With increasing urban, industrial, and agricultural water demand and projected reduced supply under climate change, allocations to the environment are critically low in many arid and semiarid basins. Consequently, many governments are striving to augment environmental flows, often through market-oriented mechanisms that involve compensating irrigated agriculture, the largest water user in most basins, for reducing diversions. A widely documented challenge with policies to recover water for the environment arises because part of the water diversion reduction can form the basis for downstream consumptive water rights or environmental flows. This article gives an empirical comparison of two incentive policies to acquire water for environmental flows for a part of the Murray-Darling Basin (MDB), Australia. One policy consists of paying irrigators and water delivery firms to make capital and management investments that improve on-farm irrigation and water-conveyance; the other policy consists of having the government buy water from irrigators on the active MDB water market. The results show that the first option results in relatively larger return flow reduction, while the second option tends to induce significant irrigated land retirement with relatively large reductions in consumptive use and small reductions in return flow. In cases where irrigation losses result in little useful return flow (e.g., evaporative loss reduction or during drought in some instances), efficiency-improving investments may provide some cost-effective opportunities. Where a large portion of loss forms valuable return flow, it is difficult to make a case for the cost-effectiveness of policies involving payments for investments in irrigation and conveyance system upgrades.
 Rising demand for fresh water by the agricultural, industrial, and municipal sectors and a finite global supply of water mean that less fresh water is available for the environment. In the 1990s, global freshwater consumption rose more than sixfold, which is twice the rate of population growth, and resulted in one third of the world's population living in countries with moderate to high water stress [United Nations Environment Programme, 1999]. The Colorado River in the United States, for instance, no longer reaches the Sea of Cortez, the Aral Sea is drying up due to a loss of inflows from the Amu Darya and Syr Darya rivers in the Aral Sea Basin, and flows from China's Yellow River no longer reach the Bo Hai Sea [Postel, 2000]. Results from recent climate change models suggest greater water stress from decreased precipitation in many arid and semiarid regions worldwide, including parts of the Middle East, Africa, Australia, and the United States [Shindell et al., 2006].
 Decreased and more variable freshwater supplies to the environment translate into stressed ecosystems and loss of habitat and species. For instance, water withdrawals from the Colorado River have led to downstream losses in the Mexican Riverine Wetlands of over 50%. Consequently, 68% of native fish species have been lost, with 15 rare or endangered fish species going extinct in the last four decades [Lemly et al., 2000]. Water diversions from the Nile River have contributed to the extinction of over 20 fish species and the loss of a fisheries industry supporting over 1 million people [Hinrichsen et al., 1998]. Water withdrawals and storage projects in Australia's Murray-Darling Basin (MDB) have resulted in substantial losses in ecologically significant floodplain forests, diversity of waterbirds, and aquatic vegetation [Lemly et al., 2000]. There is clear evidence that the health of the MDB is in decline [Murray-Darling Basin Ministerial Council, 2001] and the system is facing major threats [Cooperative Research Centre for Freshwater Ecology, 2003]. The annual flows to the sea are very low, and the situation requires immediate attention if the MDB is to be returned to a healthy working river condition [Cooperative Research Centre for Freshwater Ecology, 2003]. Recovering water from existing consumptive water users, however, may involve significant social, economic, and political challenges [Young et al., 2002] with differing distributional impacts [Gippel et al., 2002; Brennan, 2006].
 Agencies worldwide have been engaged in a variety of programs to combat losses in habitat and environmental services arising from water scarcity. Often this involves targeting water use by irrigated agriculture as a source of water. This is not surprising, given that irrigation accounts for the lion's share of water use. For example, irrigation accounts for 85% of all water extractions from the Nile River [Tanji and Kielen, 2002] and 95% of that from the Aral Sea Basin [Dukhovny et al., 2002]. Similarly, 8 of 10 L of surface water in California's Central Valley and 7 of 10 L of surface water in Australia's MDB are allocated to irrigated agriculture. Demand for irrigated agricultural production is likely to expand and intensify in order to meet a predicted 40%–45% increase in demand for food by the year 2025 [United Nations Environment Programme, 1999].
 We evaluate two incentive-based policies that are available to acquire water for environmental purposes from agricultural entities. One evaluated approach mimics entering an existing market to buy annual water rights leases or permanent water rights from irrigators. The water market approach has been gaining favor in some parts of the world. For example, nearly US$61 million has been spent in the United States on leases and purchases of water for the environment [Landry, 1998]. In Australia, in 2008, the Commonwealth Government's Water for the Future plan included a AUD$3.1 billion (henceforth all dollar amounts refer to Australian dollars) commitment over 10 years to buy water in the MDB for environmental purposes (http://www.pm.gov.au/docs/national_plan_water_security.pdf). In the 2008–2009 financial year, the government spent about $50 million in purchasing 35 GL of 75% low-reliability water access entitlements with an average price of $1131 per ML and 25% high-reliability water access entitlements with an average price of $2124 per ML [National Water Commission, 2008].
