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

Changes in drought risk with climate change

Executive Summary

1. Introduction

2. Drought Risk under Current Climate

3. Drought Risk under Scenarios of Climate Change

4. Other Issues

5. References

6. Technical Appendix

6. Technical Appendix

6.1 Drought Indicator Issues: Notes from Expert Workshop

6.1.1 Drought risk workshop

The key objective of the workshop was to propose a quantitative indicator that could be used to assess potential changes in drought incidence and severity with climate change. This required a consultative process to draw from previous New Zealand experience using drought assessment indicators, and their potential suitability for this objective.

To this end NIWA and the Ministry for the Environment hosted a drought risk workshop at Greta Point, on 10 May 2004. The workshop brought together scientists and consultants from NIWA, Ministry for the Environment, AgResearch, Crop and Food, Earthwise Consulting, Fonterra, MAF Policy, and Wrightson Consulting.

The workshop reviewed aspects of how past droughts in New Zealand had been measured and assessed, and identified many issues which needed to be considered in arriving at a 'most suitable' drought index. These issues are noted in the following section.

6.1.2 Identification of key issues

The main purpose of the drought index was to compare the risk now with the risk in say 50 years' time, eg. 'There will be 30% more droughts then than now'. The index might describe both relative and/or absolute change. Participants noted that it was also likely to have implications for future water use allocation, for example rural versus urban water quotas, and water abstraction. It might help with quantifying flow-on economic impacts from potential changes in agricultural production. The drought index might also indicate further research needs, particularly in issues like ground water abstraction, irrigation scheduling, and social costs.

In Canterbury at present, water allocations are made on the basis of rainfall volumes, as not enough is known about flow-through volumes in rivers and groundwater. It is possible that water in Canterbury is already over-allocated.

The workshop recognised that drought usually has the largest impacts of any national or international calamity on the economy of New Zealand. For example, the drought of 1997-98 was bigger economically for New Zealand than the 'Asian crisis' (although currency fluctuations also had an impact). Heavy reliance on rain-fed agriculture meant that just 21 days without rain was often enough to trigger drought like conditions.

Concern was expressed that any drought index identified for the current purpose should not be used to 're-litigate' recent drought conditions, or be applied to such purposes as redistribution of current water rights.

Although not relevant to the present aim to derive a drought risk indicator, it is relevant to quantifying the economic impacts of climate change to note that droughts are always disadvantageous for some, but are often advantageous for others. Hence economic measurements of drought must recognise that impacts are not contained within affected regions; there may be positive impacts elsewhere. For example, stock moved from a dry region may benefit farmers in higher rainfall areas; trucking companies who move the stock gain extra income.

Over the past few years farmers have adapted well to short (eg. 3-month) droughts, but prolonged droughts have more profound impacts. Of particular concern is when severe droughts lead to disposal of capital stock or specialist breeding stock. It is important to establish benchmarks or trigger points for making critical decisions. It may also be important to have an indication of recovery time - eg. the amount of rainfall required to bring the drought to an end.

A drought index is no use in isolation - it must be capable of measuring something else, for example, production. Different indices may be needed for example for different cropping systems, or to adequately represent hydrological conditions. The timing of drought is therefore important. A late season drought, after harvest, may not be important to cropping farmers. Much more data are available now, and holistic interpretations of drought need to be considered. On the other hand, pastoral, rain-fed agriculture will be dominant for a long time in terms of the national economy. However it may be important to distinguish between dry land and irrigated agriculture. It may be possible to access a water consent database to obtain location, volume, and land use. This would also provide information in water restrictions. A further possibility would be to attach land use properties to indicator climate stations (water use, farm type, catchment area).

Drought risk changes are not just meteorological - risk can also change because of socio-economic and technological changes. An ideal index would be derived from a biophysical model, but with strong linkages to social and economic outcomes.

Extreme droughts can trigger species and ecosystem changes, for example, a change from C3 to C4 grass species.

There are some advantages of having an index that is independent of land use. In this case a range of 'what-if' scenarios could be used to account indirectly for land use. Some examples would be varying soil types and root depths, key return period thresholds, and proportion of irrigated versus non-irrigated land. An extension of an irrigation usage scenario could be various water abstraction rates. It is important also to account for situations where the source of water can be different to its place of use. For example, one estimate suggests 70% of Canterbury's water resource is from rainfall in the Southern Alps.

