc pre and post processor Visual MODFLOW 4.1 which is a three-dimensional, finite difference code. This modelling package was chosen for its relatively short run times, particle tracking modules and easy to use interface. The model characterization involved taking field data describing the aquifer system and translating this information into input variables that the model code uses to solve governing equations of flow. Steady state simulations were predominantly used to allow the particle tracking module MODPATH to produce the full range of path lines and travel times of particles arriving at public supply wells. For instance, it was not practical to run

2 a transient simulation for up to 300 years to determine possible ages of water entering a well. Additionally, the advective transport calculations used to calculate particle movement in MODPATH was considered the most conservative approach for determining travel times. The area of interest or model domain encompasses what is considered a near complete representation of the aquifer system for the Christchurch City, and is based on previous hydrogeological investigations (Brown et al, 1992; NCCB, 1983; 1986; ECan, 1995, 1997, 2000, 2002 and 2003). For the Christchurch aquifer system, this is considered to cover an area of the Canterbury Plains from the Waimakariri River in the north, Intake Road in the west, the coast to the east, and Banks Peninsula in the south. The model domain covers an area of km 2 and was divided into a 200 m x 200 m grid consisting of 166 rows and 208 columns. Due to the orthogonal nature of the model domain, model cells considered outside the aquifer system, or that naturally act as no flow boundaries, (such as Banks Peninsula) were designated as inactive (Figure 1). Figure 1: Map showing Christchurch City Groundwater model domain, inactive flow boundaries and the active model grid. The input variables for the model were supplied by CCC and Environment Canterbury 1 (ECan), and included topographic data, climate data, surface water flows, hydrogeological data, and groundwater abstraction data. The topographical profile used in the model was developed from LIght Detection And Ranging data (LIDAR) supplied by CCC on a 50 m grid within the City limits. For the areas outside the city limits, a digital elevation model (DEM) was supplied by ECan, also on a 50 m grid, which was originally derived from 1: scale topographic data sets (NZMS 260 series). The distribution of hydraulic conductivity is based on the geological database maintained by ECan and also the conceptual hydrogeological understanding which is described in publications such as NCCB (1983; 1986). The broad hydrogeology of the Christchurch aquifer system adopted for this model consists of confined aquifers to the east (generally beneath Christchurch City) and unconfined to leaky confined aquifers to the west, with a general decrease in hydraulic conductivity with depth in both areas. More highly permeable strata associated 1 Environment Canterbury is the promotional name for the Canterbury Regional Council.

3 with recent Waimakariri River gravel deposits, that are incised into older gravel strata are also represented in the model. The grid discretization and hydraulic conductivity zonation are displayed in Figure 1 and Figure 2 respectively. Figure 2: Map showing the Christchurch City Council groundwater flow model hydraulic conductivity zonation in the upper layers. A total of 20 layers are utilised in the model, two for each of the five separate aquitard (confining) and aquifer units. Two layers per aquitard and aquifer were chosen to accommodate gradients in flow and physical properties, and to allow for realistic numerical computation of head changes across cells. The thickness of the layers was based on bore logs collected from both public and private bores which is then collated by ECan. The many intersections of bore logs with known aquifer/aquitard units can be laterally interpolated to produce a continuous layer across the model domain. Model areas containing more points (bore logs) are therefore more precise. In contrast, areas with few bores mean that layer information is less precise. The bottom of the model was chosen to be a sloping plane represented by the bases of screens in wells drilled into Aquifer 5. Surface water data were collected that detail the Waimakariri River recharge to the Christchurch aquifers, and flows within spring-fed streams that act as sinks, removing groundwater from the aquifer system. The most recent estimate of Waimakariri River recharge is considered to occur at a rate of 7.0 m 3 /s to 8.5 m 3 /s (NCCB, 1986) into the adjacent aquifers in a southward and vertical direction, from Intake Road to Harewood Stop bank. The river bed elevation was based on topographic contours used on 1: topographic maps, with an arbitrary width of 50 m used for all river cells. Stream bed conductance was altered to achieve the 7.0 m 3 /s to 8.5 m 3 /s leakage. Drain cells were used to represent the spring-fed streams that are found throughout the Christchurch City area including the South Branch of the Waimakariri River, Styx River, Avon River, Heathcote River and Halswell River. Most of these streams are monitored by ECan for flow and stage levels at recorder sites, except for the Otukaikino River (South Branch) for which an estimate of 3.0 m 3 /s was used (NCCB, 1986). Drain cells were applied to the model up gradient of the recorder sites and drain conductance was adjusted so that the mass flux into the drain cells equals the total flow of the recorder for These upstream reaches are the major areas of interaction with the groundwater system, whereas further east the streams are located above a relatively

