In Australia, climate variability and the predicted impact of climate change help in making seasonal rainfall less predictable and seasonal irrigation supplies more uncertain, eroding agricultural production prospects and profitability. Water allocation forecasts have economic value to irrigators for making informed cropping decisions. This study estimated the economic value of improved irrigation allocation forecasts in the Coleambally irrigation area (CIA) in southeastern Australia using a non-linear programming model. The model uses production and profit functions to estimate yield and gross margins for various water allocation levels rather than using given crop yields and gross margins. The model also captures the tactical response of improved cropping decisions made by irrigators based on water allocation announcements throughout the irrigation season. Tactical responses include changing the winter crop combinations, abandoning irrigation for a percentage of the summer crops, temporary purchase or sale of water, and deficit irrigation.
Australia has one of the most highly variable climates in the world that exposes agriculture to considerable challenges and risk (Nelson et al., 2002; Proust et al., 2007). Climatic variability occurs at widely varying temporal and spatial scales. This variability often impacts negatively on agricultural and natural ecosystems (Meinke and Stone, 2005). Irrigated agriculture, in semi-arid regions of Australia, is often dependent upon available irrigation water allocations, which is dependent upon rainfall and the availability of water stored in reservoirs. Climate change and climate variability make seasonal rainfall less predictable and seasonal irrigation supplies more uncertain, eroding agricultural production and farming profitability. Climatic uncertainty often leads to conservative farming strategies that sacrifices productivity to reduce the risk of losses in poor years (Jones et al., 2000). In addition, prolonged droughts and extreme rainfall events can cause abrupt changes in water supplies and worsen the risk pervasive in agriculture (Pannell et al., 2000).
Water is the most important input in agriculture, with the timely and reliable supply of water being a major determinant in cropping decisions (Khan et al., 2006). However, irrigators often have to make key decisions on crop acreage and input investments in the absence of reliable information on water availability. Irrigators risk foregoing their investments in inputs should actual water availability fall short of expected volumes (Isik et al., 2003). Crop insurance helps farmers in cushioning the risk. However, the weather risk markets are practically non-existent in Australia, mainly because of requirements for higher levels of market transparency, adverse selection and moral hazards, and the need for greater effort in the development of local standardized weather indices that can be used for the basis of analysis and product design (Tindall, 2008). In addition to the stressors of climate variability, new environmental water demands have priority for water allocation over agriculture, and are thus putting even further strain on the ever dwindling water resources available to farmers (Ward et al., 2006) and blurring the information available on seasonal allocations to irrigators. Uncertain water allocations also deter irrigators from making long-term investment or entering into seasonal water trading and insurance contracts (Dwyer et al., 2005). Reliable and skilful forecast can provide economically valuable information to irrigators for improved decision making for land and water management. In addition, reduction in prediction error through improved forecasts suggests potential to reduce farm risk by selection of insurance products with the purpose of increasing farm income stability (Cabrera et al., 2006). Furthermore, adjusting index insurance premiums based on improved seasonal climate forecasts may reduce adverse selection and also allow the insurance industry to offer a range of index-based products (Skees et al., 1999).
Inter-seasonal rainfall variation is important to decision making in Australian agriculture and, consequently, research efforts have focussed upon better understanding of core climate patterns and rainfall systems, and associated causal processes at a range of associated time scales. Key climate variables such as sea surface temperature (SST), El Niño-Southern oscillation (ENSO/SOI), Madden-Julian oscillation (MJO) and sub tropical ridge (STR) have been modelled statistically (probabilistically) to predict inter-seasonal rainfall (Orlandini et al., 2006; Meza et al., 2008;) with an increased degree of forecasting skill and reliability. However, recent improvements in our understanding of interactions between the atmosphere and sea and land surfaces, and with further advances in the new generation of dynamic climate forecast models such as the Predictive Ocean Atmosphere Model for Australia (POAMA), suggest a higher degree of predictability and reliability of climate fluctuations over a wider range of temporal scales (Wang et al., 2008). Forthcoming improvements in the POAMA model are addressing both better performance issues at the 3- to 6-month predictive time scales, at finer spatial resolution (eventually down to 12 km), with better incorporation of soil moisture information, which will help provide a fuller understanding of model biases and skill parameters.
Seasonal climate forecasts have been heralded as being as important to improvements in crop production as the Green Revolution (Stone and Meinke, 2005). Indeed, theoretical research has shown that, properly interpreted, probabilistic climate forecasts can lead to increases in long-term average production and decreases in production risk (Carberry et al., 2000) with investment losses minimized in poor years. Research into cropping system simulation models has demonstrated that the use of seasonal climate forecasting techniques can improve farm and cropping systems profitability (Hammer et al., 2000; Stone and Meinke, 2005). This premise is empirically tested in this study by contextualizing the forecasts as a water policy issue and computing the economic value of knowledge on water allocation forecasts to irrigators in the CIA in southeast Australia (Figure 1).