A drought index should clearly show the impact on pasture growth [It was noted that AgResearch with Crop and Food have developed a pasture growth model, which could be applied to drought scenario studies.]. However pasture growth is heavily dependent on solar radiation, and changes in radiation (and cloudiness) with climate change are highly speculative at this stage. Climate change scenarios in this study are based on changes in rainfall, air temperature, and circulation regimes. However, estimates of variability in radiation and windiness from historical data, particularly their link to changes in observed drought incidence, would be useful to test the sensitivity of future drought risk to similar variability within climate change scenarios. The level of uncertainty associated with changes in radiation (and also with the degree of change in windiness) is a component of the uncertainty in changes in drought risk resulting from climate change.

A starting point could be a model of unrestricted (by lack of moisture) pasture growth, and then modify the outcome with increasing deficit. A refinement would be to model species change that may occur as a result of climate change.

The workshop recognized the potential importance of information on changes in drought risk within boundaries of district and regional councils, to improve information for local authority decision-making.

Finally, from a science perspective, it is easy to define an index. It is however harder to ensure that it is the most suitable index for the intended purpose. The limitations of a proposed index should be clear, as also possible ways it might be improved.

6.1.3 Desirable properties of a drought index

• Universality - the index should be applicable to all parts of the country, and be suitable for both nation-wide and regional analyses. The index should also be sufficiently versatile to cope with varying thresholds or scales of severity.
• Easily interpreted
• Supported by readily available data to enable calculation of robust anomaly and recurrence statistics
• Should enable improved advice to farmers for land use planning
• Be an indication of production loss.
• Be suitable for subsequent research needs - eg. land use, social implications, water policy.
• Be based on parameters for which predictions of future change can be plausibly developed, given current knowledge.
• Be able to be linked to decision trigger points (eg. the kinds of thresholds that farmers might use to implement drought mitigation actions).
• Represent both the duration and intensity of droughts, as both are important.

6.2 Literature Review of Drought Indicators

This section outlines examples of drought indicators from the New Zealand and international experience. These examples provide a broad context against which the relationship between climate parameters and drought risk can be established. Of key interest are the features of drought indicators that most clearly define differences between drought events, particularly where they are relevant to the requirements of this study.

Drought indicators are inherently complex due to the multiple causes, processes and impacts of drought. The difficulty of obtaining relevant data series, and, more importantly, of modelling the interactions between natural processes and human responses, typically leads to drought assessments being oversimplified. In this section, some examples of drought indicators from New Zealand, Australia and the United States are given, as a background to the selection of an appropriate drought index for New Zealand.

6.2.1 New Zealand

In New Zealand, up to about the late 1980s, drought relief consideration for farmers was triggered when rainfall at representative climate stations in a drought-affected area, for a consecutive three-month period, was at a one in 20 year low. This measure was often further qualified by the additional condition that total days of soil moisture deficit, based on a daily water balance calculation, were correspondingly high.

Economic consequences of drought have been reported in various ways, including:

• Loss of gross farm income (individual and regional);
• Loss of production (various categories);
• Changes in expenditure patterns including wages;
• Loss of value added including feed stocks;
• Run down in savings;
• Changes in stock numbers;
• GDP losses. For example, the New Zealand Institute of Economic Research estimated that the 1997/98 El Niño associated drought resulted in a loss of $407 million (0.4%) of GDP (Gardiner 2001).
• Environmental consequences.

6.2.2 Australia

Following the severe eastern Australian drought of 1994/95, the Australian Government adopted a policy of Drought Exceptional Circumstances for intervention (Laughlin and Clark, 2000), based on the assessment of six criteria:

1. Meteorological conditions;
2. Agronomic and stock conditions;
3. Water supplies;
4. Environmental impacts;
5. Farm income levels;
6. Spatial scale of the event.

