Introduction

Global climate modeling and advances in the physical climate process description and resolution have brought about better estimates and a reduction of uncertainties (Cubasch et al. 2013). The Intergovernmental Panel on Climate Change’s (IPCC) 5th Assessment Report (AR5) called Coupled Model Intercomparison Project Phase 5 (CMIP5) is based on global circulation models (GCMs) and state-of-the-art Earth system models (ESMs), and includes a representation of biogeochemical cycles and high-performance computers (Flato et al. 2013). The horizontal resolution assumes that a latitude and longitude of 1° represents approximately 111 km. It also includes a vertical resolution of the atmosphere up to 95 levels, which comprises a wide range of atmospheric processes. The CMIP5 model components include atmosphere (up to 52,000 horizontal grid points), land surface, ocean (up to 110,000 horizontal grid points), sea ice, aerosols, dynamic vegetation, atmospheric chemistry, carbon cycle, and ocean biogeochemistry cycles (Flato et al. 2013; Cubasch et al. 2013).

Higher temperatures due to climate change are expected to impact evapotranspiration, which affects water demand by crops. However, reservoir water inflow depends on rainfall and, thus, water availability for irrigation and other demand sectors. Climate change and food demand may act as additional drivers for water resource conflicts due to increasing water needs, especially for irrigated agriculture.

Studies about climate change and irrigation water demand have been performed globally. Some have also simultaneously analyzed crop area growth as a result of the need to increase food production. Panagoulia and Bili (2002) found that net irrigation requirement increased for all scenarios (varying from 11 to 45%) in 45 cultivation areas and growing seasons due to higher evapotranspiration and crop coefficients. An increase in irrigated area was observed because of the change of the rain-fed system to irrigated crop. According to Panagoulia (2004), part of the water demand could be met by increasing irrigation efficiency, but a new water supply is needed in this case. Maeda et al. (2011) indicated that the total annual volume of the irrigation water required was likely to decrease in Eastern Kenya during the next 20 years. On the other hand, the tendency to expand agriculture is expected to aggravate water scarcity problems. No appreciable change in total irrigation water demand due to the shortening of the irrigation period for rice crops in Bangladesh was reported by Shahid (2011); however, climate change will increase the irrigation rate or daily use of water for irrigation, affecting the groundwater level. In addition, irrigation water demand impacted by climate change may vary by geographic location and environment peculiarities. Wada et al. (2013) reported results that ranged from slight decreases and increases to considerable increases in water irrigation demand, as higher greenhouse gas concentration scenarios were considered. The study also concluded that the magnitude of the increase depends on the degree of global warming and associated precipitation patterns. Valverde et al. (2015) concluded that climate change is expected to severely affect water requirements for irrigated agriculture in the Guadiana River basin, Portugal. A general increase in net irrigation requirements of the main representative crops was identified for the five different climate change scenarios. Hong et al. (2016) also reported an increase in the net irrigation requirement for cereals and vegetable crops in South Korea.

Some climate change assessments have also been done in the semi-arid northeastern region of Brazil. Krol and Bronstert (2007) obtained contradictory future rainfall projections. They also concluded that the agricultural water demand results could significantly vary based on the GCM applied. Gondim et al. (2012) concluded that irrigation water needs increased in the Jaguaribe basin by 7.9 and 9.1% over the period of 2025–2055 compared with the 1961–1990 baseline levels. Northeastern Brazil has historically high emigration rates due to a combination of severe drought periods and better labor opportunities in the Brazilian southeast. The impacts of climate change on the economic performance of the agriculture sector and on migration may create situations of socioeconomic vulnerability, which demand the design of appropriate adaptation policies (Barbieri et al. 2010). Regarding the Jaguaribe River basin, concerns with possible increased water demand require information to delineate adaptation policy strategies for future climate. The objective of this study was to identify climate change impacts on water availability and demand in the Jaguaribe River basin, Brazil, using rainfall seasonal evaluation performance as a screening technique to select IPCC CMIP5 GCMs, as well as to propose an adaptation policy for the irrigation sector.

