View all publications

3 Methods to estimate changes in extreme rainfall

3.1 Introduction

The first step in estimating the effects of climate change on river flood flows is to estimate the change in rainfall. Simple methods for carrying out this step have been disseminated to councils since the first edition of the Climate Change Effects manual in 2004. This chapter looks at screening and advanced tools that can be used to help river managers estimate changes in extreme rainfall due to climate change.

3.2 Screening methods

Factors used for deriving extreme rainfall information

The most straightforward method for estimating extreme rainfall for preliminary screening is given in the Ministry for the Environment’s Climate Change Effects manual (Ministry for the Environment, 2008a). That manual lists (in its table 5.1, p 65) ways to estimate a variety of climatic factors at both the screening and advanced levels. For calculating both heavy rainfall and flood, the Climate Change Effects manual suggests using a factor by which rainfall is adjusted for each 1 degree Celsius of temperature change. This factor can be used in combination with HIRDS 7 (the High Intensity Rainfall Design System, see Glossary) and local rainfall data to estimate the effect of climate change on extreme rainfall statistics (see the Stoney Creek small catchment case study in chapter 6 for a worked example). Table 1 below lists these factors for a range of average recurrence intervals and a range of durations of precipitation (sourced from the Climate Change Effects manual).

These factors can also be used with extreme rainfalls from sources other than HIRDS; for example, the map of 24-hour design rainfalls mentioned in Stormwater Rainfall-Runoff Model TP108 in chapter 4, or any other suitable source of design rainfall information.

The Climate Change Effects manual also notes that if a screening assessment using the mid-range scenario, based on table 1, does not reveal any significant impacts, best practice would indicate the need to re-run the screening using a scenario from the upper bound of possible future climate change.

Table 1: Factor of percentage adjustment per 1°C to apply to extreme rainfall, for use in deriving extreme rainfall information for screening assessments
  ARI (years)
Duration 2 5 10 20 30 50 100
< 10 minutes 8.0 8.0 8.0 8.0 8.0 8.0 8.0
10 minutes 8.0 8.0 8.0 8.0 8.0 8.0 8.0
30 minutes 7.2 7.4 7.6 7.8 8.0 8.0 8.0
1 hour 6.7 7.1 7.4 7.7 8.0 8.0 8.0
2 hours 6.2 6.7 7.2 7.6 8.0 8.0 8.0
3 hours 5.9 6.5 7.0 7.5 8.0 8.0 8.0
6 hours 5.3 6.1 6.8 7.4 8.0 8.0 8.0
12 hours 4.8 5.8 6.5 7.3 8.0 8.0 8.0
24 hours 4.3 5.4 6.3 7.2 8.0 8.0 8.0
48 hours 3.8 5.0 6.1 7.1 7.8 8.0 8.0
72 hours 3.5 4.8 5.9 7.0 7.7 8.0 8.0

Note: This table recommends percentage adjustments to apply to extreme rainfall per 1 degree Celsius of warming, for a range of average recurrence intervals (ARIs.). The percentage changes are mid-range estimates per 1 degree and should be used only in a screening assessment. The entries in this table for a duration of 24 hours are based on results from a regional climate model using the medium-high (A2) emissions scenario. The entries for 10-minute duration are based on the theoretical increase in the amount of water held in the atmosphere for a 1 degree increase in temperature (8 per cent). Entries for other durations are based on logarithmic (in time) interpolation between the 10-minute and 24-hour rates.

Note that the largest percentage increase in table 1 is 8 per cent per 1 degree Celsius of local warming, and this factor applies to the longest ARIs and the shortest durations. This upper limit of 8 per cent has theoretical support in that it is the rate of increase in the moisture-holding capacity of air as temperature increases. Studies have found that, at least in the extra-tropics and for a regional average, the 8 per cent increase agrees well with global and regional model estimates (Pall et al, 2007). However, the possibility of precipitation extremes increasing faster than this 8 per cent cannot be ruled out (Lenderink and van Meijgaard, 2008), particularly for short duration falls of one hour or less. Preliminary NIWA regional modelling results also show ‘hot spots’ where rainfall extremes are larger than 8 per cent per 1 degree of warming, 8 although further work is required to establish whether this is due to changes in the intensity or the frequency of storms. Nevertheless, our recommendation at this time is to apply the factors of table 1 for all locations. Further research is needed before we can have any confidence in the location of such hot spots.