The Criteria for Exceptional Circumstances are that:

1. The event must be both rare and severe. A rare event is one that occurs once in every 20-25 years. A rare event is severe if it is of a significant scale - measured by the number of farm businesses affected, sector impacts, size of the area affected, and overall value of lost production.
2. The effects of the event must result in a severe downturn in farm income over a prolonged period.
3. The event must not be predictable or manageable through normal risk management strategies available to farmers, or be part of a process of structural adjustment.

For example, the drought of 2002-2003 had significant impacts (Adams et al., 2002):

• 30% reduction in 2002-2003 agricultural output, equivalent to 1% of GDP
• Flow-on effects to rest of economy lowered 2002-2003 GDP a further 0.6%
• Net effect on 2002-2003 GDP was a loss of 1.6%
• Loss of 70,000 jobs, mainly in wholesale, retailing and repairs (25,000), transport (9,000), business services (12,000), agricultural services (5,500, e.g. crop spraying and harvesting) and food processing and beverages (10,000).
• Worst affected regions, in terms of Gross Regional Product, were south-west Queensland (-21%), north-west NSW (-18%), the Victorian Mallee (-16%) and northern NSW (-15%)

6.2.3 United States

Byun and Wilhite (1999) argued that most currently used drought indexes were not precise enough to detect the onset, end, and accumulated stress of drought.

They suggested four classes for the study of drought:

1. Causes - directed at understanding atmospheric processes that lead to drought;
2. Frequency and severity - directed at characterizing the probability of drought events of various magnitudes;
3. Impacts - directed at quantifying the costs and losses associated with drought, including economic, social and environmental consequences, which may be direct or indirect;
4. Responses - directed at preparedness and mitigation strategies, and focusing on means of impact reduction.

These writers noted that most drought indices were based on meteorological or hydrological variables only. They pointed out a number of aspects of these indices which could be improved, highlighting the following features:

1. Accumulated deficit. Drought indices should be calculated with the concept of consecutive occurrences of water deficiency, rather than just the departure from climatological mean for a predefined period, as was currently the case.
2. Time step. Daily units of time were essential, because a water deficit could be overcome by just a day's rainfall. Most indices were based on monthly time steps.
3. Water storage term. Drought indices should characterise both soil moisture and other water resource (eg. lakes, ground water) storage as separate features. Byun and Wilhite noted that the Palmer Drought Severity Index (Palmer 1965), and the Surface Water Supply Index (Shafer and Desman 1982) considered these two features separately; others did not.
4. Time dependent reduction function. This was needed to account for daily water resource depletion through runoff, evapotranspiration and other factors, particularly to evaluate the residual resource of rainfall that had occurred some months previously.
5. Problems with modelled or estimated data. Oversimplification of data, for example soil moisture content, was inevitable because of variability in topography and other soil characteristics. It was better to use measured parameters only, such as precipitation.
6. Lack of other information. Drought indices failed to provide good information on the duration of drought, how much deficit of water had occurred, when the drought was likely to end, and how much rainfall was needed to return to normal conditions.

The writers introduced a new concept, effective precipitation (EP). Total precipitation over a period, for example a year, defined the water resource, and the contribution to the resource of each rainfall event (the EP) was qualified by how long ago it occurred (i.e. its input value decayed over time). Drought duration was then calculated from the number of consecutive days when EP was less than a derived normal, and drought severity was taken to be the depth of the accumulated deficit. A further term, the precipitation needed for a return to normal, was also calculated. Finally the writers proposed a number of drought severity indices that could be derived using this procedure.

National Drought Mitigation Centre

The Western Drought Coordination Council (1998), supported by the National Drought Mitigation Centre, University of Nebraska-Lincoln, describe the need for environmental, economic and social information to define drought and its impacts. They add that local customisation of drought information is essential because the causes and impacts of drought vary regionally. They argue therefore that the scale of the information should be:

'representative of the area experiencing drought and comprehensive enough to adequately examine corresponding impacts'.

