• About this site|
• Site index|
• Contact us|
• Jobs

A methodology to assess the impacts of climate change on flood risk in New Zealand

Executive Summary

1. Introduction

2. Planning and decision-making context

3. Methodology

4. Assumptions and uncertainties

5. Further reading

Appendix: Results for the Buller Catchment example

3. Methodology

3.1. Overview

The methodology described in this report enables councils to establish how much the current flood risk could alter under climate change, based on the expected changes in rainfall intensity during storm events.

The methodology follows a staged approach of increasing complexity.

The essential foundation for this approach is knowledge of the flood risk under the current climate conditions, based on some representative past storm events. An increasing number of councils already use rainfall information rather than river gauge data to determine flood risk and inundation levels associated with any given AEP. This underpinning information on actual rainfall in the catchment, and how this translates into river flows and inundation levels, is required to allow robust estimates to be made on the future effects of climate change on flood risk.

The first step of the methodology is to undertake a simple screening test. This involves the uniform scaling of the amount of rain falling during a storm event within the catchment over the peak rainfall period, where the scaling factor is proportional to the assumed increase in temperature caused by climate change.

If the above screening analysis shows a potential significant change in flood risk for the area in question, councils should proceed to the second step. This is a more detailed modelling of the likely changes in rainfall for the specific catchment under a warmer climate. This can be done by using a Regional Atmospheric Modelling System (RAMS), which is sensitive to specific catchment characteristics.

The increased rainfall from either the simple screening test or the more complex weather modelling approach can then be translated via run-off and river flow models into the peak flow and risk of inundation at areas of interest. The models used to translate the increased rainfall into river flows and inundation levels are the same as would be used in a sophisticated flood risk analysis for the current climate. This approach, for the current climate at least, is increasingly being adopted by councils.

It should be recognised that rainfall is not the only parameter that is likely to change in future and that could influence flood risk. Other factors include climate variability, land-use change, sea-level and storm surge for coastal locations, and the level of development and hence protection needed. Some of these factors rely on a number of assumptions and uncertainties. The report will briefly comment on how and the extent to which these factors can be taken into account, and as a consequence how robust the results of the methodology outlined here are likely to be.

3.2. Establishing the base case (current climate) flooding event

In order to assess the likely changes in flood risk under climate change, the council will need to choose a storm or storms under current climate conditions with which changes can be compared. It is often best to choose real examples of past storms, where their impact on flooding is known. These storms should span the range of storms likely to impact on the catchment being considered. They need not be the biggest storms experienced, but ones that could be regarded as being representative of the weather situations likely to be experienced. Choosing such storms enables weather and rainfall-to-river flow models to reproduce the current conditions. This should give councils a good idea of how well the models replicate the weather and its effects on flood risk for the catchment.

Current extreme rainfall rates for particular locations, durations and average recurrence intervals can be obtained from an analysis of historical rainfall data sets from monitored sites, or from the High Intensity Rainfall Design System (HIRDS) CD-Rom, available from NIWA.

For coastal settlements, councils may wish to consider the current flooding risk posed by a combination of heavy rain with high tides and storm surges.

3.3. Which climate change scenarios?

It is impossible to predict the precise amount by which the Earth will warm over the next century, and how much the different regions of New Zealand will warm, as this depends on future greenhouse gas emissions. In addition, although all climate models show a general warming trend, they give slightly different answers with regard to the exact amount of warming. We recommend councils consider at least two temperature change scenarios covering the expected lifetime of infrastructure and other major developments. For example, mid-low and mid-high scenarios for both 2030 and 2080 could be suitable choices.

It is generally recommended that more than one mid-range scenario be used, since a narrow focus on the mid-range may underestimate the uncertainties associated with climate change projections. While some timescale needs to be chosen to establish the size of changes, it must also be recognised that climate change will not end at 2080, but will continue to have an increasing effect into the distant future.

There are a number of factors to consider when deciding which climate change scenarios are most appropriate, including;

• The duration of decisions to be made
• The nature and value of the assets which may be at risk
• The extent of the assets which may be at risk

The projected changes in average temperature for each region of New Zealand are summarised in Table 2 of Preparing for climate change (see Further Reading, below). More details can be found on pages 11 and 12 of Climate Change Effects and Impacts Assessment.

3.4. Rainfall: the simple screening test

One approach to determining how climate change will affect extreme rainfall in a community is to use a standard scaling factor as set out in Table 7 (shown below) of Preparing for climate change (see Further Reading, below). The table recommends percentage adjustments to apply to extreme rainfall for a range of average recurrence intervals. The increases in rainfall are per degree Celsius of warming. Take, for example, a 1-in-50-year event lasting 3 hours. For this example, the amount of rain expected to fall under today's climate should be increased by 7.2% for every degree Celsius of projected increase in the annual mean temperature.