3.2.2 Influence of temperature on extreme rainfall

Table 1 provides percentage adjustments to apply to extreme rainfall per degree Celsius of warming. The recommended temperature increase to use is the annual average increase for each region, shown below in tables 2 and 3 (taken from tables 2.2 and 2.3 of the Climate Change Effects manual). Tables 2 and 3 give the projected annual temperature increases at 2040 and 2090, separately for each regional council area of New Zealand, for the six IPCC illustrative marker scenarios. The table columns show the 12-model averages, along with the lowest and highest temperature changes across the 12 models, as a function of the emission scenario (note that A1T and B2 give the same result and so are shown as a single column). As an example, for the Auckland region in 2090, the annual temperature is projected to increase by an average of 1.4 degrees for the lowest B1 emission scenario (with a range between 0.6 and 2.6 degrees), and by an average of 3.0 degrees for the highest A1FI emission scenario (with a range between 1.3 and 5.8 degrees).

Table 2: Projected changes in annual mean temperature (in °C), 1990 to 2040, by regional council area
  IPCC scenario
Regional council area B1 B2/A1T A1B A2 A1FI
Northland 0.6 [0.2, 1.2] 0.8 [0.3, 1.5] 0.9 [0.3, 1.8] 1.1 [0.4, 2.2] 1.3 [0.4, 2.6]
Auckland 0.6 [0.2, 1.1] 0.8 [0.3, 1.5] 0.9 [0.3, 1.7] 1.1 [0.4, 2.1] 1.3 [0.5, 2.5]
Waikato 0.6 [0.2, 1.1] 0.8 [0.3, 1.4] 0.9 [0.3, 1.6] 1.1 [0.4, 2.0] 1.3 [0.5, 2.4]
Bay of Plenty 0.6 [0.2, 1.1] 0.8 [0.3, 1.4] 0.9 [0.4, 1.6] 1.1 [0.4, 2.0] 1.3 [0.5, 2.4]
Taranaki 0.6 [0.2, 1.0] 0.8 [0.3, 1.4] 0.9 [0.4, 1.6] 1.1 [0.4, 2.0] 1.3 [0.5, 2.3]
Manawatu–Wanganui 0.6 [0.2, 1.1] 0.8 [0.3, 1.4] 0.9 [0.3, 1.7] 1.1 [0.4, 2.0] 1.3 [0.4, 2.4]
Hawke’s Bay 0.6 [0.2, 1.1] 0.8 [0.3, 1.4] 0.9 [0.3, 1.6] 1.1 [0.4, 2.0] 1.3 [0.5, 2.3]
Gisborne 0.6 [0.2, 1.0] 0.8 [0.3, 1.3] 0.9 [0.4, 1.5] 1.1 [0.4, 1.9] 1.3 [0.5, 2.2]
Wellington 0.6 [0.3, 1.0] 0.8 [0.3, 1.3] 0.9 [0.4, 1.5] 1.1 [0.5, 1.9] 1.3 [0.6, 2.2]
Tasman–Nelson 0.6 [0.2, 0.9] 0.8 [0.3, 1.2] 0.9 [0.4, 1.4] 1.1 [0.5, 1.8] 1.3 [0.5, 2.1]
Marlborough 0.6 [0.2, 0.9] 0.8 [0.3, 1.2] 0.9 [0.4, 1.4] 1.1 [0.4, 1.7] 1.3 [0.5, 2.0]
West Coast 0.6 [0.2, 0.8] 0.8 [0.2, 1.1] 0.9 [0.3, 1.3] 1.1 [0.3, 1.5] 1.3 [0.4, 1.8]
Canterbury 0.6 [0.2, 0.8] 0.8 [0.3, 1.1] 0.9 [0.3, 1.3] 1.1 [0.4, 1.6] 1.3 [0.5, 1.9]
Otago 0.6 [0.1, 0.9] 0.7 [0.1, 1.1] 0.9 [0.2, 1.3] 1.1 [0.2, 1.6] 1.3 [0.2, 1.9]
Southland 0.6 [0.1, 0.9] 0.7 [0.1, 1.1] 0.8 [0.1, 1.3] 1.1 [0.1, 1.6] 1.2 [0.1, 1.9]
Chatham Islands 0.6 [0.2, 1.2] 0.8 [0.3, 1.5] 0.9 [0.3, 1.8] 1.1 [0.4, 2.2] 1.3 [0.4, 2.6]