The points noted below are included in their recommendations for drought indicator information.

i) Environmental information

• Precipitation - indicates which regions are most susceptible to drought, and characterises drought patterns over time in drought prone areas;
• Water supply sources - both surface and ground water, including managed (dams) and unmanaged. It's important to know when water sources are located in a different hydrological basin to where a drought occurs;
• Impacts of soil loss and sediment deposition - an example would be sheet erosion due to heavy rain following dry periods;
• Impacts on surface and ground water, from soil moisture to lakes and wetlands, including both quantity and quality of water;
• Effects on air - for example dust storms;
• Effects on wild life and plants - impact on habitats, diversity, and stress on species;
• The connection between drought and wild fires - both forest and rangeland, following extended dry periods, including both immediate and residual impacts.

ii) Economic information

• Understanding economic linkages and trends. While the impacts of drought may be felt first in the agricultural sector, there are always flow-on effects in many other sectors of the economy. In addition, the impact on rural communities and on families is often severe.
• Other economic factors - eg. awareness programmes, drought recovery loan schemes, and insurance.

iii) Social information

• Public health and safety - there are many issues here including health risks due to water shortages or contamination in dry seasons, mental and physical stress brought on by the situation, and the physical dangers posed by fires.
• Individuals' perceptions of drought - differing 'interpretations of drought characteristics may produce different attitudes and perceptions of how to deal with drought'.
• Acknowledging diversity - there is a wide range in the way drought affects people, because of diversity in social, cultural and economic circumstances.
• Government/NGO interactions - dialogue is important to determine best policy implementations for drought preparedness and relief.
• Political or government perspectives - avoidance of conflicting objectives in the management of economic sectors, where they make effective drought planning more difficult.

iv) Customized information

It is important that drought assessments and declarations are appropriate and accurate to the localities that are affected. Regional assessments of drought may be too general.

U.S. Department of Agriculture

The United States Department of Agriculture, in its report to the National Drought Policy Commission, highlighted the use by various agencies of 'trigger points' for drought assessment (USDA 2000):

'... public declarations of drought are often triggered by specific and well-defined conditions, such as a specific reservoir elevation on a specific date. In some cases, there are well-defined exit points that trigger a resumption of normal activity. These "drought triggers" become the practical definition of drought for a particular region and for specific issues. Defining these triggers is an inseparable part of planning for and responding to droughts. Once these triggers are defined, a region is much better able to estimate the costs, expected frequency, and risks of drought response.'

The Commission report further recommended that drought triggers should be both supply-type, reflecting moisture deficiencies caused by acts of nature (lack of rain, excessive temperatures), as well as demand-type, reflecting drought impacts.

Examples of current supply-type triggers used in general to define drought or trigger actions related to potential drought included:

• Precipitation less than 60% of normal for the season or present water year (used by the National Weather Service's Western Region);
• Precipitation less than 85% of normal over the past six months (used by the National Weather Service's Eastern Region);
• The Palmer Drought Index -2.0 or less;
• Consolidated drought indices at the 20th percentile or less (used by the Drought Monitor). For Federal action, more rigid triggers such as the 5th percentile drought might be appropriate, reflecting truly unusual circumstances.

Examples of demand (impact) based triggers included water supply less than 60% of normal (used by the National Weather Service's Western Region) and various crop loss thresholds, used by the U.S. Department of Agriculture.

6.3 Calculation of Potential Evapotranspiration Deficit

The water balance calculation used to derive the Potential Evapotranspiration Deficit (PED) drought index assumes that the water gains and losses to the soil profile are typically in balance. Provided water is non-limiting the balance for a given rainfall period can be written:

P = PET + Ro + D ± ΔS

Where P is precipitation, PET is potential (or upper limit) evapotranspiration, Ro is surface runoff, D is drainage loss through percolation, and ΔS is the change in water storage. For the purposes of this study, PET is calibrated for pasture water use.

In principle, for each day,

S = Sd-1 + P - PET - Ro - D

where S is the new storage, and Sd-1is the water storage for the previous day.

Field capacity water storage is defined by the available water capacity (AWC), which we have taken to be 150 mm for this study. Rainfall in excess of field capacity is assumed to be lost to the water balance by runoff and drainage.

if S d-1 + P - PET > AWC

then(S d-1 + P - PET) - AWC =(Ro + D)

As S is reduced, it becomes increasingly difficult for plants to extract water from the soil, and water transpiration decreases. Here we have used a method of estimating constrained water use by assuming evapotranspiration (ET) continues at its potential rate until half AWC is depleted, following which it ceases until further rain occurs.

if S < ½(AWC)

then ET = 0

The difference between the subsequent soil water-restricted evapotranspiration, (RET), and the atmospheric potential evapotranspiration for the period (PET), is referred to here as the potential evapotranspiration deficit (PED) and is incremented on a daily basis.