Materials and methods

Study area

Ceará State, Brazil, is located in the semi-arid northeastern region of the country. It has historically been marked by extreme climate events, including severe droughts and sporadic floods. The inter-annual variability is highly due to the El Niño/Southern Oscillation. The annual rainfall variability is high and 75% of rainfall is expected to occur in 4 months of the year (February–May). As a result, reservoir infrastructures have been spatially distributed in the state territory to meet demands during dry months and droughts, as well as to avoid floods during extremely rainy years.

The Jaguaribe River basin occupies an area of 74,621 km2, which is approximately 48% of Ceará State, Brazil. According to Souza Filho and Lall (2004), the main water demands are urban (20%) and irrigation (80%), which are concentrated mainly during the dry season (August–November). The water balance is negative in most months (January, February, and June through December period), when crop irrigation is most required. The study area includes the Castanhão Dam catch surface area, which supplies water for irrigated agriculture downstream, between 4°39′30″ and 4°40′00″S and 37°35′30″ and 38°27′00″W comprising 6415.10 km2 (Fig. 1) along a 160 km-long river reach. The observed river inflow varies strongly depending on the month and the amount of rainfall. Monthly maximum values are observed in March and April (33.7 and 48.5 m3 s−1, respectively), and minimum values are observed in October and November (1.3 and 0.8 m3 s2, respectively). The observed annual average (1971–2000) is 13.5 m3 s−1. At high temperatures (averages ranging from 23 to 27 °C), irrigation occurs all year long in the target area, ranging from 3883 ha in March to a maximum of 8778 ha in August. The crop pattern varies monthly; the main crops are banana trees, rice, pasture grass, cowpea, melon, and corn, of which cowpea, melon, and corn are cultivated from July to December. A representation of 97% of the irrigated crops in the study area is shown in Table 1. This is a crop pattern update from the study by Gondim et al. (2012) after the Tabuleiro de Russas irrigation project began operational activities.

Fig. 1
figure 1

Study area digital elevation model (m)

Full size image
Table 1 Growing months and crop areas (ha) in the study region
Full size table

Model screening

The rainfall seasonal pattern has been used as an evaluation criterion, as it is critical to the assessment of climate change impact on water resources and agriculture. The crop growing season and river flow are dependent on the rainfall temporal distribution, and a poor seasonal representation may compromise a climate change impact assessment. In addition, it is considered that the total monthly rainfall and its seasonal occurrence obtained from GCM hindcasts demonstrate a model skill to represent the present climate. Silveira et al. (2013) performed a Seasonal Evaluation (EVALs) of IPCC CMIP5 models for northeastern Brazil (0°S–10°S and 33°W–44°W) to identify the models that best represent twentieth century rainfall pattern. The observed data were based on an interpolated dataset from CRU and National Oceanic and Atmospheric Administration (NOAA). In this task, the same methodology has been applied to 25 CMIP5 models (Table 2) for Ceará State, where Jaguaribe basin is located. GCMs Seasonal Evaluations (EVALS) were based on the root-mean-square error (RMSE) and Pearson Correlation (CORREL.) statistics of the monthly rainfall percentage contribution relative to the annual accumulated amount, as characterized by the following equations:

Table 2 CMIP5 GCMs Evals.
Full size table
$${\text{EVA}}{{\text{L}}_{\text{S}}}={\alpha _{\text{c}}}\frac{{{\text{CORREL}} - {\text{CORRE}}{{\text{L}}_{{\text{MIN}}}}}}{{{\text{CORRE}}{{\text{L}}_{{\text{MAX}}}} - {\text{CORRE}}{{\text{L}}_{{\text{MIN}}}}}}+{\alpha _{\text{r}}} {\frac{{{\text{RMS}}{{\text{E}}_{{\text{MAX}}}} - {\text{RMSE}}}}{{{\text{RMS}}{{\text{E}}_{{\text{MAX}}}} - {\text{RMS}}{{\text{E}}_{{\text{MIN}}}}}}}$$
(1)
$${\alpha _{\text{c}}}+{\alpha _{\text{r}}}=1.$$
(2)