 The other incentive-based policy for generating environmental water considered involves payments for upgrades in irrigation efficiency and conveyance, defined here as investments to reduce irrigation and conveyance seepage, leakage, evaporation, and drainage losses. Many governments, including that of Australia (http://www.environment.gov.au/minister/wong/2008/pubs/sp20080429.pdf) and that of Oregon [Huffaker and Whittlesey, 2003] focus on this approach because improving efficiency appears to allow for less water use by irrigation without reduced production.
 A handful of previous studies have recognized that attempts to recover water for the environment from improvements of on-farm irrigation efficiency or reductions in irrigation conveyance losses may not lead to real water savings at the basin level if authorities recognize that what is often considered a “loss” is actually a return flow that forms the basis for downstream water rights [Hornbaker and Mapp, 1988; Caswell, 1989; Smith et al., 1996; Grafton and Hussey, 2007]. In the Nile Basin, for example, on-farm and district-level water use efficiencies are extremely low (approximately 30%), and yet the unused water in one district is used downstream by other districts and farms, resulting in an estimated basin-wide efficiency of nearly 80% [Keller, 1992; Cai et al., 2001]. Huffaker and Whittlesey [1995, 2003] showed that while policy measures providing incentives to improve irrigation efficiency may appear to conserve water by reducing withdrawals, in reality these measures redistribute water between river and aquifer with no real water savings. Ahmad et al.  illustrated that field-scale water savings due to adoption of resource conservation technologies such as zero tillage and laser land leveling have contributed to increased net water use at the system level due to field-scale savings being used to establish greater crop area on uncultivated land owned by farmers in the water-scarce Indus Basin of Pakistan. Ward et al.  and Ward and Pulido-Velázquez , in their discussion of possible water conservation options within the Rio Grande Basin, used the term illusionary water savings and stress the importance of accounting for the basin-scale hydrology.
 In the MDB empirical setting considered here, water has been tradeable independent of land since institutional reforms were put in place in 1987. As a result, the market for permanent water entitlements and annual leases in the MDB is now one of the most active in the world [National Water Commission, 2008; Kaczan et al., 2009]. Rights for water in the MDB are defined as quantity diverted, and this amount is tradeable.
 This article quantifies the potential gains from employing a policy that allows more flexibility in terms of how growers and irrigation managers may respond to financial incentives to save and sell water relative to an incentive-based policy that limits the allowable responses to irrigation and conveyance system upgrades. In addition to illustrating that the gains, both in cost savings and the level of environmental water supplied, can be large, our results emphasize two additional issues. First, for regions in which water markets currently exist, opportunities to shore up water for the environment through low-cost efficiency upgrades have likely been dissipated. As such, it is unlikely that water can be purchased for any price less than the current market price. Second, we quantify how much the expected levels of new water supplies for the environment based on gross purchases may differ from actual levels in one Australian catchment when the water balance accounting underpinning water rights ignores return flows. Our results support much of the literature (on biophysical and/or economic aspects) above by showing that very little real water gain can be expected from investments to improve water use efficiency via irrigation improvements and conveyance upgrades. Hence, the anticipated cost-effectiveness of the water savings plan will tend to be greater than the actual cost-effectiveness per unit water saved. We show that assumptions regarding the return flow fraction become less important when water savings arise due to reductions in the consumptive use (i.e., evapotranspiration) rather than nonconsumptive use (e.g., actions that reduce runoff and deep percolation). Given that the more flexible market policy scheme we evaluate generates a significant portion of environmental water through reductions in consumptive use through deficit irrigation and conversion to dryland agriculture, water purchases much better reflect true water savings than do water purchases through irrigation and conveyance system upgrades. This outcome, of course, may not hold under all hydrologic circumstances, including those characterized by very arid river systems with limited ground–surface water connectivity and those for which losses do not form any appreciable usable return flows.
2. Case Study: Incentive Policy for Environmental Water in the Murrumbidgee Catchment
 The Murrumbidgee catchment is located between the Lachlan and Murrumbidgee rivers in southwestern New South Wales, Australia. The main irrigation activities in the catchment are rice, coarse grains, citrus and wine grapes, and pasture for livestock production [Khan, 2004]. Large-scale farming of a single crop, or broadacre cropping, is the predominant user of irrigation water in the catchment; rice has historically been the major user of water, followed by pasture land for livestock and horticulture [Qureshi et al., 2007a]. However, recent droughts have changed land use and, consequently, water use in the catchment. For example, according to the Australian Bureau of Statistics, in 2006, rice was the third-largest (18%) user of irrigated land; the largest users of irrigated land were cereals (39%) and pasture and forage for dairy, beef, and sheep (34%) (http://www.abs.gov.au/AUSSTATS/abs@.nsf/ProductsbyReleaseDate/A770B096742410F1CA25755800122AA9?OpenDocument). The Murrumbidgee is a significant potential source of water for the stressed MDB environment for several reasons: it is a large water user, accounting for 27% of average annual diversion in the southern MDB, it is the largest single source of water on the active annual water lease market, with 13% of all water rights trading out of the basin in 2007–2008, and there is potential for large reductions in water diversions within the Murrumbidgee through improvements in conveyance infrastructure that reduce seepage, leakage, and evaporation [Khan et al., 2004; Khan, 2004, 2005].
 We focus on assessing the relative cost-effectiveness of these options, how much new water each option may actually provide the environment, and how the supply of new water for the environment depends on differences in the fractions of total water use that are not taken up by the crop (i.e., losses) and form useful return flows.