PED = PED d-1 + (PET - RET)

In effect, PED is approximately equivalent to the amount of water that would need to be added by rainfall or irrigation to keep pasture growing at its daily potential rate.

PED was accumulated daily for the July to June year, beginning from zero each year. Note that the soil moisture deficit carries over from one year to the next, even though PED is reset at the beginning of each July-June period. The water balance calculation was initiated on 1 January 1972, so there was a potentially non-zero starting value of soil moisture deficit at the beginning of July 1972.

PED is closely related to the frequently used 'Days of evapotranspiration deficit', which are days on which pasture is growing at less than its potential rate for a given season.

Figure 6.3.1 below shows the relationship between the potential evapotranspiration deficit and the number of days of deficit for Lincoln, for July to June seasons from 1881/82 to 2003/04. The data show that the average annual PED at Lincoln is about 400 mm, which equates to about 100 days of deficit.

6.4 Sensitivity of Potential Evapotranspiration Deficit to Rainfall and Available Water Capacity

As described in Section 3, in this study we have applied a range of projected rainfall offsets to the current climate to estimate changes in PED under future climate. Only changes in means of the underlying climate elements have been considered. Changes in modes of rainfall, such as the number of wet days per months or the daily persistence of rainfall events, were not considered.

Historical data show that there is a reasonably strong relationship between inter-annual variation in rainfall and changes in PED. Figure 6.4.1 illustrates the dependence of PED (based on an Available Water Capacity of 150mm, half of which is 'readily available' in the root zone) on rainfall during the height of the growing season (November through April), using the observed relationship at Napier for 1941-2004. The data suggest an approximately 80 mm increase in PED is likely for each reduction in rainfall by 100 mm. The data also illustrate how equivalent percentage reductions in rainfall in wet seasons have more impact on PED than in dry seasons. For example, from the equation, a 10% reduction in seasonal rainfall from 600 mm to 540 mm lifts PED from 196 mm to 244 mm, or 48 mm, while a 10% reduction from 200 mm to 180 mm of rain increases PED from 516 to 532 mm, or just 16 mm.

In a similar way, we might expect that shallow soils, where PED is already typically high under the present climate, are likely to be less impacted under climate change, relative to current PED levels, than deep soils with currently low PED. This is illustrated in Table 6.4.1.

Table 6.4.1 shows example water balance output for an idealised 10% reduction to all rainfalls for the years 1949 to 2004. Other meteorological conditions were assumed to remain constant. The water balance was run for five available water capacities as shown. Return periods were calculated using the same algorithm as in the main report (Kim et al., 2003). A 1-in-20 year PED for each available water capacity under the present climate (fourth column) would recur (given a 10% rainfall reduction) on average at intervals shown respectively in the final column. For example, a 1-in-20 year PED for an available water capacity of 150 mm would become a 1-in-15 year drought after a 10% rainfall reduction.

Throughout this report (apart from Table 6.4.1), all PED calculations are done using a uniform 150mm Available Water Capacity across the country, which is considered reasonably typical. One feature that Table 6.4.1 highlights is that return period changes are surprisingly robust across a range of water capacities. This suggests that our return period calculations are applicable to a range of soil depths.

6.5 Downscaling Potential Evapotranspiration

Potential evapotranspiration has been calculated in many different ways - see McKenney and Rosenberg (1993) for a review of eight alternative estimation methods. The fundamental climate elements involved are solar radiation, temperature, humidity and wind speed. Section 6.5.1 discusses factors influencing evapotranspiration, and notes observational relationships.

Unfortunately, the climate elements most readily available from climate models (precipitation, temperature and mean sea-level pressure) do not match the list above. The approach we have therefore taken in this study is to estimate PET variations from the available climate model data instead. Two steps are required before future scenarios of PET can be generated. Firstly, it is necessary to check the validity of replacing the 'primary' climate elements (radiation, temperature, humidity and wind speed) by those available from the models. This exploratory analysis is carried out on station data (section 6.5.2). Secondly, a downscaling procedure is needed to convert the changes at the global model grid-scale to PET changes on the 0.05° grid (section 6.5.3).