4 low permeability aquitard. The model cells representing the Waimakariri River and the spring fed streams are shown in Figure 3. Figure 3: Map showing the Christchurch City Council groundwater flow model river boundary cells, spring-fed stream drain cells, constant head cells. One of the main groundwater discharge components in the model is abstraction via wells. The most up to date groundwater abstraction data set (as of 1999) available to use in the model is maintained by ECan. This database comprises monthly abstraction rates for individual wells and details of the aquifer from which the abstraction is made. For the steady state model, the annual average abstraction rate (m 3 /day) was used based on 1999 values. A total of 937 wells are active in the model and include CCC well abstractions. Boundary conditions described here relate to fluxes of water that naturally migrate into or out of the model domain. These need to be replicated by setting the location and nature of flux boundaries in the model. The following boundaries have been applied to the Christchurch aquifer model: The Waimakariri River to the north provides a flux boundary which provides recharge to the aquifer system; The Pacific Ocean to the east and Lake Ellesmere to the south provide a constant head boundary at the down-gradient margins of the groundwater flow system; Banks Peninsula to the south-east is designated as a no-flow barrier to groundwater flow; A groundwater stream-line to the south-west is set as a no-flow boundary; The western boundary occurs within the alluvial gravel strata although the model was able to be calibrated with no through-flow occurring across that boundary. This indicates the dominance of Waimakariri River seepage into the area, relative to any western throughflow. Within the model domain, groundwater recharge sources are provided both from Waimakariri River seepage and from infiltrating rainfall. Discharge occurs: via offshore flow to the east; through-flow to the south-east; into spring-fed streams such as the Avon and Heathcote Rivers; and via pumping wells. The distribution of constant head cells, drain cells, river cells and no-flow cells in the model are shown in Figure 3.

5 MODEL CALIBRATION Calibration of a groundwater model generally involves adjusting parameters (such as hydraulic conductivity and recharge) so that the modelled values resemble as closely as possible the field monitoring data. Most of the calibration was completed manually, however an automated parameter estimation program (WINPEST) was used as a check on the range of parameters utilised, which showed similar results. Initial steady state runs appeared to support the basic principles of the hydrogeological model used to define the model. The main controls on the simulated heads were the vertical hydraulic conductivities in the aquitard units with heads remaining relatively consistent in the confined areas of the model regardless of changes in hydraulic conductivity in the aquifer units or changes in recharge. Calibration of this model mostly focused on hydraulic conductivities in the western unconfined areas and adjusting conductance terms in the drain cells and river cells to achieve measured fluxes. The values of hydraulic head collected from the field to be used as targets in the model consist of monitoring data collected by ECan and CCC from 157 wells across the model domain. All well records containing more than 12 monitoring data points per annum were included in the model except those wells with water levels indicative of shallow aquifers that are not representative of the wider aquifer system across the Plains. This provided data coverage for all aquifers with the most complete spatial coverage in the top two aquifers. Simulated piezometric contours are shown in map format in Figure 4, and in cross-section in Figure 5. Figure 4: Map showing the simulated piezometric head (equipotentials) from the Christchurch City Council groundwater flow model. Equipotentials represent heads in aquifer 1 and surrounding units and are shown in metres above sea level (masl). The relative difference between the simulated groundwater levels and the measured groundwater level or residual showed that the residuals tend to lie in positive territory, meaning that the modelled heads are slightly higher than measured heads. This is an artefact of balancing the piezometric head fitting requirements with the need to fit river seepage and spring discharge requirements. The numerical expression of the residual errors is called the objective function and is measured by the root mean squared error (RMS) and the mean error. For the steady state simulation with pumping, calibrated to average 1999 groundwater levels, the RMS is (m) and

