think climate. think change.
Department of Climate Change
Mark Howden - CSIRO Sustainable Ecosystems, Roger Jones - CSIRO Atmospheric Research
Australian Greenhouse Office - October 2001
- By 2030, there is an 88% chance of production being above current levels in 2030 (and a 12% chance of it being less than current levels). The mean increase is 3% above current levels.
- By 2070, there is a 64% chance that average national grain production will be higher than current levels (and a 36% chance it will be less than current levels). The mean increase is 1.8% higher, but the range could be between 25% below to 10% above current production.
- By 2030, there is a 32% chance of wheat crop value being below current levels (and a 68% chance of it being above). Value is based on both quantity and quality (protein content), which determines market price and end use. Potential change in value could be between a decrease of $250M/year (although the chance is very small) to an increase of $150M/year.
- By 2070 there is a 45% chance of national wheat value being lower than current levels. Potential increases are limited to about $220M/year, but potential decreases could be as much as $800M/year (although the likelihood is low).
- By 2030, value of exports could fall by an average of $82M. There is a 91% chance of exports being below current levels.
- By 2070, value of exports could fall by an average of $164M. There is a 72% chance of exports being below current levels.
- These are based on ABS mid-range population projections, and are due to increased domestic consumption of wheat as well as the climate change effects. Varying assumptions of population growth and grain demand significantly affect export values.
- cropping regions in WA: There is a high chance that productivity and value will be below current levels in both 2030 and 2070. Average declines by 2070 are expected to be $13M to $104M (3-15%). If this occurs there may be major regional economic implications.
- cropping regions around Emerald (Qld) and Wagga (NSW): There is a very small chance of production and yield being below current levels in both 2030 and 2070. Mean increases in value by 2070 are expected to be $13M to $24M (9-13%).
- other cropping regions: In South Australia, Victoria, NSW and southern Queensland there is some risk of negative impacts but a larger chance of positive impacts. Mean change in production is between 6 and 12%.
Effect of climate change and CO2 increase for the years 2030 and 2070 on percent change in production (currently 21.7Mt), value of production (currently $4.2 billion) and value of exports (currently $3.3 billion) assuming either current management practices or adapted management practices.
| Year 2030 | Year 2070 | ||
| Production | - current - adapted |
3.1 8.0 |
1.8 8.5 |
| Value of production | - current - adapted |
0.4 1.6 |
-0.8 1.6 |
| Value of exports | - current - adapted |
-2.5 0.0 |
-4.8 -0.6 |
Wheat is the major crop in Australia in terms of both value and volume. National wheat production is currently about 22Mt from an area of 11Mha. On average about 80% of the harvest is exported, worth some $3.8 billion each year with the total crop being worth about $4.2 billion. Yields are generally low (national average over the past decade of 1.75t/ha) due to low rainfall, high vapour pressure deficit and low physical and chemical soil fertility. High climate variability forces low input management to limit financial risk. Thus the Australian wheat industry is highly sensitive to climatic influences. Increases in levels of atmospheric CO2 and other greenhouse gases are considered likely to significantly change global climate, increasing temperature and changing regional rainfall patterns (Fig. 1). These climate changes will have potentially significant impacts on wheat production in Australia due to the sensitivities to climate outlined above. In addition, there will also be impacts directly from the increased CO2 levels which tend to enhance crop growth through increased photosynthetic rates and water use efficiencies, but can reduce grain protein content (e.g. Howden et al. 1999). Changes in areas suitable for cropping, changes in salinity and erosion risk may also occur.
A preliminary analysis of the implications of combined changes in atmospheric CO2 concentration and regional climate change was made for Australia by Howden et al. (1999a,b), but this was based on the CSIRO 1996 scenarios and early versions of the SRES-based scenarios. Furthermore, it did not attempt to take into account the full range of possible outcomes, which include future CO2 levels, the implications of this for global temperature change, the consequences of global temperature change for Australian regional rainfall and temperature and finally the implications of these changes for the production, quality and economics of wheat cropping systems across Australia. Each of these elements has uncertainty associated with it.
Fig. 1. Relationship between CO2 concentration and global temperature change (oC) for the SRES scenarios (IPCC 2001).
This study attempts to incorporate all these different levels of uncertainty into one analysis. The general approach is to use Monte Carlo (random number) sampling across the specified ranges of uncertainty at each level. The output of such analyses is a probability distribution - it indicates the likelihood of different outcomes. Such information is used in risk management in a large range of industries in policy development, strategic analyses and tactical decisions.
The project focuses on ten sites in the major wheat growing districts as a pathway to scaling up results for the whole of the industry. The districts and sites (Fig. 2) are:
Fig 2. Current wheat cropping areas in Australia and the sites used in this study.