6.5.1 Sensitivity of PET to changes in climate

Air Temperature

Air temperature influences evapotranspiration in several ways, principally by determining the maximum amount of moisture the air can hold, and by the amount of energy that is supplied to evaporating surfaces. Higher temperatures typically increase the evapotranspiration potential, and thus the potential for increased drought risk.

Figure 6.5.1 shows the relationship between mean temperature and potential evapotranspiration (PET) at Christchurch Airport. Although the explained variance in PET is low (~30%), there is a clear trend of higher moisture demand in years with higher temperature. The highest PET (1001 mm) occurred in the warmest year, the La Niña season of 1988-89.

The data in Figure 6.5.1 suggest that a 2°C increase in temperature may raise PET by about 10%. However, preliminary work elsewhere on the relationship between mean air temperature and PET has indicated increases in PET of about 5% with a 2°C temperature rise (e.g. McKenney and Rosenburg, 1993). The data presented in Figure 6.5.1 show that relatively high (in comparison to temperature) PET occurred in several of the El Niño years, particularly the events of 1997-98 (marked 1 in the figure), 1991-92 (4), 1982-83 (3), and 1977-78 (2). Given the typically windy nature of El Niño events in Canterbury, PET increases in these years are likely to have been at least partly attributable to increased windiness.

The highest PET (1001 mm) occurred in warmest year, the La Niña season of 1988-89. Open points are El Niño years; enumerated points are referred to in the text. The trendline is indicative only.

The relationship between temperature and potential evapotranspiration raises the possibility that the higher temperatures expected with climate change may in any case increase the risk of drought, even if rainfall does not decrease.

Wind

Wind typically plays an important role in evapotranspiration, by increasing turbulence and facilitating the movement of moisture-laden air into the drier atmosphere. Wind increases the loss of moisture from wet surfaces, but where there is no moisture, for example in a very dry paddock, little additional loss of moisture may occur. Therefore, if windiness increases with climate change, it is likely to increase moisture loss while moisture is still available (for example early in the growing season), and thus potentially hasten the onset of drought.

As noted above, changes in windiness associated with El Niño seasons may partly explain increases in evapotranspiration during those seasons, though further work is needed to separate this effect from the influence of air temperature on its own. McKenney and Rosenberg (1993) obtained a similar trend in their work at two relatively windy North American sites, where they found that a 20% increase in wind speed led to a 9% increase in PET.

Solar Radiation

Solar radiation is the main source of energy for evapotranspiration. Solar radiation incidence is likely to change if cloudiness changes with climate change. For this study an estimate has been made from historical data of the variability of incident radiation using rainfall as a proxy indicator. The estimates show that evapotranspiration is inversely dependent on cloudiness, and this has been taken into account in calculating changes in PED with climate change.

6.5.2 Use of proxy variables to describe PET variations

Many changes in climate elements are interrelated. For example, a very sunny month (anomalously high solar radiation) at some location is also likely to be a dry month (anomalously low rainfall). Thus, it seems reasonable to try using precipitation as a proxy for solar radiation in a regression equation for PET. Local wind variation is notoriously difficult to predict, particularly in New Zealand's variable terrain. On the larger scale, though, wind is related theoretically to pressure gradients, so the use of some pressure index suggests itself as a proxy for site-specific windrun.

Figure 6.5.2 shows results from attempts to predict interannual variations in PET at a number of climate sites using multiple linear regression. The length of record is variable: Napier (1950-2003), Masterton (1950-1991), Blenheim (1953-1987), Lincoln (1950-1987), and Dunedin (1991-1999). The record length is selected to avoid any site changes that could adversely affect the homogeneity of the data. Separate regression equations are estimated for each calendar month.

