Effects of Climate Change on Optimal Farm Plans in a Low Rainfall Mediterranean Environment of Australia M. John1,2 and R. Kingwell1,3 1: School of Agricultural and Resource Economics, University of Western Australia 2: CRC for Plant-based Management of Dryland Salinity, University of Western Australia 3: Department of Agriculture, Western Australia

Particularly since the early 1990s considerable scientific effort has been devoted to the study of climate change, with most analyses concluding there is a strong likelihood of substantial climate change occurring over the next 30 to 50 years[1]. Most international studies that examine the impact on agriculture of climate change due to global warming conclude that in many instances agriculture will be disadvantaged (Cline 1992, Schimmelpfennig and Yohe 1998, Evenson 1999, Rosenzweig et al 2002).  The majority of these studies have focused on the impacts on crop yields rather than animal production.  For example, Cline (1992) forecasts a 7 per cent decline in global crop yield attributable to global warming. Rosenzweig et al (2002) forecast that, in the United States alone, damage to corn production over the next thirty years caused by excessive precipitation associated with global warming will be $3 billion per annum.  Similarly Cline (1992) predicts agriculture in the United States would suffer a net annual loss of $18 billion (in constant 1990 prices).  In Europe marked regional differences in climate impacts are predicted (Olsen and Bindi 2002) with northern areas benefiting from a warmer environment supportive of a greater range of plant species, higher crop production and an expanded area of cropping.  By contrast, southern areas are forecast to experience water shortages, higher yield variability and reduced crop areas.

The Intergovernmental Panel on Climate Change Third Assessment Report (IPCC 2001c) concludes that the agricultural sector in Australia is particularly vulnerable to climate change with the strong possibility of more frequent El Nino events.  Regional differences in the agricultural impacts of climate change are predicted (CSIRO 2001; Howden and Jones 2001; Pittock 2003).  Australia’s main agricultural industry, dryland broadacre farming, is predicted to experience a range of impacts (Howden et al 2003) such that on balance “the scale of the potential negative impacts is considerably greater than that of the potential positive impacts” (p. 114, Pittock 2003). 

Forecasting the agricultural impacts of climate change has relied on various approaches.  The initial approach of the IPCC (1990) superimposed climate change on an agricultural sector devoid of coping strategies.  Understandably this approach was criticised as failing to reflect likely behavioural responses of farmers (Cline 1991, Mendelsohn et al. 1994).  It produced over-estimates of the economic impacts of climate change.  Other approaches (Nordhaus 1991, Mendelsohn et al. 1994) assumed farmers could costlessly adjust to new climate regimes and these analyses also have been criticised by Quiggin and Horowitz (2003) for failing to appropriately include adjustment costs. Being comparative static models that ignored adjustment costs they understated the cost of climate change.  At a global or broad regional scale where spatial impacts of climate change differ, and even where these impacts on the value of aggregate output sum to zero, adjustment costs may be significant.  Most approaches to measuring the agricultural impacts of climate change, although mentioning the possibility of important irreversibilities and catastrophic change (Fisher 2003) rarely explicitly include these costs.

For a farm in a specific location, the impacts of climate change broadly speaking are of two forms.  Firstly, the climate regime may alter to either diminish or improve the relative profitability of the farm business through primary impacts on crop and pasture production or subsequent climate-related market impacts on commodity prices and input costs.  Secondly, as outlined by Quiggin and Horowitz, climate change will increase adjustment costs where farm assets, or other assets directly or indirectly linked to the farm, are geographically fixed (eg grain handling) or are long-lived assets tied to an enterprise affected by the rate of climate change.

This paper explores the impacts of climate change for a farm in a low rainfall Mediterranean environment in Western Australia.  Using a comparative static framework the impact of climate change on farm profit and the optimal mix of enterprises and their tactical management is reported.  The nature of capital investment is examined and implications for adjustment costs are discussed.  The paper is in four sections.  The first describes the set of climate change predictions for the region in which the representative farm is located.  The second section provides a description of the farm model subsequently used in the analysis.  The third section presents and discusses a range of modelling results showing the impact of climate change on representative farms.  Finally, a set of conclusions is drawn.