For each site, analyses are made of the effect of change in CO2, rainfall and temperature on average wheat yields. These analyses use results from the agricultural systems model I-Wheat (Meinke et al. 1998) which is a module of the APSIM systems modelling framework (McCown et al. 1996). I-Wheat was run for a large range of feasible combinations of CO2 increase, temperature and rainfall change using modified 100-year climate records. These provide relationships (multiple regressions) of the general form:
Yield change (% from 100-year historical average) = aCO2 + bT +cR +d
Where T (oC) and R (% change) are temperature and rainfall change respectively from the 100-year average. Some regressions had non-linear terms either in place of, or in addition to, those in the above equation. A separate set of equations was developed for simulations which included management adaptations of change in variety and change in planting window to optimise management under the new climate regimes (Howden et al. 1999a)
These relationships are the basis for developing probabilistic assessments of changes in yield under future climates and CO2 levels by including the standard errors of the coefficients (a,b,c,d), the standard error of the multiple regression and the correlation between the co-efficients. Monte Carlo sampling is performed with a proprietary package (@RISK).
Change in grain nitrogen (%) was calculated as a function of yield change using similar regression techniques.
To make these relationships operational, uncertainty in the input variables (i.e CO2, temperature, rainfall) needs to be addressed. The methods to do this are described in the following section.
There is considerable uncertainty in scenarios of CO2 increase and related climate change - and wheat responds to both factors. Atmospheric CO2 levels are indicated to rise from current levels (370ppm) to between 420 and 480ppm by the year 2030 and between 525 to 720ppm by the year 2070. In the same time frames temperatures across Australia may increase by between 0.4 and 2oC by 2030 and by 1 to almost 7oC by the year 2070. Large changes in rainfall are possible with changes of + 20% by 2030 but up to + 60% by 2070 - noting that there is marked variation between regions and seasons.
The approach used is partly based on the approach of Jones (2000) which deals explicitly with the cascading hierarchy of uncertainty in relation to future global temperature change and the implications of this for regional changes in precipitation and temperature. The four key enhancements to this approach are:
to incorporate the main driver of change, CO2 concentration, as the primary source of uncertainty in the hierarchy. The rationale for this is obvious when the relationship between the two factors is considered (Figure 1) along with the dependence of all anthropogenic climate changes on alterations in concentrations of greenhouse gases - of which CO2 is the dominant one. Additionally, the importance of CO2 concentration on plant biological processes and the interaction of these changes with temperature requires that sensible links be made between the CO2 concentration and climate changes
to explicitly include uncertainty in the productivity functions which translate variations in CO2, temperature and rainfall to changes in wheat yield and nitrogen concentration
to include non-uniform probability distributions in the transfer functions which calculate regional temperature and precipitation change from global temperature change
the inclusion of economic components in the analysis.
There is a range of possible global temperature changes for any given future CO2 concentration (Fig. 1). This variation is a result of both diversity on the possible emission trajectories (including those of aerosols which reduce warming) as well as model uncertainty in translating such emissions to global temperature change. The approach developed here to deal with this is to develop regressions which relate CO2 concentration to temperature change for the 'maximum' and 'minimum' parts of the envelope of temperature change from the SRES scenarios. Sampling of possible CO2 concentrations is made using a uniform distribution between the highest and lowest values for each scenario year (2030 and 2070). A uniform distribution is used as there is no a priori reason for CO2 concentrations to have a greater likelihood of occurrence at any point within this range. The central tendency in the resultant sampling (Fig. 3) arises from the regressions used to bound the sample. Alternative sampling approaches are feasible, however comparison of these is beyond the scope of this project.
Fig. 3. Global temperature changes simulated for the year 2070 using the procedure outlined in the text.
To translate these global temperature changes to regional temperature and rainfall changes, CSIRO Atmospheric Research have synthesised the results from nine Global Climate Models. The results have been output as monthly change in temperature (oC) or rainfall (%) per degree global warming for each of the ten sites. Means of the monthly values across the growing season (typically May to October) were calculated to provide input into the wheat yield regression equations. For Emerald, Dalby and Moree, change in precipitation was calculated as the January to September mean due to the importance of summer precipitation allowing storage of soil water in the deep soils - an important contributor to yields in these regions. These changes in temperature and precipitation per degree global warming were assessed for the type of probability distribution appropriate using the monthly and mean results.
Probability distributions were fitted using a proprietary software program (BestFit). In many cases a triangular distribution was suitable although in a couple of examples, there was no central tendency for results and a uniform distribution between the maximum and minimum values was used. When the global temperature change (GTC) and change in rainfall/temperature per degree GTC are sampled according to their respective probability distributions, a sample of possible outcomes in rainfall and temperature is provided for each site (e.g. Fig. 4).
Fig. 4 Change in a) rainfall (%) and b) temperature (oC) for Dubbo for the year 2070 simulated via sampling.
