REPORT - Earthquake Commission...Increased Flood Vulnerability: Overland Flow Model Build Report Job...

REPORT Chapman Tripp on behalf of the Earthquake Commission (EQC) Increased Flood Vulnerability: Overland Flow Model Build Report

REPORT

Report prepared for:

Chapman Tripp acting on behalf of the Earthquake Commission (EQC)

Report prepared by:

Tonkin & Taylor Ltd

BMT WBM Ltd

Distribution:

Chapman Tripp acting on behalf of the Earthquake Commission (EQC) 1 copy

Tonkin & Taylor Ltd (FILE) 1 copy

August 2014

T&T Ref: 52010.150 Final Issue 2

Chapman Tripp acting on behalf ofthe Earthquake Commission (EQC)

Increased Flood Vulnerability:Overland Flow Model Build Report

Increased Flood Vulnerability: Overland Flow Model Build Report T&T Ref. 52010.150 Chapman Tripp acting on behalf of the Earthquake Commission (EQC) August 2014

Table of contents

1 Introduction 6 2 Modelling software 8

2.1 Model selection 8 2.2 Model developers 8 2.3 Software description 8 2.4 Software version 9

3 Data 10 3.1 Topography data 10

3.1.1 Catchment boundaries 10 3.1.2 LiDAR 10 3.1.3 Streams 11 3.1.4 Pipes 11 3.1.5 Temporary stopbanks 12 3.1.6 Land use (roughness) areas 12 3.1.7 Soil areas 13 3.1.8 Groundwater level 13

3.2 Boundary data 15 3.2.1 Tidal data 15 3.2.2 Rainfall data 15 3.2.3 Upstream inflows – Avon catchment 16

4 Model coverage 17 4.1 Avon catchment 17 4.2 Heathcote catchment 18 4.3 Styx catchment 19

5 Starting parameters 20 5.1 Model parameters 20 5.2 Topography scenarios 22 5.3 Infiltration losses 22 5.4 Base case starting parameters 23

6 Comparison to previous modelling 24 6.1 Comparison of peak flows to river models 24 6.2 Comparison of flood extents to river models 24

7 Sensitivity 25 7.1 Sensitivity methodology 25 7.2 Outline 26 7.3 Cell size 26 7.4 Rainfall duration hyetograph 30 7.5 Roughness (Manning’s value) 33 7.6 Soil infiltration 34 7.7 Starting groundwater level 37 7.8 Lowering roads 37 7.9 Buildings 38 7.10 Representing drainage network 39 7.11 Channel inverts 40

Increased Flood Vulnerability: Overland Flow Model Build Report Job no. 52010.150 Chapman Tripp acting on behalf of the Earthquake Commission (EQC) August 2014

7.12 Sensitivity – flooded properties 42 7.13 Sensitivity – summary 44 7.14 Sensitivity recommendations 50

8 Calibration to March 2014 event 51 8.1 Boundary data 51 8.2 Calibration datasets 53 8.3 Method 53 8.4 Calibration model with revised parameters 54

8.4.1 Representation of urban areas 54 8.4.2 Modelling of pipes as breaklines 55 8.4.3 Rainfall in the Upper Avon catchment 55 8.4.4 Infiltration rate 55

8.5 Results 55 8.5.1 Time series 55 8.5.2 Observed flood extents 57 8.5.3 Stage-flow relationship for Gloucester Street gauge 61 8.5.4 Comparison of recorded groundwater surface with the 85th percentile

surface 62 8.6 Conclusions 64

9 Base case – with revised parameters 65 9.1 Comparison to initial parameters 65 9.2 Sensitivity to downstream boundary conditions 66 9.3 Final base case parameters 69

10 Final overland model boundary conditions 71 11 Peer review comments and responses 72 12 Conclusions and recommendations 74

12.1 Sensitivity 74 12.2 Calibration 75

13 References 77 14 Applicability 78 Appendix A –Base case model with final parameters - Infiltration sensitivity run results i

1


Executive summary

An overland flow model of the whole of Christchurch was developed in TUFLOW 2D software. The model was built to augment results from Christchurch City Council (CCC) “river flood models” to provide estimates of flood depth in overland flow areas (generally remote from major drainage network elements) across the Canterbury earthquake sequence. The flood depths are applied to the assessment of Increased Flood Vulnerability (IFV).

The details of the assessment of IFV are given in the T&T (April 2014) Canterbury Earthquake Sequence: Increased Flooding Vulnerability Assessment Methodology report. T&T (April 2014) also describes the criteria for models and the application of model results. The details of the river flood models are given in the companion report T&T (August 2014b) Canterbury Earthquake Sequence: Increased Flooding Vulnerability River Modelling and Coastal Extensions.

The TUFLOW GPU engine was used for the overland flow model. A key reason for the model selection was the computation speed of the GPU engine, which enabled model sensitivity analyses to be carried out within reasonable timeframes not possible with the classic CPU engines available. Sensitivity analyses were required to give confidence in model results obtained. The TUFLOW GPU engine also enabled flood simulations to be carried out at a finer grid resolution than would have been feasible using the alternative CPU approach. The model section also considered that the application of the model would be primarily for the 1% AEP rainfall event for the assessment of IFV1.

The TUFLOW model covers the entire urban area of Christchurch City. It can be split into component models covering the individual catchments of the Avon, Heathcote and Styx Rivers if required, but can also be run as a single city-wide model. The city-wide model has benefits in extreme events as cross-catchment flows can be represented. Smaller, catchment-scale models can be run if computation speed is important for any specific investigation.

This model has been developed in 2D only. All surface waterways have been represented in the TUFLOW GPU model using dimensions taken from existing drainage infrastructure, but to a lesser degree of detail than in the river models. Simulated pipes to account for the drainage effect offered by pipes (particularly large diameter pipes) were also included in the TUFLOW GPU model to represent pipes of diameter 600 mm and greater.

The TUFLOW GPU model takes account of rainfall infiltration, which is an improvement on previous overland flow models available for Christchurch catchments that had no allowance for these hydrological losses. The simulation of rain infiltration will better align the overland flow models with the river flood models which include infiltration (albeit at a more coarse scale than what has been developed in TUFLOW). Infiltration rates were applied spatially, dependent on underlying surface soil type. Sensitivity assessments to the infiltration rates has been undertaken. In addition, the effect of potentially near-surface groundwater table has been simulated to ensure that infiltration ceases when the groundwater level reaches the ground surface level (not done in previous overland flow or river models). Groundwater flooding was not simulated in the modelling.

1 The 2% and 10% AEP rainfall events are simulated to provide information for valuation. The reliability of the model is reduced approaching the 20% AEP rainfall event as the primary piped stormwater system becomes more influential.

2


The process for developing the TUFLOW model is illustrated in Figure i and detailed in this report.

The Avon River catchment was selected for sensitivity testing as it was considered to be adequately representative of Christchurch catchments and enabled more sensitivity assessments to be undertaken with shorter run times. An exception to this was that the rainfall duration sensitivities that were conducted using the city-wide model, given that each of the three major catchments are known to respond differently to rainfall of differing durations. The conclusions reached from the sensitivity testing was applied to final model runs that cover the whole city.

Figure TUFLOW model build and testing process

Sensitivity testing was carried out over a range of parameters and model architecture elements, including the rainfall distribution (variation of temporal profile applied for constant rainfall depth for each duration and frequency), surface roughness, surface soil infiltration rates, antecedent groundwater level, representation of roads in the hydraulic model, allowance for piped networks and variation in channel invert level.

The models that have been developed exhibited relatively low sensitivity to many of the parameters for which sensitivity testing was carried out. The greatest sensitivity was exhibited for the primary pipe network capacity, which was tested by removing rainfall to represent the primary pipe network draining with a 20% AEP level of service. This sensitivity test revealed that if this pipe network does globally function to this level of service at all times, then substantially less flooding than predicted by the models would occur. Upon closer consideration this assumption was found to be flawed, mainly due to the very flat topography in Christchurch. Once tailwater levels have risen during a flood event, available hydraulic grade is minimal and it was found that pipe networks convey less and less flow as flood levels rise, to the point where submerged pipes may convey very little flow at all near to the peak of significant floods (due to a lack of available hydraulic grade). This conclusion is supported by observations of flooding.

The results were also shown to be sensitive to the infiltration rates applied. There was a large difference in model results between simulations with no infiltration and simulations with some allowance for infiltration. Accounting for soil infiltration has been the practice adopted by CCC,

3


and is the practice for the river flood models, which have been calibrated. The finding of the sensitivity analysis was that allowance for soil infiltration should be made.

Relatively low sensitivity was displayed in model results to the other parameters examined. Of particular interest was the low sensitivity in resultant peak flood levels and peak flood discharge to adopted rainfall profile, when comparisons between the CCC rainfall distributions and the nested rainfall hyetograph was made. Low sensitivity in results was also observed when the rainfall duration was increased.

The sensitivity testing has enabled a thorough understanding of the results sensitivity to model architecture and input parameters. The models can now be used with greater confidence for the assessments for which they have been built.

Subsequent to the model building process, a flood event occurred in Christchurch that was appropriate for model calibration. The rainfall event of 4/5 March 2014 has been hindcast using the TUFLOW model. The rainfall was varied spatially as collected in 18 rain gauges across the city, and modelled flood extents were compared with observed flood extents. This comparison revealed a need for inclusion of more of the piped drainage network in the model than had been previously included. This was addressed with simulated pipes in subsequent iterations of the model. After the inclusion of simulated pipes in the TUFLOW model, there was a good fit between observed and modelled flood extents in areas where sufficient observations had been collated.

Finally, the model was compared to both earlier MIKE overland flow models and MIKE river models. This comparison necessitated an amendment to the chosen tidal boundary. The earlier model adopted a static 10% AEP tidal boundary, whereas as a result of the comparison a dynamic tidal boundary timed to coincide with peak intensity rainfall for each of the three rivers was adopted for the base model final parameters. After the downstream tidal boundary was adjusted the different models produced comparable results in the tidally affected areas.

4


Glossary

Term Definition

x% AEP (Annual Exceedance Probability) flood

The flood caused by a rainfall event with a depth that has an x% probability of being exceeded during a year.

BMT WBM Developers of the TUFLOW software, and providing guidance for the build of the TUFLOW overland flow model used for assessment of IFV.

CCC (Christchurch City Council)

The local government (Council) in Christchurch, who are owners of the river flooding models used in the assessment of IFV.

CGD (Canterbury Geotechnical Database)

A library of geotechnical information in Canterbury, created after the Canterbury earthquake sequence.

CPU (Central processing unit)

The part of a computer which undertakes the majority of the computational work. Many computers will have two or four of these. See also GPU.

DEM (Digital elevation model)

The model used to represent the ground.

DHI Developers of the MIKE modelling software packages and an engineering consultancy who developed the Avon river model on behalf of CCC.

Earthquake sequence Defined as the earthquakes in the Canterbury region ongoing since September 2010.

EF (Exacerbated flood depth)

The measure of IFV. The lesser of change in flood depth ( F) and change in ground surface ( L).

EQC (Earthquake Commission)

The national insurer of residential land for natural disaster damage in New Zealand, who require the use of the flood models to assess IFV.

Flood depth The flood depth is the modelled water depth that would occur in the 1% AEP flood.

FLCA (Floor level control area)

An area designated by CCC as being susceptible to flooding in Christchurch, with special rules surrounding work undertaken within the area.

FMA (Flood management area)

An area designated under Plan Change 48 of the Christchurch City Plan as being susceptible to flooding, with special rules surrounding work undertaken within the area.

GHD An engineering consultancy. GHD has developed a model in the Styx catchment on behalf of CCC.

GPU (Graphics processing unit)

The part of the computer that processes the graphics, these often have thousands of cores, which, while not as individually fast as CPU cores, can split up large simple processes and thus run a flood model much faster than a CPU model. See also CPU.