3. Mathematical Programming Model
 To estimate the cost of providing varying levels of environmental water for scenarios A and B discussed above in the introduction section, two separate optimization problems are solved. For scenario A, the model is solved with restrictions on the activities that can reduce water use; in effect, this outcome is intended to represent the water supplier's profit-maximizing response to incentives to supply water savings via activities that reduce conveyance losses and improve irrigation efficiency. For scenario B, the least-cost solution is estimated with no restrictions on water saving activities; this solution is intended to represent the irrigator's profit-maximizing response to a water market in which the government participates so as to buy water for the environment. In effect, the government is buying permanent water rights in an already existing market and retiring those rights.
 To derive the environmental water supply from growers themselves, which we label on-farm activities, we treat the catchment irrigation sector as a single profit-maximizing entity with the objective, in the absence of an environmental water incentive, expressed as
Profits are maximized over choice of Xij, which represents hectares of crop i in production practice (technology or management option) j, and Wij represents irrigation water use level. The parameter pic is the unit price of output (yield) yij. Twelve agricultural activities which occupy about 99% of the irrigated area within the Murrumbidgee catchment are included in our model: rice, grapes, beef, dairy, sheep, oilseeds, deciduous fruits, citrus fruits, legumes, cereals, potatoes, and other vegetables. Conversion to dryland is modeled as the option of withdrawing all irrigation water from a hectare and accepting a relatively conservative average return from dryland wheat of about $100 per hectare (NSW Department of Primary Industries, http://www.dpi.nsw.gov.au/agriculture/farm-business/budgets/about). The index j is for technologies or activities at the field level that target reducing irrigation water applications and runoff. Such activities include improved technology (e.g., drip irrigation) and management practices (e.g., irrigation scheduling). Table 1 summarizes these options as well as their potential water savings and costs.
Table 1. Average Savings, Average Estimated Costs, and Potential Total Savings in the Catchment for Alternative Optionsa
Alternatively, for off-farm entitites, in ML km−2.
Alternatively, for off-farm entities, in $ km−2.
Options Confronting On-Farm Entities
Laser leveling on rice farms
Irrigation flow rate monitoring on rice farms
Irrigation management with fine tuning on horticulture farms
Conversion to drip irrigation system on horticulture farms
Conversion to center pivot irrigation system on horticulture farms
Soil moisture monitoring and irrigation scheduling
Options Confronting Off-Farm Entities
Lining/piping supply channel system
Lining on-farm channels
On-farm recycling and storage
 The unit cost to deliver water to the farm gate is κ. The term μij represents amortized capital costs and operating costs of production, and a component of cost uniquely associated with the irrigation technology and management options (j) (including labor and amortized irrigation capital costs) is represented by cij. The term cij is a function of the amount of activity so as to capture both linear and nonlinear cost effects (functions) of the efficiency improvement options. Some of these options (such as drip irrigation) are represented with linear cost functions, while others (such as soil moisture monitoring and irrigation scheduling) are represented by nonlinear cost functions.
 We develop the cost functions of the on-farm and off-farm savings options (discussed below) based on available data and discussions with local irrigation scientists. Based on the data and expert opinion, we estimate the functions by either (1) fitting curves via regression analysis when we had several observations of water savings and costs or (2) performing linear interpolation between lowest and highest cost for those options where only minimum and maximum cost and water saving parameter values were reported. For example, to establish an increasing cost function of soil moisture monitoring and irrigation scheduling (in $ ML−1 ha−1), we inferred the existing cost and water saving parameter values (given in Table 1) and established an increasing cost function for the option of soil moisture monitoring and irrigation scheduling. Adoption of this option has the potential of saving nearly 97 GL of water. A nonlinear convex (increasing cost) function was established using the range of possible water savings (i.e., 0.1–3.0 ML ha−1) and the relevant cost of water savings for this range (i.e., $1–$220 ML−1). Utilizing this function, cost values were identified against major water savings levels which varied from 0.5 to 3.0 ML ha−1. Each water savings level was multiplied by its respective cost, which was then used to estimate a cumulative cost estimate for each point as well as a cost function [Qureshi et al., 2007b]. Note that the parameters associated with deficit irrigation and the expansion or contraction of dryland farming, both of which are allowed under scenario B but not scenario A, are provided in Table 2 and below.
Table 2. Production Function Parameters Values of Selected Major Agricultural Activities
Pasture for Beef
Pasture for Dairy
Pasture for Sheep
 Yield (t ha−1) is represented by a nonlinear (quadratic) crop water production function as follows:
The term ETij is crop water requirement or evapotranspiration. The amount of irrigation water Wij (ML) required to meet a specific crop's ET (ML) will depend on the effective rainfall, EffRainj (ML), and the on-farm irrigation efficiency, IrriEffiij (a fraction), as follows:
The parameters a, b, and c in equation (2) are estimated using crop-water production functions developed from field-level data on crop water use and yield response [Bryan and Marvanek, 2004] with Food and Agriculture Organization (FAO) crop-yield relationships [Doorenbos and Kassam, 1979]. These relationships have been used to estimate parameter values of each crop-water (quadratic) production function, given in Table 2. The FAO approach has been widely used in previous research in Australia [Jayasuriya and Crean, 2000; Jayasuriya, 2004; Qureshi et al., 2007a]. The rainfall and other climate data for calculating reference crop water requirement or ETij by the Penman-Monteith method were taken from the SILO climate database (http://www.nrm.qld.gov.au/silo/). A base case on-farm irrigation efficiency of 70% in the absence of incentive payments has been used in the analysis. By incorporating flexible and nonlinear crop-water production functions, the option to deficit irrigate is allowed in those circumstances in which it is economical to apply less water than required to maximize yield.