The present study builds on the climate and forecasting model developed by Khan et al. (2005) for the Murrumbidgee Catchment in southeast Australia. The model used SST, SOI, and the ratio of SST/SOI to train an artificial neural network (ANN) model with the capability to accurately forecast the end of the irrigation season water allocation. This paper further extends the modelling work by Khan et al. (2005) for assessing the economic value of improved irrigation allocation forecasts to irrigators. Instead of a static linear relationship used by Khan et al. (2005), we develop a non-linear programming (NLP) model, which can internally derive yield and gross margins under various water allocation levels using production and profit functions rather than using given crop yield and gross margins, and thereby maximising the total gross margin (TGM).
2. Background literature
Some researchers (Pannell et al., 2000; Hamlet et al., 2002) argue that the extra value of forecasts representing risk aversion in farm models is minimal. For example, Hamlet et al. (2002) stated that existing water management decision-making processes, which have evolved in the absence of new kinds of forecast information, may have no pathway to incorporate improved forecasts. As a result, little or no benefit would result from the new information. In some cases, negative values may even be ascribed to the new information as forecast risks are not balanced by increased benefits. Others contest that the value of agrometeorological information is often substantial (Hammer et al., 2000; Phillips et al., 2002). The climate forecast can impact farming through several pathways such as acreage allocation to crops, shifts to high-valued crops, input use, costs, prices, and revenue. Farmers could adjust to climate forecasts by altering a variety of crop-based decisions, such as growing less-water-consumptive crops, planting drought-resistant varieties, altering planting times, or adjusting input usage. By interfacing ENSO/SOI forecasts and crop insurance, Cabrera et al. (2006) found that incomes are greatest and most stable for low risk-averse farmers when catastrophic insurance for cotton and peanut are selected in all ENSO phases. However, a recent study by Meza et al. (2008) reviewed the economic value of seasonal climate forecasts for agriculture and concluded that the climate forecast valuation literature has contributed significant insights into the influence of forecast characteristics, risk attitudes, insurance, policy, and the scale of adoption on the value of forecasts.
2.1. International research
Several studies have estimated the value of agricultural forecasts. For example, Adams et al. (1998, 1995) estimated that ENSO alone can explain 15–35% of global yield variation in wheat, coarse grains, and oilseeds during the past 40 years. The value of perfect (imperfect) forecasts to agriculture in southeastern United States is about US$ 145 million (US$ 96 million). Two separate studies, Meza and Wilks (2003) and Meza et al. (2003), investigated the economic value of operational forecasts using SST for two agricultural regions in Chile. The value of perfect forecasts was generally greater than zero, indicating that El Niño forecasts have considerable economic potential.
At the farm-level, Cabrera et al. (2007) estimated the value of reliable ENSO forecasts to a rain-fed peanut–cotton–corn farm in Florida ranging from US$ 3.8 ha−1 to US$ 6.8 ha−1. The ENSO forecast values were negative during neutral years (< US$ 11 ha−1), while they were positive during the strong El Niño years (US$ 26 ha−1), indicating that highly risk-averse farmers can still benefit by using El Niño forecasts (Kerr, 1998).
2.2. Australian research
Improved forecasts in Australia can offer higher value to users because of its highly variable climate. In two separate studies, Stone and Meinke (2005) and Meinke and Stone (2005) demonstrated that changed decision making through incorporation of seasonal forecasting ultimately improved the long-term performance of the cropping enterprise.
Hammer et al. (2000) estimate that the tactical management of row cotton in an area of Queensland, Australia may increase profit by 11% based on the SOI phase forecasts. The April–May SOI phase has a long-term average economic value of less than, or equal to, the economic value of 1 mm of extra plant-available soil moisture for wheat planting in southern Queensland (Robinson and Butler, 2002).
Marshall et al. (1996) found that the average value of seasonal climate forecasting assisting wheat management was AU$ 3.83 ha−1. McIntosh et al. (2005, 2007) estimated that the expected value of perfect ENSO forecasts for winter wheat and summer cotton in Moree in NSW using SOI phases is about AU$ 166 ha−1. Similarly, Abawi et al. (1995) calculated that the quality losses from weather-damaged wheat cost the Australian wheat industry AU$ 30 million annually. They predicted that the SOI status in May gives a strong indication of relative rainfall in the following spring and summer. The value of relative rainfall information to the producer is predicted to be around AU$ 12 ha−1 per year through improvements in grain quality and reduced losses. Meinke et al. (1998) demonstrated that the use of climate forecasts in managing sorghum and cotton rotations could enhance average gross margins by 14%. Petersen and Fraser (2001) demonstrated that, for wheat farms in Western Australia, a seasonal climate forecast that provided a 30% decrease in seasonal uncertainty could increase annual profit by 5%.
Overall, the value of estimates differs for the given forecasting skill/accuracy, the definition of forecast period, and the typology of farming.