Hyetograph The graphical representation of the distribution of rainfall over time.

5


IFV (Increased flooding vulnerability)

A physical change to residential land as a result of an earthquake (subsidence) which adversely affects the amenity and value that would otherwise be associated with the land by increasing the vulnerability of that land to flooding events.

Individual earthquake The effects of a single earthquake within the earthquake sequence.

LiDAR (Light Detection and Ranging)

Ground surface elevations measured using optical sensing technologies from a plane (an aerial survey).

Major earthquakes The four major earthquakes that caused measurable ground surface subsidence on 4 September 2010, 22 February 2011, 13 June 2011 and 23 December 2011.

MIKE Hydrological and hydraulic modelling software developed by DHI, and used for the River Flood modelling for IFV.

NIWA An engineering consultancy. NIWA has developed a model in the Heathcote catchment on behalf of CCC.

Overland flow model A combined 1D and 2D flood model involving a catchment wide rainfall event superimposed on a LiDAR derived DEM.

Properties The part of a land holding on which a residential building is lawfully situated that is defined as “residential land” by section 2 of the Earthquake Commission Act 1993.

River flooding model A combined 1D and 2D flood model of a river involving loading of the river and immediately adjacent floodplain.

Temporary stopbanks Temporary stopbanks are non-engineered structures, mostly installed following the Earthquake sequence, which require regular refilling. These are not included in the models.

TUFLOW Hydraulic modelling software developed by BMT WBM, and used for the Overland Flow modelling for IFV.

T&T An engineering consultancy Tonkin & Taylor Ltd.

6


1 Introduction

The report is written for Chapman Tripp acting on behalf of the Earthquake Commission (EQC) to document the river models used for the engineering assessment of Increased Flooding Vulnerability (IFV) land damage due to the earthquake sequence in Christchurch.

Increased Flooding Vulnerability is a physical change to residential land2 as a result of an earthquake which adversely affects the use and amenity that could otherwise be associated with the land by increasing the vulnerability of that land to flooding events.

Previous reports have been supplied to EQC from Tonkin & Taylor (T&T). These reports are the ‘Increased Flooding Vulnerability Assessment Methodology (T&T, Volume 1 April 2014), and the ‘Peer Review report’ (Benn et al, 2014), and are referred to as such for the remainder of this report.

The Assessment Methodology (T&T, Volume 1 April 2014) details the criteria and thresholds, and their reasoning, for IFV. In order to assess properties for IFV, criteria and thresholds are necessary to ensure a fair and reasonable outcome.

This report, the Overland Flow Model Build report, documents the construction of the TUFLOW overland flow model. A TUFLOW overland flow model is used as part of the assessment of properties for potential IFV. The report documents the underlying philosophy of the model, as well as sensitivity analyses and calibration.

The River Modelling report T&T (Volume 2 T&T 2014) documents the use of the CCC river models and the coastal extensions to assess IFV in the major rivers and tidal areas of Christchurch.

The Peer Review report (Benn et al, 2014) is a summary of the recommendations of the peer reviewers of the work undertaken by T&T. T&T has taken these recommendations into account in the development of IFV methodology and implementation.

An overland flow model of the whole of Christchurch was developed in TUFLOW software. This was built to augment results from Christchurch City Council (CCC) river flood models, especially to provide flood depth comparisons in overland flow areas (generally remote from major drainage network elements) across the Canterbury earthquake sequence.

This report is Volume 3 in a suite of 6 reports describing IFV and its implementation for EQC. The report titles are provided below and should be read in conjunction with this report:

Volume 1: Increased Flood Vulnerability: Assessment Methodology Report

Volume 2: Increased Flood Vulnerability: River Modelling and Coastal Extensions Report

Volume 3: Increased Flood Vulnerability: Overland Flow Model Build Report

Volume 4: Increased Flood Vulnerability: Implementation Report

Volume 5: Increased Flood Vulnerability: Geological Processes Causing Increased Flood Vulnerability

2 “Residential land” is used in this assessment methodology as it is defined in the Earthquake Commission Act 1993, s2(1).

7


This report is organised into the following sections:

Section 2 of this report describes the TUFLOW software;

Section 3 of this report describes the data sources and inputs;

Section 4 of this report provides a summary of the coverage of the model;

Section 5 of this report outlines the starting parameters for the model;

Section 6 of this report provides a comparison of the model results to the river flood models run by CCC;

Section 7 of this report contains the analysis of the sensitivity of the model to the various parameters;

Section 8 of this report summarises calibration of the model to the rainfall event in March 2014;

Section 9 of this report describes the base case with revised parameters;

Section 10 provides the boundary conditions for the base model final parameters

Section 11 of this report provides a summary of the peer review of the model;

Section 12 of this report provides conclusions and recommendations; and

Section 13 of this report provides references.

The process for developing the model, sensitivity analysis, calibration and refinement is shown in Figure 1-1.

Figure 1-1 TUFLOW model build and testing process

8


2 Modelling software

2.1 Model selectionT&T selected the TUFLOW GPU engine, largely because of the significant reduction in run-times over that using a classic CPU processor (such as that used in previous overland flow modelling). The faster run-times have made sensitivity testing more feasible within the project timeframe. The sensitivity testing was undertaken to improve confidence in model results. The TUFLOW GPU engine also enables for smaller grid sizes to be viable for a large city size model domain. Smaller grid sizes are desirable because they provide better (finer) resolution, which provides more accurate modelling of flowpaths. This is important for the use of the model results by EQC, who require the models to resolve flooding changes on a property by property basis.

The TUFLOW model is well respected in Australasia, where it is used as a tool for identifying flood hazard and for planning flood mitigation works. The TUFLOW GPU engine is also accepted by the UK Environment Agency for flood modelling as a result of a 2D hydraulic modelling benchmarking exercise (Environment Agency 2013).

2.2 Model developersAssistance and review were provided by BMT WBM, the developers of the TUFLOW software.

2.3 Software descriptionTUFLOW is a suite of numerical engines for simulating urban waterways, rivers, floodplains, estuaries and coastlines. There are three main numerical engines under the TUFLOW software suite:

TUFLOW 2D grid based and linked 1D network solver;

TUFLOW GPU, a 2D only grid based solver using the parallel processing power of the modern GPU for fast simulations; and

TUFLOW FV, a 2D/3D flexible mesh finite volume solution. For the modelling outlined in this report, the TUFLOW GPU engine has been used. TUFLOW GPU is an explicit solver of the full 2D Shallow Water Equations. TUFLOW GPU utilises the multiple cores on graphics card(s) to provide faster runtimes than the standard CPU based TUFLOW.

The solution scheme computes the volume flows across cell boundaries and is volume and momentum conserving. The scheme utilises a sub-grid scale eddy viscosity model and the default method is a Smagorinksy approach.

TUFLOW GPU is available in 1st, 2nd and 4th order time integration (the default is 4th order which is used in the modelling outlined in this report). Either an adaptive timestep or fixed timestep can be utilised, with an adaptive timestep applied in this work. TUFLOW GPU is available in both single and double precision versions. TUFLOW GPU is a cell centred scheme and one elevation per 2D cell is used.

The TUFLOW GPU engine has the following capabilities:

Direct rainfall (rain on grid);

Soil infiltration (initial/continuing loss, Horton loss method or Green Ampt);

Manning’s roughness, including depth varying Manning’s curves;

9


Ability to model additional form losses as Form Loss Coefficient x V2/2g, used to model structure losses;

Ability to modify cell flow width; and

Ability to modify cell storage.

2.4 Software versionTUFLOW GPU version 2014-03-AG (64 bit Single Precision) has been used in this modelling. This is the latest version at the time of writing.

10


3 Data

3.1 Topography data

3.1.1 Catchment boundariesCatchment boundaries were initially derived based on catchment boundaries used in previous modelling, as supplied by CCC. Refinements to these boundaries were made following initial runs of the TUFLOW model, where cross-boundary flow was observed. Furthermore, in some initial runs edge-ponding was visible in results, which indicated areas where runoff from the target catchment was flowing to the adjacent catchment. In some cases an overflow boundary was delineated to allow this flow to exit the model domain, and the catchment boundary was therefore not required.

Where the catchments have extended beyond the area covered by LiDAR (Light Detection and Ranging) data, some allowance has been made by lumping inflows at certain specific locations.

3.1.2 LiDARBare earth digital elevation models (DEMs) at 2 m resolution have been created for each topographic scenario pre- and post- each major earthquake event. Five DEMs have been created:

Pre-September 2010;

Post-September 2010 (Pre-February 2011);

Post-February 2011 (Pre-June 2011);

Post-June 2011 (Pre-December 2011); and

Post-December 2011. The DEMs have been created using LiDAR surveys by a combination of binning and linear interpolation. For each 2 m x 2 m cell containing point data, the mean elevation of the points is assigned to the cell. Cells containing no point data are assigned an elevation linearly interpolated from the nearest points. All DEMs are clipped to the respective survey extents (data area). LiDAR surveys are combined in some cases, as some surveys have reduced extents. The survey extents that were flown targeted the areas with observed land damage for the previous earthquake. In these cases (Post-September 2010 and post-December 2011), the DEM is combined with the DEM of the previous survey (Pre-September 2010 and post-June 2011 respectively) by replacing cells that have no data with data from the corresponding cell of the previous DEM. LiDAR extents are shown in Figures A5 to A9 in T&T April 2014.

There are DEM (LiDAR-derived) data limitations to the cell size used for 2D models. In particularly, the pre-September 2010 LiDAR that is based on LiDAR surveys in July 2003 has a lower LiDAR measurement density of approximately 2 points per 5 m by 5 m cell. Whereas, LiDAR surveys from post-September 2010 onwards have LiDAR measurement density of approximately 17 to 30 points for each 5 m by 5 m cell. Therefore, the accuracy of DEMS used for comparative purposes across the earthquake sequence is limited by the quality of the pre-September 2010 LiDAR.

All DEMs and LiDAR surveys are to Lyttleton Vertical Datum 1937 and New Zealand Transverse Mercator 2000. Refer to T&T April 2014 for a more detailed description of the LiDAR including its accuracy.

11


3.1.3 StreamsThe model in-bank bathymetry is derived from the 1D MIKE 11 network of the MIKE FLOOD models (CCC models). The network files from these models were provided by DHI, GHD and NIWA and the channel thalweg (channel invert) was converted into a breakline in TUFLOW (i.e. the horizontal location was defined from the channel centreline, and the thalweg level along this alignment was then “burned into” the DEM). In many cases, the MIKE 11 model horizontal alignment did not align perfectly with the actual channel location and manual changes were made to align the stream lines with both the DEM and aerial photos.

Additional channels not included in the MIKE11 models were added based on CCC GIS. This occurred after calibration runs of the 4 and 5 March flood where it was determined that flooding was predicted in some areas where flooding was not expected to occur, which was attributed to the absence of these channels (and pipes) in the model. The breaklines that model channels source from MIKE11 models and CCC GIS are shown in Figure 4.1-4.3.

The levels of the channels are based on the latest available cross-section data. In some locations the pre-earthquake cross sections are the latest available information. In other locations there are post-earthquake surveys. The cross-sections used for the TUFLOW modelling for all events were converted from the latest MIKE FLOOD models.

The widths of channels are based on cross sections and from estimates from the aerial photos. The methodology used in TUFLOW is to drop the entire width of the channel to the level specified in the breakline (based on the thalweg level), which resulted in a rectangular cross section instead of the actual cross section with side slopes. This means that the channel network volume is larger in the TUFLOW model than in reality. An increased roughness allowance to counter this effect was made. It is recognised that this approach reduces accuracy for the channel and immediate surrounds, but is a trade-off with other advantages including better (finer) cell resolution. The application of the TUFLOW model in conjunction with the river models allows the strengths of both models to be used for IFV assessment. It is considered that, provided that adequate allowance is made for conveyance within the main channels, the overland flow path results will be relatively insensitive to the absolute accuracy in the main channels. The sensitivity to this was assessed by raising and lowering the channel (refer Section 7.11).