In the southwest of Western Australia, where broadacre agriculture in a major industry, climate change is forecast to most likely cause:

• a decrease in winter rainfall and probably autumn rainfall, implying a later start to growing seasons,

• a decrease in spring rainfall.  Combining the autumn, winter and spring rainfall changes with expectations of increased evaporation suggests growing seasons will be shorter.

• warmer winters and springs, combined with higher CO₂ concentrations that may increase crop yield, even under declining rainfall. However, grain protein may decrease under higher CO₂ levels.

• rising temperatures, particularly in spring (Foster 2002) that will affect all agricultural crops via potentially large changes in heat or chill accumulation and the frequency of temperature extremes.  A potential benefit for grain and horticulture crops would be the reduced risk of frost.  However, conversely more hot days during grain or fruit filling could reduce yields.  Tree crops are particularly sensitive to temperature trends because of the longer lead times associated with their establishment and development compared with annual crops.  Those currently growing at the warm margin of their climatic range will likely face reduced cold accumulation and increased heat stress.  Adaptation will require a wider range of varieties to be available.

• a lower rate of spread of salinity due to a drier, more evaporative climate.  However, daily rainfall events over summer are projected to become more intense, and this might lead to increased episodic recharge.

• an increased need for on-farm water storage due to changes in rainfall and evaporation.

• more frequent extreme weather events (eg consecutive days of extreme heat, extreme thunderstorms) that will affect rural and urban communities and potentially cause crop and stock losses.

• altered risks of damage from pests and diseases.  Higher temperatures are favourable to many insects, though their ultimate activity will be dependent on any changes to summer rainfall.  A warmer climate might also favour many plant diseases.

• less irrigation water for agriculture.  About 45 per cent of WA’s water use is for agriculture.  Irrigation water use will come under increasing pressure from competing uses such as public water supply and industrial use.  Agriculture will also be under scrutiny in regard to its efficiency of use of water, its water allocations, its receipt of cross-subsidised water supplies (e.g. Comprehensive Water Scheme) and its impacts on water quality through salinisation and nutrient export.

These predictions are based on a range of studies including CSIRO (2001), IPCC (2001a, 2001b), GATF (2002) and Pittock (2003). Figures 1 and 2 are the CSIRO (2001) projections for temperature and annual rainfall changes across Australia for 2030 and 2070.  Of particular relevance to broadacre farmers in the south-west of Australia is the forecast of substantial decline in annual rainfall, particularly towards 2070.  Pittock (2003) reports that average annual average rainfall in the south-west of Australia will tend toward a decrease in the range -20% to +5% by 2030 and between -60% to +10% by 2070.

The south-west of Australia currently often supplies around half of Australia’s wheat production, 40 % of Australia’s wool production and 70% of Australia’s grain legume production. Almost all this is exported.  Hence, the production ramifications of climate change in the south-west of Australia have implications not only for its regional economy but also for the national economy.

To examine the potential ramifications, at a farm-level, of the potential impacts of climate change, a farm model representative of a region in the south-west of Australia is constructed.  The model incorporates key biological and economic ramifications of climate change.

The farm model constructed is based on discrete stochastic programming that is a programming formulation of a decision tree (Cocks 1968, Rae 1971).  It requires the explicit specification of management choices and their possible consequences.  The states of nature or event forks are usually represented by a relatively small number of discrete outcomes (Anderson et al. 1977).  Discrete stochastic programming models are mathematical programming models that allow for uncertainties in right-hand side terms, technical coefficients and the objective function (Hazell and Norton 1986, Hardaker et al. 1991).  These models are finite horizon dynamic programming problems with discrete random events and continuous choice variables (Featherstone et al. 1993).

The model constructed is known as MUDAS (Model of an Uncertain Dryland Agricultural System) and it captures the main elements of a representative farm of the Merredin region of Western Australia (see Figure 3).  The region comprises about 1050 farms almost all of which are mixed enterprise businesses mainly producing wheat, wool and sheep for export or local consumption.

In the region the proportion of arable area sown to wheat is often around 50 per cent and the proportion remaining as pasture used almost exclusively by sheep is around 40 per cent.  The remainder of the arable area is sown to other crops, mostly barley and lupins.

The Merredin region is characterised by a Mediterranean climate.  Cramb et al. (1991) list three characteristics of such a climate.  Firstly, rainfall is concentrated in winter, with summer being almost dry.  Secondly, summers are warm to hot while winters are mild.  Thirdly, there is high solar radiation, especially in summer. 