6 the mean squared is (m). Additionally, the normalised root mean squared is 3.98 %. The software developer, Waterloo Hydrogeologic Inc, considers a model with less than 10 % normalised root mean squared error to be a calibrated model. Figure 5: Cross-section of simulated head equipotentials, and backward path line distributions from the Estuary Christchurch City pumping station. Once satisfactory model calibration has been achieved, there remain a large number of parameter combinations that could provide a similar calibration. Therefore, a calibrated model is simply one of a family of simulations that provide a reasonable range of parameters and responses under differing conditions. For the purposes of this model, verification consisted of testing the model responses to changes in rainfall and abstraction under both steady state and transient conditions. The pumping wells were switched off in the model and run under both steady state and transient conditions, with respective variations in effective rainfall recharge that correspond to seasonal extremes (60 mm/year and 473 mm/year) also applied to the model. The resulting changes in head were compared with previous estimates of seasonal water table fluctuations (NCCB, 1983) with the simulations showing good correlation with measured ranges. Generally, the simulated head fluctuations were greatest west of the confining layers when recharge inputs were altered. Conversely, the simulated head change without pumping showed relatively small change, with most of the fluctuation occurring in the north and southern parts of Christchurch city. Overall, positive correlations can be made between the measured (NCCB, 1983) and simulated fluctuations produced by the model. Mass balance describes the total water budget of the model. The total mass of water in and out of the model should be equal. However, small discrepancies can occur if the model is unstable. Mass balance discrepancies of no more than 0.01% were encountered in the transient simulations. Mass balance calculations also describe the changes in flux throughout different simulations. For example, during the simulation without pumping, river recharge inputs decreased slightly, with more discharge to the modelled surface water features, the coast and Lake Ellesmere area. Similarly, when rainfall recharge was increased, the river recharge decreased, with increases in discharge to the surface water features, coastal, and Lake Ellesmere area. MODEL SENSITIVITY A quantitative sensitivity analysis, whereby the numerical sensitivity of those parameters that affect the model the most is assessed, has not been carried out. However, a qualitative description of those parameters is provided by the following: In the eastern part of the model where the aquifer system is confined, vertical hydraulic conductivity (Kz) of the aquitard layers was critical in calibrating head levels in the aquifer units. Once this was achieved, heads in that eastern part of the model remained relatively stable despite large changes to Kx and Ky in the aquifer units, and during changes in recharge in the western part of the model, however, flows to the drain cells were relatively sensitive to such changes;

7 The western part of the model is relatively sensitive to changes in recharge and hydraulic conductivity, especially where groundwater levels came close to the bottom of layer 4. The modelling of an unconfined system in the west was much more difficult to control than the confined, layered arrangement in the east. For example, if groundwater levels dropped below layer 4 in the west, river recharge would stop due to dry cells surrounding the river boundary cells. This in turn would force groundwater levels to drop elsewhere in the model. In reality, the Waimakariri River is thought to recharge the wider Plains area via a shallow perched aquifer system that cascades outwards from the river producing very steep gradients within 1 km of the main recharge zone. This could not be modelled due to the coarse grid discretization in these areas of the model, therefore the known rate of recharge (7.0 to 8.5 m 3 /s) was applied to the model from the river boundary cells to those layers representing aquifer 1. PARTICLE TRACKING A component of the modelling software (MODPATH) is the ability to trace water flows backwards in time and space. Backwards-tracking particles were released from model cells surrounding a selection of 31 separate public drinking water supply wells or pumping stations in the two layers representing a single aquifer unit from which the water is abstracted. The particles allow a representation of advective transport through the groundwater flow system. For each well or pumping station, a total of 200 particles were released, with 100 in each of the two layers that represented an aquifer unit. This is the maximum number of particles that can be released as a single group in each respective layer. The resulting path line distributions are shown in Figure 5, Figure 6, Figure 7 and Figure 8. Figure 6: Map showing the simulated path line distributions for selected Christchurch City Council public supply wells/pumping stations.

