We consider the production of three crops (maize, tomato and potato) under four environments (humid, sub-humid, semi-arid and arid), three hydrologic years (wet, normal and dry year) and three soil types (loam, silty clay loam and sandy loam). We distinguish 20 management packages, whereby each management package is defined as a specific combination of management practices: irrigation technique (furrow, sprinkler, drip or subsurface drip); irrigation strategy (full or deficit irrigation); and mulching practice (no, organic or synthetic mulching).

We develop the marginal cost curves (MCCs) for WF reduction in irrigated crop production in four steps (Fig. 1). First, we calculate the WF of growing a crop under the different environmental conditions and management packages using the AquaCrop model (Raes et al., 2013). Second, the total annual average costs for the management packages were calculated. Third, we constructed plausible WF reduction pathways starting from different initial situations. A WF reduction pathway shows a sequence of complementary measures, stepwise moving from an initial management package to management packages with lower WFs. Finally, MCCs for WF reduction were deduced based on reduction potential and cost-effectiveness of the individual steps. This approach does not aim to represent a cost–benefit analysis from an agro-economic perspective. Reduced costs through water savings are included, but monetary benefits to the farmer through increased yield or product quality are not included. In this way, the approach fully focusses on costs to save water. Yield increases do have a direct impact on final results by reducing the WF per unit of product.

Each management package is a combination of a specific irrigation technique, irrigation strategy and mulching practice. We consider four irrigation techniques, two irrigation strategies and three mulching practices. From the 24 possible combinations, we exclude four unlikely combinations, namely the combinations of furrow and sprinkler techniques with synthetic mulching (with either full or deficit irrigation), leaving 20 management packages considered in this study.

The four irrigation techniques differ considerably in the wetted area generated by irrigation (Ali, 2011). In the analysis, default values from the AquaCrop model are taken for the wetted area for each irrigation, as recommended by Raes et al. (2013). For furrow irrigation, an 80 % wetting percentage is assumed to be representative for a narrow bed furrow, from the indicative range of 60 to 100 %. For the sprinkler, drip and subsurface drip irrigation techniques, wetted areas by irrigation of 100, 30 and 0 %, respectively, are used.

Two irrigation strategies are analysed: full and deficit irrigation. Irrigation requires two principal decisions of scheduling: the volume of water to be irrigated and timing of irrigation. Full irrigation is an irrigation strategy in which the full evaporative demand is met; this strategy aims at maximizing yield. Its irrigation schedule is simulated through automatic generation of the required irrigation to avoid any water stress. The irrigation schedule in the no water stress condition is crop-dependent: the soil moisture is refilled to field capacity when 20, 36 and 30 % of readily available water (RAW) of the soil is depleted for maize, potato and tomato, respectively (FAO, 2012). This scheduling results in a high irrigation frequency, which is impractical in the case of furrow and sprinkler irrigation. To circumvent such unrealistic simulation for the case of furrow and sprinkler irrigation, the simulated irrigation depths are aggregated in such a way that a time gap of a week is maintained between two irrigation events.

Deficit irrigation (DI) is the application of water below the evapotranspiration requirements (Fereres and Soriano, 2007) by limiting water applications particularly during less drought-sensitive growth stages (English, 1990). The deficit strategy is established by reducing the irrigation supply below the full irrigation requirement. We extensively tested various deficit irrigation strategies that fall under two broad categories: (1) regulated deficit irrigation, where a non-uniform water deficit level is applied during the different phenological stages; and (2) sustained deficit irrigation, where the water deficit is managed to be uniform during the whole crop cycle. In the analysis of simulations, the specific deficit strategy that is optimal according to the model experiments and for yield reduction not exceeding 2 % is used. AquaCrop simulates water stress responses triggered by soil moisture depletion using three thresholds for a restraint on canopy expansion, stomatal closure and senescence acceleration (Steduto et al., 2009b).

