A SWAT modeling approach to assess the impact of climate change on consumptive water use in Lower Chenab Canal area of Indus basin

Climate change is one of the central driving forces affecting food and water resources at global scale (Brown & Funk 2008; Godfray et al. 2010; Yang et al. 2012). With increasing interest towards climate change adaption, there is a strong need to increase crop productions, specifically with limited water resources, in order to meet the current increasing food demands (Emam et al. 2015). On the other hand, the identification of climate change hotspots are also required for mitigating the adverse impacts of climate change on crop production (Yang et al. 2010). Understanding the spatio-temporal patterns of consumptive water use (CWU) is one of the most suitable options for formulating the optimal water allocation plans with available water resources under changing climate (Liaqat et al. 2015; Azmat et al. 2016).

CWU is generally known as the water footprint or virtual water content in the form of actual evapotranspiration (ET_a) (Eriyagama et al. 2014). CWU is a key component for simulating hydrological cycles and scheduling irrigation water demands in the agricultural fields. It is known as a complex physical process of land atmosphere interaction which is normally influenced by various hydro-metrological conditions such as solar radiation, and land surface characteristics (Allen et al. 1998). The complex ecosystem and vegetation heterogeneity makes it difficult to estimate CWU accurately. Knowledge of accurate CWU at the regional scale requires detailed information on various composite elements such as cropping patterns, climatic parameters, hydraulic properties of underlying soils and irrigation practices (Gowda et al. 2008; Li et al. 2009). The accurate assessment of these composite elements is of paramount importance as these parameters vary significantly both in space and time for large irrigation schemes.

CWU has been traditionally quantified by multiplying the reference evapotranspiration (ET_o) estimated from weather stations with crop coefficient values at field scale (Allen et al. 1998). Several other conventional techniques exist for computation of CWU as a point measurement using a lysimeter approach, Energy Balance Bowen Ratio and Eddy Covariance (EC) systems, and water balance establishment over the entire basin (Gowda et al. 2008; Li et al. 2009; Corbari et al. 2014). A number of scientific limitations make it difficult to extrapolate these point scale measurements to large spatial scale (Liaqat et al. 2015). In past decades, substantial efforts have been made to retrieve CWU primarily from remote sensing datasets at various spatio-temporal scales. The major benefit of remote sensing based methods is that the CWU can be derived directly as a residual of energy balance which excludes the need of quantification of other complex hydrological processes (Byun et al. 2014). However, the difficulties associated with the aforementioned composite elements and underlying heterogeneous irrigation conditions, especially in arid to semi-arid regions of the world, makes it extremely difficult to rely on a single method. Moreover, these methods cannot simulate the impact of climate change on CWU. The soil and water assessment tool (SWAT; Arnold et al. 1998), on the other hand, is a physical-based, semi-distributed model which has the capability of predicting the effects of climate change on water balance components.

The Lower Chenab Canal (LCC) irrigation scheme of the large Indus basin irrigation system (IBIS), used in this study, is facing poor conditions for crop production due to major challenges of limiting water resources, primarily caused by inefficient water use, groundwater depletion, climate change, and rapid urbanization and population growth among several other factors (Laghari et al. 2012). As water is a scarce commodity of the Indus basin, there are several provincial, national and trans-boundary conflicts which are ultimately adding to the complexity of managing this scarce resource (Cheema et al. 2014). Due to limited water availability, LCC normally experiences the huge difference between available water supplies and potential crop water demand that make the accurate quantification of CWU indispensable. Moreover, high climate sensitivity in the region emphasizes the need for careful planning and monitoring of CWU for sustainable water resources management (Ahmad et al. 2009; Liaqat et al. 2015). In order to address and meet these challenges of deficit water resources, comprehensive robust counterplans are needed which could utilize spatial modeling approaches for the accurate assessment of CWU, especially for regions with complex and heterogeneous vegetation conditions.

In this study, we therefore selected the SWAT model for the assessment of CWU on monthly, seasonal (Rabi – October to March and Kharif – April to September) and annual basis for entire LCC irrigation scheme, canal command areas (CCAs) of LCC, and different hydrologic response units. The simulations were performed at grid level in the LCC irrigation scheme from 2005 to 2012. Furthermore the impact of climate change on CWU is also determined from 2013 to 2020. A high spatial resolution land use land cover (LULC) map, required for implementing the SWAT model, was generated using a remote sensing approach. Surface Energy Balance Algorithm (SEBAL) was also used to estimate ET_a in order to calibrate and validate the SWAT model. The outcomes of this study will provide guidelines to the policy makers in the region to maximize crop production based on detailed information of CWU both in time and space and to formulate sustainable policy in view of the given future impact of climate change on CWU.