 The maximization problem is solved subject to the constraint that overall water use is less than total available water, TWat,
and overall land use is less than total available land, Tland,
Notice that equation (4) includes water lost to conveyance as represented by the number of units of conveyance technology k, Xk, multiplied by the amount (ML) of water lost per unit of k, Lk. Cost, price, maximum yield, and maximum ETij as well as initial land (ha) and water allocation (ML) are summarized in Table 3. Costs and prices of individual commodities were obtained from the Australian Bureau of Agricultural and Resource Economics (http://www.abare.gov.au/publications_html/acs/acs_07/acs_07.pdf) and the Australian Bureau of Statistics (http://www.abs.gov.au/AUSSTATS/abs@.nsf/ProductsbyReleaseDate/A770B096742410F1CA25755800122AA9?OpenDocument) and other sources, as well as from state agricultural departments. Information on maximum crop yield and crop water requirement per hectare and on irrigated land use for each agricultural activity was obtained from previous studies [Bryan and Marvanek, 2004; Qureshi et al., 2007a]. The water allocation data were calculated from a combination of simulation runs from the MDBC river operations model, MSM-BigMod (A. Close, personal communication, 2005), and land use and crop water requirement information is from Bryan and Marvanek . Five states of nature of the simulated distribution of water allocations reflecting very dry, dry, average, wet, and very wet conditions are considered, each associated with a probability of 20% [Qureshi et al., 2007a]. Furthermore, upper and lower bounds were placed on each individual crop to reflect crop specific land suitability within the catchment. Baseline crop acreage data are for 2001 with a range that allows acreage responses in excess of those observed in 2006 when permanent water prices exceeded $2500 ML−1.
Table 3. Economic and Agronomic Parameter Values of Major Agricultural Activitiesa
Variable Costs ($ ha−1)
Capital Costs ($ ha−1)
Fixed Costs ($ ha−1)
Price ($ t−1)
Maximum Yield (t ha−1)
Maximum ET (mm)
Abbreviation is as follows: ET, evapotranspiration.
Pasture for beef
Pasture for dairy
Pasture for sheep
 The incentive payment to acquire water from irrigators is introduced in equations (6) and (7) by modifying equation (1) to include an activity WM defined as selling a unit (ML) of water in return for a government payment of r ($/ML) and modifying equation (4) so that the sum of water applied and sold to the government for the environment must be less than or equal to the total available water:
As emphasized in section 1, accurate estimation of water supplied to the environment requires accounting for return flows. The following equation is an accounting relationship describing the fate of water applied as irrigation; part of the applied water is consumed by crop, CWij (ML ha−1):
 Typically, some fraction of the applied water not removed through crop evapotranspiration forms useful return flows. The variable fij represents return flows as defined by the amount of applied water above and beyond crop consumption that returns to the system for use by downstream irrigators or for the environment. It is calculated as a fraction, α, of the difference between applied and consumed water. The value of α is typically less than 100% because only some of the water that is not consumed reemerges in the system as water usable for environmental or consumptive purposes. The value (1 − α)(Wij − CWij) represents the amount of applied water not consumed that does not form usable return flows. This includes losses such as drainage or runoff into ground or surface water sources that are of poor quality or inaccessible.
 The potential supply of environmental water from on-farm sources, Son, for a given water price is expressed in the following equation as the difference between the least cost levels of irrigation water use in the absence and presence of government payments after adjusting for return flows:
The least-cost levels for the choice variables in the absence of a government payment (equations (1) and (4)) are denoted as Xij0 and Wij0, whereas the least-cost levels in the presence of such payments (equations (6) and (7)) are denoted as Xij1 and Wij1. Equation (9) is used to trace out the on-farm environmental water supply curve by solving for varying levels of government payment, r.
 In addition to the possibility of on-farm water use reductions, it is possible for entities other than growers (e.g., irrigation managers) to reduce water losses and hence provide water to the government. We label these entities off-farm sources, and their water savings activities include reduction in losses from conveyance (both from the main supply channel and those on-farm channels) and recycling of water on-farm. The estimated scope and cost of available measures for conveyance loss reduction in the Murrumbidgee are shown in Table 1 also.
 To estimate the amount of environmental water a conveyance firm may wish to supply through reductions in conveyance losses or through recycling, we solve an optimization problem by choosing the levels of Xk to maximize the following equation:
where Πoff represents profit to the water conveyance firm from selling Soff (ML) of conveyance loss reductions to the government at a unit price of r. There are other objectives such water handlers might be driven by, such as cost minimization. For these purposes of our study, we assume profit maximization with the understanding that in such drought conditions with a dwindling grower base from which to assess fees, water handlers may very well take advantage of revenue- and profit-generating activities if and when they arise. The total cost per improving water conveyance is represented by the last term in equation (10), which is simply the unit cost per conveyance structure, ck, times the units of structure k, Xk. We assume that the baseline structure is in place and therefore treat its cost as zero.