Mulching is the process of covering the soil surface around a plant to create good-natured conditions for its growth (Lamont et al., 1993; Lamont, 2005). Mulching has various purposes: reduce soil evaporation, control weed incidence and its associated water transpiration, reduce soil compaction, enhance nutrient management and incorporate additional nutrients (McCraw and Motes, 1991; Shaxson and Barber, 2003; Mulumba and Lal, 2008). The AquaCrop model simulates the effect of mulching on evaporation and represents effects of soil organic matter through soil hydraulic properties influencing the soil water balance. Soil evaporation under mulching practices is simulated by scaling the evaporation with a factor that is described by two variables (Raes et al., 2013): the fraction of soil surface covered by mulch (from 0 to 100 %); and a parameter representing mulch material (fm). The correction factor (CF) for the effect of mulching on evaporation is calculated as CF=(1-fm×mc) with mc being the fraction of the soil covered by mulch. We assume a mulching factor fm of 1.0 for synthetic mulching, 0.5 for organic mulching and zero for no mulching as suggested by Raes et al. (2013). Further we take a mulch cover of 100 % for organic and 80 % for synthetic materials, again as suggested in the AquaCrop reference manual (Raes et al., 2013).

The water footprint (WF) of crop production is a volumetric measure of freshwater use for growing a crop, distinguishing between the green WF (consumption of rainwater), blue WF (consumption of irrigation water or consumption of soil moisture from capillary rise) and grey WF (water pollution) (Hoekstra et al., 2011). The green and blue WF, which are the focus in the current study, are together called the consumptive WF. To allow for a comprehensive and systematic assessment of consumptive WF, this study employs the AquaCrop model to estimate green and blue evapotranspiration (ET) and crop yield (Y) to calculate blue and green WF of crop production.

We use the plug-in version of AquaCrop 4.1 (Steduto et al., 2009a; Raes et al., 2011) and determine the crop growing period based on growing degree days. The AquaCrop model simulates the soil water balance in the root zone with a daily time step over the crop growing period (Raes et al., 2012). The fluxes into and from the root zone are runoff, infiltration, evapotranspiration, drainage and capillary rise. The green and blue fractions in total ET are calculated based on the green to blue water ratio in the soil moisture, which in turn is kept track of over time by accounting for how much green and blue water enter the soil moisture, following the accounting procedure as reported in Chukalla et al. (2015).

AquaCrop simulates actual ET and biomass growth based on the type of crop grown (with specific crop parameters), the soil type, climate data such as precipitation and reference ET (ETo), and given water and field management practices. We estimate ETo based on FAO's ETo calculator that uses the Penman–Monteith equation (Allen et al., 1998). The model separates daily ET into crop transpiration (productive) and soil evaporation (non-productive).

Evaporation (E) is calculated by multiplying the reference ET (ETo) by factors that consider the fraction of the soil surface not covered by canopy, and water stress. When the soil surface is soaked by rainfall or irrigation or when soil moisture is beyond a level called readily evaporable water (RAW), the evaporation rate is fully determined by the energy available for soil evaporation (Ritchie, 1972). When soil moisture drops below RAW, the so-called falling rate stage, the evaporation is determined by the available energy and hydraulic properties of the soil field. Experimental studies in different environments have shown that the AquaCrop model reasonably simulates evaporation, transpiration and thus ET (Afshar and Neshat, 2013; Saad et al., 2014).

The crop growth engine of AquaCrop estimates the biomass by multiplying water productivity and transpiration and computes yield by multiplying biomass by the harvest index. Water productivity is assumed to respond to atmospheric evaporative demand and atmospheric CO2 concentration (Steduto et al., 2009a).

We express the WF of crop production in two ways. The green and blue WFs per unit of land (m3 ha-1) are calculated as the green and blue evapotranspiration over the growing period of a crop. The green and blue WFs per unit of production (m3 t-1) are calculated by dividing green or blue evapotranspiration over the growing period of a crop (m3 ha-1) by the crop yield (t ha-1). The crop yield in terms of dry matter per hectare as obtained from the AquaCrop calculations is translated into a fresh crop yield (the marketable yield) per hectare. The dry matter fractions of marketable yield for tomato, potato and maize are estimated to be 7, 25 and 100 %, respectively (Steduto et al., 2012). The variability of green and blue WF is presented by calculating the standard deviation of the estimated WFs across different environments, hydrologic years and soil types.

The overall cost of a management package includes initial capital or investment costs (IC), operation costs (OC), and maintenance costs (MC). Investment costs include costs of installing a new irrigation system and/or buying plastics for synthetic mulching. Operation costs refer to costs for irrigation water, energy and labour. Maintenance costs include labour and material costs. Both OC and MC are expressed as annual cost (USD ha-1 yr-1).