The basic datasets requirement for SWAT includes information on surface topography, soil features, LULC classification, meteorological parameters and stream flow series (Table 1). A freely available 90 m resolution digital elevation model (DEM) from the Shuttle Radar Topography Mission (SRTM) project governed by US National Aeronautics and Space Administration (NASA) was used in this study to classify the underlying terrain characteristics. A digital soil map generated by the Water and Soil Investigation Division (WASID) of Pakistan was used after the simple classification of soil properties for the study region. In addition to the database of soil properties, the hydrological parameters for different sub-catchments are simulated in SWAT by utilizing the predefined LULC classes (Luzio et al. 2002). The predefined sub-catchments in the LCC irrigation scheme were used for delineation by considering LCC as an artificial watershed. Seven sub-catchments in LCC are the original routing of available canal networks, i.e. CCAs of branch canals described above under ‘Study area’ (Figure 1). Daily stream flow records at the head of each of the seven branch canals covering a time span of 2005–2012 were collected from the Programme Monitoring and Implementation Unit, Punjab Irrigation Department (PMIU-PID), Pakistan. Discharge data are used as an irrigation source in the SWAT model to devise irrigation scheduling for different crops on the basis of depth over area ratio. In addition, daily climatic records (2005–2012) of minimum and maximum air temperature, precipitation, wind speed, relative humidity, and solar radiation converted from sunshine duration were obtained from the Pakistan Metrological Department (PMD). These climatic variables are generally used for the estimation of reference ET in SWAT model (Xie & Cui 2011).

An up-to-date LULC map available from a recent study (Awan & Ismaeel 2014) on ground water recharge in LCC irrigation scheme was used in SWAT model simulations. They used the moderate-resolution imaging spectroradiometer (MODIS) products of normalized difference vegetation index (NDVI) for identification of different land cover classes using phonological approaches with high accuracy. In comparison to the ground-based data collected through field surveys from entire LCC irrigation scheme, they reported about 80% accuracy for 12 different LULC classes obtained at 250 m spatial and 8-day temporal resolution by using combined NDVI data of aqua and terra satellites of MODIS. The dominant cropping patterns in the LCC area are wheat-fodder, fodder-fallow, wheat-rice and fodder-maize with relative proportional coverage of approximately 28%, 15%, 12% and 11%, respectively. The other classes such as forest, orchards, natural grass, water and bare or urban settlements constitutes an area of less than 8% of the entire LCC irrigation scheme. Sugarcane is also grown on an area of about 5% specifically in CCA associated with the Jhang branch, while cotton with a rotation of wheat or fodder (14% of the area) is frequently cultivated in Lower Gugeera and Burala CCA. Fodder is the most commonly single grown crop in this region. A similar proportion of different crops as described above was used in the SWAT model for the current study. However, the readers are referred to Awan & Ismaeel (2014) for a further detailed description of the methodology and results of LULC classification.

The calibrated SWAT model with the same environment was run for another period of 3 years (2010–2012) to validate the results. Figure 3(b) shows that the intercept value (0.83) between model estimations is slightly decreased compared to those obtained during the calibration (Figure 3(a)) period (0.96). However, the remaining statistics during the validation period was relatively better yielding high R² of 0.93, NSE of 0.89, and RMSE of 12 mm month^–1 with a PBIAS value of less than 3%. Since the SWAT model is known to perform better in relatively wet conditions with a good quality dataset (Xie & Cui 2011), wet conditions were more prevalent during the selected validation period (2010–2012), as this study region faced two mega floods during 2010 and 2012 due to large numbers of rainfall in the monsoon season (Hashmi et al. 2012; Yu et al. 2013). This fact is perhaps clear in Figure 3(b) where the scatter points during the Kharif season were more aligned with the 1:1 line (Figure 3(b)) compared with those of the Rabi season. It also revealed that estimation of CWU from SWAT were slightly larger than SEBAL during the Kharif season while most of the CWU values were lower than SEBAL during the Rabi season. SEBAL and its similar type of energy budget models such as METRIC are known as sensitive to the hydrological extreme (dry and wet) conditions (Liaqat & Choi 2015), which are generally used for their internal calibration to remove systemic biases. Therefore, a small error in the manual selection of dry and wet anchor pixels could considerably change the final ET_a outputs of the SEBAL model (Long & Singh 2012). This could be one of the reasons for slightly contrasting seasonal results from both SWAT and SEBAL models in the current study. The obtained difference (PBIAS ≤5.2%) between SWAT and SEBAL simulations for the entire period was within the allowable limit of ±15% as described above (Arnold et al. 2012), thus we used the developed SWAT model to determine the impact of climatological conditions on the variations of water use.