As shown in Figure 3 average annual rainfall decreases rapidly in a north-easterly direction from the south-west corner of Western Australia.  Many parts of the Merredin region receive an average annual rainfall of around 300 mm.  The town of Merredin, for example, receives an average annual rainfall of 330 mm. Winter rainfall is on average about 4 times greater than summer rainfall.  Moreover, because of high temperatures in summer and associated high rates of evaporation the lesser rain that falls in summer is often ineffective in promoting and sustaining plant growth during summer.  Accordingly the sowing of crops does not occur in summer.  Rather crop sowing commences in late autumn through to early winter.  These plantings are made possible by the prospect of mild winters with continuing rainfall that promotes plant growth through until maturity of the crops in early summer.  Almost 70 per cent of average annual rainfall is during the traditional growing season of May to October.

Much of the land on farms in the Merredin region is cleared.  The region was first opened up for agriculture very early this century.  Initially farming took place on valley floors and slopes where soils were more fertile.  However, farming spread to the less fertile soils of the plateaux with the discovery of trace element treatment of soils, the availability of superphosphate and the introduction of leguminous pasture species better suited to these soils (Janes 1981).

In the last forty years the increased mechanisation of dryland agriculture has caused rapid increases in farm size and labour productivity (Ockwell 1990).  The shift toward greater reliance on mechanisation has meant that typically a farm in the Merredin region now is owner-operated with not more than one other permanent labourer.  Casual or contract labour is hired for only a few months of the year to assist in main tasks such as seeding, harvesting and shearing.  Most farm operations are highly mechanised and typical farm size is now around 3500 hectares.  Almost all farms maintain an inventory of crop sowing, harvesting and grain storage equipment; together with sheep handling and shearing gear.  This inventory is occasionally replenished by the purchase of new or previously owned and used gear.  The timing and level of these purchases is often a function of seasonal, liquidity and taxation considerations.

Crops, mainly wheat and lupins, are sown from late April until early July, depending on seasonal conditions, and are harvested from late November until very early January.  Weed control for cropping is by tillage and chemical spraying using pre- and post-emergent herbicides.  Phosphatic and nitrogenous fertlizers are applied to most cereal crops.  These crops are harvested and the grain is transported by the farmer or contractors to off-farm storage.  A portion of the harvested lupin grain is normally retained on-farm for subsequent use as a feed grain in autumn when paddock feed is of poor quality and sheep require supplementary feeding.

Almost all farms maintain a sheep enterprise.  These animals are raised for wool and sale as live sheep for export or meat production.  The main sheep breed in the region is the Merino.  These are large framed animals which produce adult fleeces in the range of 20 to 22 micron, each fleece weighing between 4 to 6 kilograms.  Lambing is in late autumn or winter and shearing is in spring and autumn.  Young wethers are often sold for export as live animals while ewes are kept for wool and lamb production, eventually being sold for mutton.

Sheep are run on annual pastures during winter and on a combination of crop residues and dry annual pastures in summer (Ferguson 1981, Bell and Ralph 1993).  The pastures contain volunteer annual grasses and herbs, with annual legumes introduced in many situations.  In autumn often feed quality in the paddocks deteriorates to such an extent (Belotti et al. 1993) that supplementary feeding, often with lupins or oats, is required (Rowe 1992).  The quality and quantity of paddock feed, especially autumn feed, affects sheep liveweights' (Tomes and Fairnie 1981, Purser and Southey, 1984) and restricts a farm's carrying capacity.  Depending on seasonal conditions and rotational and grazing history a farm's stocking rate may vary from around two to four dry stock equivalents (DSE) per hectare of winter pasture.

Crops and pastures are commonly grown in rotation.  On the sandier soils of the plateaux, for example, cereal/lupin rotations and cereal/pasture rotations have emerged as profitable options.  On the soils of the valley slopes and floors, cereal rotations that include pasture phases are common.  Less common on these soils are continuous cereal rotations or cereal rotations that include an infrequent phase of field peas.  On soils adjacent to salt lakes barley is the preferred cereal due to its greater salt tolerance.  Because most farms possess a mix of soil types, each with different production characteristics and management requirements, usually a range of rotations is evident on any one farm.  However, farmers rarely rigidly adhere to rotational sequences.  Rotations tend to be altered in response to changes in seasonal weather and commodity prices.