Notes: The average change, and the lower and upper limits (in brackets), are provided separately for the IPCC’s six illustrative marker scenarios. The B2 and A1T scenarios produce the same warming, and so are shown in a single column.

Table 3: Projected changes in annual mean temperature (in °C), 1990 to 2090, by regional council area
  IPCC scenario
Regional council area B1 B2/A1T A1B A2 A1FI
Northland 1.3 [0.6, 2.7] 1.7 [0.7, 3.5] 2.1 [0.9, 4.1] 2.5 [1.1, 5.0] 3.0 [1.3, 5.9]
Auckland 1.4 [0.6, 2.6] 1.8 [0.7, 3.4] 2.1 [0.9, 4.0] 2.5 [1.1, 4.9] 3.0 [1.3, 5.8]
Waikato 1.4 [0.6, 2.5] 1.8 [0.7, 3.3] 2.1 [0.9, 3.8] 2.5 [1.0, 4.7] 3.0 [1.3, 5.5]
Bay of Plenty 1.4 [0.6, 2.5] 1.8 [0.8, 3.3] 2.1 [0.9, 3.8] 2.5 [1.1, 4.7] 3.0 [1.3, 5.6]
Taranaki 1.4 [0.6, 2.4] 1.8 [0.7, 3.2] 2.1 [0.9, 3.7] 2.5 [1.1, 4.5] 3.0 [1.3, 5.3]
Manawatu–Wanganui 1.3 [0.6, 2.5] 1.7 [0.7, 3.3] 2.1 [0.9, 3.8] 2.5 [1.0, 4.7] 3.0 [1.2, 5.5]
Hawke’s Bay 1.3 [0.6, 2.4] 1.7 [0.7, 3.2] 2.1 [0.9, 3.7] 2.5 [1.0, 4.5] 3.0 [1.2, 5.4]
Gisborne 1.4 [0.6, 2.4] 1.8 [0.8, 3.2] 2.1 [0.9, 3.6] 2.5 [1.1, 4.5] 3.0 [1.3, 5.3]
Wellington 1.3 [0.6, 2.3] 1.7 [0.8, 3.1] 2.1 [0.9, 3.6] 2.5 [1.1, 4.4] 3.0 [1.3, 5.2]
Tasman–Nelson 1.3 [0.6, 2.3] 1.7 [0.8, 3.0] 2.0 [0.9, 3.5] 2.5 [1.1, 4.3] 2.9 [1.3, 5.1]
Marlborough 1.3 [0.6, 2.3] 1.7 [0.8, 3.0] 2.0 [0.9, 3.5] 2.5 [1.1, 4.3] 2.9 [1.3, 5.0]
West Coast 1.3 [0.7, 2.2] 1.7 [0.8, 2.9] 2.0 [1.0, 3.4] 2.4 [1.2, 4.1] 2.9 [1.4, 4.9]
Canterbury 1.3 [0.7, 2.2] 1.7 [0.9, 2.9] 2.0 [1.1, 3.4] 2.5 [1.3, 4.2] 2.9 [1.6, 5.0]
Otago 1.3 [0.8, 2.1] 1.7 [1.0, 2.8] 2.0 [1.2, 3.2] 2.4 [1.4, 3.9] 2.8 [1.7, 4.6]
Southland 1.3 [0.8, 2.0] 1.6 [1.0, 2.7] 1.9 [1.2, 3.1] 2.3 [1.4, 3.8] 2.8 [1.7, 4.5]
Chatham Islands 1.3 [0.6, 2.7] 1.7 [0.7, 3.5] 2.1 [0.9, 4.1] 2.5 [1.1, 5.0] 3.0 [1.3, 5.9]

Notes: The average change, and the lower and upper limits (in brackets), are provided separately for the IPCC’s six illustrative marker scenarios. The B2 and A1T scenarios produce the same warming and so are shown in a single column.