3. Study setting
The benefits of increased knowledge in water allocation forecasts were estimated for the CIA, located in the Murray–Darling Basin (MDB) in the southern New South Wales (Figure 1).
Water for the CIA is stored in Burrinjuck and Blowering Dams and is diverted to the area via the Murrumbidgee River at the Gogeldrie Weir. The average rainfall in the CIA is 400–450 mm year−1. There are over 360 landholdings with a total area of 79 000 ha, and a total bulk license of 629 Gl of water. Over recent years, due largely to the impact of the prolonged drought, water deliveries have declined steadily from 629 Gl during 1996/1997 to 389 Gl during 2005/2006 (Figure 2). Available water supplies are allocated on a priority basis; first to high security water and then to general security water. High security licenses are fewer with town water supplies having the highest security of all consumptive water licenses followed by permanent crops. Irrigators with high security water usually receive close to full entitlement, at 95% during 2005/2006. The general security water represents most of the CIA's farming business.
Land use pattern shows the diverse nature of agriculture in the region (Table I). Crops, including cereal and oilseeds, rice, and annual pastures for grazing and dairy stock are the major land use activities.
In the MDB, a rigorous regulatory framework is used for water allocations. The irrigation season for irrigators in the CIA is from August to the following June. The amount of water that irrigators have access to is based on seasonal allocation announcements made by State Water. An allocation is the percentage of the licence holder's entitlement and is set each year on the basis of the amount of storage in the two main reservoirs, and the minimum expected inflows. The two dams, Burrinjuck and Blowering Dams, have a holding capacity of 1026 and 1631 Gl respectively.
In 1995, the MDB Commission imposed a cap on water diversions. The cap was set at the volume of water that would have been used with 1993/1994 levels of irrigation development, assuming similar climatic conditions for the year in question. Further restrictions on water diversions for irrigators occurred in 1999–2000 when environmental flow rules were enacted, reducing the supply to irrigators by a further 4–5% of entitlement.
The water cap and environmental flow regulations have meant lower water allocations for agriculture, especially for the general security water users. For instance, the average pre-cap long-term general security water allocation was about 112%, while it fell to 72% post-cap (CICL, 2006). The end-of-year general security water allocations over the past 3 years were well below average, from the record low of 40% in the 2003/2004 drought, followed by 37% in 2004/2005 and 53% in 2005/2006 (Table II). Recently, during 2007/2008, the general security allocations have been as low as 15%. Over time, contraction in water supply/allocations has meant higher uncertainty for irrigators.
Table II. CIA general security water allocation for 1990/1991 to 2005/2006
The water licence holders participate in spot water markets and permanent trade in water entitlements (Bjornlund and Rossini, 2007). In the early years of water trading, CIA was a ‘net importer’ of temporary water (Khan et al., 2009), however, in the last 5 years, CIA has been a net exporter of temporary water (Figure 3). This can be attributed to reduced allocations leading to insufficient water for a reasonable summer-cropping program and a significant increase in the price of spot water transfers.
3.2. Water allocation forecast
The first irrigation allocation announcement for general security water users in the CIA is at the beginning of the irrigation season around July–August. This announcement is usually conservative based on high reliability that there is a 99% chance that announced allocations will be available. This judgement is based on the level of risk of a supply shortfall and should be made by individual licensees, and not the government. General security allocations build gradually over the irrigation season as inflows into storages and rivers occur. The pattern of general security allocation for the 2005/2006 period (Figure 4) with a general security irrigation allocation of 53% yields 0.53 Ml per share initially, instead of 1.0 Ml per share.
Researchers have used several techniques to forecast the end of the irrigation season (January) water allocation. Conventional modelling techniques such as probability analysis, regression analysis, and time series analysis have been used. However, recently advanced techniques such as ANN are increasingly being employed to more accurately forecast water allocation (Khan et al., in press).
The probability distribution for general security water allocations helps in characterizing the uncertainty in irrigation water allocations. The probabilities of the modelled general security water allocations can be summarized using a cumulative density function that is based on 100 years of simulated data derived from the IQQM model (Figure 5). For instance, there is a 100% probability that the January allocation will exceed 55%, an 80% probability that the January allocation will exceed 85%, and only a 54% probability that the January allocation will exceed 95%.
The regression analysis also provides an effective technique to forecast general security water allocations. Using non-linear regression analysis, the relationship between October allocation levels and the following January allocation is represented in Figure 6. A high value of R2 (0. 93) indicates that October allocation level could be used as a predictor for end-of-season allocation levels.
Owing to the inherent complexity, the non-linearity of the allocation environments and difficulty in building linear relationships between water allocations of winter and summer periods, the ANN model was deemed as the most appropriate tool for water allocation forecasting (Khan et al., 2005). The neural networks, a simplified model of the biological neuron system, ‘learns’ by mistakes and iteratively minimizes errors through practice. Once it completes its learning, it can be assigned to forecast unseen inputs to predict unavailable outputs (Cancelliere et al., 2002). An application of the ANN model to the January allocation shows good predictive capabilities (Figure 7) with a high correlation coefficient (r = 0.994) between actual and modelled water allocations.