Usually farm businesses maintain equity levels of over 85 percent and have off-farm investments that can be liquidated if the farm business experiences financial difficulties. In each weather-year farm activities are financed through the cash balances of the farm business and by reliance on commercial bill or overdraft facilities acquired from a bank.  Major purchases of equipment or additional land are mostly funded through medium-term bank loans.

A key assumption in discrete stochastic programming is that some decisions are made after a state of nature is observed (Hazell and Norton 1986).  This implies that the farm manager has scope for either avoiding losses in some circumstances or profiting from an unfolding event. Discrete stochastic programming approximates the sequential nature of decision-making in farming.  However, data and model size restrictions invariably cause much simplification of the representation of the actual decision-making process and uncertainty surrounding outcomes.

A schematic decision tree showing examples of continuous choice variables and discrete random events is shown in Figure 4.  Examples of discrete events or states of nature are illustrated in Figure 4 as diamond shapes and are weather conditions in summer and autumn.  In the illustration summers are classed as being wet or dry whereas autumns are classed as very dry, usual or wet.  The two weather possibilities for summer followed by the three weather possibilities for autumn generate 6 (=2x3) possible states of nature.

Facing these states of nature the farmer in this much simplified example needs to make a series of sheep and crop management decisions that are represented as star shapes.  There are decisions about sheep flock size and structure, the buying and selling of feed, the buying and selling of sheep and the areas to commit to crop or pasture.  The discrete states of nature define the universe of outcomes as perceived by the farmer (Featherstone et al. 1990).

To overcome the curse of dimensionality associated with models that extend over several time periods, the model (MUDAS) employed in this analysis was constructed to focus on unfolding weather-year risk.  Discussions with farmers and an examination of farm production data revealed that the dryland farming system of the Merredin region is greatly affected by weather, particularly the amount and pattern of rainfall during a production period.  Accordingly, the construction of the farm model initially focused on defining and describing weather-year states that importantly affected farm management.

Many aspects of weather influence crop and pasture growth (e.g. McCown 1973, de Wit 1986, Ritchie 1991).  Anderson (1991) points out that modelling these aspects has led operations researchers to experience the “curse of dimensionality where there are so many aspects to deal with quantitatively that clear analytical insight is difficult.” (p.4).  He adds, “progress can only readily be made through considerable simplification in order for empirical work to advance on the crude representation of climate as it is contemporarily measured.” (p.4). 

To lessen dimensionality problems associated with representing weather events and their effect on pasture and crop growth, discrete weather states were defined in MUDAS based on their potential to affect farm management, particularly the management of wheat and sheep as these are the dominant enterprises in the Merredin region.  Formal discussions with farmers and later with agricultural advisers at Merredin identified the most important weather events affecting farm management.  Of main importance to all farmers and their advisers were rainfall events.  This emphasis placed by farmers on rainfall events, rather than temperature or wind events, was not surprising because in the dryland farming region of Merredin rainfall is often the main limiting factor for wheat and pasture yields (Cornish 1985, Anderson 1992, Stephens et al. 1994).  The Merredin region’s farming system is solely rainfed; only receiving on average around 230 mm in the growing season and its far eastern parts are identified as marginal wheat-growing lands because of the paucity of growing season rainfall. 

Historical rainfall data for Merredin, growers' sowing rules and output from crop growth simulation models were used to generate discrete states of nature or types of weather-year. The years 1907 to 2002 were classified according to summer or early rainfall (high or low); the commencement of sowing wheat on clay soil (early, mid or late); the initial nature of sowing opportunities (patchy or clean[2]); the duration of sowing opportunities (continuous or discontinuous[3]) and the nature of growing conditions following crop establishment (favourable, unfavourable).  This classification generated 48 possible types of weather-year (2x3x2x2x2). 