The scaling factors are mid-range estimates. They take into account the extra rainfall likely as a result of the extra moisture holding capacity of the air. They do not take account of local catchment characteristics, nor do they reflect that storms are likely to be more intense due to the heat released by this moisture and the increase in pole to equator temperature gradient in our region.

Based on the estimates of increased rainfall from the above table, a screening assessment of the likely change in flood risk is possible. For example, the table above suggests that for a 1°C warming, 3-hour rainfalls with 50 year ARI are likely to increase by 7.2%. This figure can be used to increase the rainfall from past storms and the effects modelled. Hydrological models can be used to estimate the river flow expected from these bigger storms, and inundation modelling can be used to assess the likely increased areas and depths of flooding. These steps are outlined in sections 3.7 and 3.8. An increasing number of councils are already using such more sophisticated models to assess their current flood risk. An example of this approach is outlined in the appendix.

The simple screening test described in this section allows a preliminary and relatively low-cost assessment of whether climate change could significantly alter the flood risk for a particular location. However, a number of simplifying assumptions have been made in deriving the figures in the table, and no recognition has been given for particular catchment characteristics that could modify the preliminary findings. A more detailed assessment, as described below, is often warranted and necessary.

3.5. Rainfall: take a detailed look

If the simple screening test indicates that climate change could significantly affect an important or large-scale council function or service, we recommend a more complex approach using weather and flooding models. This will provide a truer and more detailed picture of the increased intensity of rainfall. For instance, the models may show that extremely heavy rain is concentrated more in particular locations, while nearby areas get lighter rain than would be expected under the current climate. The characteristics that make up the catchment could also be considered. This will influence both the weather and rainfall-to-river flow models.

Any weather model chosen for this work needs to be capable of reproducing the weather on scales that are relevant to the catchment. For example, in order to account for the effect of the Southern Alps, the model needs a grid spacing of around 5 km to correctly calculate the uplift of the air over the steep hills. Modern weather models are capable of replicating New Zealand weather systems in sufficient detail. For example, NIWA has used the Regional Atmospheric Modelling System (RAMS) to replicate the rainfall over the country, including the Southern Alps and Bay of Plenty.

High-resolution weather models often take their starting conditions from models of global weather. The global weather models start with observations of the weather and compute the movement of air, heat, and moisture, using physical principles, in order to simulate and forecast the weather. They form the basis of all modern weather forecasting.

Weather models are run operationally in many centres to produce forecasts of rainfall. NIWA has pioneered the use of this forecast rainfall as input into rainfall-to-river flow models for New Zealand. The results of pilot studies show that useful information on forecast river flows can be produced for many areas. For example, the flows for the Bay of Plenty flood event in July 2004 were well forecast by such a system.

The high-resolution weather model should first be tested to check it can replicate historical storms under the current climate. Then the model's starting conditions can be changed to reflect future environments of the storms. For example, the temperature of the air and sea can be raised to reflect the environment expected in say 2050. This will provide a better estimate of likely changes in heavy rainfall than the simple screening analysis undertaken in the previous section.

The results from the weather model can then be fed through a rainfall-to-river flow model which delivers inundation levels for any given rainfall event. The modelling steps required to achieve this are outlined in sections 3.7 and 3.8. It should be noted that those steps, and the models required to provide quantitative information, are not specific to climate change, but are in increasingly common use around the country to model the flood risk under current climate conditions based on rainfall data.

3.6. What else could change?

Factors such as a rise in mean sea level, change in risk of storm surge, or change in river run-off caused by possible future change in land use, may heighten the risks of flooding even more. The additional factors which need to be considered depend on the specific catchment and likely future developments. It is outside the scope of this report to provide information on how to model changes in flood risk due to land-use change, or increasing development.

For coastal regions, an inundation model should include storm surge and projected sea level rises. The guidance manual, Coastal Hazards and Climate Change, recommends that in developing scenarios, councils use at least the most likely mid-range scenario for sea-level rise. It recommends staff use a figure of 0.2 m by 2050 and 0.5 m by 2100 when considering sea-level rise in projects or plans.

3.7. From rainfall to river flow

Having designed a suitable weather event, and applied the likely effect of climate change to the rainfall, this information should be run through a suitable rainfall-to-river flow model. This section explains how that is done.

The changes we see in peak flow, as a result of the modelling outlined in this section, will produce a multiplication factor which can be applied to the inundation modelling of a 2% AEP event or other 'design storm'. The inundation modelling is outlined in the following section 3.8.

Once rainfalls have been estimated for each climate change scenario, from the high-resolution weather modelling (section 3.5) or a simple screening assessment (section 3.4), they have to be turned into river flows to provide the amount and rate at which floodwaters will spill onto areas likely to be flooded. A rainfall-to-river flow model works out such things as how much of the rainfall seeps into the ground and how much rapidly runs off the land into river channel. The model can be checked ('validated') against measured flow data for existing conditions to ensure it accurately reflects reality.