Indices \({\alpha _{\text{c}}}\) and \({\alpha _{\text{r}}}\) assume values between 0 and 1, according to the chosen CORREL and RMSE influence. In this work, 0.5 was considered to correspond to equal weights for each of the statistics. This decision may be explained by the choice to identify GCMs with equilibrium between adequate correlations with rainfall pattern in the study region and simultaneously low root-mean-square error. EVALS assumes values between − 1 and 1, ranging from perfect anti-correlation to no correlation to perfect correlation. A relative evaluation involved the minimum and maximum correlation and RMSE among models (CORRELMIN, CORRELMAX, RMSEMIN, and RMSEMAX), which allowed us to obtain the seasonal evaluation performance.

Bias removal of climate projections

The climate projections used to estimate the future water demand and availability in the Jaguaribe basin were generated by the IPCC CMIP5 GCMs for the base period of 1971–2000, as well as for 2025–2055. The choice of this future time slice is based on focusing on a near-term future to implement strategic policies for improving the sustainability of the irrigation sector and to guarantee rural employment for the next several decades. This coincides with suggestions of Taylor et al. (2009) to perform future studies when planning a CMIP5 climate change experimental design, specifically focusing on the 2026–2035 decade.

Rasmussen et al. (2012) reported a method of correcting the climate model bias by comparing the delta change (Hay et al. 2000) to distribution-based scaling (Piani et al. 2010). The conclusion was that, when using the delta change method, irrigation water demand is significantly underestimated, whereas low stream flow is overestimated. These results were due to the inability of this method to account for changes in rainfall and reference evapotranspiration inter-annual variability.

Statistical correction via the gamma cumulative distribution function was performed on the monthly mean precipitation time series from the IPCC-CMIP5 models using the methodology applied by Block et al. (2009). A gamma distribution was fitted to the observed monthly precipitation data (CRU) for the Castanhão Dam catchment basin to identify the probabilistic parameters that represent the monthly rainfall frequency distribution. In addition, a gamma distribution was fitted to the monthly precipitation time series obtained from the GCM hindcasts (from 1971 to 2000) for the basin. The GCM hindcasts gamma probability distribution was obtained to identify observed rainfall with the same probability, which corresponds to bias-removed present rainfall. Future rainfall bias-removed corresponds to the observed rainfall with the same probability as GCMs’ future rainfall gamma distribution. Considering the lack of a historical temperature measurement data set, minimum and maximum temperatures from IPCC-CMIP5 GCMs were statistically corrected using temperature data from CRU (Harris et al. 2014), which is defined by the following equation:

$${t^{\prime}_{\text{p}}}=\frac{{{t_{\text{p}}} - {{\bar {t}}_{\text{h}}}}}{{{\text{st}}{{\text{d}}_{{\text{th}}}}}}{\text{st}}{{\text{d}}_{{\text{ob}}{{\text{s}}_{\text{t}}}}}+{\bar {t}_{{\text{obs}}}},$$
(3)

where \({t^{\prime}_{\text{p}}}\) is the minimum or maximum mean monthly temperature bias-removed; \({t_{\text{p}}}\) is the monthly minimum or maximum mean monthly temperature from the GCM projection (2025–2055) time slice to be corrected; \({\bar {t}_{\text{h}}}\) is the minimum or maximum mean monthly temperature from the GCM hindcasts (1971–2000) time slice; \({\text{st}}{{\text{d}}_{{\text{th}}}}\) is the standard deviation of the minimum temperature or maximum temperature from the GCM hindcasts (1971–2000) time slice; \({\text{st}}{{\text{d}}_{{\text{ob}}{{\text{s}}_{\text{t}}}}}\) is the standard deviation of the minimum temperature or maximum temperature from the CRU (1971–2000) time slice; \({\bar {t}_{{\text{obs}}}}\) is the minimum or maximum mean temperature from the CRU (1971–2000) time slice.