Many of the theoretically possible types of weather-year are yet to be observed or have only 1 or 2 observations. The weather-year classes yet to be observed can be omitted from the farming system model.  Further, the weather states with only 1 or 2 observations in the 98 year sequence can defensibly either be overlooked or combined with the closest more common weather-year class.  For example the weather state with little summer and early autumn rain, a late start to sowing on clay soil, patchy sowing opportunities and discontinuous sowing opportunities could be combined with the much more frequent weather state involving the same conditions apart from sowing opportunities being continuous.  The farm management in either weather state is unlikely to differ.  In a late start to crop sowing usually much less wheat is sown, particularly on the clay soils (Kingwell et al. 1993).  The fact that sowing opportunities may be continuous or discontinuous is unlikely to affect sowing decisions, particularly on the clay soils.  The late start to sowing and the resulting likelihood of low grain yield are the main stimuli to reduced areas sown to crop rather that the duration of the sowing opportunity. 

Ignoring weather states that occur with very low probability admittedly means that the full variability in weather states is not represented.  However, the more frequent weather states are those more likely to influence farm management and, as already mentioned, some of the less frequent weather states, even if represented, do not affect farm management decisions.  Combining the much less frequent classes of weather states into their neighbouring more frequent weather-states means that the nomenclature of weather states changes to include terms such as mostly “clean” or mostly “continuous”.  The final grouping of weather-years and associated probabilities of occurrence for two climatic periods are given in Table 1.  The classification yielded 11 weather-year states, A to K.

(a) Bureau of Meteorology data for Merredin (b) CSIRO climate simulation data for Merredin

The objective function in MUDAS is profit maximisation as illustrated in equation (1):         

                                                                            (1)

where x_t is the return to management and capital at terminal state t,

            S_t is the probability of occurrence of ending at terminal state t,

            n is the number of terminal states and

            n equals 11 as there are 11 types of weather-year.

The maximisation of expected farm profit [4] is achieved through selection of an optimal set of farm activities.  These activities draw upon the farm's resources of soil areas, feeds, finances, machinery and labour.  Included in the set of optimal activities are decisions about rotation selection on each soil class, adjustments to crop and pasture areas in certain weather-years, livestock numbers and flock composition, livestock feeding and husbandry in each type of weather-year, machinery and labour use in each weather-year, agistment and grain storage, fertiliser and stocking rate decisions and working capital requirements.  The activity options available to the farm manager are represented as column entries in a data matrix.  The resource and logical limits to activity selection are represented as row entries in the same matrix.

The selection of the set of profit-maximizing activities is also constrained by other managerial considerations.  For example, social mores and leisure preferences cause most farmers to invest in harvest labour and machinery such that harvesting can finish in very early January.  Typically, many farmers holiday off-farm in mid-January for a few weeks.  Also soil conservation concerns restrict the removal of feed by sheep.  Heavily grazed paddocks are susceptible to wind and water erosion.  Lastly, animal welfare concerns limit the loss of liveweight by sheep.  Farmers’ preferences are to avoid their sheep being classed as in poor condition.  These sorts of managerial preferences are reflected in activity coefficients and right-hand side terms of the matrix.  The discrete stochastic programming tableau contains around 1470 columns (or activities), 1180 rows (or constraints) and 28,500 matrix elements.

To illustrate the impact of climate change upon a broadacre farm operating in low rainfall Mediterranean environment, two climate change scenarios are investigated.  The first climate scenario was a climate regime representative of recent farming experience; the period 1975 to 2004.  Probabilities of weather-year classes in this period were derived from a CSIRO data set generated for the Bureau of Meteorology site number 10093 (Merredin Research Station).  The probabilities are shown in the last column of Table 1 and were derived from CSIRO data on daily rainfall, daily maximum temperatures and daily minimum temperatures for 10 simulation runs of 30 years representative of the period 1975 to 2004 in Merredin.  The second climate scenario was based on the probabilities of weather-year classes derived from another CSIRO data set generated for the same site except that it simulated future climate forecast for the period 2035 to 2064. The same weather-year classification criteria were applied to each data set to generate probabilities of occurrence of the various weather-year classes. 

The data that formed each climate scenario were incorporated in a wheat simulation model, validated for the Merredin region (Robinson 1994) to generate estimates of wheat yield on two main soil classes, assuming no varietal or technological change through time.  The means of the wheat yield distributions on each soil class within each weather-year class for the two climate scenarios were compared and differences in the means were used as estimates of impacts of climate change on grain yields, as shown in Table 2.  Percentage differences in means of plant biomass production were also used as indicators of the impacts of climate change on pasture production across the range of soil classes. The pattern of production from oil mallee trees and deep-rooted saltbush species was assumed unaffected by the forecast change in climate.