The new dynamic model, the POAMA, which is a state-of-the-art seasonal to inter-annual seasonal forecasting system based on a coupled ocean/atmosphere model and ocean/atmosphere/land observation assimilation system, is being currently developed. However, POAMA products have not been operationalized as yet; with all available products for experimental purposes only. The model has shown substantial skills in preliminary forecasting depending on time and location. No experimental products are currently available for the CIA.
4.1. Conceptual framework
The study develops a non-linear programming model, which models yield and gross margins under various water allocation levels using production and profit functions, rather than using given crop yields and gross margins. This provides estimates of how much economic benefit irrigators can gain from improved end-of-irrigation season allocation forecasts. The model captures the tactical response of improved cropping decisions made by irrigators based on allocation announcements throughout the irrigation season. Tactical responses include changing the winter crop mix, abandoning irrigation for a percentage of the summer crop mix, and the temporary purchase or sale of water. The model maximises the total gross margin, subject to constraints such as water allocation, cropped area, available labour, water delivery, water trading restrictions, and various cropping rotational constraints.
The assumption underlying the analysis was that, with uncertain seasonal water supplies, farm productivity and gross margins are lower as land, water, inputs, and management resources may be allocated less efficiently. The model calculated the opportunity cost of foregone cropping income due to ill-informed cropping decisions. The ill-informed cropping decisions were hypothesized to have emanated basically from a lack of knowledge on irrigation water allocation forecasts, ceteris paribus.
The calculation involved a two-step process. Firstly, the NLP model was run to calculate the maximum TGM based on the irrigator's perceived end-of-season (January) allocation (i.e. the total water supply for the irrigation season was unknown but was estimated by the irrigator). The areas of summer crops chosen by the NLP model become the summer crop areas sown by the farmer. Secondly, to capture the tactical response to the actual end-of-season allocation announcements, the NLP model was run for the second time with total water supply known, forecast by the ANN model, and with the additional constraint that the total area of each summer crop remains unchanged from the first NLP run (i.e. the area of each water-stressed and unstressed summer crop in NLP1 was equal to the area of each summer crop sown in NLP2).
Although summer-cropped areas remain unchanged, the NLP2 differs from NLP1 in three respects. Firstly, water supplies are known; secondly water can be traded; and thirdly, the model assumes that the irrigation required to complete winter crops derived in NLP2, in the following spring, is carried over to the next period, thus impacting the TGM. The opportunity cost of ill-informed cropping decisions was then derived directly by the difference between gross margins of the two model runs: i.e. gross margin with perceived water allocation minus gross margin with forecast water allocation.
To determine the opportunity cost to CIA irrigators of ill-informed cropping decisions, the model was run under alternative scenarios that encompass various initial (October) allocation levels, perceived end-of-season (January) allocation levels by irrigators, and actual end-of-season allocation levels. Decision analysis using the expected regret criterion was applied to ascertain the minimum opportunity cost (or $ regret) that could be expected as a result of an allocation forecast.
A ‘with’ and ‘without’ forecast knowledge framework was used to compute the economic value or benefits of improved knowledge from water allocation forecasts to the irrigators. It was based on the concept of net knowledge benefit, which was measured as the difference between the weighted average of expected values with and without the increased knowledge from forecast. The weights were based on the percentage of CIA irrigators who would participate in each decision alternative, with and without knowledge of water allocation forecast.
4.2. Decision analysis
Decision analysis allows an individual or organization to select a decision from a possible set of decision alternatives when uncertainties exist regarding the future. The goal is to optimize the resulting return or payoff in terms of some decision criterion (Lawrence and Pasternack, 1998).
As noted earlier, initial water allocation announcements are made at the beginning of the irrigation season and are then revised upwards as the irrigation season proceeds (Figure 4). Irrigators have to make a decision on their mix of summer crops around October with the uncertainty of their total water allocation for the rest of the cropping year. An irrigator has to make the decision on whether to base the crop mix on the existing allocation at the start of the season or, alternatively, at some higher level, depending on the level of risk the irrigator is prepared to take. For each decision alternative, there is an associated payoff in terms of achievable net returns. The payoff for each decision alternative is best represented in a payoff table where the columns correspond to the decision alternatives and the rows correspond to the possible future events (also known as states of nature). The states of nature of a payoff table are defined (Lawrence and Pasternack, 1998) so that they are mutually exclusive (at most one state of nature will occur) and collectively exhaustive (at least one state of nature will occur).
The payoff table provides a basis to compute the payoffs from each decision alternative. This analysis uses the expected regret criterion to optimize the payoff from the payoff table. The optimal decision is the one with the minimum expected value on the calculated ‘opportunity cost’ or ‘$ regret’ value corresponding to each payoff. This involves a four-step process:
1.Compute the best value (maximum payoff) for each state of water allocation.