Breaklines have also been used to preserve key hydraulic controls. Where required, additional topography modifiers were created based on the DEM to ensure that hydraulic controls, especially culverts, allow conveyance of water. For example, where major culverts were known to exist, the DEM was checked to ensure that breaklines were of suitable depth to allow flow.

3.1.4 PipesDuring calibration runs of the 4 and 5 March flood event it was determined that flooding was predicted in some areas where flooding was not expected to occur. This was particularly apparent in Riccarton and Spreydon, areas where there are few open channels and large diameter pipe networks are extensive. In these areas, overland flows developed due to the absence of modelled stormwater networks. Simulated pipe networks were trialled and found to give more realistic flooding in these areas. For consistency, simulated pipes were adopted for all areas of the city.

Simulated pipes were applied to the TUFLOW model for all pipe 600 mm diameter or greater. Simulated pipes are an approximation of pipe conveyance capacity made by insertion of rectangular open channels of dimension calculated to deliver equivalent conveyance. Simulated

12


pipes were adopted because the actual drainage network includes pipes and catchpits that cannot readily be modelled in TUFLOW GPU. Simulated pipes allow inflow/outflow along their length, similar to what occurs with regular spacing of catch pits and downpipes. Refer to Figure 4-1 to Figure 4-3 for the extent of breaklines that represent simulated pipes.

The pipes are modelled by lowering the cell elevation by 0.886 times the pipe diameter. The cell flow width is restricted to 0.886 times the diameter. The value of 0.886 was chosen as this gives an equivalent flow area when representing a circular pipe using a “flat” 2D cell.

In addition to provision of equivalent flow area, Manning’s n roughness was adjusted to ensure a proper scaling of conveyance is maintained through the simulated pipe representation. This roughness scaling was based on pipe full hydraulic radius being compared against 2D flow depth. By doing this, it emerged that once the flow area approximation has been made (as described above), the roughness scaling increased from n = 0.013 for a concrete pipe to n = 0.03 for the simulated pipes.

3.1.5 Temporary stopbanksAs an immediate response to land deformation that resulted from the Canterbury earthquake sequence, temporary stopbanks were built along the downstream reaches of the Avon River. We understand their purpose was to prevent tidal flooding. Part of the criteria for the IFV assessments was to remove the influence of these from the model (refer to T&T (April 2014) for a more detailed description of this criteria.

The horizontal alignment of these temporary stopbanks was provided as a shapefile line. This line was buffered by an appropriate width and the DEM was adjusted to flatten these temporary stopbanks, based on interpolation from adjacent DEM values. The line was manually rectified where it did not appear to align with the location of the stopbank in the DEM. This line was buffered by 4 m each side based on the observed width of where stopbank levels in the post-earthquake appeared to be consistent with the levels in the pre-earthquake DEM, indicating the installation of temporary stopbanks.

3.1.6 Land use (roughness) areasLand roughness adopted in the model was based on land use. Land use was obtained from Landcare Research Land Cover Database version 3 (LCDB3). This data is available free from the Land Research Information Systems portal (https://lris.scinfo.org.nz). Details of specific roughness values applied to different land use is provided in Table 5-2.

Roughness along roads was superimposed, as this is not detailed in LCDB3. The road centrelines were supplied by DHI, GHD and NIWA. These were buffered according to the road hierarchy (for the DHI and GHD supplied road centrelines), or the number of lanes (for the NIWA data). Road widths adopted are as follows:

5 m for private and single lane roads;

10 m for local, minor arterial roads, collector and two lane roads; and

20 m for major arterial roads, motorways and four lane roads.

Once buffered, these road areas had Manning’s n roughness of 0.020 applied which superimposes the roughness from the LCDB3. In this way roughness of roads was differentiated from surrounding land.

13


3.1.7 Soil areasThe infiltration modelling (See Section 7.6) is based on the Landcare Research S-Map soil data used to set spatially varying infiltration losses across the model. This data is freely available from the LRIS portal, the June 2013 update has been used. In each area the dominant surface soil type was used.

There are a few areas in which this data does not cover the entire model. This occurs mainly in the Port Hills areas. Where no soil classification was available from in the soil layer, no soil infiltration was applied. Because this occurs mainly in the Port Hills areas, where surface soils are relatively shallow and slopes are steep, it is suggested that infiltration over such areas would rapidly approach zero in response to significant rainfall, and that this assumption is not unduly conservative.

3.1.8 Groundwater levelGroundwater levels over the entire city were applied in the model as either the 50th or 85th percentile phreatic surface. These surfaces were derived from a separate analysis of recorded groundwater levels in a large number of bores across the city that are screened in the near-surface aquifer (unconfined). The influence of this on model results has been examined through sensitivity testing. The reason for inclusion of groundwater levels in the model is to account for the effect of high groundwater rising to ground level during rainfall and causing infiltration to cease.

The data for the groundwater elevations were provided on a 25 m x 25 m grid. The gridded elevations are based on data from almost 1000 shallow monitoring wells in the Christchurch Region collected since September 2010. The documentation describing the methodology, data sets, and assumptions used to derive the groundwater elevations are provided in an updated version of a GNS Science report dated 2013 (van Ballegooy et al, 2013). Figure 3-1 shows the depths to groundwater level for the 85th percentile post-earthquake case.

14


Figure 3-1 Post-December 2011 earthquake 85th percentile depth to groundwater

15


3.2 Boundary data

3.2.1 Tidal dataThe downstream boundary conditions used for sensitivity modelling are static levels based on the extreme tide levels found in the CCC Waterways, Wetlands, and Drainage Guide (CCC WWDG). These levels are found in Appendix 1 of the CCC WWDG and are shown in Table 3-1 below. The tide levels used for each catchment are Bridge Street for the Avon, Ferrymead for the Heathcote, and Styx for the Styx.

Table 3-1 Tidal data from CCC WWDG

Section 9.1 describes a subsequent amendment to the tidal boundaries where a 10% AEP dynamic tide is adopted for the final base case model.

3.2.2 Rainfall dataThe depth-duration-frequency data for the design rainfall events was taken from the CCC WWDG. T&T calculated a nested storm hyetograph (profile) that was compared in sensitivity runs against the triangular rainfall hyetograph that is described in the CCC WWDG.

The CCC storm hyetograph is triangular with a peak at 0.7 of the storm duration, as described in the CCC WWDG. The nested storm hyetograph is a Chicago-type shape, with the peak shifted to 0.7 of the storm duration.

A 1% Annual Exceedance Probability (AEP) event was selected for simulations. The resulting rainfall hyetograph used in the sensitivity modelling are:

24 hour with nested hyetograph;



18 hour with CCC hyetograph for Avon (same as the CCC river model for Avon);

24 hour with CCC hyetograph;

30 hour with CCC hyetograph for Heathcote (same as the CCC river model for Heathcote); and

48 hour with CCC hyetograph for Styx (same as the CCC river model for Styx).

16


3.2.3 Upstream inflows Avon catchmentThe Avon catchment extends westward, of the model 2D area. Allowance for inflow for this part of the catchment was made by inclusion of some nominal steady inflows. These were assessed based on locations of known stock water race inflows and from lumped catchment hydrology. The location of these inflows are shown in Figure 4.1.

The resulting inflows to the models are small in comparison to flood flows and are likely to have little influence on model results. Part of the reason for this is that the surface soils to the west of the city are generally very pervious, exhibiting high infiltration rates, and little surface runoff from west of the city is able to flow to the major drainage network that runs through the city.

17


4 Model coverage

4.1 Avon catchmentThe Avon catchment is depicted in pink in Figure 4-1. The main channels from MIKE 11 (Section 3.1.3), which have been added into the model as breaklines, are shown in blue. Significant channels and culverts based on CCC GIS, which have been modelled as channels (Section 3.1.3) are depicted in red. Pipes which have been modelled as simulated pipe (Section 3.1.4) are depicted in purple. Temporary stopbanks (Section 3.1.5) are shown in yellow. The location of the upstream inflow (Section 3.2.3) is depicted in green.

Figure 4-1 Avon catchment

18


4.2 Heathcote catchmentThe Heathcote catchment is depicted in light green in Figure 4-2. The main channels from MIKE 11 (Section 3.1.3), which have been added into the model as breaklines, are shown in blue. Significant channels and culverts based on CCC GIS, which have been modelled as channels (Section 3.1.3) are depicted in red. Pipes which have been modelled as simulated pipes (Section 3.1.4) are depicted in purple.

Figure 4-2 Heathcote catchment

19


4.3 Styx catchmentThe Styx catchment is depicted in orange in Figure 4-3. The main channels from MIKE 11 (Section 3.1.3), which have been added into the model as breaklines, are shown in blue. Significant channels and culverts based on CCC GIS, which have been modelled as channels (Section 3.1.3) are depicted in red. Pipes which have been modelled as simulated pipes (Section 3.1.4) are depicted in purple.

Figure 4-3 Styx catchment

20


5 Starting parameters

5.1 Model parametersThis section details the starting parameters used in the TUFLOW model for comparison to river model results and as the base model for sensitivity assessments. The key hydraulic parameters used in the TUFLOW model are outlined in Table 5-1 below.

Table 5-1 Summary hydraulic model parameters

Parameter Value

Model cell size A range for cell sizes from 2 m to 10 m have been utilised, with 5 m considered as the base case. The cell size is discussed in Section 7.3.

Timestep The TUFLOW GPU model utilises an adaptive timestep. Rather than specifying timestep, a maximum Courant number is specified in TUFLOW. The adopted value (Adopted Maximum Courant Number = 1.0), this is the default value for TUFLOW GPU simulations. A test was performed to ensure that the results were consistent with lower Courant criteria. For this test a maximum Courant number of 0.8 was applied and the results compared, these were found to be consistent.

Viscosity The default viscosity approach in TUFLOW GPU is a Smagorinsky method. The default Smagorinsky coefficient of 0.20 has been adopted for this modelling.

Manning’s n The Manning’s values are defined for the land use areas defined in the model. The values used in the modelling are outlined in Table 5-2. The sensitivity of the model results to Manning’s roughness is discussed in Section 7.5.

Fraction impervious The land use areas are assigned a fraction impervious, which is used when calculating soil infiltration. The values used in the modelling are outlined in Table 5-2.

21


Table 5-2 Manning’s values and impervious fractions for land use types

Land use description Manning's value Fraction impervious

Built-up area 0.100 0.3

Urban parkland/open space 0.033 0.0

Surface mine 0.028 0.0

Dump 0.060 0.0

Transport infrastructure 0.016 0.0

Coastal sand and gravel 0.025 0.0

River and lakeshore gravel and rock 0.028 0.0

Alpine gravel and rock 0.039 0.0

Lake and pond 0.020 0.0

River 0.035 0.0

Estuarine open water 0.022 0.0

Short-rotation cropland 0.100 0.0

Vineyard 0.070 0.0

Orchard and other perennial crops 0.050 0.0

High producing exotic grassland 0.050 0.0

Low producing grassland 0.090 0.0

Herbaceous freshwater vegetation 0.100 0.0

Herbaceous saline vegetation 0.100 0.0

Fernland 0.160 0.0

Gorse and broom 0.125 0.0

Manuka and or kanuka 0.100 0.0

Broadleaved indigenous hardwoods 0.100 0.0

Mixed exotic shrubland 0.080 0.0

Grey scrub 0.080 0.0

Major shelterbelts 0.120 0.0

Afforestation (not imaged) 0.200 0.0

Afforestation (imaged, post-LCDB 1) 0.340 0.0

Forest harvested 0.160 0.0

Pine forest – Open canopy 0.100 0.0

Pine forest – Closed canopy 0.200 0.0

Other exotic forest 0.150 0.0

Deciduous hardwoods 0.125 0.0

Indigenous forest 0.150 0.0

22


Road layer 0.020 1.0

Default land use type 0.050 0.0

5.2 Topography scenariosThe model is configured to run five (5) different topography scenarios, which are pre- and post- each of the major earthquakes. The five topography scenarios where there are DEMs (see Section 3.1.2) are:

Pre-September 2010;

Post-September 2010;

Post-February 2011;

Post-June 2011; and

Post-December 2011. All five scenarios use the same in-bank bathymetry data (Section 3.1.3). The additional breaklines to represent key hydraulic controls (Section 3.1.3) differ for each scenario, based on the DEM.