(a) Numbers in brackets are the average annual rainfall received in years classified into each particular weather-year class, based on the period 1907 to 2002.
(b) These changes are based on those generated for the clay soil classes.  Similar changes in pasture yields apply in many weather-year classes.

The impact of climate change on the distribution of weather-year probabilities and crop and pasture yields is shown in Table 2.  In general, there is a shift away from wetter weather years that typically generate higher crop and pasture yields.  There is also a marked increase in the frequency of weather-year class C.  The main features of such years are an early start to crop sowing followed by dry conditions in spring that adversely affect crop yield but less so pasture production.  The variation in impacts of climate change on crop and pasture yields is due to the specific nature of weather-years, their yield outcomes as generated by the crop simulation models and their classification into a class of weather-year.  In some years the combination of warmer temperatures and reduced winter rainfall enhances crop growth as in these years winter rainfall is not the main limit to crop growth.  In other years where winter rainfall is already low, any further reduction in rainfall or increase in temperatures to reduce the effectiveness of rainfall can reduce crop production.

The following analysis that draws on the data in Table 2 is best viewed as a preliminary investigation of the possible impacts of climate change on a broadacre farm operating in low rainfall Mediterranean environment. Accurately describing the current farming system is a difficult enough task. Forecasting the variety of farm-level impacts associated with climate change is even harder, due to uncertainty about the nature of the climate change and further uncertainty about production impacts that will be a product of management responses and complex biological and physical relationships that will operate in field conditions not yet observed.  Hence, forecasts of impacts often rely on scientific judgement underpinned by process and simulation models that provide insights about the nature and performance of physical and biological systems in possible future environments.  Complex though these models can be, often they cannot describe all the interactions, irreversibilities and eco-system processes that ultimately will play a role in influencing production outcomes.

In this analysis a main assumption is made that climate change in south-western Australia primarily will alter the distribution of weather-year types and their associated crop and pasture yields through changes in rainfall and temperature distributions.  In the low rainfall environment of Merredin, the changes in rainfall and temperature are assumed to be the main determinants of changes in crop and pasture yields.  Other outcomes of climate change that may impact on crop and pasture yields are overlooked.  This is a strong assumption for it means a host of other climate change impacts are not represented in the farm model including:

1. altered weed and pest burdens and plant and animal disease incidence. Changes in their status are possible outcomes of climate change (Penuelas and Filella 2001, Muriel et al. 2000, Fuhrer, 2003).

2. innovation in agricultural technology.  Technological change in response to climate change is likely. By illustration, improving the harvest index and water use efficiency of crops will ameliorate some of the downside yield risk of climate change.  Developing deep-rooted plant species that more efficiently harvest soil moisture will also reduce the impact of reduced growing season rainfall.

3. long term processes such as acidification or salinisation that interact with climate change to affect land quality and resulting production possibilities.  In this analysis land that is already affected by salinity is assumed to remain so and no spread of saline area is assumed.

4. changes in commodity or input price relativities due to climate change.  In this analysis there is no explicit account of these price changes. The farm management issue addressed in this analysis is how can a farmer maximise profit by selecting a mix of strategic and tactical decisions that draw on existing technologies, resources and enterprise options when faced with a different distribution of weather-years and associated changes in crop and pasture yields caused by climate change.