 The overall supply of environmental water from off-farm sources, Soff, which is a function of Xk, is expressed as follows:
where Xk are units of conveyance structure of type k and Lk are the losses per unit of conveyance structure k. We assume that the baseline conveyance structure is represented by “0” and is the structure that has the highest conveyance losses per unit of structure. Water savings are represented by the reduction in lost conveyance water relative to baseline losses, this difference multiplied a factor, 1 − αk, which represents the fraction of conveyance loss reduction that would not otherwise have been useful return flows. The supply curve for environmental water from reductions in conveyance losses is traced out by solving equation (10) over varying levels of r.
3.1. Modeling Alternative Policy Options
 To evaluate the incentive policies for generating water for the environment as introduced in section 2, we formulate the model in the following manner. To represent the water market policy option, scenario B, all on-farm and off-farm water saving activities (discussed above) are included in the model. This scenario is intended to mimic the results of a government entering the water market (albeit assuming zero transactions costs) and reallocating the purchased water to the environment. A price is identified that represents the government payment, and the model chooses the profit-maximizing water use and water savings; by varying this price, a marginal cost curve for generating different levels of water savings can be generated.
 The analysis for government paying for the efficiency improvement policy option, scenario A, is intended to mimic the costs to the government of paying irrigators and water delivery firms to implement specific technologies or management strategies that improve on-farm irrigation and water conveyance efficiency, reducing seepage, leakage, and evaporation. In particular, the set of activities and management options to generate water savings is limited to irrigation and conveyance system water savings technologies and management investment options outlined in Table 1. The marginal cost curves are generated by choosing the least-cost strategy from the limited set of activities for any particular subsidy (water price) level per amount (ML) of water saved. The difference between scenarios A and B, then, is that scenario B allows growers to generate water savings to sell to the government through engaging in deficit irrigation and conversion from irrigated to dryland production and scenario A does not.
3.2. Accounting for Uncertainty of Understood Return Flows
 As discussed above, water saved through efficiency improvements reduces return flows. The net savings from any action will depend on a number of factors, including before and after irrigation efficiencies, the fraction of applied water consumed by the crop, and the fraction of drainage that eventually returns to the river or groundwater in a form that is available for consumptive or in-stream use. For the base case, we assume that 25% of nonconsumptively used water forms useful return flows. This is consistent with best available information on catchment water balance, which indicates that about 8% of total diversion reemerges downstream [van Dijk et al., 2006], and an assumption that 70% of total applied water in the catchment is utilized and represents ET, while the remaining 30% is drainage loss. Unfortunately, return flows are difficult to measure, and thus there is considerable uncertainty associated with the return flow estimates. To gain insight into the importance of return flow assumptions, sensitivity analysis is performed. In the sensitivity analysis, both greater (50%) and lesser (10%) fractions of nonconsumptively used diversions that form return flows are analyzed.
3.3. Model Assumptions
 Due to the large variations in water required for environmental use each year at particular sites, including years when no such water is required, purchasing annual rather than permanent allocations through seasonal or temporary trading is less costly [Scoccimarro and Collins, 2006]. This line of reasoning suggests that the advantage of using temporary water markets is that the cost of water would reflect seasonal scarcity, with less water being purchased for the environment in dry years. However, it is to be noted that generally there is great uncertainty in temporary markets. Thus the available water and the costs associated with sourcing water through markets are likely to vary considerably. Therefore, the focus of our analysis is on permanent water allocations rather than temporary water markets, and we have converted annualized price into permanent asset value (i.e., perpetual or permanent water entitlement) assuming a 7% discount rate and a 100-year time horizon.
 Furthermore, there can be substantial information and administrative transaction costs associated with entering a market, for both buyers and sellers [Stavins, 1995]. Given that the MDB water market has been functioning for several years and has involved trades of about 25% of the volume of all water rights in 2007–2008, including large volumes of water purchased for the environment by the government, it seems evident that this option involves relatively low transaction costs. We suspect that transactions cost might be higher for an irrigation efficiency improvement payment scheme, as this would require agreements with farmer or water conveyance firms, monitoring of practice implementation, and so on. Consequently, we believe that more accurate accounting for transactions cost would tend to favor the water market approach. However, for simplicity we have assumed a well-functioning water market without transaction costs and with profit-maximizing irrigators.
 A final issue not treated in the current analysis is the information challenge involved in choosing the location for irrigation efficiency investments. The presence of an active water market along with declining allocation levels for irrigation are leading to downsizing and abandonment of some conveyance capacity. A government will never be able to perfectly anticipate where investment in reducing conveyance loss will become redundant because water trade leads to eventual asset abandonment. This suggests that if such a strategy is used, it should be sequenced to follow major structural adjustments to significantly reduced allocations.