Figure 2 shows the average annual investment cost of irrigation techniques and their lifespan. The data are derived from different sources as specified in Appendices A and B. Investment costs that were reported as one-time instalment costs were converted to equivalent annual costs based on a 5 % interest rate and the lifespan of the techniques. The average annual maintenance cost per irrigation technique – including costs for labour and material – is assumed to be equivalent to 2 % of the annualized investment costs (Kay and Hatcho, 1992).

The average annual investment costs of 1112 USD ha-1 for synthetic mulching are based on the sources as specified in Appendix C. We further assume average operation and maintenance costs of 140 USD ha-1 yr-1 for synthetic mulching and 200 USD ha-1 yr-1 for organic mulching.

The operational cost related to the use of irrigation water is calculated from the amount of irrigation water applied and an average unit price of water (0.09 ± 0.02 USD m-3, Appendix E). The amount of irrigation supply is calculated by dividing the irrigation volume applied at field level simulated by AquaCrop by the application efficiency (Phocaides, 2000). Application efficiency, the ratio of actually applied to supplied irrigation water, is different per irrigation technique (Table 1). The operational cost related to energy use for sprinkler, drip and subsurface drip irrigation is calculated as the total energy demand over the growing season multiplied by the cost of energy (Appendix F). The total energy demand (kWh) is calculated as follows (Kay and Hatcho, 1992): seasonalenergydemand=I×h367η, where I is the volume of irrigation water to be pumped in a crop season (in m3), h the pressure head (in m) given in Table 1 and η the pump efficiency. The pump efficiency can be between 40 and 80 % for a pump running at optimum head and speed and is assumed to be 60 % here (Kay and Hatcho, 1992). Energy required to transport surface water to the field or to pump up groundwater is not included in the estimates.

The operational cost related to labour is calculated as the required labour hours per irrigation event times the number of irrigation events times the cost of labour per hour. The number of irrigation events in the crop growing period is simulated with AquaCrop. The required labour hours per irrigation event are shown in Table 1 and the cost of labour per hour is given in Appendix D.

Uncertainties in the cost estimations are represented by their standard deviation. The standard deviations in the investment and maintenance costs and operational costs for water, energy and labour were systematically combined in calculating the standard deviation for the total cost estimation.

After having calculated the total cost and WF associated with each management package, the MCC for reducing the WF per area or per unit of crop in irrigated agriculture is developed in two steps.

Identify alternative WF reduction pathways by arranging plausible progressive sequences of management packages from a baseline management package to a management package with the smallest WF.

The marginal cost (MC) of a unit WF reduction when shifting from one management package to another is calculated as MC of a unit WF reduction=(TC2-TC1)+(R1-R2)WF1-WF2. We consider both the additional annual cost of the new management package compared to the previous one and the reduced revenue due to crop yield reduction that may result from the new management package. In the equation, TCx refers to the total annual cost of management package x, Rx to the revenue from crop production when applying management package x, and WFx to the water footprint of management package x.

The MCC shows how subsequent WF reductions can be achieved in the most cost-effective way by moving from the baseline management package to another package, and further to yet another package and so on. It shows both cost and WF reduction achieved with each step. With each step, the marginal cost of WF reduction will increase.

The WFs were calculated for four locations (UK, Italy, Spain and Israel), three hydrological years (wet, normal and dry years) and three soil types (loam, silty clay loam, and sandy loam). The input data on climate and soil were collected from four sites: Rothamsted in the UK (52.26∘ N, 0.64∘ E; 69 m above mean sea level), Bologna in Italy (44.57∘ N, 11.53∘ E; 19 m a.m.s.l.), Badajoz in Spain (38.88∘ N, -6.83∘ E; 185 m a.m.s.l.) and Eilat in Israel (29.33∘ N, 34.57∘ E; 12 m a.m.s.l.). These sites are characterized by humid, sub-humid, semi-arid, and arid environments, respectively. Daily observed climatic data (rainfall, minimum and maximum temperature) were extracted from the European Climate Assessment and Dataset (ECAD) (Klein Tank et al., 2002). Wet, normal and dry years were selected from 20 years of daily rainfall data (observed data from the period 1993 to 2012). Daily ETo values for the wet, normal and dry years were derived using FAO's ETo calculator (Raes, 2012). Soil texture data, which are extracted with a resolution of 1×1 km2 from the European Soil Database (Hannam et al., 2009), are used to identify the soil type based on the soil texture triangle calculator (Saxton et al., 1986). The physical characteristics of the soils are taken from the default parameters in AquaCrop. For crop parameters, by and large we take the default values as represented in AquaCrop. However, the rooting depth for maize at the Bologna site is restricted to a maximum of 0.7 m to account for the actual local condition of a shallow groundwater table (average 1.5 m). The main components of the average annual cost per management package have been collected from the literature. We use crop prices per crop and per country averaged over 5 years (2010–2015) from FAOSTAT (2017); the costs for water, labour and energy are averaged over data for Spain, Italy and the UK, i.e. from three of the four countries studied here. An overview of the costs and their sources is presented in Appendices A to F. In presenting the WF estimates per management package, we show averages over the different cases as well as the range of outcomes for the cases (different environments, hydrologic years and soil types). To develop the MCCs, we use the averages.