Temporary stopbanks were removed for the four post-quake topography scenarios. The topography scenario which does not remove the stopbanks is the pre-September 2010 scenario because they did not exist at this time. For the other four scenarios, the stopbanks are flattened to the level around the edges of the stopbanks.

5.3 Infiltration lossesThe Horton loss model was used in accordance with guidance given in the CCC Waterways, Wetlands and Drainage Guide (WWDG 2011).

The Horton approach utilises the equation:

= + ( )

Where f0 is the initial infiltration rate in mm/h, fc is the final (indefinite) infiltration rate, t is time in hours and k is the Horton decay rate. For the TUFLOW implementation, the time (t) is the period of time that the cell is wet.

For the base case with starting parameters the values adopted are summarised in Table 5-3. The porosity values are based on Rawls et al (1983). The porosity values are not directly used in the calculation of the infiltration rates using the Horton method. These are used to calculate the rise in groundwater level caused by the infiltrated water.

Table 5-3 Horton infiltration parameters

Predominant soil type

CCC infiltration type

Initial infiltration rate (f0) (mm/hr)

Ultimate infiltration rate (fc) (mm/hr)

Horton decay rate (k) (1/s)

Porosity

Clay Poor 1 1 1.50E-03 0.385

Loam Moderate 5 2.5 1.00E-04 0.434

Sandy loam Free 10 5 3.00E-05 0.412

23


Silty loam Moderate 5 2.5 1.00E-04 0.486

For the soil infiltration simulations, TUFLOW allows a starting groundwater level to be specified. When a starting groundwater level is specified the infiltrated water is added to the starting groundwater level in each cell, based on the porosity of the soil. Once the groundwater level reaches the ground level, no more soil infiltration losses are allowed.

The base case is the 85th percentile groundwater level. The 85th percentile groundwater level is exceeded 15% of the time based on recorded data.

5.4 Base case starting parametersThe base case used for comparison to previous modelling (Section 6) and sensitivity assessment (Section 7) below have the following starting parameters:

1% AEP, 24 hour rainfall with nested profile;

10% AEP, static tide level downstream boundary;

5 m cell size;

Post-December 2011 topography;

Manning’s n values as outlined in Table 5-2;

Horton infiltration with parameters outlined in Table 5-3;

Groundwater level set to the 85th percentile; and

Only four major pipes modelled with breaklines, others are not modelled.

24


6 Comparison to previous modelling

6.1 Comparison of peak flows to river modelsThe peak flows at a number of locations from the base case with starting parameters (Section 5.4) were compared to the CCC river flood modelling. For the TUFLOW modelling, this was using the base case (see Section 5.4) model configuration.

The TUFLOW base case uses a nested storm approach, whilst the river flood models use a CCC hyetograph. The sensitivity of the model results to the hyetograph shape/length is described in Section 7.4.

The results are shown in Table 6-1. The flows in the TUFLOW models for the base case starting parameters are generally comparable to those in the MIKE FLOOD models which have had some calibration.

Table 6-1 Comparison of peak flows to river models

Maximum flows (m3/s)

Location TUFLOW1 MIKE FLOOD

Avon at Gloucester 49.7 50.0

Heathcote at Bridge Street 45.1 51.4

Styx at Tide Gates 14.9 26.2

Styx at Sheppard’s Drain 23.0 15.7 1 Base case with starting parameters (Section 5.4)

It is noted that the TUFLOW model predicts higher flows on the Styx at Sheppard’s drain, but lower flows at the Tide Gates. The Sheppard’s Drain site is upstream of Tide Gates. This indicates that the TUFLOW model is predicting more attenuation of the flood wave; this could be due a number of reasons, including the downstream boundary condition or the representation of the storage in the lower areas. Furthermore, given that the TUFLOW model does not include representation of the tide gates (which prevent flow upstream), the peak water depth in the Lower Styx Ponding Area (immediately upstream of the gates) is able to adjust more quickly to changes in downstream tidal level with a consequent smaller level difference across the gates. This smaller level difference equates to lower peak flow at this location. Conversely in the MIKE FLOOD model, which does contain representation of the tide gates, a larger level difference is shown in the model, which drives the larger discharge.

6.2 Comparison of flood extents to river modelsThe flood extents of the results of the revised base case (Section 9) were compared to the river model. It was observed that the flood extents of the TUFLOW model were more extensive, so the parameters were again revised. This is described further in Section 9.

25


7 Sensitivity

7.1 Sensitivity methodologyThis chapter describes the method and simulations performed in order to assess the sensitivity of the model results to the key input parameters. The sensitivity simulations are run for the base case starting parameters (refer Section 5.4) by varying the key input parameters one at a time.

The first step of the sensitivity simulations was to run the model with a range of cell sizes. The cell size sensitivity is presented in Section 7.3. A cell size of 5 m was adopted for the remaining sensitivity runs.

Following this the model was simulated for a range of parameter variations. The sensitivity tests assessed the following parameters:

1. Rainfall duration/hyetograph shape; 2. Roughness (Manning’s n value); 3. Soil infiltration; 4. Starting groundwater level; 5. Lowering of roads; and 6. Representing drainage network.

A matrix of the sensitivity simulations undertaken with the 5 m cell size is presented in Table 7-5.

Running the sensitivity simulations for a single catchment model had a shorter computational time than running the city-wide model. This approach was taken to allow for a larger range of simulations to be performed. For the sensitivity simulation presented below, the Avon catchment was used. As discussed and agreed with the international peer review panel on the 28 February 2014, the Avon catchment was selected for the following reasons:

Largest number of potential IFV properties;

Mix of urban and rural areas; and

Catchment size. Figure 7-1 shows a hydrograph comparison for both the Heathcote and Avon Rivers. For each comparison, the models were run as a single catchment and compared with the results for a combined catchments. The figure shows a close correlation in each catchment between the two models and demonstrates that there is no requirement to run a combined model for all sensitivity runs. Based on our previous assessments of the Styx catchment where there are a low number of potential IFV properties (approximately 550, of which 500 are river model results), there is less value in undertaking time consuming sensitivity runs for the Styx.

26


Figure 7-1 Comparison of individual vs combined model runs

7.2 OutlineThe results in this section are presented in a variety of methods:

Histograms of the difference in peak water level of raster outputs (Section 7.3 - 7.11);

Flooded properties (Section 7.12);

Long profiles (Section 7.13);

Tabular summaries of peak water levels and flows at key locations (Section 7.13);

Flow time series (Section 7.13); and

Fuzzy maps (Section 7.13).

7.3 Cell sizeThe choice of cell size for a 2D hydraulic model is influenced by a number of factors including the resolution of the topographic data, the width of the flow paths and the computational runtime.

The cell sizes tested were 2 m, 3 m, 4 m and 5 m. It was a finding of previous overland flow modelling and peer review of the results of that modelling, that 10 m cell size was too large for adequate representation of overland flow paths. For each case the difference in peak water level was compared to the 5 m results and a histogram of the difference created. This distribution gives an indication of the range and skew of the data. The histograms are shown in Figure 7-2, Figure 7-3 and Figure 7-4. The difference between 5 m and 4 m cell is approximately normal; whereas the difference between 5 m and 3 cells and 5 m and 2 m cells is slightly skewed with smaller cells producing higher water levels.

All differences are within the chosen thresholds of 0.2m and 0.1m and are therefore not an issue for the intended use.

27


Figure 7-2 Histogram of the differences in peak water level, and cell sizes


28



Peak modelled water levels were taken for the different scenarios. The location of the comparison points is presented in Figure 7-5 and the water levels are shown in Table 7-1. Comparison points were all selected as being along the river, as these locations are where upstream effects have essentially been integrated to result in the river level. If there was sensitivity to cell size being shown in the model results, it was expected that this sensitivity would show up most within the major river channels where river level is made up from the overland flow from adjacent non-river areas, upstream flow in the river to the observation point and downstream tailwater level. The range in predicted water levels in Table 7-1 is generally within 20 mm. It is noted that for Point A, the location is immediately upstream of a small constriction and the smaller cell sizes provide a higher water level, which is confined to the channel.

29


Figure 7-5 Location of water level measurements

Table 7-1 Peak water level comparisons for different cell sizes

Cell size scenario

Peak water level (m RL) at comparison points

A B C D E F G H

2 m 9.61 9.11 6.50 4.99 2.80 2.80 2.74 2.17

3 m 9.34 9.10 6.49 4.99 2.80 2.80 2.74 2.17

4 m 9.22 9.12 6.48 5.03 2.79 2.79 2.73 2.17

5 m 9.17 9.11 6.49 4.99 2.81 2.80 2.75 2.17

Range (m) 0.44 0.02 0.02 0.04 0.02 0.02 0.02 0.01

The primary output of the model is peak water level and a spatial plot of the difference in peak water level is presented Figure 7-6. This does show the maximum difference between 5m and higher resolution models occurring within the major drainage channels. There is generally good consistency between the 5 m model and the finer resolution models. A spatial plot of maximum velocity for the 5 m simulation is presented in Figure 7-7, at this resolution roads and streets are acting as expected flow paths. In this plot the areas where maximum velocity exceeds 0.5 m/s are clearly shown to be aligned along major drainage channels or roads, which is an expected result.

Given DEM (LiDAR-derived) data limitations (refer Section 3.1.2) and the lack of notable sensitivity to smaller grid sizes, a cell size of 5 m was selected as reasonable for subsequent simulations. The benefit from a 5 m cell size is faster run times, which enable more extensive sensitivity and calibration.

Having established a cell size of 5 m this cell size was used for subsequent sensitivity testing.

30


Figure 7-6 Difference in peak water levels for cell size minus cell size

Figure 7-7 Velocity distribution for cell size

7.4 Rainfall duration hyetographThe base case modelling uses a 24 hour nested rainfall hyetograph. To test the sensitivity of the model results to this, the following hyetographs were tested for the Avon catchment:

31


1% AEP, 24 hour nested rainfall hyetograph (base case);

1% AEP, 36 hour nested rainfall hyetograph (Scens 1);

1% AEP, 48 hour nested rainfall hyetograph (Scens 2); and

1% AEP, 24 hour CCC rainfall hyetograph (Scens 3). The flow time series presented in Figure 7-20 shows that the predicted hydrograph shapes for the rainfall hyetographs is similar. The nested rainfall hyetograph results in a higher peak flow for the same 24 hour hyetograph duration. The peak flow for the 24 hours nested rainfall hyetograph is 1.0 m3/s (2.2%) higher than the peak flow for the 24 hour CCC rainfall hyetograph. The histogram presented in Figure 7-10, shows the peak water levels are also generally higher for the nested rainfall hyetograph.

For the nested storms, the longer duration 36 hour and 48 hour durations results in higher peak flows. For the 36 hour nested hyetograph the peak flow is 1.6 m3/s (3.5%) higher than the 24 hour nested hyetograph (refer Table 7-6). For the 48 hour duration nested hyetograph the peak flow is 1.7 m3/s (3.8%) higher than the 24 hour event.

The histograms in Figure 7-8 and Figure 7-9 show that the longer duration (36 and 48 hour respectively) generally provided higher peak water levels than the 24 hour duration storm.

The longer duration 48 hour nested hyetograph includes the longest time of concentrations that occur in the three-catchments and therefore is will be adopted for the final simulations. The sensitivity assessment shows that the 48 hour nested hyetograph has the highest flows and generally the highest peak water levels.