The IPCC fifth report introduced the concept of effective radioactive forcing (anthropogenic plus natural), measured in W m−2 (Myhre et al. 2013). Four future emission scenarios called Representative Concentration Pathways (RCPs) (RCP2.6; RCP4.5; RCP6.0; RCP8.5) were developed. In addition to the scenarios designed in the Special Report on Emission Scenarios (SREs) (Nakicenovic et al. 2000), in the third and fourth reports, these four scenarios also include a more consistent approach to short life gases, land-use change, and radioactive driving force stabilization by 2100 (low 2.6 W m−2, medium–low 4.5 W m−2, and medium–high 6.0 W m−2 or high 8.5 W m−2) (Cubasch et al. 2013).

Estimating Castanhão Dam water inflow

Present and future water availability in the Jaguaribe basin was estimated using the Soil Moisture Accounting Procedure (SMAP) rainfall—river flow hydrological model and bias-removed rainfall as the input climate data set. The SMAP model was calibrated against monthly stream flow from 1912 to 1969. The correlation coefficient (CORREL.) between observed and modeled values was 0.90. Validation was performed over the 1974–1996 period (CORREL. = 0.95), indicating a very strong model performance. The SMAP model runs a monthly time step containing two reservoirs (subsurface and groundwater) and four parameters (soil saturation capacity; subsurface flow; recharge coefficient; a base flow recession coefficient), which makes it appropriate for application to big basins (Block et al. 2009).

Estimating reference evapotranspiration

According to Allen et al. (1998), crop evapotranspiration (ETc) refers to the evaporation demand for crops that are grown under optimal soil water, excellent management, and environmental conditions, and that achieve full production under the given climate. ETc depends on the reference evapotranspiration (ETo) and crop development stage measured by the crop coefficient (Kc). Crop water needs (CWN) are equal to crop evapotranspiration, also called the net irrigation requirements (NIR). Irrigation systems provide water to crops, but part of this water is lost by evaporation, run-off or leaching. Silva et al. (2007) applied the irrigation water needs (IWN) term, hereinafter referred to as the amount of water to meet ETc plus expected water losses by irrigation, which may be associated with irrigation system technology and its operational and management procedures. Bias-removed maximum and minimum temperatures from IPCC-AR5 GCMs were applied to estimate reference evapotranspiration, using the limited climate data model by Penman–Monteith (Allen et al. 1998). The evaluation and processing steps to assess water demand and availability are represented in Fig. 2. The complete mathematical model by FAO-Penman–Monteith (Allen et al. 1998) requires a complete data set of climate variables from a reference station to estimate reference evapotranspiration, which is given by the following equation:

Fig. 2
figure 2

Coupled Model Intercomparison Project Phase 5 (CMIP5) Global Circulation Models (GCMs) evaluation criteria and processing to assess water demand and availability. ETo, reference evapotranspiration; SMAP, Soil Moisture Accounting Procedure

Full size image
$${\text{E}}{{\text{T}}_{\text{o}}}=\frac{{0.408\Delta ({\text{Rn}} - G)+\gamma \frac{{900}}{{T+273}}{u_2}({e_{\text{s}}} - {e_{\text{a}}})}}{{\Delta +\gamma (1+0.34{u_2})}},$$
(4)

where ETo is the reference evapotranspiration (mm day−1); Rn is the net radiation at the crop surface (MJ m2 day−1); G is the soil heat flux density (MJ m2 day−1); T is the average daily air temperature measured at a height of 2 m (°C); u2 is the wind speed at a height of 2 m (m s−1); es is the saturation vapor pressure (kPa); ea is the actual vapor pressure (kPa); es − ea is the saturation vapor pressure deficit (kPa); Δ is the slope of the vapor pressure curve (kPa °C−1); γ is the psychometric constant (kPa °C−1).