For the first five panels of Figure 6.5.2, the predictors are: precipitation, temperature, and "wind". Three curves are plotted according to what wind measure was chosen: windrun at the site (which we would assume to be the most reliable), the "Z1" pressure index (anomalous pressure difference between Auckland and Christchurch), and both Z1 and "M1" (anomalous pressure difference between Hobart and Chatham Island). These pressure indices were tested because they have been widely used in New Zealand climate analysis since originally devised (Trenberth, 1976), are predictors in previous downscaling work (Mullan et al., 2001), and can be readily calculated from model grid data (either historical analyses or future projections).

Last panel (lower right) shows effect of using rainfall as proxy for solar radiation at Lincoln.

Results show that the PET explained variance is maintained fairly well when site windrun is replaced by the pressure index Z1 (except for the short record site of Dunedin). There are even instances where the local evapotranspiration is better estimated using Z1 than the local wind (Masterton).

The bottom right panel of Figure 6.5.2 shows a similar intercomparison for the Lincoln site (where long-term solar radiation records are available) using predictors: temperature, windrun, and either radiation, rainfall, or neither. It is clear that it would be valuable to have solar radiation as a predictor, particularly over the summer months when PET is highest. However, having rainfall is better than nothing.

6.5.3 Multiple linear regression downscaling of PET

The PET on the New Zealand 0.05 grid covers the period 1972-2003. Monthly anomalies (as %) were modelled by multiple linear regression using for predictors: precipitation (%) and temperature (C) at the same gridpoint, and Z1 and M1 pressure indices calculated from NCEP-NCAR reanalysis mean sea-level pressure data. (All the predictors are anomalies from their respective monthly climatologies). Figure 6.5.3 shows the explained variance (on the dependent data) for spring and summer months, aggregated into seasons. It is particularly satisfying that the explained variance is highest (and very significant, statistically) in the eastern drought-prone regions where our estimate of PET is most critical. The explained variance is lowest in the winter season (not shown) when absolute levels of PET are very low.

The individual predictor coefficients (maps not shown) look physically sensible. Over most of the country, the precipitation coefficient is negative (more precipitation meaning less solar radiation and therefore lower PET), and the temperature and Z1 coefficients positive (increased PET for higher temperature and stronger winds). For the drier eastern regions of New Zealand, the temperature regression coefficients suggests a range of 2-8% increase in PET per 1°C increase in local temperature, which is consistent with the single variable regressions noted in section 6.5.1. In eastern regions, the wind regression coefficients (not shown) suggests a 1-2% increase in PET per 10% increase in mean westerly wind speed, which is somewhat lower than the observational result from McKenney and Rosenberg (1993).

Future scenarios of monthly PET change are estimated by applying the regression equations to model projections of mean changes in the four predictors (temperature, rainfall, Z1 and M1).

6.6 IPCC 25% and 75% scaling of GCM patterns

The CSIRO Mark2 and Hadley Centre HadCM2 global climate models on which our scenarios are based were idealised transient simulations using 1% per year compounding carbon dioxide concentration. Their global average surface air temperature projections lie in the middle of the IPCC range (Figure 6.6.1, where the IPCC low/high envelope is taken from the IPCC Third Assessment, Cubasch et al., 2001). Rather than use the model changes directly, they were rescaled to take some account of uncertainty in model projections and emission scenarios. See Appendix 2 of Wratt et al. (2003) for further discussion of rescaling the local climate change projections.

Mullan et al. (2001) downscaled climate projections for New Zealand from six global climate models. Although only two models are studied in this report, all six models are used in the rescaling procedure, which is as follows. From the global temperature changes of the six models (four only beyond 2050), we determine the scaling factor that reproduces the IPCC envelope - that is, what factor makes the 'coldest' model match the IPCC lower bound, and what factor makes the 'warmest' model match the upper bound. This was the procedure used in Wratt et al. (2003) to generate the extreme IPCC range for New Zealand changes, where all the model changes were multiplied by these 'lowest' and 'highest' factors to represent the full spectrum of possible changes. For example, for the 1990 to 2080s change, these extreme scaling factors are approximately 0.55 and 1.44 for the suite of available global models.

In this report, we focus on 'low-medium' and medium-high' scenarios instead of the IPCC extremes (for reasons mentioned in section 1.3). We define these scenarios as arising from factors one-quarter and three-quarters of the way between the extreme factors. These points are denoted as the IPCC 25 percentile and 75 percentile scaling factors.