Figure 7-8 Histogram of the change in peak water level for 36 hour nested storm

32


Figure 7-9 Histogram of the change in peak water level for 48 hour nested storm

Figure 7-10 Histogram of the change in peak water level for 24 hour CCC storm

33


7.5 Roughness (Manning’s value)The base case Manning’s values are the best estimate of Manning’s values. Two simulations testing the sensitivity of adopted Manning’s n value were performed:

Roughness decreased by 10% for all land use areas (Scens 4); and

Roughness increased by 10% for all land use areas (Scens 5). The decrease/increase in Manning’s n was applied to all land use types outlined in Table 5-2.

The peak flows in the model increase by 3.9 m3/s (8.6%) for the decreased roughness simulation; whereas for the increased roughness the peak flow decrease by 3.5 m3/s (7.7%). The hydrograph in Figure 7-20 shows that decreasing the roughness causes the peak flow to occur earlier in the simulations, and the increased roughness delays the flood peak.

The histograms in Figure 7-11 and Figure 7-12 show the change in peak water levels for the decreased and increased roughness respectively. From these it can be seen that whilst the flows in the main channel do change, the peak water levels are generally within 20 mm of the base case.

Given the relatively small range in resulting flow and water levels with changes in roughness values, and in the absence of calibration or comparison data, the recommended Manning’s values are those presented in Table 5-2.

Figure 7-11 Histogram of the change in peak water level for 10% decrease in roughness

34


Figure 7-12 Histogram of the change in peak water level for 10% increase in roughness

7.6 Soil infiltrationFor the base case models with starting parameters the Horton infiltration is included using the lower range estimates of parameters from the CCC WWDG (CCC, 2011) refer Section 5.3.

To test the sensitivity of the model to Horton infiltration parameters two sensitivity simulations were performed:

Soil infiltration using Horton approach and upper range parameters (Scens 11); and

No infiltration (Scens 6). The parameters in the base case are outlined in Table 7-2. The parameters for the sensitivity with higher infiltration are outlined in Table 7-3. It should be noted that the initial infiltration parameter is changed but the final parameter remains the same. The porosity values are unchanged from the base case.

Table 7-2 Horton loss parameters Base case

Predominant soil type CCC infiltration type

Initial (mm/hr)

Ultimate (mm/hr)

Decay (1/s) Porosity

Clay Poor 1 1 1.50E-03 0.385

Loam Moderate 5 2.5 1.00E-04 0.434


35



Table 7-3 Horton loss parameters Higher infiltration

Predominant soil type CCC infiltration type

Initial (mm/hr)

Ultimate (mm/hr)

Decay (1/s) Porosity

Clay Poor 5 1 1.50E-03 0.385

Loam Moderate 10 2.5 1.00E-04 0.434



The results show that the model is sensitive to the infiltration rate applied. The hydrograph in Figure 7-20 shows for no infiltration there is a 36% increase in the peak flows compared to the base case; whereas with an increase in the infiltration rate there is a 4% decrease in peak flow (also refer to Table 7-6).

A histogram of the difference in peak water levels for the increased infiltration is presented in Figure 7-13, which shows reductions in depths by up to 50 mm. A histogram of the difference in peak water levels for the no infiltration losses is presented in Figure 7-14 and shows larger differences of up to 250 mm.

It is recommended to use the infiltration rates as adopted in the base case. Justification for selection of the lower end infiltration rates is that floods typically occur in winter months in Christchurch, during which times evapotranspiration is low, groundwater level can be high and there is frequently antecedent wet conditions. The infiltration rates are considered further in Section 8 where the model parameters are calibrated for the 4 and 5 March 2014 flood.

It should be noted that the event simulated is a nested rainfall hyetograph, which includes a high-intensity rainfall burst within a longer duration of lower intensity sustained rainfall. Thus the effect of soil infiltration is being tested on a range of intensities nested within the design rainfall hyetograph adopted. However, the high-intensity rainfall is always preceded by lower intensity rainfall, which is not always the case in reality.

36


Figure 7-13 Histogram of the change in peak water level for higher infiltration

Figure 7-14 Histogram of the change in peak water level for no infiltration

37


7.7 Starting groundwater levelTo test the sensitivity of the model two simulations were undertaken:

85th percentile groundwater land (base case); and

50th percentile groundwater land (Scens 7). The 50th percentile groundwater level represents a lower starting groundwater level.

A histogram of the difference in peak water level is shown in Figure 7-15. The change in peak modelled flow at Gloucester gauge is -0.1 m3/s (refer Table 7.6). The model results are relatively insensitive to the change in starting groundwater level. There are some localised areas in which the groundwater level is limiting the amount of infiltration; however, for the majority of the model the starting groundwater level is not severely limiting the infiltration amount.

As an example, with a soil porosity of 0.4 and a cumulative infiltration of 100 mm, if the groundwater level is greater than 250 mm below the surface the entire 100 mm can be infiltrated.

The recommended starting groundwater level is the 85th percentile. Despite being relatively insensitive to this, the 85th percentile groundwater level is more conservative than the 50th percentile, but is possibly more representative of “typical” winter conditions when flooding is more likely.

Figure 7-15 Histogram of the change in peak water level for 50th percentile groundwater level

7.8 Lowering roadsBased on the recommendation of the peer reviewers, a sensitivity simulation was performed in which the road areas were lowered by 300 mm (Scens 8). This can be included in model

38


schematisation to increase the conveyance along the roads where the model grid size is unable to accurately represent the conveyance in these areas.

The results indicate the by lowering the road layer by 300 mm there is more flow down the roads. Scenario 8 results in a decrease in predicted peak flow at the Gloucester gauge of -4.1 m3/s (-9%) (Refer Figure 7-8). A histogram of the difference in peak water levels is presented in Figure 7-16. The histogram shows that the lowering of roads has a tendency to lower the peak water level by up to 250 mm; however, there are some areas where the peak flood levels increase. It is considered that artificial road lowering may result in under-estimation of flooding and flood extents on residential property.

For cell sizes of 5 m or less, the conveyance on the roads should be represented adequately with the model grid. Also the road have been given a lower roughness, refer Section 3.1.6. Therefore, the recommended approach is to use the road elevations from the DEM with the representation of roads with lower roughness.

Figure 7-16 Histogram of the change in peak water level for lowered roads

7.9 BuildingsThe sensitivity of inclusion of buildings was considered but not tested due to there being no available GIS layer of the building polygons. If this is to become available there are various methods for representing buildings which can be incorporated into the model with minimal effort:

Increase the roughness in these areas (can be depth varying);

39


Raise these to make low platforms, to allow rainfall to runoff these areas but to reduce conveyance;

Reduce the area of the underlying cells to reduce the storage;

Add a form loss (applied as a v2/2g loss);

Reduce the flow widths through the buildings; and

Any combination of the above. The majority of the flows are not through building areas, rather are through the waterways with overland flow paths along roads. Therefore, modelling the buildings directly may not influence the results dramatically.

The Manning’s n value for the urban areas is set to a high value of 0.1, to reflect the resistance to flow through these areas. This aims to simulate the flow impediments such as fences and buildings in urban areas. Note the Manning’s n value was further considered at the calibration stage, refer Section 8.4.1.

7.10 Representing drainage networkThere is not the data or ability to model all of the urban drainage networks. A sensitivity test was made by reducing the rainfall depth to account for the nominal capacity primary drainage layer. For this simulation it was assumed that the first 4 mm/hr is conveyed by primary pipe network draining with a 20% AEP level of service.

Given the shape of the hyetograph, a constant loss of 4 mm/hr significantly reduces the total rainfall depth, even for the nested rainfall profile adopted. A histogram of the difference in peak water level when compared to the base case simulation shows lowering of the peak water levels, refer Figure 7-17. This indicates that the local drainage does have a role, but

This sensitivity test revealed that if this pipe network does globally function to this level of service at all times, then substantially less flooding than predicted by the models would occur. Upon closer consideration this assumption was found to be flawed, mainly due to the very flat topography in Christchurch. Once tailwater levels have risen during a flood event, the available hydraulic grade is minimal and it was found that pipe networks convey less and less flow as flood levels rise, to the point where submerged pipes may convey very little flow at all near to the peak of significant floods (due to a lack of available hydraulic grade). This conclusion is supported by observations of flooding.

Therefore, the urban drainage network is likely to become limited in effectiveness when the water surface in the river rises. The condition of the drainage network is also not well known between and after the earthquakes. Therefore, the recommended approach is to not use a rainfall loss to represent the local drainage network.

The sensitivity analysis does highlight that the reliability of the model is reduced approaching the 20% AEP rainfall event as the primary piped stormwater system becomes more influential.

As part of the calibration process outlined in Section 8, the effect of the drainage network was further investigated.

40


Figure 7-17 Histogram of the change in peak water level for rainfall losses to represent the drainagenetwork

7.11 Channel invertsAs described in Section 3.1.3, the channels are modelled as a breakline with the channel invert from the surveyed cross-sections in the MIKE FLOOD models. This is a simplistic representation of the in-bank bathymetry. There is no detailed cross-section survey of the main channels after each intermediate earthquake event, therefore the sensitivity to changes in the cross sections was tested.

To test the sensitivity of the modelled results to the channel inverts two additional simulations were performed, in these the inverts of the main channels (where the inverts were derived from the MIKE FLOOD modelling) were modified. The two simulations were:

Raising the channel invert for the main channels by 0.5 m (Scens 12); and

Lowering the channel invert for the main channels by 0.5 m (Scens 13). Histograms of the differences in peak water level are presented in Figure 7-18 and Figure 7-19 for the raising and lowering cases respectively. The flow and long profile results are presented in Section 7.13. Whilst there is a small localised effect, the modification does not significantly change the results in the overland areas.

This effect is due to there being relatively few model cells within the main river channels. When assessing the effect of channel lowering on model results, clearly the effect will be most notable in model cells within the main channels. The lack of sensitivity to the channel lowering indicates that overland flow depths are not controlled by river level, but are more a function of the specific parameters controlling the overland flow (such as depth, slope, roughness) in the overland flow

41


areas. This satisfies the intended purpose of this testing, to give confidence that by having a coarse representation of the major river channels in the model, there is little effect on the predicted flood depths in the overland flow areas.

It is recommended that channel inverts converted from the post-earthquake MIKE FLOOD models be used for all cases.

Figure 7-18 Histogram of the change in peak water level; Main channel invert +0.50 (+500 mm)

42


Figure 7-19 Histogram of the change in peak water level; Main channel invert -0.50 (-500 mm)

7.12 Sensitivity flooded propertiesA sensitivity of the number of flooded properties has been undertaken for selected scenarios. Whilst it would be preferable to carry out the sensitivity to numbers of potential IFV properties, it is not possible without modelling all the intermediate events as these are used to identify potential IFV (refer T&T, April 2014).

All land parcels using the LINZ database for Christchurch were overlaid with the flood extent raster for each scenario. The properties were counted for flood depths greater than 100 mm. 100 mm flood depth was chosen because this is the approximate accuracy of the LiDAR that describes the land settlement associated with any single earthquake event. The parcels in this database were not filtered by land use i.e. it does not distinguish between residential, commercial, parks or industrial use. Instead it provides a count of the total number of different land parcels that have an intersection with any part of a 5 m x 5 m cell containing greater than 100 mm depth of flooding.

The results for the selected scenarios are presented in Table 7-4. The results show that the number of properties affected by 100 mm flood depth or greater is not sensitive to the selected scenario.

43


Table 7-4 Number of property parcels with greater than 100 mm flood depth

Scenario Parcel count

Base 63,924

Scenario 2 48 hour rainfall duration 63,953 (+.05%)

Scenario 4 Roughness -10% 63,500 (-0.7%)

Scenario 6 Infiltration on 63,516 (-0.6%)

44


7.13 Sensitivity summaryThis section summarises the results for the sensitivity simulations outlined in Sections 7.4 through to 7.11. Sensitivity scenarios are shown in Table 7-5.