The meteorological data required to estimate the Penman–Monteith ETo consist of air temperature, air humidity, wind speed, and radiation. A climate data set containing maximum and minimum temperatures, actual and saturation vapor pressures, liquid radiation, and wind speed allows for the estimation of the Penman–Monteith FAO ETo by applying a source of equations, the so-called limited climatic data model procedure suggested by Allen et al. (1998). Popova et al. (2006) in Bulgaria, Jabloun and Sahli (2008) in Tunisia, and Sentelhas et al. (2010) in Canada have validated this model for each of the sites. For the study region, the validation procedure has been done by Gondim et al. (2012).

Estimating irrigation water needs (IWN)

The irrigation water needs (IWN) (Eq. 5) in the basin were estimated using the approach applied by Gondim et al. (2012), where weighted crop coefficient (WKc) estimations for the irrigated crops in each particular month (Table 1) were obtained by multiplying each Kc (Table 3) by the respective crop area divided by the total monthly irrigated area. The meaning of WKc is the evaporation demand coefficient of a basket of crops weighted by the area to be used to estimate all crop evapotranspiration by month. The weighted monthly irrigation efficiency (WEf) (Table 4) was calculated similarly, using the average field measured efficiency (Ef as obtained by field evaluation) for each irrigation system adopted by farmers, multiplied by the area of each irrigation system, and then divided by the total irrigated area of each month, as shown in Table 4. Thus, WEf refers to the weighted-average Ef of all systems in operation by month. The area occupied by each crop varied each month, as did the irrigation systems, which resulted in different monthly WKc and WEf values:

Table 3 Crop coefficients (Kcs) of each growing crop by month and respective monthly weighted Kc
Full size table
Table 4 Total area of irrigation systems type in operation each month and respective irrigation field efficiency (Ef), monthly weighted efficiency (WEf), and monthly improved (WEf)
Full size table
$${\text{IWN}}=\frac{{{\text{E}}{{\text{T}}_{\text{o}}} \cdot {\text{WK}}{{\text{c}}_i}}}{{{\text{WE}}{{\text{f}}_i}}} - {P_i}$$
(5)

where ETo is the Penman–Monteith FAO reference evapotranspiration (mm month−1); WKci is the weighted crop coefficient for month i (dimensionless); WEfi is the weighted irrigation efficiency for month i (dimensionless); Pi is the average monthly effective rainfall for month i (mm).

The four factors for projected climate and hydrological variables in the RCP4.5 and RCP8.5 scenarios are (1) rainfall (mm), (2) reference evapotranspiration (mm) estimated from minimum and maximum temperature (°C) by the Penman–Monteith limited data model (Allen et al. 1998), (3) irrigation water needs (IWN, mm), and (4) river flow (Q, m3 s−1). CRU represents measured climate data. Maximum and minimum temperature and rainfall bias-removed input data (Table 5) were obtained from GCM hindcasts and model future projections output. Climate change impact analysis was based on anomalies, which were calculated by the difference between the twenty-first and twentieth century time slice annual averages, divided by the twentieth century annual average. Anomalies may then assume negative or positive values (signs). When GCM future projections of a certain climate variable disagree, this disagreement may be attributed to uncertainties surrounding modeling the future climate. According to Cubasch et al. (2013), model uncertainty is an important contributor to discrepancies in climate predictions and projections. Some use this term to represent the range of behaviors observed in ensembles of climate models (model spread). Causes of this range may be the uncertainty in greenhouse gas future emissions (scenarios), the model uncertainty used to represent climate, internal climate variability, modeling drivers forcing, and the initial and boundary conditions used in model runs.

Table 5 Mean climate variable input data (bias-removed)
Full size table

In addition, an irrigation water demand scenario was run by applying improved irrigation efficiency (Table 4), along with possible gains by irrigation system management performance, based on each irrigation system with adopted technology.