The 25% and 75% scaling factors are given in Table 6.6.1. Figure 6.6.1 shows the rescaling schematically. In practice, all the temperature changes are for 30-year averages (2020-2049 and 2070-2099), not individual years. This method may seem unnecessarily complicated. However, we cannot scale the model projections so that the individual model global temperature matches the 25% and 75% points of the IPCC temperature range, since this could push other models outside the IPCC extreme bounds. (Such a scaling is clearly wrong at the extremes).

Stars on the vertical bar at 2085 indicate schematically the IPCC 25 percentile and 75 percentile for the 2080s. Note that in practice (see text) we use a 30-year period, not a single year, and calculate the quartiles of the extreme low to high scaling factor range, not quartiles of the temperature change.

For the CSIRO and Hadley models, the scaled global-average temperature increase by the 2080s lies between 1.8°C (the 25% scaling) and 2.9°C (the 75% scaling). For the year 2100, this would correspond to a range from about 2.3°C to 3.6°C. This is slightly below the quartiles of the widely quoted 1.4°C to 5.8°C IPCC range at 2100 (i.e., 2.5°C and 4.7°C) because the CSIRO and Hadley models have a lower global climate sensitivity than some of the other models used in the IPCC Third Assessment.

6.7 Natural Variability of Drought

6.7.1 Variation in Drought Risk with El Niño-Southern Oscillation

The years used in the composite (eg, 1972/73 is first El Niño year) are chosen according to the state of the El Niño-Southern Oscillation over Dec-Feb season.

New Zealand climate varies from year to year, and also decade to decade, and therefore variation in drought incidence can be expected too. One of the main causes of interannual variation in New Zealand climate is the El Niño-Southern Oscillation (ENSO), and the 31-year data set can be used to quantify drought variation with ENSO. The result is shown in Figure 6.7.1, where the July-June PED is composited separately over 10 El Niño years (left panel) and 9 La Niña years (right panel). ENSO events do not coincide exactly with our July-June PED year, of course. However, the highest evapotranspiration and greatest PED deficit accruals occur over summer, and therefore we have characterised the ENSO 'year' according to the tropical Pacific sea surface temperature anomaly (in the so-called Niño-3.4 region) over this season, which is the time of year that ENSO events typically reach their peak (Gordon, 1995).

Figure 6.7.1 shows that there is an average PED deficit over most of the country in El Niño years, which is especially marked in eastern areas of both Islands. During La Niña years, the two deficit regions that stand out are Wanganui-Manawatu in the North Island, and coastal Otago in the South Island. Otago is notable for experiencing a deficit under both El Niño circulation (more westerly and therefore drier in the east) and La Niña circulation (more anticyclonic and therefore drier over the lower South Island). This nonlinearity of ENSO response, where El Niño and La Niña conditions are not opposite, was first noted by Mullan (1995).

6.7.2 Decadal Variation in Drought Risk

Coherent variations in New Zealand climate over decadal and longer timescales have been identified (Salinger and Mullan, 1999). One of the factors that appears to contribute to this decadal variation is the Interdecadal Pacific Oscillation (Mantua et al., 1997). Three phases of the IPO have been identified during the 20th century: a positive phase (1922-44), a negative phase (1946-77) and another positive phase (1978-98). The pattern associated with the positive phase is higher sea surface temperatures in the tropical Pacific (more El Niño-like) and colder conditions in the North Pacific. Around New Zealand, the sea temperatures tend to be lower, and westerly winds stronger.

The 31-year gridded data set is too short to examine IPO variations, beginning as it does in 1972. However, longer records of PED are available from some sites in the NIWA Climate Database. Figure 6.7.2 (Figure 7 in Phase 1 report, Porteous (2004)) shows the mean change in July to June PED between the 1950/51-1977/78 period and 1978/79-2002/03 period. Although the figure must be considered preliminary because of the limited number of sites used, the pattern of increasing dryness in the east for the most recent positive IPO phase is consistent with other information on how the Interdecadal Pacific Oscillation affects New Zealand climate (Salinger and Mullan, 1999; Salinger et al., 2001; Wratt et al., 2003).