Figures 3 and 4 show the WF per area and WF per unit of crop, and the annual average costs corresponding to 20 management packages.

For each combination of a certain mulching practice and irrigation strategy, the consumptive WF and the blue WF in particular decrease when we move from sprinkler to furrow to drip and further to subsurface drip irrigation. Under a given irrigation strategy and mulching practice, the WF in m3 ha-1 in the case of subsurface drip irrigation is 6.2–13.3 % smaller than in the case of sprinkler irrigation. The annual average cost always increases from furrow to sprinkler and further to drip and subsurface drip irrigation. Under a given mulching practice and irrigation strategy, the cost in the case of furrow irrigation is 58–63 % smaller than in the case of subsurface drip irrigation. The cost of furrow irrigation is small particularly because of the relatively low investment cost, which is higher for sprinkler and even higher for drip and subsurface drip irrigation. The operational costs, by contrast, are higher for sprinkler and furrow than for drip or subsurface drip irrigation, because of the higher water consumption and thus cost for sprinkler and furrow. Sprinkler has the highest operational cost because it requires a high pressure head to distribute the water (hence the higher energy cost).

Under a given irrigation technique and mulching practice, DI always results in a slightly smaller WF in m3 ha-1 (in the range of 1.6–5.7 %) and lower cost (in the range of 4–14 %) as compared to full irrigation (FI). The decrease in cost is due to the decrease in water and pumping energy. The WF of crop production always decreases in a stepwise way when going from no mulching to organic mulching and then to synthetic mulching, while the costs increase along the move. This cost increase relates to the growing material and labour costs when applying mulching (most with synthetic mulching), but the net cost increase is tempered by the fact that less water and pumping energy will be required.

Figure 5 shows the scatter plot of the 20 management packages, the abscissa and ordinate of each point representing the average annual cost and average WF, respectively, of a particular management package. In this graph, the blue arrow indicates the direction of decreasing WF and costs. The points or management packages connected by the blue line are jointly called the Pareto optimal front or non-dominated Pareto optimal solutions. Moving from one to another management package on the line means that WF will decrease while cost increases, or vice versa, which implies that along this line there will always be a trade-off between the two variables. “Best solutions” may be identified using the MCC when policy goals are specified, for instance a certain WF reduction target in m3 t-1 or m3 ha-1 is to be achieved, or the largest WF reduction is to be achieved with a given limited budget. Each management package that is not on the line can be improved in terms of reducing cost or reducing WF at no cost for the other variable, or even WF reduction and cost decrease can be achieved simultaneously.

In developing a new irrigation scheme or renovating an existing one in a water-scarce area, it would be rational to implement one of the management packages from the Pareto optimal set if the goal is to arrive at a cost-effective minimization of the WF of crop production. In an existing farm, where the management package is not in the Pareto optimal set, there can be alternative pathways towards reducing the WF. This involves a stepwise adoption of complementary measures that eventually leads to a management package in the Pareto optimal set.

Figure 6 shows alternative WF reduction pathways from the two most common baseline management packages: full irrigation and no mulching with either furrow or sprinkler irrigation. The figure shows four WF reduction pathways from the baseline with furrow irrigation and two pathways from the baseline with sprinkler irrigation. In all pathways, the WF of crop production is continually reduced by changing one thing at a time, i.e. either the irrigation technique, the irrigation strategy or the mulching practice. In some cases, a step may be accompanied by a cost reduction, but in the end most steps imply a cost increase. Logically, all pathways end at a point at the Pareto optimal front.