2.Compute the regret for each decision alternative by calculating, for each state of water allocation, the difference between its payoff value and the best payoff value.
3.Estimate the expected regret for each decision alternative by multiplying the probability for each state of water allocation by the associated regret and then summing these products.
4.Select the decision alternative that has the minimum expected regret. Minimizing the expected regret reduces the opportunity cost of ill-informed cropping decisions and thus optimizes the objective function of maximizing the total gross margin.
4.2.1. Non-linear programming model
An optimal farming plan for the whole CIA was generated with the aim to maximize the aggregate gross margin (total revenue less variable costs) from crop production and water trading subject to several land, water, technical, and administrative constraints. The model is represented as
where i are the number of crops grown in the whole CIA, r represents irrigation technology; Pi is output prices (AU$ t−1) of crop i, Yir is crop yield (t ha−1) grown with technology r; Xir is the area devoted to different crops (ha) with technology r; Cir is the cost of variable inputs other than water costs (AU$ ha−1) with technology r; Ps is surface water price (AU$ Ml−1); Pg is the groundwater pumping cost (AU$ Ml−1); Wi is the crop water use (Ml ha−1) of i crop with r technology; and γ is the percentage of water from surface water (γ = 1 implies just surface water was used); Pt is the temporary water trading prices (AU$ Ml−1), and Wt is the quantity of water trading, buying Wt < 0 or selling Wt > 0 in Ml.
The yield of main crops in response to water use was obtained by developing crop production functions. Crop yields for the various crops were estimated using the yearly Griffith rainfall data for the years 1962–2001 and applying irrigation of specified amounts at set dates during the growing period. Total water inputs, i.e. irrigation plus rainfall–crop yield production functions, were derived for various crops using the SWAGMAN–Destiny model. The production functions (Table III) were derived by fitting the following non-linear curve using ordinary least squares (OLS) regression analysis.
Table III. Estimated water yield production function for the CIA
where, Yir is the yield of crop i with technology r, Wir is the total water used by crop i with technology r, and βi are coefficients of total water use, β0 is a constant, and εi is the error term. An example for maize production function is presented in Figure 8.
The value of temporary traded water was estimated using a regression model involving monthly average trade water price and water allocation data from the year 2001/2002 to 2005/2006. The resulting high R2 value (0.93) indicates that water allocation is the key factor determining water market price (Figure 9). The following estimated function was used in the NLP Model:
where A is the general security water allocation (%) in a given month.
4.2.2. Model constraints
22.214.171.124. Surface water constraint.
Total surface water use must not exceed the corresponding announced water allocation for the year, as shown below:
where NCWR is the net crop water requirement (Ml ha−1) of crop i with technology r, SW is the total surface water entitlement for CIA (Figure 2), ω is the general security water announced allocation, which is a fraction of the total surface water entitlement.
126.96.36.199. Groundwater constraint.
Groundwater licenses/withdrawal of water should not increase above the minimum sustainable yield, as represented through
where GWsy is the sustainable groundwater, which is based on a groundwater sharing plan. Groundwater use of 206 000 Ml during 2000–2001 was lower than the announced allocation (471 200 Ml) for the same year (Kumar, 2002).
188.8.131.52. Land constraints.
Land allocated to various crops under different irrigation technologies must not exceed the total available cultivable area during the summer and winter seasons, as show in
where TA is the total cultivated area available, which is about 79 000 ha in CIA.
184.108.40.206. Allowable area constraint.
Management considerations, market conditions, machinery capacity of the farm, and climatic conditions restrict the minimum or maximum land acreages under certain crops such as rice to meet the regulations on local land use in the area. For instance
where µminand µmax are minimum and maximum fractions of the cultivated area allowed under crop i and technology r respectively.
220.127.116.11. Water market constraints.
Water markets in Australia are subject to a set of rules and regulations. These involve placing limits on where water can be traded and the mechanisms for establishing the price, as well as limiting the maximum tradable volumes (Khan et al., 2009). For instance
where Φmax is a fraction of the total allocation that can be traded in temporary water markets.
18.104.22.168. Non-negativity constraint.
The non-negativity constraints are given in Equation (9).
This constraint ensures that the solution remains feasible.
22.214.171.124. Model Assumptions.
The following assumptions were used in the NLP model:
Allocation water charges: AU$ 35.65 Ml−1 (includes fixed and variable charges)
The optimization models have a short-term focus and were estimated under the assumption of a relevant output price range and relatively inelastic demand for water.
Average yield (kg ha−1), crop water use (Ml ha−1) and gross margin (AU$ ha−1) are shown in Table IV.
It is desirable when dealing with optimization problems that the global optimum be found. To search for the global rather than local optimum, we used the median values of crop water requirements (Table IV) as starting values. The non-linear optimization was carried out using Matlab 7.1 software (Matlab, 2005).