There is not much variation across New Zealand in the rate of warming, unlike the strong gradients seen in the precipitation changes. Figure 7 shows the information from table 3 averaged over all of New Zealand. It is apparent from this that three of the climate models are outliers: one is colder and two are warmer than the cluster of the remaining nine models. For purposes of consistency, if the rainfall pattern from a particular model is being used in a scenario risk assessment, then it is desirable to use the appropriate temperature change from that same model.

Figure 7: Change in New Zealand average annual temperature to 2090 (°C), by IPCC emissions scenario

Figure 7: Change in New Zealand average annual temperature to 2090 (°C), by IPCC emissions scenario

Note: The vertical coloured bars highlight the range over the 12 climate models considered, with stars indicating individual models and circles to indicate outliers.

Figure 7: Change in New Zealand average annual temperature to 2090 (°C), by IPCC emissions scenario
The figure shows the change in New Zealand average annual temperature to 2090 (°C), for 6 IPCC emissions scenarios, for each of the 12 global climate models. These results show that the range of temperatures increase from around 1.2 deg C for the B1 scenario to around 3 deg C for the A1FI scenario. The individual models show different values, ranging from the B1 scenario having values ranging from 0.6 to 2.4 deg C, and the A1FI scenario having values from 1.5 to 5.0 deg C. Two sets of model results are highlighted. One shows a model that is consistently lower than other results, by around 0.4 Deg C in the B1 scenario, to 0.7 deg C in the A1FI scenario. The other one shows a model that is consistently higher than other results, by around 0.6 Deg C in the B1 scenario, to 1.3 deg C in the A1FI scenario.

3.3 Advanced methods

In order to estimate changes in flood flows, you need to start with rainfall time-series that have been adjusted to take account of climate change influences. A wide range of approaches are possible, from statistical and empirical adjustments to observed rainfall data, through to numerical model simulations of future climate.

3.3.1 Weather generators

A weather generator is a stochastic model 9 for simulating a daily time-series of linked climatic elements – commonly rainfall, maximum and minimum temperature, solar radiation and wind run (Wilks, 1992; Thompson and Mullan, 2001). The weather generator is first tuned to current site data, and then adjustments for future climate change can be made to the climatological means and standard deviations, and to other parameters such as the frequency of a wet day. This approach has been used widely in agricultural modelling, where simultaneous variations in a number of climate elements in addition to rainfall are required.

3.3.2 Empirical adjustment of daily rainfall data

Most of the methods that estimate changes in flood magnitude due to climate change require rainfall information at a daily or finer time resolution. The rainfall changes previously developed in the Climate Change Effects manual (section 2.2.2) provide information at a monthly timescale and are not suitable on their own for assessing changes in flood risk. A method for empirically adjusting a daily rainfall time-series is described here, which uses the scenarios of mean rainfall change and also adjusts the distribution to increase the most extreme daily amounts.

Step 1 (below) adjusts the daily data using monthly rainfall offsets. The distributional adjustment in steps 2 to 4 have the effect of decreasing the number of days per year when rain falls and allocating more precipitation into the upper tail of the rainfall distribution. The formula in step 2 is based on analysis of extreme rainfalls at a few grid points in the Wellington region for one of the regional climate model runs (see figure 8). The formula changes the frequency of rain days – reducing the number of rain days by decreasing the probability of a rain day by 1.75 per cent per degree Celsius increase in annual-average temperature. Further work is required to clarify how appropriate this formula is across all of New Zealand.

