Chapter 2 provides quantitative and qualitative projections of future changes in a number of key climate variables for New Zealand. This appendix provides additional background on the downscaling and model projections, the level of agreement between the models and changes in extreme precipitation.
The global climate models used in the IPCC Fourth Assessment, 2007 tend to have improved spatial resolution relative to those of the Third Assessment, 2001. Yet the resolution is still too coarse to show much detail over New Zealand. For the 12 General Circulation Models retained for the downscaling exercise, the grid-point spacing varies from 1.125° to 3.75° in longitude, and 0.56° to 2.5° in latitude at the equator81. Of the five models rejected of the original 17, three had the coarsest resolution of all (up to 5° longitude grid spacing).
Clearly, it is necessary to ‘downscale’ the General Circulation Model to represent the effects of New Zealand’s complex topography. The statistical downscaling algorithm used is exactly the same as that applied in the previous edition of this Guidance Manual (Ministry for the Environment 2004), and the same as that described in Mullan et al (2001). The difference occurs in the observational datasets that were used. In the previous edition the downscaling algorithm was applied to individual climate station time series, evaluated at 58 temperature sites and 92 rainfall sites. For this edition, the downscaling is applied to a gridded dataset that covers all of New Zealand with 0.05° latitude–longitude boxes. There are approximately 11,500 grid points over the New Zealand land mass. The gridded data were developed by interpolating observed daily rainfall plus maximum and minimum temperatures from some hundreds of daily reporting sites (Tait et al 2006). The gridded data begin in January 1960 for rainfall, but not until January 1972 for temperature.
The downscaling procedure uses monthly anomalies over the period 1972–2003 to develop regression equations for precipitation and mean temperature. For each climate element, the grid-point anomaly is related to three predictors: the large-scale zonally-averaged anomaly over 160–190°E at the same latitude as the grid point, and the anomalous components of two wind indices known as the ‘Trenberth Z1’ and ‘M1’ indices (Trenberth 1976). The large-scale anomaly field, and the Trenberth indices that are derived from pressure differences (Z1=Auckland minus Christchurch, M1=Hobart minus Chatham Islands), are well-defined at the scale of the global climate model. These observed predictors are evaluated from NCEP re-analysis data82 in the New Zealand region.
The regression equations are seasonally stratified and use the three contributing months in each season. As an example, Figure A3.1 shows the explained variance in the winter for both temperature and precipitation. Generally, the winter months have the lowest explained variance in temperature and the highest explained variance in precipitation, compared to other seasons. This seasonal pattern in the explained variance in precipitation is what we might expect: in winter, precipitation is more strongly determined by large-scale weather systems (which are detected in the latitude-averages and wind indices), whereas in summer, much of the rainfall might be convective in nature and quite local to the observing site.
Figure A3.1: Explained variance of downscaling regression equation for the winter months (June, July, August), for mean monthly temperature (TMea, left panel) and monthly precipitation (Rain, right panel).
A useful amount of interannual variability is explained by this approach: more than 50% of the variation in monthly temperatures is explained at most locations, except at high altitude in the South Island. The underlying patterns with respect to Z1 and M1 variation are as we would expect: for example, more rain in the west and less in the east under positive Z1 anomaly (more westerly); and lower temperatures under positive M1 anomaly (more southerly).
The regression relations are formulated so that the departure of the local anomaly from the latitude-average anomaly is calculated from the anomalies in the wind indices. In simple terms, the circulation anomaly imposes spatial structure on the broad-scale change. Thus, if there is very low explained variance in the regression at some location, the climate change at that point will effectively be the same as the latitude-average evaluated at the model grid scale. In applying the regression to the future projections, the changes in circulation (Z1, M1 indices derived from the model mslp field) and in latitude-average climate (from model precipitation or temperature field), relative to the base period of 1980–1999, replace the observed monthly anomalies.
One further adjustment is made before downscaling the future changes. Mullan et al (2001) noted that there can be a systematic bias in the model simulation of the current climate: eg, the westerlies might be too strong over New Zealand in the model relative to observations, and this will cause the monthly variability and longer-term trends to also be too large. Thus, the changes (eg, 1980–1999 to 2030–2049) in the wind indices are scaled by a factor that makes the model interannual variance match the observed variance in the respective index. It turns out that all the models but one (cnrm_cm3) over-estimate the variation in the west–east wind component, and about half the models underestimate the variation in the north–south component.