Climate change impacts were quantified by comparing the present climate and hydrological variables (1971–2000) to future projections (2025–2055 time slices) for the RCP4.5 and RCP8.5 scenarios. GCMs that presented the five highest scored EVALS were selected for evaluation by this impact assessment.

Estimating future water shortage

The annual water-deficit index (I) is defined as a performance indicator using annual water availability (S) and irrigation water demand (D) as follows:

$$I=\frac{{S - D}}{D},$$
(6)

where S is the total annual water storage (m3) and D is the total irrigation water demand (m3). I ≥ 0 indicates that no water-deficit exists. A lower I value indicates less water availability. I ≤ 0 indicates agricultural water deficit (Moursi et al. 2017).

Results

Seasonal evaluation

For the EVALs, Fig. 3 shows a three-dimensional representation of the results for RMSE (X-axis), EVALs (Y-axis), and CORREL (Z-axis). Models with high RMSE and low CORREL are located to the right of the X-axis and at the bottom of the Z-axis. Models shown in blue have high CORREL and low RMSE, and are located to the left of the X-axis at the top of Fig. 3, as a consequence of having high EVALs. Five GCMs exhibited EVALs indices above 0.90, which were the Community Earth System Model 4 (CCSM4), Hadley Centre New Global Environmental Model 2-Arctic Oscillation (HadGEM2-AO), Community Earth System Model 1-Biogeochemical (CESM1-BGC), Beijing Climate Center-Climate System Model 1.1 (BCC-CSM1.1), and Geophysical Fluid Dynamics Laboratory-Coupled Model 3 (GFDL-CM3), which presented EVALS of 1.000, 0.948, 0.947, 0.913, and 0.903, respectively. This result suggests that these models are appropriate CMIP5 GCMs for running hydrological studies on the rainfall output dataset for Ceará State.

Fig. 3
figure 3

CMIP5 GCMs’ EVALs for Ceará State, Northeast of Brazil

Full size image

Among all 25 CMIP5 models used for the study region, the five mentioned GCMs performed the best for rainfall pattern. This GCM screening criterion may be more appropriate for application to the previous hydrological studies, because future projections for rainfall are more uncertain than those for temperature. Model selection was based on the assumption that the models should be able to run more likely future projections, and anomaly sign disagreement was smoothed by bias-removing techniques using the cumulative gamma distribution function. Models selected for future climate data were submitted to bias correction, excluding model over- and underestimation values, compared to the observed data.

Climate variable projections

All five selected CMIP5 models projected future temperature increases (positive sign) for both the RCP4.5 scenario and the RCP8.5 scenario (Fig. 4). The observed differences among the models are related to the magnitude of the change, showing positive sign anomalies in all months of the year. Figure 5 provides both reference evapotranspiration and rainfall changes for the five selected CMIP5 models. Due to higher temperatures occurring in the future, the five selected GCMs projected only positive change signs for ETo, with average increases ranging from 2.3 to 6.3% for the 2025–2055 time slice relative to the 1971–2000 baseline period (scenarios RCP4.5 and RCP8.5). This percentage represents annual ETo increases ranging from 42 to 112 mm. In regard to rainfall, uncertainties still remain surrounding the projections of the CMIP5 models, as positive and negative anomalies (%) were observed in the RCP4.5 scenario and only negative anomalies were observed in the RCP8.5 scenario (Fig. 5). In other words, differences exist in the sign change depending on the model, scenario, and month of the year. Three of the five models (CCSM4, HADGEM2-AO, and CESM1-BGC) projected annual decreases (from − 8.5 to − 15.2%), and two models (BCC-CSM1.1 and GDFL-CM3) projected increases (varying from 4.8 to 12.5%) (RCP4.5 scenario). All five selected GCMs projected decreases (varying from − 1.6 to − 12.5%) (RCP8.5 scenario), as observed in Fig. 5. The changes represent decreases ranging from 81 to 144 mm annually or increases from 46 to 118 mm (RCP4.5). The RCP8.5 scenario showed decreases ranging from 17 to 139 mm annually.