Step 1
Adjust the daily data using the monthly rainfall offsets (Climate Change Effects manual, table 2.4 for 2040, table 2.5 for 2090). The change in monthly climatological rainfall is then calculated (eg, a 10 per cent increase in a monthly climatology of 100 millimetres means an extra 10 millimetres). The monthly climatological rainfall is estimated by averaging over many years so that the resulting monthly totals represent the current state of the climate. The monthly change in rainfall so obtained (eg, 10 millimetres) is then expressed as a percentage change for the current month. For example, if the current monthly total is 120 millimetres, then the percentage change is 10 millimetres divided by 120 millimetres, which equals 8.3 per cent. This percentage change is then applied to each rain day in the month. This step does not change the number of rain days in the record or alter the inter-annual variance in monthly rainfalls.

Step 2
Allow for the changes in frequency of rain days. This reduction in low-rainfall days helps to balance the increased rainfall extremes in step 3. Thus, if:

  • NW = number of rain days
  • NT = total number of days in a year (ie, 365.25)
  • ΔT = warming

then the number of rain days will change from NW to (rounded down to the nearest integer) NW – 0.0175 * ΔT * NT.

This corresponds to about seven fewer rain days per year per degree of warming. The reduction is applied to days with the lowest rainfall by ranking all rain days in an ascending order and setting the calculated number (0.0175 * ΔT * NT) of lowest rainfall days to zero rainfall.

Step 3
After applying steps 1 and 2, calculate the rainfall percentiles P from the adjusted daily data (all months and years combined). Note that the percentiles are calculated over rain days only (ie, ignoring dry days). The percentile values are then changed according to the formula:

Change in daily rainfall (in % per °C) = 6.15 * [1. – ln (100–P)/2.3 ].

This formula gives zero change at percentile P = 90, + 8 per cent per degree Celsius change at P = 99.5, and about –6 per cent per degree Celsius change at P = 0. For P > 99.5, the change is capped at +8 per cent per degree of local warming (taken as the change in annual-average temperature). This 8 per cent per degree value is widely recognised as the rate at which the water vapour saturation level increases in the atmosphere (the Clausius–Clapeyron relationship), and is the upper limit recommended in the Climate Change Effects manual for adjusting return periods of extreme rainfall. Apply these percentage changes in rainfall to the results of step 2.

Step 4
Recalculate the total rainfall over the whole period (all months and years included) after step 3, and check to see the total rainfall is still consistent with the total in step 1, after the prescribed scenario changes are applied. If it is not consistent, then adjust all daily rainfalls by the factor required for consistency (eg, if the total rainfall is 130 millimetres after step 1, but the total rainfall is 137.2 millimetres after step 3, then multiply all rain days by the factor 130 millimetres divided by 137.2 millimetres). If this adjustment leads to some daily rainfalls dropping below 0.1 millimetres (the threshold for a measurable rainfall), then reset rainfall on these days to the minimum, 0.1 millimetres, and adjust all daily rain days, with rainfall greater or equal to 1.0 millimetres, by multiplying them with a consistency factor (less than 1).

Figure 8: Percentage change in rainfall amount as a function of the percentile in the distribution of daily rainfall: NIWA regional climate model data (averaged over several grid squares in the Wellington region) with local warming of 2.5°C (blue line) and idealised rainfall distributional-adjustment model (red line)

Figure 8: Percentage change in rainfall amount as a function of the percentile in the distribution of daily rainfall: NIWA regional climate model data (averaged over several grid squares in the Wellington region) with local warming of 2.5°C (blue line) and idealised rainfall distributional-adjustment model (red line)

Notes: The horizontal co-ordinate is minus the natural logarithm of (100-P), where P is the percentile in the distribution of daily rainfall amounts. This coordinate scale accentuates the high-end rainfalls. Note that the model rainfall changes are approximately linear over a wide range in this co-ordinate space.