8 M35/2249 M35/6791 M35/2249 Figure 7: Simulated backward path line distributions from six Christchurch City pumping stations (oblique view from east). The corresponding changes in flow path lines or travel times were analysed for six public supply wells that have reliable age determinations. The areal extent of the path line distribution for these six wells is shown in Figure 7. This assessment of ages is most useful to compare the relativity of ages between wells, rather than the absolute values of age distribution produced by the model. It is also interesting to observe the change in age distribution with different model stresses, which is helpful in explaining what physical processes affect the groundwater flow to a specific well and how these processes vary across the model domain. For example, wells M35/2249 (Avonhead) and M35/6791 at the same location in the model, abstract from Aquifer 1 and Aquifer 5 respectively. The Aquifer 1 abstraction, showed older fractions of water particles during times of no abstraction, and low recharge. In Aquifer 5, under high recharge conditions, the water ages tend to be younger than under low recharge conditions, although older during the pumping scenario. The well at Harewood (M35/1653) abstracting from Aquifer 1, shows a mixture of younger and older water during the pumping scenario compared to the non pumping scenario. The particle tracking software provides an indication of the range of different flow paths that reach a well, as demonstrated by the cross-section in Figure 5 and the three-dimensional view shown in Figure 8. These various pathways explain the distribution of water ages that make up any particular groundwater sample. A major limitation of using MODPATH to simulate groundwater flow paths is the early termination of path lines due to weak sinks or boundary conditions (such as pumping wells). Due to the large number of pumping wells or weak sinks in the model, very short path lengths were encountered that are unfeasible in a real aquifer system. Therefore, simulations without pumping confirmed that these groups of particles were artefacts of the model and would justify their exclusion from simulated estimations of age distributions. However, including these groups of shorter path lines gives a more conservative (younger) estimate of the ages produced by the model. Another artefact of the model relates to the layer in which the particles are released. Generally, large contrasts in age are observed between particles released in the lower layer of an aquifer unit and the upper layer of an aquifer an aquifer unit. Scenarios without pumping showed virtually no mixing of particles released in either layer with some mixing in scenarios with pumping. This is an unavoidable constraint of the orthogonal

9 (finite-difference) modelling grid, whereby the simulated orientation of flow between cells is at right angles to cell boundaries. Figure 8: Three dimensional view of simulated backward path lines (shown in light blue) that migrate from selected public supply wells in the aquifer system beneath Christchurch. In the background, Banks Peninsular is shown; with aquifer units shown in dark blue; and aquitards shown in red. Visual representation of the modelling outputs is very useful when trying to communicate complex processes in the subsurface to people that have limited experience in understanding such information. Groundwater flow is sometimes a very abstract concept to many audiences and basic three dimensional displays such as path lines heading toward a well like that shown in Figure 8 increase communication of model outcomes. Additionally, having all the factors controlling groundwater such as aquifers and aquitards, with areas of recharge, well abstraction integrated into one coherent display assists the end users with a clear understanding of the elements and the scale of an aquifer system for which management decisions are made. CONCLUSION 1. The model has produced a verifiable output that can be related to measured data. The output also relates to seasonal and pumping stresses such as rainfall and abstraction variation. 2. The model can be used to show how groundwater moves beneath Christchurch and how different pathways for groundwater relate to relative groundwater ages. 3. The model provides an integrated tool for water supply management enabling a more robust rationale for supply monitoring and targeting of key risk wells. 4. Visualisation of hydrogeological configurations and groundwater flow paths using a numerical model communicates clearly the processes that can occur in aquifer systems used for drinking water supplies.

10 REFERENCES Brown L and JH Weeber 1992: Geology of the Christchurch urban area; Institute of Geological and Nuclear Sciences, Geological Map 1. ECan 1995: Groundwater availability guide for the Waimakariri-Rakaia plains; Canterbury Regional Council Report U95/57, 15 p. ECan 1997:Christchurch West Melton groundwater; Environment Canterbury Report U97/28/1, 36 p., plus appendices. ECan 2000: Christchurch West Melton groundwater investigation; Environment Canterbury Report U00/39, 50 p. ECan 2002: Christchurch West Melton groundwater quality; Environment Canterbury Report U02/47, 141 p. ECan 2003: The Christchurch artesian system; Environment Canterbury Report U03/47, 47 p. MoH 2005: Proposed Drinking Water Standards for New Zealand 2005, Ministry of Health, Wellington. NCCB 1983: Interim report on the groundwater resource of the Central Plains; Report published by North Canterbury Catchment and Regional Water Board, Christchurch. NCCB 1986: Christchurch artesian aquifers; Report published by North Canterbury Catchment and Regional Water Board, Christchurch, 159 p. NZMS 260: New Zealand Map Sheet 260 series; 1: Land Information New Zealand. ACKNOWLEDGEMENTS We would like to acknowledge Christchurch City Council for commissioning this work, in particular Mike Kininmonth and Bruce Henderson. Many thanks also go to the council staff that supplied data for this project. We would also like to acknowledge Environment Canterbury for their assistance in providing invaluable input data for the model from their vast databases. In particular, we would like to thank David Scott for his time and expert advice throughout the development of the model.