4. Results and Discussion
Figure 1 presents estimates of environmental water supply as a function of price for the two modeled scenarios. In Figure 1a, we assume that 25% of the runoff and deep percolation flows reenter the system for use downstream. As shown, the least-cost option is scenario B. Up to a water price of approximately $1000 ML−1, or a supply of nearly 35 GL, the least-cost approach to supplying water to the environment is through low-cost efficiency and conveyance upgrades. Above $1000 ML−1, however, the efficiency upgrades associated with scenario A become very expensive relative to the additional responses allowed under scenario B, that is, deficit irrigation and conversation to dryland agriculture. Furthermore, as is evident in Figure 1a, the ability to generate over 200 of the 352 GL of possible water savings (assuming that center pivot and drip irrigation cannot be used on the same parcel of land) through the use of irrigation and conveyance system upgrades is exorbitantly expensive. Yet, up to approximately 800 GL of water could be purchased through the use of a water market at prices less than $3000 ML−1.
 To better understand the manner and extent to which the on-farm and off-farm technologies are adopted under scenario A, Table 4 provides the amount of water saved under each technology for two hypothetical market prices, $1500 and $3000 ML−1. For a market price of $1500 ML−1, all of the water savings, 57.2 GL, are provided by on-farm entities. Better irrigation flow rate management on rice farms provides 47% of the savings (27 GL), with irrigation management on horticulture farms and better soil moisture monitoring and irrigation scheduling providing the remaining 28% (16 GL) and 25% (14.1 GL), respectively. As the market price increases to $3000 ML−1, all of the on-farm technologies are employed, while the off-farm entities engage in substantial on-farm recycling and storage to save nearly 32% of the 177.4 GL of total water supplied. Only 1 GL of water is supplied via upgrading the supply channel system.
Table 4. Volume of Water Saved Through Efficiency Improvement Options
Table 5 provides a comparison, again using $1500 and $3000 ML−1 water prices, of how the adoption of different agricultural activities varies across the two scenarios. The top section of Table 5 lists the agricultural response, both in terms of total hectares and applied water, for each cropping activity; similar information is listed in the bottom section of Table 5, but for scenario B. Baseline hectares and water use also is provided in the first two columns of numbers. The last row in each section highlights the fact that at a price of $1500 ML−1, nearly 40 GL more water can be generated for the environment under scenario B relative to scenario A, and for $3000 ML−1, nearly 700 GL more water can be generated under scenario B relative to scenario A.
Table 5. Irrigated Area and Water Use Responses Under Alternative Market Prices/Subsidies
 Inspection of Table 5 illustrates how the differences in savings come about. For instance, comparing the baseline to the solution under a $1500 ML−1 price, we see that there is no change in acreage and a slight reduction in water use as growers implement on-farm technologies to save water that increase their irrigation efficiency. Alternatively, under the market option (bottom section of Table 5), both deficit irrigation and conversion to dryland irrigation are noticeable. Some rice acreage (approximately 60,000 hectares) is converted into dryland agriculture to save water. Most of the other savings is derived from deficit irrigation on pasture lands.
 Under scenario A, water use reductions are observed on most crops as water use efficiency levels increase with the adoption of the on-farm technologies. Under scenario B, however, the least-cost solution involves substantial reduction in irrigated hectares (mostly rice) into dryland agriculture, from 102,637 to 41,003 hectares at $3000 ML−1. This conversion, coupled with deficit irrigation, results in a reduction of 778 GL of water, or approximately 56% of the water that was being applied at $1500 ML−1. Similar to the situation with regard to acreage, most of the water reduction occurs on rice crops, which uses 723 fewer GL of water.
 To summarize these initial findings, the total supply attainable under an irrigation and conveyance upgrade subsidy is much less than by sourcing water on the market. The main reason for this result is that under the market scheme, growers have more options to generate real water savings to sell to the government than might be generated under the efficiency improvement policy alone. Indeed, the most water that could be generated for the environment under the efficiency improvement at any price below $6000 ML−1 is approximately 198 GL; alternatively, a more flexible market scheme can generate an estimated 930 GL at those price ranges. The additional water saving options generated by growers under scenario B are primarily the result of switching to dryland agriculture and deficit irrigation. While the two scenarios are similar in their ability to generate the first 50 GL of water savings, further savings under the efficiency improvement scheme are limited and quite expensive relative to the market scheme.
 In the Australian context considered here, active markets already exist for trade in water among irrigators. Consistent with expectations based on economic theory, our results suggest that significant levels of water purchases for environment purposes are likely at current water market prices, even in the absence of any additional incentive policy. Average water market prices in June 2008 were $1590 ML−1 according to the Waterfind Analysis of the Federal Government Buyback (http://www.waterfind.com.au/docs/WaterfindBuybackAnalysis.pdf), and recent trends indicate that the market price may reach close to $2000 ML−1 or above. As shown in Figure 1a, for a market price of $1500 ML−1 it would be profitable for irrigators to adopt practices leading to one third of all practical potential efficiency savings (57 of 188 GL potential savings). There is little incentive in this active market for growers to offer up irrigation efficiency savings in exchange for a government payment that is below the existing water market price. Given the current property rights arrangements in the MDB, these savings would belong to irrigators, who have the right to use it for additional irrigation or to sell it to other irrigators. Under these property rights arrangements, this water would not be available for the environment unless a change in property rights is made.