Figure 8: Percentage change in rainfall amount as a function of the percentile in the distribution of daily rainfall: NIWA regional climate model data (averaged over several grid squares in the Wellington region) with local warming of 2.5°C (blue line) and idealised rainfall distributional-adjustment model (red line)
The figure shows the percentage change in rainfall amount as a function of the percentile in the distribution of daily rainfall. The NIWA regional climate model data (averaged over several grid squares in the Wellington region) with local warming of 2.5°C is shown in blue. An idealised rainfall distributional-adjustment model is shown in red line. The plots show a percentage change in rainfall for frequent events (-4 on the -log(100-P) scale) of around 10%, to an increase in rainfall for rare events (+0.7 on the -log(100-P) scale) of around 20%. Above this value the percentage change is depicted to be constant in the idealised plot.

 

We can now illustrate this method in table 4. Note that our example has just 30 days of data, of which many have zero rainfall, whereas in practice we would have several decades of data. In step 1 below, we apply a hypothetical 10 per cent increase in monthly climatological rainfall of about 100 millimetres to the current monthly total of 120 millimetres, so the total rain for the month is now 130.0 millimetres (ie, 0.10 * 100 millimetres + 120 millimetres). In step 2, we apply the 1.75 per cent formula to find that our 30 days should have 9.48 (rounded down to 9) rain days instead of 10; that is, we need one fewer rain day. So we rank the rain days in ascending order and find the smallest daily rainfall (the shaded 0.5 millimetres on day 12) and set it to zero. The total rain for the month is now 129.5 millimetres.

In step 3, we compute a percentile for each non-zero daily rainfall (after the rain day reduction in step 2), then compute the percentage change per degree for rainfall extremes, assume 1.0 degree Celsius annual average warming, and apply each percentage change to the corresponding day’s rainfall. The total rain for the month is now 132.7 millimetres. In step 4, we scale the rain for the total period (in this case only one month) by 130.0 millimetres divided by 132.7 millimetres, to obtain a total rain of 130.0 mm, consistent with the total rain after applying the monthly climate change scenario in step 1. None of the non-zero rainfalls have been reduced below 0.1 millimetres (the threshold for measurable rainfall) by this scaling, so the process stops.

Table 4: Applying the empirical adjustments of rain
Day of month 1 2 3 4 5 6 7 8 9 10 11 12 13 30 Total NW*
Climatological monthly total (average over many years) 100.0  
Rain (mm) 60.0 0 14.0 13.0 12.0 0 7.0 6.0 4.0 2.0 1.5 0.5 0 0 120.0 10
Step 1 Adjust by 10% for climate change impacts    
Rain (mm) 66.0 0 15.2 14.1 13.0 0 7.6 6.5 4.3 2.2 1.6 0.5 0 0 130.0 10
Step 2 Reduce to account for changes in frequency    
Ranking in ascending order 10   9 8 7   6 4 3 2 1         9.48**
Rain (mm) 66.0 0 15.2 14.1 13.0 0 7.6 6.5 4.3 2.2 1.6 0*** 0 0 129.5 9
Step 3 Calculate percentiles and adjust to give changes in intensity    
Ranking in ascending order 9   8 7 6   5 4 3 2 1          
Percentiles for non-zero rain 100   87.5 75.0 62.5   50.0 37.5 25.0 12.5 0.0          
Per cent change per degree 8.0   -0.6 -2.5 -3.5   -4.3 -4.9 -5.4 -5.8 -6.2          
Rain (mm) 70.2 0 15.1 13.7 12.5 0 7.3 6.2 4.1 2.0 1.5 0 0 0 132.7 9
Step 4 Apply distributional changes    
Rain (mm) 68.8 0 14.8 13.5 12.3 0 7.1 6.1 4.0 2.0 1.5 0 0 0 130.0 9

* NW indicates the number of rain days.
** This is the target number of rain days when using the 1.75 per cent reduction.
*** The shaded 0 mm value was set to zero to reduce the number of rain days.

3.3.3 Analogue selection from observed data

Whereas the previous two methods are limited to adjusting daily rainfalls, analogue selection can be applied to the rainfall data of any temporal resolution. The approach here is to select a subset of past rainfall data that has specific characteristics anticipated in a future climate; for example, choosing rainfall data from the hottest years, or periods with a certain mix of circulation types, such as a negative IPO. Analogue selection has been applied to future New Zealand rainfall (Sansom and Renwick, 2007) and to hydrological studies (Yates et al, 2003).

