In 2013 the Brazilian government took a step towards the consolidation of a national irrigation policy through the enactment of Law 12,787 (http://www.planalto.gov.br/CCIVIL_03/_Ato2011-2014/2013/Lei/L12787.htm, last access: 25 November 2019), with two of the objectives being to encourage the expansion of irrigated areas and to increase productivity on an environmentally sustainable basis. According to Law 12,787, policy implementation would have to be based on regional and national plans estimating expansion potential and indicating suitable areas for the prioritisation of public investments. However, to date, a national plan has not yet been developed and the official study available to support the plan is expected to be fully reviewed in 2019 (FEALQ-IICA-MI, 2015). Underlying policy goals include striving for equitable socioeconomic development (VanWey et al., 2013), for a continued large role of biofuels in national energy production and for a strong agricultural sector serving national and international demands of commodities such as soybean (Dalin et al., 2012). One of the governing principles in this policy is the sustainable use and management of land and water resources for irrigation, thereby not negatively affecting communities or sacrificing water resources, unique ecosystems, and the services they provide (Alkimim et al., 2015; Castello and Macedo, 2016; Lathuillière et al., 2016).

The extent to which irrigation is a suitable measure to achieve these goals is debated in the literature. Both Fachinelli and Pereira (2015) and Scarpare et al. (2016) find that in the Paranaíba river basin, covering about 25 % of the Brazilian Cerrado biome, irrigation increases sugarcane yield, in particular in projected expansion areas, but this increase is also in the central region of the basin where sugarcane production is already established. Irrigation shows the potential to reduce costs, thereby enhancing the economic viability of sugarcane expansion. Yet both studies caution not to compromise available water resources and hence to restrict irrigation practices to areas where water is sufficiently available, which, according to Scarpare et al. (2016), generally corresponds to most of the central and western portions of that basin. In a study on the Amazon region Lathuillière et al. (2016) identify that the best land–water management would be one that intensifies agricultural production by expanding cropland into pasture and considering irrigation while avoiding conflicts with downstream users such as electricity producers and reducing pressure on aquatic ecosystems in the Amazon basin. The expansion of rain-fed agriculture in southern Amazonia is known to reduce the water vapour supply to the atmosphere (Lathuillière et al., 2018). Lathuillière et al. (2018) note that this effect could slow down or be reversed by an increase in the vapour supply to the atmosphere following widespread irrigation, but this is not without consequences on surface or groundwater resources.

The Cerrado in central Brazil with a savannah climate is a region with both a strong trend over the last several years of advancing large-scale agribusinesses for agriculture and livestock and the potential for more sustainable land management (Dickie et al., 2016). For example, Alkimim et al. (2015) propose that it is possible to expand sugarcane production in Brazil by converting existing pasturelands into cropland without further environmental losses, whereby they estimate that an area of 50 Mha is moderately or highly suitable for sugarcane production. In another study, Strassburg et al. (2014) assess that current productivity of Brazilian cultivated pasturelands is one third of its potential and that increasing the productivity to one half of the potential would suffice to meet national demands for meat, crops, wood products and biofuels until at least 2040, thereby avoiding the additional conversion of natural ecosystems. Sparovek et al. (2015) analyse comprehensive scenarios with a spatially explicit land-use model for Brazilian agriculture production and nature conservation. They find that a substantial increase in crop production, using an area 1.5–2.7 times the current cropland area, is feasible with much of the new cropland being located on current pastureland.

Land use and land management affect the utilisation of water resources, so every strategy and decision with respect to land is also a strategy and decision with respect to water. This holds for both the precipitation-supplied water stored in the soil matrix (termed green water) and the water in streams, lakes, wetlands and aquifers (termed blue water) (Falkenmark, 1995). While Brazil may be considered well-endowed with water resources, these resources are unevenly distributed across the country. Hence, efficient, sustainable and equitable strategies must be developed, thereby considering the spatially and temporally varying water availability. To that end, Getirana (2016) points out that ineffective energy development and water management policies in Brazil have magnified the impacts of recent severe droughts, which include massive agricultural losses, water supply restrictions and energy rationing.

Metrics of water scarcity and stress have evolved from simple threshold indicators to holistic measures characterising human environments and freshwater sustainability (Damkjaer and Taylor, 2017). The Brazilian national water agency ANA (Agência Nacional de Águas) uses the availability of blue surface water in operational management, whereby the river discharge, partly delivered by regulated reservoir flows, is compared to water withdrawals. ANA distinguishes water scarcity classes based on the risk of river flow to fail to support environmental services (ANA, 2015).