Table IV. Yield, crop water use, and gross margin per hectare in the CIA
5.1. Economic value of forecasts with perfect knowledge
The economic value of an end-of-season allocation forecast with perfect knowledge was measured in terms of the change in total gross margins associated with a change in forecast knowledge. Assuming that irrigators have perfect knowledge of the end-of-season allocation level, for an end-of-season allocation of 100%, the potential total gross margin for the CIA estimated by the NLP model is AU$ 47.6 million. As allocations decrease, total gross margins for the CIA decrease in a non-linear pattern, with a 1% decrease in allocation resulting in a AU$ 130 983 fall in the total gross margin (Figure 10), assuming that all model parameters remain constant other than water allocation levels and the price of temporary traded water.
The marginal value of water can be used as an indicator of the maximum price that irrigators in the CIA collectively would be prepared to pay to secure additional water at the beginning of the irrigation season given the average crop prices. In the CIA, the marginal value for irrigation water, the increase in the total gross margin for an extra megalitre of water at a particular allocation level is AU$ 26 Ml−1 for a 100% allocation, and AU$ 75 Ml−1 for allocation levels between 50 and 80% (Figure 10). The marginal value rises sharply for allocation levels of 10 to 30% with the highest value $ AU251 Ml−1 at 20% allocation; however, marginal values decline to a very low level, $ 48 Ml−1, at above 0% allocation levels.
The rapid rise in the marginal value was mainly because of high water market prices, which provide irrigators opportunities to further increase their farm income by taking out water planned for low value crops and selling it in water markets, rather than buying. The rapid decline in the marginal value of water when allocation levels fall below 10% is probably due to land and crop rotational constraints. For instance, minimum pasture requirements for sustaining livestock force irrigators to buy water at higher prices even though it would be economically inefficient. On the other hand, the low marginal values at high water allocations are attributed to land use policy. For instance, the maximum area planted to rice offsets further productive use of increased water supply. The sharp decline in the marginal value of water after 10% of water allocation (Figure 10) denotes the suboptimal use of the production resources.
The economic value/benefit of an end-of-season allocation forecast is the difference between the value of agricultural production with and without the forecast knowledge. An approximate measure of the benefit of an end-of-season allocation forecast can be derived from Figure 10. For example, if CIA irrigators on average base their cropping decisions on an end-of-season allocation of 60% with given current knowledge, production will equate to AU$ 34.68 million, but if irrigators act upon an accurate forecast of a 70% allocation, production will be AU$ 38.67 million—a net economic value of forecast to the CIA of AU$ 3.79 million. This assumes that the forecast is early in the season so that irrigators can choose the crop mix that maximizes the total gross margins, and that cropping decisions remain unchanged throughout the irrigation season regardless of any changes in allocation levels.
The estimate of economic value forecasts under this scenario are somewhat unrealistic, as it does not account for any tactical adjustments in farm management as the end-of-season allocation becomes more certain. More realistic scenarios to assess the opportunity cost of agricultural production from cropping decisions based on allocation announcements would need to consider the timing of the allocation announcements, irrigators' perception of the end-of-season allocation, the level of risk that irrigators are prepared to take when allocating crop areas to perceived allocation levels, and tactical responses to better information and/or changed allocation conditions. A more realistic scenario to estimate the economic value of forecasts with imperfect knowledge is given below.
5.2. The economic value of forecasts with imperfect knowledge
When estimating the economic value of an end-of-season allocation forecast based on imperfect water allocation knowledge, two situations can be hypothesized:
5.2.1. Pessimistic water allocation outlook
This results in foregone agricultural productivity by irrigators who underestimate end-of-season allocations. A risk-averse farmer may base cropping decisions on existing initial allocation levels and may not be prepared to predict further increases in water allocation. As a result, cropping decisions are most likely not to be the optimum ones in terms of gross margins. Thus, if further increases in water allocation are announced, this will result in foregone agricultural productivity.
5.2.2. Optimistic water allocation outlook
This implies a loss in agricultural productivity (i.e. yield loss, loss or failure of crop, and non-optimal yearly crop mix) when irrigators overestimate end-of-season water allocations. This occurs when a farmer is prepared to take some level of risk based on cropping decisions on some perceived level of end-of-season allocation. As a result, the irrigator may overestimate the actual end-of-season allocation and therefore is required to use various tactical responses to the cropping program such as purchasing temporary water, increasing the timings between irrigations, or even adopting deficit irrigation practices to overcome the shortfall in their water requirements, resulting in lost agricultural productivity.