 Furthermore, if we assume that the market already has responded to the $1500 ML−1 price, that is, supplies available at market prices of less than $1500 ML−1 have been exhausted, the potential for generating additional environmental water differs dramatically across the two policies. Consider a uniform price of $3000 ML−1. Under the efficiency improvement scheme (scenario A), this increase in price is estimated to generate at most an additional 120 GL of water for the environment. In contrast, offering a uniform price of $3000 ML−1 in a water market, and again assuming that water supplies available at market prices of less than $1500 ML−1 have been exhausted, can generate up to 778 GL of water.
 Both the potential water savings from any of the above policies and their expected costs are sensitive to assumptions about return flows. Figure 1b provides an illustration of the extent to which a government might overestimate supply and underestimate cost if it fails to consider this issue in a water balance framework and, consequently, overlooks the role of return flows. Overlooking return flows might lead one to conclude that at a price of $2700 ML−1, about 188 GL of environmental flows could be generated under scenario A. Yet, a comparison with Figure 1a suggests that when return flows are acknowledged, only about 150 GL of water will be supplied. Failure to account for return flows leads to an overestimate of water savings at the price of $2700 by 33 GL (or 25%) and a substantial underestimate of the cost per unit, as it would take a unit price of around $5100 to acquire 188 GL of water for the environment. The lesson here is that overall system water use efficiency, which includes the role return flows play elsewhere in the system, is greater than irrigation efficiency; consequently, changes in irrigation efficiency do not translate into equal changes in system efficiency.
 While Figure 1a presents one estimate of how runoff and deep percolation flows contribute to return flows and system efficiency, we now consider upper- and lower-bound estimates to better understand the possible extent to which these programs may (or may not) generate environmental flows. Figures 1c and 1d present the results from assuming a lower and upper bound on the fraction of runoff and deep percolation flows that return to the catchment. In Figure 1c, we assume that only 10% of the nonconsumptive water returns to the system. Consider first the implications for an irrigation efficiency investment incentive policy. With a smaller fraction of runoff and deep percolation becoming return flows, improvements in irrigation efficiency result in greater net additional water for the environment. The lower return flows assumed in Figure 1c result in irrigation efficiency measures generating nearly 30 GL of additional environmental flows at a price of $3000 ML−1 compared with the outcome illustrated in Figure 1a, which assumes a higher return flow fraction. In contrast, Figure 1d illustrates the cost-effectiveness of acquiring environmental flows when 50% of the runoff and deep percolation flows return to the system (i.e., high system efficiency relative to the base case results shown in Figure 1a). Here, irrigation efficiencies generate 50 GL less water for environmental flows at a price of $3000 ML−1 relative to the base case environmental flows in Figure 1a. Our results suggest that the cost of supplying water for the environment would be 60% greater under a 50% rather than a 10% return flow fraction. A comparison of Figure 1c with Figure 1d illustrates that when return flows are higher, more runoff and deep percolation stay within the catchment and the ability to generate environmental water flows through irrigation efficiency improvements diminishes.
 In contrast, purchasing water through a water market that allows growers more flexibility in how they supply water (e.g., deficit irrigation) was found to be much less sensitive to return flow assumptions than the more restrictive options of scenario A. For example, at a price of $3000 ML−1, assuming a 50% return flow fraction rather than a 25% fraction leads to an overestimate of supply by 5% only, while under scenario A the overestimate is approximately 16%. Similarly, and again at a price of $3000 ML−1, assuming a 10% return flow fraction rather than a 25% fraction leads to an underestimate of supply by 8% under scenario B yet an underestimate of 27% under scenario A.
 The degree to which water acquired through any particular program approximates true water savings will depend largely upon whether those savings arise from reductions in consumptive use (i.e., evapotranspiration) or nonconsumptive use (e.g., runoff). Consider Figure 2, which shows how gross water purchases, reductions in runoff and deep percolation, and reductions in consumptive use vary under different price points for each scenario. At a price of $900 ML−1, there is very little difference between the two policies (as observed in Figure 1a); gross water purchases arise through reductions in runoff primarily. Yet, for prices above $1500 ML−1, Figure 2 suggests that runoff reductions are still the primary source of supply for environmental water under scenario A yet play less of a role in gross water purchases under scenario B. Indeed, as shown under the more flexible market option, as water prices increase, reductions in consumptive use begin to mirror increases in gross water purchases while reductions in runoff and deep percolation become less and less significant. This relationship explains why overlooking return flows is likely to create more of a divergence between expected and actual water savings under scenario A relative to scenario B. Under scenario A, even though diversions are reduced, so too are return flows. A reduction of, say, 10 volume units of water in diversion might be accompanied by a reduction of, say, 5 units in return flows, thus making available for the environment only 50% of the anticipated benefit. In general, the lower the fraction of the return flows, the greater is the benefit to the environment.