In studying possible expansion of irrigated areas, as encouraged by the Brazilian government under Law 12,787, this paper addresses the trade-off between the choice of the level of blue-water scarcity that is deemed acceptable and the feasible expansion of the irrigated area complying with that limitation. In addressing this issue, we restrict the analysis to irrigation expansion on cropping areas in the production year 2012, representing the situation just before Law 12,787 came into effect in 2013.

Our assessment entails the following steps: i.

the spatially explicit calculation of green- and blue-water consumption for the main crops cultivated in Brazil for both rain-fed and irrigated production systems,

Precipitation, maximum and minimum temperature, solar radiation, relative humidity, and wind speed data for the production year 2012 were obtained from Xavier et al. (2016), who developed a daily gridded dataset for Brazil with a 0.25∘×0.25∘ resolution of these meteorological variables based on 3625 rain gauges and 735 weather stations. In order to determine the required soil properties, data on bulk density, organic carbon content, and fractions of sand, silt and clay have been extracted from the ISRIC (International Soil Reference and Information Centre) SoilGrids1km database (Hengl et al., 2014).

Saturation and residual water content θs and θr (m3 m-3) and the parameters α and n of the van Genuchten function (van Genuchten, 1980) were estimated using the level 3 pedotransfer function of Tomasella et al. (2000) for Brazilian soils, under the assumption that coarse- and fine-sand fractions have an equal share of the total sand content. The field capacity and wilting point were determined as soil water content at -33 and -1500 kPa, respectively, following van Genuchten (1980). Soil types were determined using the nomenclature of the United States Department of Agriculture (USDA). Data on the harvested area and yield of nine main crops for the production year 2012 as provided by IBGE (Instituto Brasileiro de Geografia e Estatística) were utilised in this study. The crops considered are cotton, rice, sugarcane, Vigna spp. and Phaseolus spp. bean, cassava, corn, soybean, and wheat. Combined those nine crops account for 96 % of harvested area (ha), 98 % of production mass (tonne) and 90 % of production value (Brazilian real) in Brazil in the year 2012 (IBGE, 2012). Planting and harvesting dates for the sub-regions considered were taken from Conab (2015). For some crops, multiple harvests per year are considered, following information provided by IBGE. Catchment-scale data on surface water supply were obtained from the ANA GeoNetwork (http://metadados.ana.gov.br/geonetwork/srv/pt/main.home, last access: 25 November 2019). An overview of the underlying data is given in Table 1.

In order to assess water consumption of the potential expansion of irrigation, impacts on water scarcity and limits to irrigation expansion under scarcity thresholds, we applied a site-specific crop water balance model at the catchment scale. To this end, high-resolution gridded data on climate and soil were combined with statistical information on irrigation management to run a countrywide daily crop water balance model for 166 842 sub-catchments in Brazil to determine rain-fed and irrigated water requirements. The crops considered were cotton, rice, sugarcane, Vigna spp. and Phaseolus spp. bean, cassava, corn, soybean, and wheat.

In the present study the biophysical boundaries of the said strategy have been specified in a quantitative manner by comparing the potential water demand to fully cover the water demand of rain-fed areas by irrigation with the available river flows. The underlying environmental and agronomic data were carefully selected to account for the high spatial variation of hydrological conditions across Brazil. A few choices and the resulting implications require further attention.

With respect to the choice of a water availability indicator, Q95 has been selected in order to provide a conservative water availability scenario. This is important due to the high variability of hydrological conditions in Brazil and to account for dry periods over time. Moreover, choosing Q95 complies with the indices utilised by the Brazilian water agency ANA and decision makers.

The selection of crop-specific parameter sets was an important aspect in the design of such a study. Crop coefficients and the length of growing seasons of the individual crops studied here have been assembled from a well-recognised source (Allen et al., 1998; i.e. parameters implemented in the Food and Agriculture Organization of the United Nations (FAO) CROPWAT model), a Brazilian study (Hernandes et al., 2014) and regional information for Brazil, as provided by Companhia Nacional de Abastecimento (Conab) (https://www.conab.gov.br/, last access: 25 November 2019). We acknowledge that further spatial differentiation is desirable, should reliable data be available. We have chosen the procedures put forth by Allen et al. (1998), as their high level of robustness, transferability and repeatability have been shown (Pereira et al., 2015). Moreover, in a large-scale irrigation requirement study for the Murray–Darling basin, Multsch et al. (2013) report that the choice of the potential evapotranspiration calculation method outweighs the role of the local refinement of crop coefficients. Lastly, planting dates are known to change based on the onset of the rainy season (Arvor et al., 2014), which is strong evidence for the use of a window of planting dates based on precipitation regimes different regions. To address this, the actual and region-specific crop calendars (Conab, 2015) were utilised for the determination of crop water requirements to account for varying conditions in different parts of Brazil.