Table 7-5 Sensitivity scenarios

Rain hyetograph Roughness Infiltration Ground-water level

Road lowering

Buildings

Drainage network by rainfall reduction

Channel inverts

Sens

itivi

ty

num

ber

24-h

our n

este

d

36-h

our n

este

d

48-h

our n

este

d

24-h

our C

CC

Best

est

imat

e

Min

us 1

0%

Plus

10%

Low

er b

ound

Upp

er b

ound

50th

%ile

85th

%ile

No

Yes

No

Yes

No

Yes

Best

Raise

d

Low

ered

Base case x x x x x x x x

Hyetograph

1 x x x x x x x x

2 x x x x x x x x

3 x x x x x x x x

Roughness 4 x x x x x x x x

5 x x x x x x x x

Infiltration 6 x x x x x x x x

Groundwater 7 x x x x x x x x

Road lower 8 x x x x x x x x

Buildings 9 x x x x x x x

Drainage 10 x x x x x x x x

45


No infiltration 11 x x x x x x x

Raised inverts 12 x x x x x x x x

Lowered inverts 13 x x x x x x x x

46


The flows at the Gloucester gauge for the sensitivity simulations are depicted in Table 7-6 (peaks), and Figure 7-20 (time series).

Table 7-6 Peak flows for the Avon at Gloucester

Scenario number

Sensitivity scenarios Maximum flows (m3/s) Change from base case (m3/s)

- Base case 45.5 N/A

1 36 hr nested hyetograph 47.1 1.6

2 48 hr nested hyetograph 47.2 1.7

3 24 hr CCC hyetograph 44.5 -1.0

4 10% decrease in Manning's n 49.4 3.9

5 10% increase in Manning's n 42.0 -3.5

6 Higher infiltration 43.3 -2.2

7 50th percentile groundwater 45.4 -0.1

8 Lowered roads 41.3 -4.1

9 Buildings N/A N/A

10 Drainage losses 9.95 -35.5

11 No infiltration 61.8 16.3

12 Raised river inverts 45.5 0.0

13 Lowered river inverts 46.1 0.7

47


Figure 7-20 Flow time series Avon at Gloucester for sensitivity runs

A long section profile of the peak water level is shown in Figure 7-21 (location) and Figure 7-22 (long section profile).

Figure 7-21 Long section location

0

5

10

15

20

25

30

35

40

45

50

55

60

65

70

0 6 12 18 24 30 36 42 48 54 60

Flow

s (m

3 /s)

Time (hours)

Base

Scen1 (36hr)

Scen2 (48hr)

Scen3 (24hr CCC)

Scen4 (low n)

Scen5 (High n)

Scen6 (High Infil)

Scen7 (50%ile GWL)

Scen8 (Lowered Rd)

Scen 10 (Drainage losses)

Scen11 (No Infil)

Scen12 (Raised Inverts)

Scen13 (Lowered Inverts)

48


Figure 7-22 Long section profiles for sensitivity runs

Peak modelled water levels were taken for the different scenarios. The location of the measurements is depicted in Figure 7-5 and the water levels are shown in Table 7-7. The mean and standard deviation are also presented for each of the comparison points.

0

2

4

6

8

10

12

14

16

18

20

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 22000 24000 26000

Peak

Wat

er L

evel

(m)

Chainage (m)

Base Case

Scen1 (36 hr Nested)

Scen2 (48 hr Nested)

Scen3 (24 hr CCC)

Scen4 (10% decrease in Manning's n)

Scen5 (10% increase in Manning's n)

Scen6 (Higher Infiltration)

Scen7 (50 percentile GWL)

Scen 8 (Lowered Roads)

Scen10 (Drainage Losses)

Scen11 (No Iniltration)

Scen 12 (Raised Inverts)

Scen 13 (Lowered Inverts)

49


Table 7-7 Peak water level comparisons for different scenarios

Sensitivity scenario Peak water level (m RL) at comparison points

A B C D E F G H

Base case 9.49 9.22 6.48 4.91 2.8 2.8 2.73 2.09

Scen1 9.49 9.22 6.49 4.93 2.85 2.85 2.78 2.13

Scen2 9.48 9.22 6.49 4.93 2.86 2.86 2.79 2.14

Scen3 9.36 9.19 6.46 4.89 2.79 2.79 2.72 2.09

Scen4 9.47 9.21 6.47 4.92 2.81 2.81 2.74 2.1

Scen5 9.51 9.23 6.48 4.89 2.79 2.79 2.72 2.08

Scen6 9.46 9.21 6.45 4.87 2.76 2.75 2.68 2.07

Scen7 9.49 9.22 6.47 4.91 2.8 2.8 2.73 2.09

Scen8 9.48 9.21 6.41 4.81 2.8 2.8 2.74 2.11

Scen10 8.72 8.9 5.46 3.75 1.91 1.93 1.89 1.83

Scen11 9.73 9.3 6.67 5.14 3.09 3.08 3.02 2.3

Scen12 9.74 9.25 6.51 4.96 2.88 2.87 2.81 2.10

Scen13 9.28 9.18 6.45 4.87 2.74 2.73 2.66 2.08

Mean (m) 9.44 9.20 6.41 4.83 2.76 2.76 2.69 2.09

Standard deviation (All) (m)

0.25 0.09 0.29 0.33 0.27 0.26 0.26 0.10

Standard deviation (no Scenario 10) (m)

0.12 0.03 0.06 0.08 0.09 0.09 0.09 0.06

A fuzzy map (Figure 7-23) was produced, showing the number of sensitivity runs in which a cell is shown to be flooded. For the fuzzy maps a value of 1.0 indicates that it was wet in all simulations, a value of 0.5 was wet in 50% of the simulations. The fuzzy shows red and blue areas least sensitive areas, whilst yellows is the most sensitive. The fuzzy map shows that there are relatively few areas where high sensitivity in results is shown for the sensitivity assessments carried out. Therefore, the overall overland flow results are relatively insensitive to the selected parameters and can be used with confidence.

50


Figure 7-23 Fuzzy map Green areas most sensitive, blue and red areas least sensitive

7.14 Sensitivity recommendationsRecommendations from the sensitivity testing undertaken include the following:

5 m cell size;

48 hour nested rainfall hyetograph;

Manning roughness values as per Table 5-2 (base case);

Infiltration, using Horton’s method and the values presented as the base case, consistent with the CCC WWDG;

85th percentile groundwater level;

Do not lower the roads in the DEM;

Do not reduce rainfall to account for pipe capacity; and

Use channel inverts converted from the post-earthquake MIKE FLOOD models.

51


8 Calibration to March 2014 event

In order to give greater confidence in the modelling, the model was calibrated to the flood event of the 4 and 5 March 2014. The model was simulated with observed rainfall and tidal boundaries and the predicted flood results were compared with the following:

River flow and stage gauges; and

Observed flood extents. The boundary data is described below in Section 8.1 and the results are presented in Section 8.2.

The model was simulated for the 72 hour period from midnight on 3 March until midnight on 6 March (03/03/2014 00:00 to 06/03/2014 00:00). The majority of the rainfall occurred in the 36 hours between midnight on 4 March and midday (12pm) on 5 March.

A preliminary analysis of this flood (T&T, May 2014) indicates that the flood frequency in the Avon at Gloucester Street was between a 5% AEP and 2% AEP event. The rainfall intensity varied across the city and in the Flockton Basin area of St Albans the frequency was between a 10% and a 5% AEP for durations of 6 to 48 hours.

8.1 Boundary dataRainfall data from the 17 rain gauges located across the city was utilised. This has been applied to the TUFLOW model as distributed rain-on-grid rainfall. The distribution was based on Thiessen polygons drawn about each gauge. Within each polygon the recorded hyetograph for closest recording station is applied to all cells in the polygon. The rainfall gauges and Thiessen polygons are presented in Figure 8-1. The rainfall data is reported with at 15 minute intervals with the exception of Site 325615 (Kyle Street EWS) which was provided as hourly. The cumulative rainfalls for the gauges are presented in Figure 8-2.

The recorded tidal water levels are used in the model as downstream boundaries conditions, the boundary time series are presented in Figure 8-3.

For the groundwater levels the 50th percentile groundwater phreatic surface was used given that the rainfall event occurred at the end of the summer period when groundwater levels are not expected to be particularly high.

52


Figure 8-1 Rain gauge locations and Thiessen polygons used for distributed rainfall calibrationevent

Figure 8-2 Cumulative rainfall at rain gauges for and March 2014 calibration event

53


Figure 8-3 Tidal boundaries for and March 2014 calibration event

8.2 Calibration datasetsThe calibration data of observed flooding was sourced from a number of locations. The primary sources of information were:

NIWA’s Neon monitoring system;

ECan’s website for river flows and stages; and

T&T, CCC and NIWA observed flood maps. In the sections below the modelled results are compared to the observed time-series data, extents and stage-flow relationships.

It is noted for the Avon at Gloucester flows, the raw data extracted from the NIWA Neon monitoring website shows the peak flow value as 71.86 m3/s. This value is much greater than the reported value of 28 m3/s obtained from the ECan website. It is our understanding that the flow was measured (by accurate gauging during the event) at 24m3/s and the peak flow was estimated to be 28 m3/s (the measurement did not occur at the peak flow). For the flow hydrographs presented below, the flows have been multiplied by 0.39 (28 / 71.86) to match the peak flow of 28 m3/s.

8.3 MethodIn order to perform a number of iterations of the model, the Avon model was simulated in isolation. The Avon was chosen for a number of reasons:

The flow was gauged during the flood event giving greater confidence in the flows/volumes;

Largest number of potential IFV properties; and

54


Catchment size is adequately representative for the greater Christchurch urban area. Once a good fit was achieved at the Avon gauge the entire model (Styx, Avon and Heathcote Catchments) were modelled using the same parameter set.

8.4 Calibration model with revised parametersIn Figure 8-4 the recorded flows (blue) for Avon at Gloucester are compared to the TUFLOW model results for the base case with sensitivity recommendations (green) (Section 7.14). The total volume under the recorded hydrograph excluding the base flow (of ~1 m3/s) is approximately 1.4 Mm3. The volume for the model for the base case with sensitivity recommendations is approximately 2.0 Mm3 and the flood is still receding. This indicated that there is significantly more volume for base case with sensitivity recommendation when compared to the gauged flow. The primary reason for the discrepancy is likely to be due to the infiltration parameter assumed.

As well as the additional volume in the modelled results, it was noted that the model is slower to respond than the recorded flow. A slower responding model may mean that modelled flood peaks occur later than in reality. The TUFLOW model contains no representation of surface water connection with groundwater, and it is possible that in some of the more permeable areas this connection may have some effect on total runoff volume. It is less likely to affect flood peak due to the short duration over which a flood peak may occur, but could contribute to the reason for there being a greater volume in the modelled result than shown by the recorded result.

A number of simulations were performed to better match the recorded data. Only the key simulations are described here.

The following parameters were changed for the calibrated model:

Manning’s n was applied as depth varying in urban areas;

Fraction Imperviousness was increased from 0.30 to 0.50 for urban areas (the predominant land use in the Avon model);

Pipes were included as breaklines;

The College of Education (325507) hyetograph was used for the upper catchment;

Infiltration rate was increased from an initial condition of 0 mm/hr up to a maximum constant loss rate of 4.5 mm/hr (limited by the depth to groundwater) was used for all soils; and

Base flow of 1 m3/s was included. There is a discussion below on each of these changes with results described in Section 8.5.