3.3.4 Downscaling global model results

Estimates of changes in rainfall can be made by downscaling the results from global climate models (GCMs). GCMs do not have the resolution to simulate very intense convection, but one approach that could be used in New Zealand is to apply adjustments to observed rainfall probability distributions, guided by distributional changes predicted by the GCMs (Semenov and Bengtsson, 2002). The statistical distribution of daily rainfall can be fitted to a ‘gamma distribution’, and GCMs analysed to evaluate how the shape and scale factors of this distribution change under a warming scenario. Similar parameter changes can then be applied to the rainfall distribution at a site. This approach is conceptually similar to that discussed in section 3.3.2 and is not discussed further here. (Refer to Appendix A3.3 in the Climate Change Effects manual for possible applications of this method.)

3.3.5 Mesoscale weather model

A mesoscale weather model 10 (eg, the Regional Atmospheric Modelling System, Cotton et al, 2001) can be used to simulate the effects of climate change on rainfall. Such models can be used for case studies or weather forecasting and have detailed representations of the physical processes that influence precipitation. In a case study the model is used twice: once to simulate an event that could or has occurred under the current climate, and a second time with an increased air temperature consistent with the warming from a climate change scenario. The warmer air is able to hold more moisture and so more precipitation is possible. A model of this kind provides detailed information on the location, structure and timing of precipitation across the catchment or region of interest, rather than assuming a uniform percentage increase in rainfall. An example of this type of weather modelling is included as part of the Westport case study in chapter 6.

3.3.6 Regional climate model

NIWA has developed a capability for modelling global and regional climate based on the UK Met Office’s unified model framework (Drost et al, 2007). A global climate model (GCM) developed at the UK Met Office, known as HadCM3, has been used to generate boundary conditions in the New Zealand region and hence to run a regional climate model (RCM). The RCM simulates all the atmospheric processes that are important in the creation of heavy rainfall events and can allow these processes to change under global warming.

The RCM does not itself simulate river flows, but the impact of climate change on flood flows can be quantified for major catchments throughout New Zealand by coupling bias-corrected climate information from the RCM into a catchment hydrological model (eg, TopNet, described further in section 4.3). The RCM can provide all the climate inputs required by these models, such as surface temperature and rainfall, at hourly resolution, for both the current and future climate. The RCM-TopNet simulations can potentially provide scenarios of flood risk (changes in flood frequency and magnitude, and seasonality) for all major river catchments of New Zealand.

Simulations of the RCM suitable for this purpose already exist. These include a current climate experiment for 1970 to 2000, which makes use of observed sea surface temperatures, and two future climate experiments for 2070–2100. These future experiments use two very different greenhouse gas emission scenarios, known as B2 and A2 (low and moderately high) to span the uncertainty in possible future emissions. The RCM can also be forced with historical observed data, for which two reconstructions are available. This data comes from weather models that incorporate as many observations as possible, and the RCM then interpolates from the regional circulation to the local climate. This allows for the direct comparison of RCM output with observations of individual flood-causing weather systems, without the need to rely on long-term averages.

The quantification of uncertainty in the model outputs is a key step in making the research results relevant to users who are involved in assessing insurance risks, making investments in flood protection works and planning land use. Simulations of the RCM-TopNet model driven by observed circulation reconstructions can be compared with rainfall and river flow observations to allow detailed bias corrections to be established for use in future climate runs. The changes observed in these future runs could then be used to estimate changes in flood risk and the uncertainty in those projections. Although the RCM is at a high resolution of 30 kilometres with regard to climate studies, it has a relatively low resolution when compared to individual river catchments. However, TopNet is capable of adjusting for this problem.

The main weakness of the RCM-TopNet methodology lies in the reliance on only one – or possibly two – GCMs when making future projections and so it is unable to capture any uncertainty due to the use of different models (unlike the use of statistical downscaling, see section 3.3.4). A significant advantage is the capturing of realistic changes in regional rainfall extremes with a resolution of one hour, although in some areas of the country the changes may be significantly greater than those given by the screening method given in section 3.2.