Not all alternative WF reduction pathways from a specific baseline are equally cost-effective. In both cases it makes much sense to move from full to deficit irrigation first, because that reduces the WF and cost at the same time. Next, it is best to move from no to organic mulching because the cost-effectiveness of this measure is very high, which can be measured in the graph (Fig. 6) as the steep slope (high WF reduction per dollar). Finally, the most cost-effective measure, in both cases, is to move towards drip irrigation in combination with synthetic mulching. One could also move to drip irrigation and stay with organic mulching, which is also Pareto optimal; the cost of this will be less, but the WF reduction will be less as well. However, moving to drip irrigation in combination with synthetic mulching is more cost-effective (higher WF reduction per dollar) than moving to drip irrigation while staying with organic mulching.

For both baseline management packages, we have drawn the MCCs in Figs. 7 and 8 for the case of maize. The curves are shown both for reducing the WF per area (Figs. 7a and 8a) and the WF per unit of product (Figs. 7b and 8b). From these curves, we can read the most cost-effective measures that can subsequently be implemented. For each step we can read in the graph what the associated marginal cost is and what the associated WF reduction is. In both cases, the first step goes at a negative cost, i.e. a benefit, while the next steps go at increasing marginal cost. Each step is shown in the form of a bar, with the height and width representing the cost per unit WF reduction and the WF reduction, respectively. The area under a bar represents the total cost of implementing the measure.

For tomato and potato we find similar results as for maize, as shown by the data presented in Appendix G.

In this section, we elaborate a practical application of an MCC for WF reduction, using a selected case with a certain WF reduction target given a situation where the actual WF needs to be reduced given a cut in the WF permit. The future introduction of WF permits to water users or WF benchmarks for products in water-scarce areas is likely if the sustainable development goals (SGDs) are to be met, particularly SDG 6.4, which reads “by 2030, substantially increase water-use efficiency across all sectors and ensure sustainable withdrawals and supply of freshwater to address water scarcity, and substantially reduce the number of people suffering from water scarcity”. Here we will illustrate how an MCC for WF reduction can help in achieving a certain WF reduction goal.

An MCC for WF reduction – ranking measures according to their cost-effectiveness in reducing WF – can be used to estimate what measures can best be taken and what is the associated total cost to achieve a certain WF reduction target. For farmers, it will not be attractive to go beyond the implementation of those WF reduction measures that reduce costs as well, but from a catchment perspective further WF reduction may be required. An MCC will show the societal cost associated with a certain WF reduction goal. Governments, food companies and investors can make use of this information to develop incentive schemes for farmers and/or investment plans to implement the most cost-effective measures in order to achieve a certain WF reduction in a catchment or at a given farm.

In a hypothetical example, the WF in the river basin exceeds the maximum sustainable level. Agriculture in the basin consists of irrigated maize production with a current consumptive WF on the farms of 6380 m3 ha-1. The farms apply sprinkler and full irrigation and no mulching. In order to reduce water consumption in the basin to a sustainable level, the river basin authority proposes various measures including a regulation that prohibits land expansion for crop production and the introduction of a WF permit to the maize farmers that allows them to use no more than 5200 m3 ha-1. This means they have to reduce the WF of maize production by 1180 m3 ha-1. Figure 9 shows how the MCC for WF reduction can help in this hypothetical example to identify what measures can best be taken to reduce the WF by the required amount and what costs will be involved.

As shown in the figure, we best implement deficit irrigation first (providing a total benefit of 189 USD ha-1, which is the net result of a USD 231 gain from saved water and a USD 42 loss from crop yield decline), followed by organic mulching (with a total cost of 72 USD ha-1). The third and last step to finally achieve the required WF reduction can be to implement drip irrigation combined with synthetic mulching on 25 % of the maize fields (at a total cost of 366 USD ha-1). The other 75 % is then still with sprinkler and organic mulching, but the combined result meets the target. Alternatively, because in this particular case the cost-effectiveness of moving to drip irrigation with organic mulching is close to the cost-effectiveness of moving to drip irrigation with synthetic mulching, one could move in the third step in 100 % of the fields to drip irrigation with organic mulching, which would result in a WF reduction of 1176 m3 ha-1. In order to meet the full target, a small percentage of the total fields would need to implement synthetic mulching in addition.