Flood peaks are caused by the rainfall that does not seep into the ground and that moves quickly across the land to the river channels. The rate at which the water runs off the land depends on:

• The steepness of the land.
• The type of land surface present in the catchment, e.g., water runs off smooth surfaces, such as concrete or asphalt parking lots, much more quickly than off ploughed fields.
• How wet the ground surface is when the rainfall occurs.
• The ability of the land to store water on vegetation, in the soil and in depressions in the ground.

The water that seeps into the ground does not usually reach the river in time for it to add to a flood peak. However, it pre-conditions the ground so that a future rainfall event of the same size as one just experienced can cause much more runoff to the river channel. This is because 'antecedent' rainfall fills up the storage in the soil and if a second event occurs before the stored water has had chance to drain, the new rainfall must run off to the river channel. The response of the soil depends on:

• The type of the soil. Sandy soils drain quickly and so their 'memory' of previous events is soon lost. Clay-like soils drain more slowly and so a previous rainfall can have a marked effect on how much of a later event runs off and how much infiltrates.
• The condition of the soil. By this is meant whether the surface is compacted, as in a mob-stocked field, or open as a result of ploughing.
• The steepness of the slope on which the soil lies.
• The depth to any impervious layer beneath the soil since this will impede the downward movement of water and may force some of the seepage back towards the surface.
• The type of vegetation growing in the soil. Trees tend to have deeper rooting systems than grass and can accordingly extract more water from deeper layers of the soil and transpire it back to the atmosphere. Thus the same soil under tree-type vegetation tends to dry out more quickly following a rainfall than an equivalent grass covered area.
• The temperature. In some parts of New Zealand moisture in the surface layers can freeze during winter and then, if a rainfall event occurs, the surface becomes impervious and large amounts of the rainfall will quickly reach the river channels. This situation can be further complicated if there is snow lying on the ground, since incoming warm rainfall may melt the snow and effectively cause an event worse than the rainfall event on its own.

Thus calculation of how much rainfall reaches a river is a complex procedure and rainfall-to-river flow models seek to use available data to make due allowance for all the various factors that can affect it.

Once the water running off the ground into the river channels has been calculated, a further calculation is needed to work out how long it will take the water to find its way down the channel network. The time for the water to reach a downstream location such as the breakout point for inundation flooding depends on:

• The steepness of the various parts of the river channel network.
• The roughness of the channel bed. Smooth channels transmit water quickly, whereas rough and weed infested channels delay the flow of water.
• The density of the channel network. A catchment with many kilometres of channel per square kilometre of ground surface will respond differently to a catchment with only a few kilometres of channel per square kilometre of ground surface.
• The shape of the catchment. Roughly circular catchments will have different flow characteristics when compared to elongated catchments.
• The direction in which a storm moves relative to the catchment. In elongated catchments a storm that travels from the headwaters to the outlet will add more to a flood peak than a storm that moves up a catchment from its mouth.
• 'Catchment resonance' or the way several upstream tributaries may all deliver their peak runoff to a main channel at about the same time and so compound the flood flow arriving at a down stream breakout point.

To cope with all the above sources of variation, a rainfall-to-river flow model must be 'spatially distributed'. What this means is that the model must be able to:

• Accept rainfall that varies in both space and time.
• Calculate the surface runoff from each sub-area that contributes runoff to a river channel, taking into account things like the steepness, soil, and vegetation characteristics, and antecedent wetness of the sub-area.
• Combine the flows from these sub-areas.

A model able to accommodate the above characteristics will also be able to take into account factors that change with climate. So, for example, if as a result of climate warming more forest is converted into pasture, then this information can be put into the model by changing the vegetation type in those sub-areas where conversion occurs. Similarly, if the rainfall patterns change (e.g., become more intense in some areas and less intense elsewhere), then this can also be taken into account. Likewise, a suitable model should also be able to incorporate possible changes to river vegetation, (e.g., more willows growing in the channel will effectively roughen the channel and slow the passage of a flood).

Ideally, rainfall-to-river flow models should take account of the potential effects of climate change on conditions in the catchment, such as land use changes. In some cases, however, the effects will be small compared to the overall uncertainities involved in such a complex prediction exercise.

Councils are increasingly beginning to use such sophisticated rainfall-to-river flow models to better understand and quantify the flood risk for their communities under the current climate, based on rainfall observations in relevant catchments. These models, where data are available and models have been tested and implemented, can therefore readily be used to estimate the effects of climate change at a modest additional cost.

3.8. From river flow to flooding

As inundation modelling is computationally slow and expensive, the approach suggested here is to apply the weather and riverflow models to a range of storms but to limit the inundation modelling to just one 'design storm'.