As we are going to simulate national yield changes and subsequent national economic implications, it is important that for each sampling, the possible relationship between the changes in rainfall and temperature is incorporated. For example, we would anticipate a priori that adjacent sites (e.g Wongan Hills and Katanning) are likely to have similar climate changes within a sampling whilst sites distant from each other others (eg Katanning and Dalby) may be largely independent. To incorporate this effect, correlation matrices were developed between all sites for both precipitation and temperature using the monthly values for change in precipitation and temperature per degree global temperature change.
Changes in site yield (t/ha) are scaled to regional productivity (tons) using the average regional Australian Bureau of Statistics (ABS) production statistics for the past decade and the change in yield under a given global change scenario. For example, the production is summed from the ABS sub-Divisions for which Katanning is broadly representative and the total is multiplied by the change in yield calculated from the multiple regressions for that sampling of CO2, rainfall and temperature change. The same occurs for all other regions. These regional values are then aggregated to give national production.
Crop value ($/ton) is calculated as a function of grain N concentration (%N) based on several years data (Howden et al. 1999a):
Value ($/ton) = -66.395x3+435.6x2-851.36x+656.81
Where x is calculated percent nitrogen (N%) in the grain derived as discussed previously. For simplicity, the same equation is used for all States although minor differences occur.
Regional crop value is then calculated from the regional productivity and the revised crop value. This is then compared with baseline values (i.e. with no global change). Aggregate national values are the sum of regional crop value.
Wheat is currently grown to in all States and Territories and occupies effectively its full biological and economic range. Historical experience in Australia (e.g. Meinig 1962) demonstrates that significant decreases in production potential are likely to result in reduction of the area planted to wheat, particularly in the dry margins. For example, if productivity decreased by 40% marginal areas would become unviable, core cropping areas may become marginal and only currently high rainfall areas would remain as core cropping areas. A full study of this possibility is far beyond the scope of this analysis. However, it may be a significant factor in those scenarios which have marked reductions in potential yield. To incorporate this component we have included a simple relationship between yield reduction and regional production (effectively area sown) such that there is no reduction in area sown for a 10% reduction in yield potential (i.e. assumes that there is full adaptive capacity for small changes) but that after this, there is a decline in area which varies with site (Fig 5). Sensitivity analyses show that at the national level this inclusion has only a moderate effect, extending the negative 'tail' of the results. However, at a regional level, the implications of such change are likely to be marked. The assumed reductions in area for a halving of productive potential are given in Table 1 and are based on assessment of average yield and sown area of ABS sub-divisions.
Fig. 5. Change in productive area assumed with decline in production potential under a given Monte Carlo sampling. The three lines represent different levels of sensitivity.
Table 1 The ten sites in the study and the assumed reductions in area cropped with decrease in production potential by 50% from current levels.
| Site | Reduction (%) | Site | Reduction (%) |
| Geraldton | 50 | Wagga | 30 |
| Wongan Hills | 50 | Dubbo | 50 |
| Katanning | 30 | Moree | 50 |
| Minnipa | 50 | Dalby | 40 |
| Horsham | 40 | Emerald | 80 |
We do not incorporate expansions in cropping areas for scenarios with increased production potential for two main reasons. Firstly, expansion into the marginal zones is likely to be limited by suitable soil types (Walker 1982; Reyenga et al. 1999; Howden et al. 2001) with the possible exception of the Mitchell grass region of southern Queensland and northern NSW (Howden et al. 1999a). Secondly, establishing cropping in previously uncropped areas results in large-scale loss of soil carbon (typically 40-50% within the top 1m) and hence greenhouse gas emissions. There remains some possibility that such emissions may be the subject of some form of carbon accounting allied with a credit/permit/tax system that may make expansion less favourable economically.
Currently about 78% (17 Mt) of the Australian wheat crop is exported with domestic consumption comprising of 2.2 Mt used for food, 1.9Mt used for livestock feed and 0.5 Mt used for seed. For scenarios for the year 2030 and 2070 we calculate export volumes as production minus domestic consumption (food, feed, seed). Food consumption is a product of population and per capita consumption. ABS scenarios of population were used (Trewin 2001) for Low, Mid and High growth scenarios for 2030 and 2070. Per capita consumption of wheat having grown rapidly over the past decades seems to have levelled off at about 115kg/year and this is assumed to stay stable. Consumption of wheat by livestock has been increasing irregularly over the past decades. These trends seem likely to continue with trends towards increased consumption of poultry, pigs, aquaculture species and lot-fed beef along with increased concentrate consumption by dairy cattle to boost production. We assume that grain consumption by these livestock will plateau at the high levels (1.87 Mt) experienced over the past three years.
Hence, for any given climate change scenario, export volumes and value can be calculated and change in these from current baseline values can be assessed.