The economic value of forecasts without perfect knowledge was estimated for the pessimistic and optimistic water allocation outlooks. The payoff matrix in Table V illustrates the opportunity cost (or $ regret) of agricultural production for the CIA from cropping decisions based on the perceived and actual end-of-season allocation when the October allocation is between 50 and 60%. The columns correspond to the possible decision alternatives (the perceived January allocation on which cropping decision are made) and the rows correspond to the possible future events or states of nature (these being the actual January allocations). The matrix contains the payoffs resulting from a particular decision alternative when the corresponding state of nature occurs. For example, if the irrigators in the CIA based their cropping decisions on a (optimistic) perceived January allocation of 70%, but the actual January allocation was 60%, the minimum opportunity cost after tactical responses by the irrigators is AU$ 8.9 million. However, if the actual January allocation was 75% against the (pessimistic) perceived January allocation of 70%, the minimum opportunity cost after tactical responses by the irrigators is AU$ 1.6 million. Thus, the lost gross margin (AU$ 8.9 million) far exceeds the foregone gross margin (AU$ 1.6 million) despite tactical responses implying that farmers would lose their investments in variable costs should actual water allocations fall short of the perceived allocation.
Table V. Payoff matrix when October allocation is between 50 and 60% in the CIA
Actual end-of-season allocation (%)
Actual allocation range (%)
Probability of exceedance of water allocation (%)
Farmers estimated January allocation (AU$ million)
Expected value (AU$ million)
Losses from inaccurate irrigation allocation outlooks are avoidable. A zero opportunity cost or regret occurs when irrigators correctly estimate the January allocation. This is because correctly estimating the actual end-of-season allocation is the optimum decision alternative for each state of nature. Alternatively, cropping decisions based on an allocation level that is either above or below the actual end-of season allocation can result in lost agricultural production. Nevertheless, the losses far exceed those when water allocations are overestimated as resources and inputs are overly committed to execute the ill-informed cropping plans. A risk-averse farmer would attempt to reduce such losses by choosing the lowest expected regret value.
The expected regret criterion was used to determine the decision alternative with the lowest overall payoff. The probability of exceedance estimates were used to compute the regret value for the future water allocation events (actual end-of-season allocation). These estimates are based on the allocation data simulated by the hydrological modelling tool called Integrated Quantity and Quality Model (IQQM) developed by the NSW Department of Land and Water Conservation, Australia (1999). The expected value for each decision alternative was calculated by multiplying the probability of exceedance of each water allocation (Figure 5) by the associated payoff and then summing these products, as explained earlier. Using the expected regret criterion, the decision maker would select the decision alternative with the lowest expected value. For example, when the initial water allocation is between 50 and 60%, CIA irrigators would achieve the lowest expected value (i.e. opportunity cost of lost agriculture production) of AU$ 0.23 million if they collectively made their cropping decisions based on a 65% end-of-season water allocation (Table V). This is an allocation level that is only 5% greater than the initial allocation level.
The calculation of the benefits of increased knowledge from the forecast was based on the net knowledge benefit, which is the expected value of the knowledge benefit for each initial allocation percentile band. The knowledge benefit for each initial allocation percentile band is the difference between the weighted average of expected values with and without the increased knowledge of forecasts. Weightings are based on the percentage of the irrigators who would participate in each decision alternative, with and without the knowledge of the better forecasts. The level of risk that irrigators are willing to take in the CIA in making their cropping decisions based on allocation information remains unknown. However, it was assumed that they are relatively risk averse and that one-third of irrigators will base their cropping decisions on perceived end-of-season allocations that are 0, 5, and 10% above the initial allocation level. With the forecast information, it was assumed that all irrigators will base their cropping decisions on whichever perceived end-of-season allocation had the lowest expected value.
The total benefit of forecast knowledge is the sum of the expected value benefit for each initial allocation percentile band (Table VI). The total benefit attributed to the forecast knowledge for the CIA was estimated to be AU$ 305 949 per year, or AU$ 3.97 ha−1. The net knowledge benefits of forecast knowledge are greater at the lower end of allocation level due to the relative water scarcity; but, for allocation levels exceeding 80%, the net knowledge benefits are near zero due to less water scarcity as well as area restrictions on crops with higher gross margins such as rice. The zero knowledge benefit when initial allocations exceeded 90% implies that there was enough flexibility in the tactical response options to achieve the potential total gross margin for all decision alternatives.
Table VI. Expected value benefit of better forecast information in the CIA
Start of season allocation (%)
Knowledge benefit (AU$ million)
Probability of start-of-season allocation (%)
Net knowledge benefit (AU$ million)
5.3. Sensitivity analyses
The robustness of the results was tested using a sensitivity analysis by looking into the change in knowledge benefit due to changes in the percentage of irrigators for each decision alternative. The sensitivity analysis therefore helps in accounting for the level of risk (unknown) that irrigators are willing to take in the CIA in making their cropping decisions based on allocation information. With forecast knowledge, it was assumed that all irrigators will base their cropping decisions on the perceived end-of-season allocation that has the lowest expected regret value.