8.4.1 Representation of urban areasThe application of roughness to the 2D model has effects on both the rainfall-runoff response and on the overland flow response. In the TUFLOW model buildings have not been specifically represented, with land use being the driver for differing surface roughness. The intention is that overland flow is driven by surface roughness, with higher roughness being used for areas that are more built up and pose greater restriction to overland flow. Application of higher roughness in these areas has the collateral effect of slowing the rainfall-runoff response, which is not how the actual surfaces perform in reality.

The urban Manning’s n value of 0.1 was found to be too high and was artificially slowing the runoff from flowing off urban areas (‘Built up area’ in Table 5-2). A depth varying Manning’s n was

55


used as this allows for a low roughness to be used at shallow depths to represent the rainfall runoff (e.g. roofs and driveways). At higher depths, an increased roughness is applied to represent overland flow through urban areas, where fences and building provide an impediment to flow. The depth varying roughness is outlined in Table 8-1. An increase in the fraction impervious from 0.3 to 0.5 was found to better match the observed flow hydrograph.

Table 8-1 Depth varying Manning’s values for urban areas

Depth Manning's n

Less than 50 mm 0.015

50 mm – 100 mm The value varies linearly from 0.015 to 0.05

Greater than 100 mm 0.05

8.4.2 Modelling of pipes as breaklinesThe absence of the drainage network (pipes) was found to be artificially slowing the catchment response. In reality, the rain that flows into the pipe system is transported very quickly to the rivers and creeks. In addition, areas of the city that are served predominantly by pipe systems e.g. parts of Spreydon and Woolston had an over representation of flooding compared to the observed flooding.

To address these issues all pipes greater than 600 mm were represented as simulated pipes using breaklines (refer Section 3.1.4).

8.4.3 Rainfall in the Upper Avon catchmentBased on the Thiessen polygons, the upper catchment the College of Education (325507) hyetograph has been applied to the upper sections of the Avon hydraulic model. The Templeton rainfall gauge is located to the West (Upstream) of the Avon hydraulic model; this station reported a cumulative rainfall depth of 82.4 mm compared to 120.8 mm at the College of Education site. This indicates that the rainfall was decreasing to the west. For the calibration run the College of Education (325507) hyetograph was applied in the Thiessen polygon for the Upper Avon catchment.

8.4.4 Infiltration rateAs discussed earlier in Section 8.4, the modelling for the base case with sensitivity recommendation showed that the volume of runoff predicted was higher than observed when using the infiltration rates based on the CCC guidelines. An initial infiltration rate of 0 mm/hr increasing to a constant loss rate of 4.5 mm/hr (limited by the depth to groundwater) for all soils was found to give a good fit to the observed data.

8.5 Results

8.5.1 Time seriesThe modelled and recorded flow are presented in Figure 8-4. The modelled results for the base case with sensitivity recommendations are presented in green, after the calibration exercise the model results are shown in orange.

56


It can be seen in Figure 8-4 below that with these changes (orange line) the calibration model responds more quickly to the rainfall. And the volume is more consistent with the observed volumes. This gives a better calibration to the recorded data. The peak flow estimate is 30.6 m3/s, which is within 10% of the recorded value of 28 m3/s. We consider that the observed peak flow may be too low because the flow-stage relationship for the gauge may be less accurate for high stages when the flow is out of bank and spreads across flat surrounding land.

Figure 8-4 Flow time series Gloucester gauge for the and March 2014 calibration event

The whole model (Styx, Avon and Heathcote catchments) were then simulated after making amendments based on the calibration of the Avon catchment. Flow data was not available for recording stations on the Styx and Heathcote rivers, and therefore levels have been compared. The stage at Buxton Terrace in the Heathcote catchment is presented in Figure 8-5. The stage at Radcliffe Road in the Styx catchment is presented in Figure 8-6.

The model shows a good calibration for the Buxton Terrace gauge on the Heathcote River, with the model showing a good response to the timing and levels. As discussed earlier, the in-bank bathymetry is simply represented in the TUFLOW model.

The Styx was not as closely correlated between recorded and modelled river levels at Radcliffe Road. Furthermore, calibration of the CCC Styx river model has also been difficult. It is noted from T&T’s experience that the gauge site is subject to weed growth and can be inaccurate.

57


Figure 8-5 Stage time series Heathcote at Buxton Terrace the and March 2014 calibration event

Figure 8-6 Stage time series Styx at Radcliffe Road the and March 2014 calibration event

8.5.2 Observed flood extentsThe observations of flooding for the 4 and 5 March 2014 flood calibration event are shown in Figure 8-7 to Figure 8-12. Of note is that the observations are visually based (no topographic survey was carried out), the extent of the observations was limited by resources, and that the observations were made after the peak of the flooding. Also note that the absence of observed flooding does not mean that flooding did not occur, but that flooding may not have been

58


recorded by an observer (as these people could not be everywhere). Therefore, comparison of predicted with observed is only useful in areas where flooding was observed. For the flood calibration event of the 4 and 5 March 2014, we consider that the modelled flood extents are generally consistent with the flood observations.

Figure 8-7 Comparison of calibration model maximum depth with observed flood extents forand March 2014

59


Figure 8-8 Comparison of calibration model maximum depth with observed flooding for andMarch 2014 Flockton Basin

Figure 8-9 Comparison of calibration model maximum depth with observed flooding for andMarch 2014 Bexley

60


Figure 8-10 Comparison of calibration model maximum depth with observed flooding for andMarch 2014 Avondale

Figure 8-11 Calibration run of maximum depths compared to combined T&T, CCC and NIWAobserved flooding map for the and March 2014 event

61


Figure 8-12 Calibration run of maximum depths (right) compared to combined T&T, CCC andNIWA observed flooding map (left) for the and March 2014 event Flockton Basin.

8.5.3 Stage-flow relationship for Gloucester Street gaugeComparisons between the recorded stage and flows (i.e. the gauge rating curve) and the modelled stage flow relationship were made. These are presented in Figure 8-13. It can be seen that the values match well on the rising limb of the flood, but there is a significant “looping” or hysteresis in the falling limb for the modelled relationship. This hysteresis is to be expected and is observed in natural systems for the flood recession. As a result there is likely to be a lesser flow on the flood recession, than for the same stage when compared to the rising limb. This is because of the flatter hydraulic gradient of the river that occurs for the flood recession.

This comparison gives confidence that the in-bank representation using breaklines is providing a realistic representation of the conveyance in the Avon River. It also explains why the falling limb of the observed hydrograph is not replicated so well by the model (Figure 8-4), because it indicates that the observed hydrograph is probably under-predicting flows as the stage-discharge relationship as the flow gauge does not account for this hysteresis.

It should be noted that the TUFLOW model was never intended to be used for instream channel hydraulics, and that the goodness-of-fit observed is some justification of the notion that peak flood levels are frequently driven more by backwater effect than by channel hydraulics.

62


Figure 8-13 Stage-flow relationships for Gloucester gauge for the and March 2014 calibrationevent

8.5.4 Comparison of recorded groundwater surface with the 85th

percentile surfaceWe have prepared figures comparing the 85th percentile modelled groundwater surface with boreholes that were monitored on 7 March 2014 (Figure 8-14). On average across the monitored boreholes the difference in depth to groundwater between the two data sets is 0.07 m with a maximum measured difference of 1.18 m. Given that the sensitivity results show that the peak flood levels are insensitive to the level of the groundwater we consider that the 85th percentile is a reasonable level to adopt for the 1% AEP rainfall event.

63


Figure 8-14 Comparison of recorded depths to groundwater with 85th percentile modelled surface

64


8.6 ConclusionsComparisons of the recorded data and modelled results for the 4 and 5 March 2014 flood event have been presented in this section. With the revised parameters, the model is providing and acceptable calibration to the observed data. The calibrated model with revised parameters are considered acceptable and will be used for final design simulations.

65


9 Base case with revised parameters

9.1 Comparison to initial parametersFollowing the calibration of the hydraulic model, the results from the base case with starting parameters (refer to Section 7) was compared to the results from the base case using the revised parameters from the calibration (refer to Section 8).

The peak flows for the base case with revised parameters and the base case with initial parameters and the river model are presented in Table 9-1 and the time series are presented in Figure 9-1. The base case with revised parameters provides a higher peak flow, which occurs more quickly following rainfall. This is consistent with the model changes from the calibration.

Table 9-1 Peak Flows at Gloucester gauge on the Avon river for base casesimulations

Simulation Maximum flows (m3/s)

Base case with initial parameters 45.5

Base case with revised parameters (final) 52.6

MIKE FLOOD with revised parameters 50.0

Figure 9-1 Flows at Gloucester gauge on the Avon River for base case simulations

66


The number of properties affected by flooding greater than 100 mm for the base case with revised parameters is 59,969 (-6%) compared to the base case with initial parameters that had 63,924 (refer to Table 7-3).

9.2 Sensitivity to downstream boundary conditionsWhen the flood extents of the TUFLOW model with the revised parameters were compared to the river model, it was found that the TUFLOW results had more extensive flooding in the lower Avon area than was expected. The static tide level used in the base case in the Avon was shown to be above ground level for a large portion of the downstream area, as shown in Figure 9-2. The result of this was that there were backwater effects from flow into the area, as there was no way for water to discharge out of the model, which resulted in unrealistically high water levels.

Figure 9-2 Areas below tide level in the Avon catchment

Various tidal scenarios were tested for the Avon catchment:

A static mean level of sea (MLOS) of 9.363 mRL Christchurch Drainage Datum (CDD);

A static low water (MLWS) of 8.488 mRL CDD;

A static low water (MLWS) of 8.488 mRL CDD with an 18 hour triangular hyetograph storm;

A dynamic tide level centred around MLOS with an amplitude of 1.75 m (centred around 9.363 mRL CDD; and

A dynamic tide level with a peak at the 10% AEP tide level (centred around 9.954 mRL CDD). As a result of the tidal sensitivity analysis a dynamic tide level with a peak at the 10% AEP tide was chosen for the final base case. This boundary condition was chosen for the following reasons:

This approach is consistent with the CCC modelling Specifications (GHD 2012);

This is the approach adopted in the MIKE rainfall runoff (river) models; and

67


This approach (as compared with a static high tide) allows the model to drain at low tide cycles (as would occur in reality).

The time series of the boundary condition was altered slightly from the dynamic tide centred around 9.954 mRL, to the time series used in the river models, which include storm surge in the peak. An example of this for the Avon is shown in Figure 9-3.

Figure 9-3 Adopted tide based on river model compared to tide used in sensitivity test for the Avon

The MIKE FLOOD river models have timed the peak river flood flow to coincide with the peak tide level. It is not possible for the overland model to synchronise the peak river flow with the peak tide because the peak flows in the three rivers modelled occur at different times. Whilst the model could be split into three separate flow models to allow the synchronisation of peak river flow and peak tide; given the cross catchment boundary flow/flood issues this would create, we have preferred to use a combined city-wide model and to take the following approach. We have synchronised the peak rainfall intensity with the peak tide level. Figure 9-4, Figure 9-5, and Figure 9-6 show the outcome of this method. Whilst the peak water level of the models do not align with the actual peak of the tide, the peak water levels trend from upstream to downstream (top lines to bottom lines) towards aligning with one of the other peaks in the downstream boundary. Also as the overland flow model will be used with the river model then the higher water levels at the coast in the river models will prevail.

68


Figure 9-4 Water level time series compared to peak tide Avon

Figure 9-5 Water level time series compared to peak tide Heathcote

69


Figure 9-6 Water level time series compared to peak tide Styx

The time series data of the tidal boundaries have been based on time series provided by DHI, GHD and NIWA. This data extended for various periods of time, and for different return periods:

DHI time series extended sufficiently before and after peak tide for use in the TUFLOW model, for the three required return periods for modelling (10%, 50% and mean AEP);

GHD time series extended sufficiently before and after peak tide for use in the TUFLOW model, but was only for the 10% AEP tide (not 50% or mean tide) – the tides were scaled based on the ratio between the peak tides of the various return periods in the table in the CCC WWDG; and

NIWA data did not extend sufficiently before peak tide (after peak there was sufficient data), but was available for the three required return periods. The data was extrapolated for the TUFLOW model.