Chapter 2 presents of the downscaled precipitation and temperature changes under the SRES A1B emissions scenario. Maps are shown for the annual average and seasonal average changes, as an average over the 12 climate models.
The range of projected changes, particularly when rescaled to the full IPCC emission range (Tables 2.2–2.5 in chapter 2), is often quite large. Indeed, for rainfall the results can sometimes suggest anything from a large decrease to a large increase in the amount of precipitation. We appreciate that this can make it difficult for a user to decide on an appropriate value to use in a risk assessment. The purpose of this section is to help the reader make these decisions by providing more information on the level of model agreement. In particular, while there are situations where model agreement is low, there are also instances (eg, a particular season or part of the country) where agreement is high.
Figures A3.2 to 3.5 show projected changes in annual mean temperature or rainfall, at 2040 and 2090, separately for each of the 12 climate models analysed. The same colour scale is used in the 12 panels within each figure, to facilitate comparison between model projections.
The models tend to be fairly consistent through time in the relative amplitude of their respective changes and the spatial pattern. In the temperature projections, four of the models (gfdl_cm20, miroc32_hires, mpi_echam5, mri_cgcm232) tend to indicate that the North Island will warm faster than the South Island, whereas another four models (cccma_cgcm3, csiro_mk30, ncar_ccsm30, ukmo_hadgem1) show larger temperature increases in the South Island. The remaining four have a more complex pattern or little north–south gradient in the rate of warming. Two models (csiro_mk30, mpi_echam5) indicate noticeably less warming than the majority, and one (miroc32_hires), markedly more.
The projected changes in precipitation depend very much on whether the projected westerlies will increase or decrease across New Zealand. Nine of the models show a marked drying in the north and east of the North Island, and down the eastern coastal part of Marlborough and Canterbury, in the annual mean.83
An examination of Tables 2.2 and 2.3 in chapter 2 shows that the projected future warming is least in the spring season. This feature was not apparent in the previous New Zealand scenarios (Ministry for the Environment 2004). Figure A3.6 shows this result is very consistent across the 12 General Circulation Models.
In Figure A3.6, the darkest red colour means 11 or 12 models have spring as the season of least warming. The darkest blue colour means just 0, 1 or 2 models choose spring. The second to darkest blue means three models choose spring, which is the number expected by chance if there is no systematic difference with season in the rate of warming. It is clear that there is strong agreement on the spring season showing the least warming. By 2090, only Northland shows no preference for the spring season. There is no season preferred for the most warming84 (not shown) – at least, not for the country as a whole (Figure 2.4 shows of all four seasons, winter to have the greatest warming in the eastern South Island).
Figures 2.7 and 2.8 present maps of the annual temperature increase for each of the 12 models individually at 2040 and 2090. It is apparent that one model (miroc32_hires) indicates markedly more warming than any of the other 11. This is actually the main reason why the upper end of the projected warming is substantially higher in this updated edition than in the earlier edition of the Guidance Manual (Ministry for the Environment 2004).85
Figure A3.7 shows the situation across all regional council regions. The upper star on each vertical bar corresponds to the warming projected by the miroc32_hires model. If we consider the two extreme models as outliers in the distribution of projected temperature increase, then the average increase over the remaining 10 models is only slightly smaller than the 12-model results of Tables 2.2 and 2.3. However, the range of projected increases in annual temperature by 2090 is dramatically reduced for North Island regions: for example, in Northland the 12-model range of about 3°C (+1.0 to +4.0°C warming) decreases to less than 1°C (+1.6 to +2.4°C) for the A1B scenario shown in Figure A3.7.
Note: 1990–2090 changes are averaged over all grid points within each of 15 regional council areas (as listed in Table 2.2). Regions are abbreviated by a three-letter code, but occur in the same order (north to south) as in Table 2.2. Vertical coloured bars show the range over all 12 models, and stars the 12 individual model values.
81 1° latitude corresponds to a distance of approximately 110 km.
82 See reference to Kalnay et al. (1996) in Appendix 2.
83 The seasonal changes can, in some instances, be the opposite of the annual change (see Figures. 2.6 and 2.7, and Appendix 3.2.3).
84 In the corresponding Tables 2.2 and 2.3 of Ministry for the Environment (2004), winter appeared to be the season of most rapid warming in all regions (although by only a few tenths of a degree at most by the end of the 21st century).
85 For example, the upper end of the annual warming range at 2090 (2080–2099 average) in Northland is now +5.9°C (Table 2.3), compared to +4.0°C to the 2080s (2070–2099) in the previous edition (Ministry for the Environment, 2004).