The content of blue soil water and the blue-water fluxes could be further separated into blue water originating from irrigation water and blue water originating from capillary rise, as for example in Chukalla et al. (2015), to track which fractions of ET originate from rainwater, irrigation water and capillary rise, respectively.

An important aspect when assessing water scarcity caused by agricultural water consumption is return flows, e.g. due to evapotranspiration recycling (Berger et al., 2014) or water losses in irrigation systems (Pereira et al., 2002; Jägermeyr et al., 2015). We neglect evapotranspiration recycling effects in the present study, since great care has been taken to subdivide the study area into sub-catchments with sizes where this effect does not play a significant role. The calculated blue-water consumption represents net water requirements, which only includes evapotranspiration by crops and from soils.

Determination of water scarcity was carried out here using the consumption-to-availability ratio. Two aspects require further discussion: the effect of environmental flow requirements and of non-consumptive losses. Environmental flow requirements (EFR) were not included here. Considering EFR results in a reduction of blue-water availability (Boulay et al., 2018; Hoekstra et al., 2012), the water scarcity levels determined here would increase. It is challenging to determine the level of environmental flow requirements for a given region (Hoekstra et al., 2012). Such an analysis is beyond the scope of the current study. A broad range of methods is available in the literature (e.g. Abdi and Yasi, 2015; Hoekstra et al., 2012; Książek et al., 2019; Richter et al., 2012; Smakhtin et al., 2004; Tennant, 1976). In future work it is recommended to select an adequate method to determine EFR and to include such EFRs to carry out a detailed assessment of the impacts of different potential cropping systems on the water cycle, thereby including a quantification of land and water resource trade-offs in the context of agricultural intensification, as suggested by Lathuillière et al. (2018). Losses, e.g. at the field scale and during transport, were considered by adjusting the scarcity levels. The adjustment was based on the work by Wriedt et al. (2009), who estimated gross irrigation demands in the European Union and Switzerland to be 1.3–2.5 times higher than field requirements, depending on the efficiency of transport and irrigation management. To consider these non-consumptive losses, the scarcity levels in the current study were adjusted from those originally used by ANA (2015) (acceptable below 5 %, comfortable between 5 % and 10 %, worrying between 10 % and 20 %, critical between 20 % and 40 %, and very critical above 40 %) using a central factor of 2. Applying the lower (1.3) or higher (2.5) bound found by Wriedt et al. (2009) would result in higher (3.8 %, 7.7 %, 15.4 % and 30.1 %) and lower (2 %, 4 %, 8 % and 16 %) scarcity thresholds, respectively, than those employed here using the factor of 2 (2.5 %, 5 %, 10 % and 20 %).

It is critical to note that pumping river water for irrigation, as investigated here, likely has impacts on natural systems and should be carefully evaluated, thereby considering water management measures. In addition, the effect of land conversion requires attention. Recent studies show the potential effects of future land use and land cover change scenarios in the Amazonian region of Brazil on the hydrological regime in the region (Abe et al., 2018; Dos Santos et al., 2018). The results of the spatially explicit quantification regarding water resources of this study add information on several aspects as explained below.

Based on the assessment of crop water consumption as fraction of water availability (in terms of Q95) and classifying the results regarding water scarcity for Brazil, the following can be concluded:

Avoiding critical water scarcity on currently irrigated land. In order to avoid critical water scarcity, irrigation must be discontinued on 54 % of the area (2.3 Mha) for an acceptable water scarcity level, on 45 % (1.9 Mha) for a comfortable water scarcity level and on 35 % (1.5 Mha) for a worrying water scarcity level of 4.3 Mha of currently irrigated land (not considering multiple cropping).

Expanding irrigation can be an effective measure to increase agricultural production. Using a spatial explicit modelling tool for the sensible, forward-looking and sustainable planning of expansion areas can be achieved by avoiding an expansion in areas where high water scarcity would be the consequence. This applies in particular to the Cerrado biome. Moreover, the temporal variations regarding crop water requirements have been addressed by process-oriented modelling with respect to the local cropping calendar. This work provides a sound basis for further assessment of water management strategies in order to achieve the nationwide development and implementation of sustainable agricultural policies.