9.3 Final base case parametersThe final base parameters for the TUFLOW model are therefore:

1% AEP Rainfall, 48 hour nested storm;

10% AEP dynamic tide downstream boundary, based on river model;

5 m cell size;

Manning’s n values as outlined in Table 5-2, except urban areas, which had depth varying Manning’s

Fraction impervious of 0.5 for urban areas;

Initial infiltration rate of 0 mm/hr, increasing to a constant infiltration rate of 4.5 mm/hr, limited by the depth to groundwater;

70



All pipes greater than 600 mm diameter modelled as simulated pipes using breaklines.

71


10 Final overland model boundary conditions

The boundary conditions that have been applied to the overland flow model for the assessment of IFV are summarised below. The basis for the boundary conditions are explained in T&T (April 2014).

Type Model Source Version File Name Date Received

(yyyy/mm/dd) Rainfall

Downstream Boundary

Topography (LiDAR)

Ove

rland

Flo

w m

odel

s

Pre-Sept T&T 064 EQC_All_PreSep_05m_100y_48hNest_DynamicTide10_064_d_Max 2014/04/30 Rainfall duration of 48h, total rainfall depths of: 167.27 mm Avon 182.40 mm on “flats” 160 mm on “subcatchment 14” 207.99 mm on “hills” 165.05 mm on Styx

10% AEP peak.

10.829 mRL at Avon, 10.768 mRL at Heathcote, 10.843 mRL at Styx

Pre Earthquake (Combined 2003 CCC, 2003 & 2005 WDC, 2008 SDC)

Post-Sept

T&T 064 EQC_All_PostSep_05m_100y_48hNest_DynamicTide10_064_d_Max 2014/04/30 Post September 2010 overlaid on Pre Earthquake

Post-Feb T&T 064 EQC_All_PostFeb_05m_100y_48hNest_DynamicTide10_064_d_Max.tif

2014/04/30 Post February (Combined March 2011 & May 2011)

Post-June

T&T 064 EQC_All_PostJune_05m_100y_48hNest_DynamicTide10_064_d_Max 2014/04/30 Post June (September 2011)

Post-Dec T&T 064 EQC_All_PostDec_05m_100y_48hNest_DynamicTide10_064_d_Max 2014/04/30

Post December (February 2012 overlaid on Post June)

72


11 Peer review comment and responses

The TUFLOW model has undergone significant peer review. The peer review comments and the responses by T&T are shown in Table 11-1.

In addition, the International Peer Review Panel requested additional sensitivity runs on infiltration, which are included in Appendix A.

Table 11-1 Peer review comments and responses

Peer review comment T&T action

Check if sensitivity in Avon catchment is also representative of sensitivity in Styx and Heathcote.

T&T added note of this in report Section 7.1.

Check differences between Styx river model and overland flow model results.

The issue was at flap gate, where overland flow model did not include time varying tide data, but river model did. Comparison point changed to a non-tidally influenced area.

Reviewers were interested in the impact of the sensitivity runs on number of properties affected by IFV.

It is not possible to do full IFV assessment on sensitivity runs, but number of properties affected by flooding for each sensitivity run are provided (3).

Calibration of model using the 4/5 March 2014 flood event would be hugely beneficial.

Calibration to March flood was undertaken. Model was revised accordingly. Summary is provided in Section 8.

Consider sensitivity to river cross sections, as only one set of cross sections has been used.

River cross section sensitivity undertaken, summary in Section 7.11.

T&T to ask CCC for information as to when river cross section surveys were undertaken.

T&T requested to CCC. Council provided database. However, the model is insensitive to river cross sections.

Table of boundary conditions and inputs needed for overland flow models as well as river models.

Section 10 for Overland boundary conditions. River boundary conditions to be provided in a separate report.

Flow extraction line does not extend across the entire flooded width. Thus, the flows in Table 6-1 may be under represented.

Extraction line extended to rectify error.

Groundwater sensitivity questioned. Groundwater sensitivity undertaken in Section 7.7.

Comparison of hydrographs with river models. Not undertaken as hydrographs no available. Less important due to additional calibration to 4/5 March 2014 flood.

Validation runs to include comparison to hydrographs only include channel flow. Stage-flow relationship thus may not be accurate above bank full.

Noted.

Compare stage-flow relationship in model to rating curves.

See Section 8.5.3.

73


Provide evidence for why primary drainage does not carry much flow and can be excluded.

Sensitivity run undertaken, and primary drainage network pipes >600 mm have been included.

Include comment on flood frequency analysis in report.

Analysis of the 4/5 March 2014 flood frequency found in Section 8.

Update parcel count for revised base case. Not available – revised based case was only run for the Avon catchment.

A chart for each catchment showing hydrographs (water level or flow) at two or three locations (upper/mid/lower) along the main waterway within the primary areas of potential IFV, along with the tidal boundary, to show the timing of the flood peak with the timing of the tidal cycles. If possible, it would also be useful to see any river modelling hydrographs you have on the same charts.

Provided in Section 9. We do not have river model hydrographs for the same locations. Given that the river hydrographs are for different duration storms we do not consider there is value in adding them to the same charts.

74


12 Conclusions and recommendations

12.1 SensitivityInitial sensitivity assessment was carried out on grid size, so that a constant grid size could be used for all subsequent sensitivity assessments. It was found that a grid size of 5 m was adequate for these assessments in that no significant additional accuracy could be gained by making use of a finer grid, and that run times could be minimised by making use of the 5 m grid.

Relatively low sensitivity in the model results was found for a variation in adopted rainfall temporal profile (hyetograph) for the 1% AEP event. The insensitivity to rainfall duration is due to the nested profile. As a result, it is not of significance whether or not final runs are done across the three catchments to the same rainfall duration, or if this is varied according to catchment critical duration as previously assessed by CCC.

The adopted Manning’s roughness correlated with those from several recognised references. The variation in Manning’s roughness was found to have small effects in model results.

Significant model sensitivity was found in surface infiltration, not so much in the values used but in whether or not allowance for any infiltration was made. Previous overland flow modelling was carried out without any allowance for infiltration, and it was found that significantly more flooding would result under this assumption when compared against the results from simulations where allowance for infiltration was made. Relatively low sensitivity was found when the infiltration values were tested across reasonable ranges for the various surface soil types.

Low model sensitivity was found when groundwater level at the start of a simulation was varied between the 50th and the 85th percentile levels. This is probably due to groundwater levels being closest to ground surface beneath the finer-grained soils that exhibit the lower infiltration capacities and which are located in the lower, downstream areas of the catchments.

Lowering of all roads within the model domain by 300 mm was found to have a significant effect on model results. Doing this would increase conveyance along roads while reducing flooding on adjacent property. With use of a 5 m, or finer model grid, it was found that this artificial road lowering was not realistic and would probably result in under-estimation of flooding and flood extents on residential property.

Accounting for the conveyance capacity of the buried drainage network was made by subtracting what would be a 20% AEP rainfall intensity from the 1% AEP rainfall, which results in a significant difference in model results. Upon inspection, it was found that in reality the effectiveness of the pipe network diminishes with the severity of the flood, as tailwater levels rise to the extent that there is little hydraulic grade in many pipes.

Examination of the fuzzy map produced from the sensitivity assessments reveals that, with the exclusion of sensitivity test number 10 (the reduction in rainfall to allow for pipe network flow), found the overall overland flow results are relatively insensitive to the selected parameters and can be used with confidence.

Recommendations that emerged from the sensitivity testing (that are unchanged by the subsequent calibration) include the following:

5 m cell size;

Use adopted Manning roughness values;

75


Adopt the 85th percentile groundwater level for further simulations;

Rely on the accuracy of results within the areas indicated in the fuzzy map, and treat other areas with some caution, based on field verification;

Do not lower the roads in the DEM; and

Do not reduce rainfall to account for pipe capacity.

12.2 CalibrationSubsequent to the model building process a flood event occurred in Christchurch that has been used for model calibration. The rainfall event of 4/5 March 2014 has been hindcast using the overland flow model. For the calibration runs the rainfall was varied spatially based on the recorded rainfall at 18 rain gauges across the city, and modelled flood extents were compared with observed flood extents.

The observed flood extents suggest that in areas where the existing primary drainage network comprises large diameter pipework, additional pipe representation is required by approximation of the piped network conveyance capacity with terrain modifiers in the model.

The calibration runs also identified that the model with parameters recommended from the sensitivity runs had higher water levels and flows, predicted a later flood peak and had too much runoff volume. The calibration required modifications to a number of parameters these form the basis of further recommendations for the model:

Increase the fraction impervious to 0.5 for urban areas;

Infiltration rate = 0 mm/hr initial, 4.5 mm/hr as a constant loss;

Set a depth varying Manning’s value for urban areas, reducing the roughness at shallow depths;

Simulated pipes included for pipes greater than 600 mm diameter;

Templeton rainfall used for the upper catchment; and

Base flow increased by 1 m3/s at Hagley Park. A review of the base model with revised parameters showed that that a static tidal boundary created significantly more flooding in the downstream reaches of the Avon. Therefore a dynamic tide timed to coincide with peak rainfall intensity for the three rivers has been adopted as the base model final parameters.

The final parameters are as follows:

1% AEP Rainfall, 48 hour nested storm;

10% AEP dynamic tide downstream boundary, based on river model;

5 m cell size;

Manning’s n values as outlined in Table 5-2, except urban areas, which had depth varying Manning’s

Fraction impervious of 0.5 for urban areas;

Initial infiltration rate of 0 mm/hr, increasing to a constant infiltration rate of 4.5 mm/hr, limited by the depth to groundwater;


76


All pipes greater than 600 mm diameter modelled as simulated pipes (equivalent channels) using breaklines.

77


13 References

Benn, J, Smart, G, Syme, W, 5 May 2014. EQC Increased Flooding Vulnerability Damage Peer Review Final Report Part 1 (Draft 4)

Christchurch City Council. 2011. Waterways, Wetlands, and Drainage Guide.

Environment Agency August 2013 Benchmarking the latest generation of 2D hydraulic modelling packages Report – SC120002

GHD July 2012 Christchurch City Council Stormwater modelling specification for flood studies

Landcare Research Land Cover Database version 3 (LCDB3). Land Research Information Systems portal (https://lris.scinfo.org.nz)

Rawls, W. J., D. L. Brakensiek and N. Miller, (1983), "Green-Ampt Infiltration Parameters from Soils Data," J Hydraul. Div. Am. Soc. Civ. Eng., 109(1): 62-70.

S. van Ballegooy, S. C. Cox, R. Agnihotri, T. Reynolds, C. Thurlow, H. K. Rutter, D. M. Scott, J. G. Begg, I. McCahon. 2013. Median water table elevation in Christchurch and surrounding area after the 4 September 2010 Darfield Earthquake. GNS Science Report 2013/01, 66 pages plus 8 appendices, ISSN 1177-2425, ISBN 978-1-972192-34-4.

T&T April 2014 Volume 1: Increased Flood Vulnerability: Assessment Methodology Report

T&T 2014 Volume 2: Increased Flood Vulnerability: River Modelling and Coastal Extensions

Report

T&T 2014 Volume 3: Increased Flood Vulnerability: Overland Flow Model Build Report

T&T 2014 Volume 4: Increased Flood Vulnerability: Implementation Report

T&T 2014 Volume 5: Increased Flood Vulnerability: Volume 5: Increased Flood Vulnerability: Geological Processes Causing Increased Flood Vulnerability

Tonkin & Taylor. May 2014. Christchurch Rainfall Analysis

REPORT - Earthquake Commission...Increased Flood Vulnerability: Overland Flow Model Build Report Job...

Documents

Transcript of REPORT - Earthquake Commission...Increased Flood Vulnerability: Overland Flow Model Build Report Job...