Fig. 4
figure 4

Temperature anomaly

Full size image
Fig. 5
figure 5

Rainfall and evapotranspiration anomaly (%)

Full size image

Future water availability and demand

The sign changes of the annual river inflow (%) into Castanhão Dam follow rainfall behavior uncertainties, varying by GCM and by scenario, as represented in Fig. 6. The HadGEM-AO and CESM1-BGC models projected that annual river run-off would become lower in both scenarios (− 43.7 and − 50.3%) for RCP4.5 and (− 43.4 and − 26.8%) for RCP8.5. The CCSM4, BCC-CSM1.1, and GDFL-CM3 models projected values of − 48.9, 19.8, and 3.5%, respectively, for RCP4.5 and of 3.8, − 43.0, and − 8.4%, respectively, for RCP8.5. Most scenarios projected future reductions; this lower water availability should be taken into consideration in future policy adaptation and strategy design.

Fig. 6
figure 6

Streamflow anomaly (%)

Full size image

Projected increases in ETo and rainfall anomalies resulted in greater annual IWN for the 2025–2055 future time slice for all five models and applied scenarios. Average annual increases ranged from 3.1 to 16.7% (RCP4.5 scenario) and from 2.8 to 16.7% (RCP8.5 scenario), representing a 42–248 mm irrigation water demand annually (Fig. 7). When applying an improved irrigation efficiency scenario, water availability could meet irrigated agriculture demand (IWN) (Fig. 7) according to all five GCMs (RCP4.5 scenario and RCP8.5 scenario) if greater irrigation system efficiencies (IEf) are reached (Table 4) in the river basin. Annual water demand changes varied from − 1.9 to − 14.7% (RCP4.5), and from − 1.7 to − 35.2% (RCP8.5), representing a − 29 to − 211 mm and − 26 to − 520 mm annual demand, respectively.

Fig. 7
figure 7

IWN anomaly (mm) by model, scenario, and improved irrigation efficiency (IEF)

Full size image

Lower future water availability is projected, as detected by less positive water-deficit index (I) in all models except for the BCC-CSM1.1 and GDFL-CM3 models in the RCP4.5 scenario and by all models in the RCP8.5 scenario. No water deficit is projected (I > 0) when only agriculture water is studied (Fig. 8).

Fig. 8
figure 8

Water-deficit index by model (1971–2000) and (2025–2055), RCP4.5 and RCP8.5

Full size image

Discussion of results

Woznicki et al. (2015) applied ten bias-removed GCMs to conclude that the Kalamazoo, Michigan basin water balance depended strongly on the sign and magnitude of the rainfall and temperature obtained from the GCMs. They also reported uncertainties about the future, and recommended that the sign and magnitude of the rainfall and temperature obtained from the GCMs should be considered in water resource storage and supply. Even though this task has been developed only with the selected rainfall seasonally evaluated GCMs, uncertainties about future projected rainfall and, consequently, basin water balance still persist.

The magnitude of the IWN change was not great enough to cause a collapse in water supply considering agriculture demand only, unless it was associated with food demand and irrigated area increases. In this case, an adaptation policy should be implemented to achieve a sustainable water supply to meet demands.

Improved irrigation water use efficiency (IEf) (Fig. 7) has been demonstrated to play an important role in adaptation to future higher levels of irrigation water demand. Elliot et al. (2014) also reported that efforts to increase water use efficiency could compensate for climate change without further exploiting water resources in rivers due to water loss reduction. Similar conclusions were reported by Panagoulia (2004) in Greece. Therefore, adaptation measures should be applied to irrigation water use efficiency improvement at the field level to adopt water use technologies and for improvement of water resource management. System operation, delivery of precisely estimated quantities of water as needed by users, and installation of water meters should be considered. As reported by Rehana and Mumjumdar (2013), assessment studies on climate change impacts on irrigation demand should help in developing adaptation policies for reservoir operations.