Should the irrigators of the CIA be totally risk averse, they will base their cropping decisions on the existing initial water allocation. As a result, there will be a knowledge benefit of AU$ 206 304 from the forecast (Scenario 8 in Table VII). The more the risk that irrigators take when estimating the end-of-season allocation, the greater is the knowledge benefit from the forecast. If all irrigators are prepared to take some level of risk in estimating the water supply for the year, for instance 10% above the announced initial allocation, the net knowledge benefit to the CIA becomes AU$ 738 195 per year (Scenario 1 in Table VII). As farmers are generally risk averse, the actual knowledge benefit to the CIA would probably be somewhere between the totally risk averse value of AU$ 103 152 (AU$ 1.33 ha−1) (Scenario 6 in Table VII) and the marginally risk value of AU$ 738 195 (AU$ 9.58 ha−1) (Scenario 1 in Table VII). The sensitivity analysis confirms that the knowledge benefits from the forecasts are positive, the estimates are robust to risk-averse behaviour for a battery of allocation levels, and the water allocation forecasts have substantial economic value in aggregate terms.
Table VII. Sensitivity of net knowledge benefit to percentage of irrigators for the CIA each decision alternative
Percentage of irrigators for each decision alternative (percentage above the initial allocation)
Net knowledge benefit
Total (AU$ million)
In Australia, especially in the MDB, a highly variable climate and the looming threat of climate change has intensified the scramble for water. Alongside irrigation, the water demand exceeds supply, and new environmental pressures are emerging for reallocating more water to the environment. Consequently, seasonal water allocations are becoming less and less predictable. Irrigators are bearing the brunt of their ‘double exposure’ to climate change and climate variability, with their historical water allocations reduced to near zero levels in 2006/2007, resulting in lost investment and profitability given the limited or no information on future water allocations. Improved water allocation forecasts can deliver significant economic benefits to irrigators through more informed and risk-averse cropping decisions.
Against this backdrop, the economic value of knowledge of irrigation water allocation forecasts was evaluated, taking a large irrigation area in the southern MDB as an example. An NLP model was used to maximize gross margins, subject to various resource constraints (including cropped area, crop rotations and state restrictions on the area under certain crops, water allocation levels, and knowledge from information on water allocation forecasts). The economic analysis of irrigation allocation forecasts shows that the potential total gross return for the irrigators at a 100% allocation level was AU$ 47.56 million (AU$ 1 = US$ 0.77) provided that irrigators have perfect prior knowledge of the end-of-season allocation level. For every 1% fall in allocation, the potential total gross margin fell by around AU$ 130 983. The actual total gross margin would fall below its potential total gross margin due to production inefficiencies from uncertainty of the total water supply for the irrigation season, unless there is perfect knowledge about water allocations. The perfect knowledge scenario is however somewhat unrealistic, as it does not account for the tactical responses in farm management practices due to relative water shortages.
A more realistic scenario to estimate the economic value of forecasts would be the imperfect knowledge case with (1) a pessimistic water allocation outlook and (2) an optimistic water allocation outlook. The modelling results show that if irrigators based their cropping decisions on an (optimistic) perceived January allocation of 70%, but where actual January allocation was 60%, the minimum opportunity cost after tactical responses by the irrigators was AU$ 8.9 million. However, if the actual January allocation was 75% against the (pessimistic) perceived January allocation of 70%, the minimum opportunity cost after tactical responses by the irrigators was AU$ 1.6 million. This suggests that losses far exceed those when water allocations are over-estimated by irrigators, as resources and inputs are overly committed to execute the ill-informed cropping decisions. Better allocation forecasts can help reduce such losses. The knowledge benefits from the forecasts are greater at the lower ends of allocations due to relative water scarcity, but, for allocation levels exceeding 90%, the knowledge benefits are near zero due to higher water availability. This suggests that investments in water allocation forecasts and related agrometeorological information could be a very useful water policy tool, especially when water allocations are in the low-to-middle range. When water allocations are plentiful, economic value of forecasts may be low.
Farmers respond to predictions of drought or forecasts of low water allocations and signs of water stress through adaptations of their practices to climatic variability. The adaptations are costly and require significant investment. Responses by farmers may also generate new opportunities and markets. For instance, some farmers in the CIA sold hay for the first time during the 2006–2007 drought and earned higher gross margins than from rice crops in a pre- or post-drought year. Others switched from surface to drip irrigation for orchard production, while the pruning of citrus blooms from outer canopy layers saved water and improved fruit quality and price. The sale of ‘saved water’ and gains in gross margin from premium price were enough to payback the investment immediately. Until now, seasonal climate forecasts have not played a role in the structure and pricing of index-based weather insurance derivatives and contracts. Efficiently designed insurance schemes can help take advantage of climate forecasts and reduce financial risk for insured farmers and insuring companies, as well as promote risk reduction.
The analysis presented here therefore tells only part of the story as it does not include farmers' responses to climate variability or the impact of prices and government support measures. The analysis can be extended by considering these aspects and also incorporating state or federal farm programs such as the Australian Government drought assistance package. Nevertheless, these results support the argument that irrigation service providers must establish agrometeorological and water forecasts systems in major irrigation system.