3.4 Summary

  • A number of more advanced methods for estimating extreme rainfall have been discussed, including weather generators, empirical adjustments, analogue selection from observed data, downscaling of global models, mesoscale weather models and regional climate models.
  • These methods provide estimates of how climate change may affect extreme rainfall. Each method, in its own way, converts forecasts of climate change into time-series of rainfall. They differ in their complexity, data requirements, and reliability of results.
  • More complex methods require greater expertise to carry them out, so it is recommended that you consider three factors when deciding how to develop extreme rainfall forecasts:
    • what climate data is available?
    • what accuracy and precision do we need?
    • do we have access to the expertise and computing facilities to undertake the analysis and modelling?
  • Table 5 contains a summary of the advanced methods for estimating rainfall, and will help you to choose the most appropriate method.

Table 6:Summary of advanced methods for estimating flood flows as a guide to selecting the appropriate method

Table 5: Summary of advanced methods for estimating rainfall as a guide to selecting the most appropriate method
Method Description Advantages Disadvantages Data and climate change requirements
Weather generators (WGs) Statistical and empirical models of local weather features Easy to make multiple simulations or generate very long time-series of daily weather sequences.

Match daily variability well.

Can handle multiple weather parameters (typically rainfall, max/min temperature, solar radiation and wind run) with realistic cross-correlations.
Accuracy of simulations depends on realism of assumed underlying distributions.

Have insufficient low-frequency (year-to-year) variability, a problem known as overdispersion.

Not always straightforward to adjust parameters for future climate.
Observed station data to fit WG parameters to current climate.

Future simulation time-series, or some other method to generate WG parameter changes, for future climate.
Empirical adjustments Adjusts historical rainfall records with monthly climate change projections Very easy to apply, and provide daily output.

Can adjust rainfall distribution (eg, greater extremes) as well as mean rainfall changes.
Not physically based (although adjustments may be guided by physics models such as a regional climate model). Observed station time-series.

Future scenarios of monthly rainfall changes (and annual temperature change if making adjustments to the distribution).
Analogue selection from observed data Mimics climate change by comparing with selected historical events Straightforward in principle.

Selecting a number of analogues (eg, best N) will give a distribution of resulting climates.
The various ways to select analogues can lead to different answers.

Analogues selected independently for different sites may not be consistent with each other.
Observed data for analogue selection.

General climate change trends.
Downscaling of global models Converts GCM results to local results based on local statistics Relatively straightforward to apply to many GCMs once downscaling approach has been decided. Output is monthly.

Statistically based approach will misrepresent some physical realities.

Climate trends may move outside the range of observations.
Current observed data to generate downscaling relationships.

Global model data (usually monthly) for future scenarios.
Mesoscale weather models (RAMS) Used to simulate the effects of climate change on rainfall for a small region Physically based.

Output data obtained at high spatial and temporal resolution.
Computer intensive.

Usually run for short-period case studies only.
Three-dimensional weather data fields for initialisation of model.
Regional climate models (RCMs) Used to simulate more detailed climate change at the regional scale Based on physical equations of the climate system.

Output data obtained at high time and space resolution for virtually any weather and climate variable of interest.
Very computationally intensive; produces large amounts of data.

Requires running a global climate model to generate boundary conditions around the RCM domain (atmosphere + sea surface).
Three-dimensional weather and climate data fields for the entire simulation period of the model.

Note: Data requirements specifically related to climate change are underlined.


7 HIRDS is currently being updated (March 2010), with the addition of a new feature to enable the rainfall adjustment factor to be included automatically.

8 There is a lot of noise in a 30-year simulation, and this affects estimates of extremes. Another model run starting with slightly different initial conditions but forced by the same emission scenario and with similar warming, will show different ‘hot spots’.

9 A stochastic model is a tool for estimating probability distributions of potential outcomes by allowing for random variation in one or more inputs over time.

10 A numerical weather prediction model designed to simulate phenomena with horizontal scales between a few kilometres and several hundred kilometres.