Despite the future increase in rainfall projected by some GCMs, they are not enough to compensate for water demand increases for irrigation due to the higher projection of ETo. This occurs, because the rainfall increase is mostly observed in the rainy season, whereas irrigation demand mostly occurs in the second semester of the year. The availability of water supply when needed shows a dependence on storage infrastructure and efficient management. Ashofteh et al. (2013) also projected increases in water demand for irrigation (16%) using the HadCM3 model for both positive and negative rainfall projection signs for the Aidoghmoush River, Azerbaijan. Mainuddin et al. (2015) also reported increases in irrigation water demand using two models with different rainfall signs in Bangladesh. A similar situation was reported by Zamani et al. (2016) in a study on the effects of climate change on agricultural water requirements in Iran. Regular changes and increases in temperature and an irregular change in precipitation (either decreasing or increasing) were expected in the future compared to the base period. Increases in the amount of the net water requirement and water demand volume for irrigation were projected in the future, as well.

To provide a precise time for and quantity of irrigation, efficient water use strategies should also focus on meteorological station availability for ETo estimation, Kc (determined by crop and development stage), adoption of soil water retention practices and dissemination of technical information to farmers.

The most important source of water demand increases to consider in the Jaguaribe River basin is the expansion of monthly irrigated fields, which may be observed by comparing the maximum of 5957 ha in October, as reported by Gondim et al. (2012), and now 8778 ha in August. Mainuddin et al. (2015) also reported that the increasing water demand for irrigation due to climate change may be less significant than the impact of food demand increases, which are drivers for irrigated crop area expansion.

Aside from climate change, water supply capacity assessment of the Castanhão Dam should address key points such as other water use sectors, irrigated agricultural expansion regulations limited to water availability, identification of possible reservoir system operation failures to meet demand, and adaptation measures to prepare farmers for future challenges. This should include the need to share water resources and avoid conflicts between users. Other water use sectors should be included in future studies to analyze water sustainability and continuation of the current agriculture water use behavior may no longer be sustainable.

Conclusion

Five climate change models were selected by rainfall seasonal evaluation performance criteria. These models were used to run a climate change impact assessment on irrigation water availability and demand. Climate change is projected to increase temperature and reference evapotranspiration and thus water resource demand for irrigated agriculture in the studied basin, according to all five selected CMIP5 GCMs.

Even though CMIP5 has brought about improved GCMs, climate models still disagree on future projections for rainfall, as observed in the different signs of future changes. Uncertainties in the future climate will persist until models are improved and there is a better understanding of the greenhouse gas response to climate change, especially to rainfall compared to future temperature.

It has been demonstrated that even in cases where rainfall has a positive sign, the increased magnitude is not sufficient to compensate for water demand increases caused by higher ETo. This may be because these observed rainfall increases are expected to occur during the rainy season, which does not relieve increasing demand in the second semester (dry season) of each year.

Uncertainties were also observed in average river flow, as positive and negative sign anomalies were projected by different models and scenarios. Most scenarios project future decreases, which implies less water storage and availability.

When assessing the future climate in addition to the irrigation area expansion in the basin, it is expected that available water will become even scarcer. Comparing the water demand increase caused by rainfall and ETo magnitude change to the irrigated area expansion, the latter is a relevant driver of water stress.

If current practices persist, the available water may not meet future irrigation demands. Sustainable water strategies should address climate change and food demand increase scenarios; water scarcity is the main challenge for the agriculture sector and water managers in semi-arid regions. Sustainability of irrigated agriculture is at risk in the Jaguaribe River basin, and climate change can be a negative influence.

It is possible to reduce future water demand and to compensate for demand increases caused by climate change by achieving improved irrigation water use efficiency, as demonstrated by all models and the output of the studied scenarios.

Even though the water-deficit index (I) is positive, when only agriculture water is considered, I is expected to become lower in the future, indicating a lower water availability for agriculture.

Future studies to identify critical failures in water supply for irrigation and reservoir-stored water availability should consider other water use sectors for delineating policy aimed at water regulatory rules to prioritize access by category users.