The design storm can be chosen from past events, or built from an understanding of the riverflow statistics. For most applications, this design storm will be a 2% AEP flood.

In this approach, we take the increase in peak flow from the rainfall-to-riverflow modelling, and use it as a multiplication factor for increasing the peak flow of a design flood event under climate change. An inundation model takes this information combined with other aspects, such as tide and storm surge, to show not only where flooding would occur, but how deep and fast floodwaters could flow through the community. This section explains the factors involved in inundation modelling.

While the process for converting rainfall into river flows is reasonably well-established and computationally fast, converting river flows and sea conditions at the mouths of rivers into accurate flood inundation levels is comparatively new. It is also computationally slow because to derive an accurate estimate, calculations have to be done at a fine spatial scale. In many cases, the size of the calculation grid can be as small as 7 m by 7 m and the resultant models can use several million computational grid points. While this level of refinement may seem unduly high it is necessary if accurate results are required. Without this level of refinement and the field data on ground levels, it is not possible to ascertain if, where and when the stopbanks will be breached. Since only a small section of stopbank needs to be overtopped to cause inundation, and the location of this section is critical to the flooding, the elevations of the stopbank must be accurately measured and modelled in sufficient detail to be sure that the model correctly predicts where the failure would occur.

Note that in the inundation simulations, stopbank failure by overtopping is the only mechanism considered. The model needs to treat such overtopping failures correctly, i.e., it allows for the gradual erosion of the stopbank over time. Failure by other mechanisms such as slumping following piping of water under the stopbanks is currently beyond the ability of the present generation of inundation models.

When a stopbank fails, the water that breaks through flows in all directions. Modern inundation models are able to model the spread of this water to predict the areas that will be flooded, and the depths and speed of this flooding.

For an inundation model to produce reliable results, it must use high quality data. Several types of data are needed to run the model. These are:

• High-resolution topographic data for the areas at potential risk of flooding. These data come from LIDAR surveys, or digital airborne photogrammetry. Survey grade global position system (GPS) measurements may also be used. Information on the elevation of the near shore seabed and channel bed beneath the river water surface is also required. These data come from bathymetric survey of the seabed, and river channel measurements. The topographic data are critical to the success of inundation modelling.
• Information on the roughness of the ground. This affects how fast the water flows across the ground. Classification of the ground cover types and use of suitable look-up tables provide these data.
• Information on the expected flood flows issuing onto the flood plain. This information comes from the rainfall-to-river flow model, and is in the form of a flow hydrograph, i.e., a graph of flow against time.
• Information on the downstream conditions that block the exit of the flood to the sea. This information is a mixture of mean sea level, the state of the tide expected during the flood and storm surge, i.e., the extra height of the sea caused by the low barometric pressure over the sea effectively "sucking" water up on to the land. Both mean sea level and storm surge are expected to be affected by climate change.
• Information on special features of any particular flood plain situation, e.g., culvert and bridge dimensions and soffit levels.

Councils are increasingly making use of the tools available to map flood plains and inundation levels for the purpose of better understanding and managing flood risk for their communities for the current climate. The same tools and models can be used to model the effects of climate change.

3.9. Validation

All the model results for current conditions can be checked ('validated') against measurements. For example, the weather models can be validated against rain gauge data. However, as gauges measure rainfall only at a point and the weather models are aiming to represent rainfall over an area, it may in fact be better to validate the catchment average rainfall by comparing the rainfall-to-river flow model output with measured flow. Further, it would be useful to compare the inundation expected from the inundation modelling with what occurred during a known event. This would then give confidence in the ability of the models to reproduce reality.

3.10. What to do with the results?

The steps outlined above will allow councils to produce estimates of the inundation levels associated with future 2% and 1% AEP levels under a warmer climate. It is equally possible to estimate the change in AEP for a given inundation height, that is, how much more frequent the current 1-in-50 year flood will be in a warmer climate. This information is relevant under the Building Act, Resource Management Act, Local Government Act, and regional policies and plans.

Undertaking a climate change study will be most cost-effective when done in conjunction with a study of the current flood risk. The findings of such an assessment can be applied whenever new developments are planned, or river control and flood protection schemes are reviewed or upgraded, to ensure the same minimum level of protection can be maintained over the lifetime of the development in question. Alternatively, councils may be able to plan for later upgrades of flood defences as flood risk increases. In that case, however, it is important to ensure future development still makes such staged upgrades possible.

Producing detailed projections of likely flooding with as much accuracy as is possible will provide councils with a basis for community consultation and informed decision-making consistent with the regulatory framework, resource constraints, and community expectations.

Computer models and techniques are being refined all the time, as are climate change predictions, so councils may wish to review the findings from time to time.