 Understanding the role return flows play in overall water supply becomes even more critical under policy proposals in which a government shares water savings with irrigators (i.e., of the total amount saved, a fraction is allocated to the environment and the remaining amount is allocated back to irrigators). Previous government policy in Australia proposed that farmers be paid to adopt more efficient irrigation practices, with half of the reductions in diversions allocated back to irrigators and the other half reverting to the government for environmental flows. The Government of Victoria Food Bowl Modernisation project, alternatively, proposed that one third of the efficiency savings be kept by irrigators, one third go to its capital city, Melbourne, for municipal industrial use, and one third be retained for the environment [Victorian Government, 2007]. When water savings are split between irrigators and the environment and there are high rates of return flows, efforts to generate water for the environment through increases in irrigation efficiency can actually reduce net water available for the environment substantially. This is because return flows already go back to the environment. If the efficiency improvement simply stops the return flows (e.g., by lining a canal to prevent leaks) and directs the “savings” to the environment, it is a straight substitution with no net gain for the environment. If the apparent “saving” is shared between the environment and the irrigators, as is proposed, the environment may actually receive less water than it was originally allocated. For example, suppose we have a crop that needs 70 units of water and an initial irrigation efficiency of 70%. The grower is allocated water so as to apply 100 units, 70% of which is consumed by the crop and 30% runs off the field. Of the runoff portion, a fraction (say 90%, or 27 ML) reappears downstream. If the irrigation efficiency increases to 80%, the grower applies 70 units to meet the crop needs, and 20 units becomes runoff. Following the same 90% return flow assumption, 18 units reappear downstream. When initial return flows are accounted for and deducted from the total savings, the net water for the environment is only 1 unit, that is, (10 + 18) − 27 = 1. If 50% of the saved water is returned to farmers for expansion in irrigation, then the net savings is −4 units, that is, (5 + 18) − 27 = −1. However, most actual cases are unlikely to have such a perverse outcome, which arises only when the apparent saving is shared and is wholly or nearly wholly from leaks that become return flows; yet it is possible in some cases. The observation that savings may often be much less than is apparent is certainly true in general.
 This research investigated the cost-effectiveness and ability of two incentive based policy options available to government to address dwindling environmental flows within the Murray-Darling Basin. Consistent with a long history of research on alternative policy instruments for environmental management [e.g., Dales, 1968; Tietenberg, 1985], we find the use of a flexible water market to be less costly than the use of an irrigation efficiency technology incentive payment alone and to be able to generate nearly five times the amount of environmental flows. The additional cost savings arise due to government taking advantage of low-cost water reduction strategies such as deficit irrigation and land fallowing after the realization of an initial level of irrigation efficiency improvements.
 Hence, expenditures on acquiring environmental flows and the ability to purchase substantially more environmental flows both favor government participation in a water market. The potential exists to induce irrigators to supply some water for the environment with irrigation efficiency technology incentive payments. Yet, in regions for which an active water market exists, such low-cost supplies have most likely been sold on the market and do not exist; hence, efforts to shore up environmental water would require governments to pay prices above the current market price. This is certainly the situation in the MDB, which has had an active water market since the late 1980s, when land use was formally separated from water use and water markets were formed. There is evidence that when incentive approaches targeting water at below-market price have been tried in Australia, the government has struggled to reach its environmental water supply goals as few willing sellers came forward [Wahlquist, 2009] (http://www.theaustralian.news.com.au/story/0,25197,25092093-11949,00.html). Our results suggest that the ability to generate water savings through irrigation efficiency improvements is limited to approximately 143 GL at most, when one considers that there is no incentive to sell the government water for less than the market price for water. Furthermore, these 143 GL of potential savings come with a high price, potentially up to $6000 GL−1, since low-cost irrigation efficiency options have already been implemented in response to existing water market incentives. In contrast, government could expect to source a significant amount of water from markets if it paid above market prices. Our results suggest that for a uniform market price of $3000 ML−1, an additional 733 GL of water for environmental flows could be generated; alternatively, the irrigation efficiency upgrade option is limited to 144 GL of additional environmental flows if a price of $3000 ML−1 were offered (and assuming the market has already cleared by $1500 ML−1).
 We have also stressed and illustrated the importance of acknowledging the physical laws of water balance that exist when considering catchment level management. Policies that overlook the contributions of runoff and deep percolation to return flows may fall far short of generating sufficient flows. Our analysis shows that as the return flow fraction increases, so do the benefits of going to the market for water as compared with making incentive payments directed at uptake of technologies to improve efficiency. The reason for this outcome is that water markets provide opportunities to purchase the consumptive part of water diversions (i.e., evapotranspiration reductions), which results in less water being on-farm without significantly reducing return flows. On the contrary, water purchased through investment in efficiency improving technology primarily results in reductions in the nonconsumptive portion of the applied water, some of which would otherwise have been return flows. Finally, we emphasized that when incentive programs involve water savings being split between irrigators and the environment and there are high rates of return flows, efforts to generate water for the environment through increases in irrigation efficiency can actually reduce net water available for the environment substantially.
 This paper was produced as part of the CSIRO Flagship Program Water for a Healthy Country. John Ward, Jim McColl, Ahmed Hafi, and Mobin ud-Din Ahmad gave valuable comments and provided useful suggestions. We acknowledge Anna Lukasiewicz for collecting useful information about water savings and allocations and Karin Hosking for the editorial changes. The comments and recommendations given by anonymous reviewers, the editor, and the associate editor have helped us in improving the manuscript quality and are gratefully acknowledged.