5. a CO₂ fertilisation effect.  There is a large literature on the possible impacts of increased levels of CO₂, especially regarding crop yields (eg Wheeler et al. 1996, Brown and Rosenburg 1997, Howden et al. 2001, Laurila 2001, Newman et al. 2001). Laurila (2001) predicted that the combined effects of higher concentrations of CO₂ and higher temperatures would raise spring wheat yield in Finland by 6%.  However, Olesen and Bindi (2002) reported that Mediterranean areas in Europe would experience lower yields, greater yield variability and a reduction in traditional cropping. Wheeler et al (1996) predicted a decline in winter wheat yield in southern England following a doubling of CO₂ concentration and a 0.8 deg C increase in mean growing season temperature.  Zalud et al. (2000) forecast that barley yield in the Czech Republic would decline by 14% as the beneficial effects of a doubling of the CO₂ concentration was overwhelmed by other adverse temperature related impacts.  
   One of the few studies on pasture production, as influenced by climate change, is that of Newman et al. (2001) who found that temperature and CO₂ concentration favoured production of legume and grass fodder species.
   In Australia there are few published studies providing crop or pasture yield forecasts associated with climate change. Howden et al (2001) examined agriculture in north-east Queensland and concluded that although cropping benefited from the yield-enhancing effects of greater CO₂ concentration, a higher frequency of El Nino conditions could lessen crop production.  The CSIRO (2000) submission to the federal government inquiry into Australia's response to global warming outlined the national wheat production ramifications of various climatic scenarios. Although a doubling of CO₂ levels would boost grain production by 20%, temperature changes of more than 2ºC and reductions of annual rainfall by 20% would largely negate the beneficial effects of CO₂ fertilisation.  In the particular case of the Merredin region we assumed that the forecast increases in temperatures, combined with reduced rainfall, would in many weather-year classes negate any beneficial CO₂ fertilisation effect.
   Nicholls (1997) examined historical wheat yields in Australia and, by relating their trends to temperature changes and CO₂ concentration, concluded that the major beneficial influence of climate change was higher daily minimum temperatures that reduced frost incidence.  Over the period 1952 to 1992 average wheat yield in Australia rose by 0.5 t/ha to be 1.6 t/ha.  Nicholls concluded that about 30 to 50% of the increase in wheat yield was attributable to improved climatic conditions, especially higher daily minimum temperatures.  The higher CO₂ concentration and changes in rainfall had minimal impacts.  His findings are supported by Fuhrer (2003) who in a review of agroecosystem responses to combinations of elevated CO₂, ozone and global warming concluded that the dominant influence on the ecosystem responses would be the climate shifts with their associated changed weather patterns rather than the CO₂ fertilisation effect.

6. altered death rates in sheep flocks.  It could be that warmer winters may lessen lamb deaths due to less wind chill and frosting, while hotter summers may stress animals, reduce their growth rates and increase death rates due to heat stress.  Such possible changes are overlooked.  The main impacts on sheep production are assumed to occur through alteration in pasture yields and subsequent feed availability and feed quality.

The first set of results compares an optimal farm plan in the absence of climate change with one when climate change exists.  Table 3 lists key aspects of optimal farm plans.  The main difference is that climate change is forecast to lower farm profit substantially by 44%.  Over the 50 year period during which the climate change is forecast to unfold this decline in profit equates to an annual decline of around 1.2% per annum. Several other smaller changes accompany the decline in profit.  The area sown to crops declines by 5%, in spite of the area of lupins expanding by 36%.  The stocking rate declines by 17% and sheep numbers decline by 3%.  There is a change in flock structure that allows 3% more sheep to be sold.  More lupin grain is fed to sheep and there is a very large increase in sheep agistment in some weather-year classes during late winter and spring.  There is a less frequent switching of pasture area to cropping on clay soil classes in certain weather-year classes potentially suiting crop production.  However, optimal farm plans remain characterised by occasional large alterations in the area sown crop.  In the most favourable weather-year classes over 70% of the farm is devoted to crop while in many other years around 40% of the farm remains in crop.  There is a 14% reduction in the area planted to oil mallees, while the area devoted to saltland pastures increases by 74%.

Table 3. Optimal farm plans with and without climate change

(a) Expected area of pasture switched into cropping in certain weather-year classes

The effect of climate change is to produce a distribution of weather-years that on average are less conducive to crop and pasture production, causing lower crop and pasture yields. The end result is a small switch in the mix of enterprises away from crops toward pastures, yet overall farm profit declines substantially.  Wheat production and wheat yield decline, sheep numbers fall as does the stocking rate, while more grain feeding and agistment is required.  The opportunity to undertake profitable tactical increases in crop areas on clay soils in certain classes of weather-year is also reduced as these favourable years, as a group, become less frequent.

As shown in Table 4, across the 7 classes of soil that form the farm, there is no alteration in land use, in response to forecast climate change, on soil classes 1, 5 and 7.  However, on all other soil classes there are changes in land use.  Pasture is no longer grown on the sandy soils (soil classes 2 and 3) but rather there is an expansion in crop rotations, especially those including lupins.  Pasture is introduced on the duplex soil (soil class 4) and additional areas are sown to saltland pasture on the valley floor clay soil (soil class 6).