Further, we have determined the significance of air temperature and rainfall corresponding to variations in CWU on a seasonal and annual basis (Table 2). The results showed that the changes in CWU were significantly (p-value <0.05) controlled by the rainfall during the Rabi season but by air temperature during the Kharif season. The values of Pearson linear correlation, R, were also high (≥0.83) for both variables at their significant level. However, the analysis at annual scale revealed that only temperature was significantly impacting the CWU with R values approaching 0.80, while showing non-significant results for rainfall with relatively poor correlation (0.49). The high positive correlation between CWU and air temperature means a high potential of available energy exists in this region to maximally support the vaporous process. In spite of the effects of other factors such as irrigation practices and cropping pattern, this finding indicated the importance of changes in these variables, which in turn largely influence the CWU of large LCC irrigation schemes. Overall, the relatively huge fluctuations in these variables would significantly affect crop production by influencing water demand and supply, and have the capacity to convert a water production area (rainfall > CWU) into a water scarce area (rainfall < CWU).

CWU from the hydrological operated catchments usually depends upon available irrigation supplies and meteorological conditions as explained above under ‘Effect of current metrological variables on CWU in LCC irrigation scheme’. Due to relative homogeneous cropping patterns and system design of the LCC, irrigation water supplies should remain consistent in forthcoming decades. However, the fluctuations in meteorological conditions specifically based on mean precipitation and air temperature could significantly impact the CWU in this region. Based on this assumption, both climatic variables were simulated using Representative Concentration Pathways (RCP 4.5 and RCP 8.5) scenarios for the period 2013–2020 on a daily basis using NorESM global circulation model (Bentsen et al. 2013). We reported a decrease of –11% using RCP 4.5 and an increase of 3% using RCP 8.5 for temperature while an increase of 70–75% in rainfall was observed under both scenarios (RCP 4.5 and RCP 8.5) compared with the reference values of the LCC region in our previous study (Awan & Ismaeel 2014). In order to determine the future impacts of climate change on CWU in this study, the SWAT model was also implemented by using future scenario simulations of rainfall and air temperature from 2013 to 2020 while other variables in SWAT parameterization were kept similar as used during the calibration and validation period (2005–2012). The response of SWAT outputs to future simulated variables was analyzed on a monthly and annual scale.

The average monthly changes in CWU would range from –40 to 67% under RCP 4.5 and from –17 to 80% under the RCP 8.5 scenario compared with the reference values during the validation period (Table 3). The maximum increase (>60%) in CWU would occur during the months of October and November for both scenarios. This increase could be attributed to the observed regional climate shifts that gradually changes the cycle of cool and warm conditions, which in turn could largely disturb the atmospheric system and global energy budget cycle (Ravelo et al. 2004). During the hot months (May–August) of Kharif season, which also represents the period of monsoon, the average monthly CWU would increase more than 100 mm. This increase can be attributed to high rainfall expected during the monsoon period. The decrease in CWU could be generally noticed during the later period of Rabi (January–March) or during the starting period of Kharif (April–May) season. The highest decreases of –40 and –17% in CWU would occur during February and April for RCP 4.5 and RCP 8.5, respectively. These decreases can be due to low rainfall and temperature values in these months. Overall, the results simulated from SWAT in this study showed that the annual average CWU would be 76 and 79 mm under RCP 4.5 and RCP 8.5 scenarios which would be approximately 7% and 11% higher by the end of 2020 than the actual CWU simulated during the validation period, respectively.

The methodology proposed in this study successfully demonstrated the use of a hydrological model and satellite remote sensing for not only assessing the current use of available water resources for agriculture in large irrigated areas, but also to assess the impact of climate change on CWU in the LLC area of the Indus basin. The SWAT model was successfully calibrated and validated with CWU determined by SEBAL (R² = 0.87 ± 0.06, NSE = 0.83 ± 0.06). Results of the modelling on a mean monthly basis during the base period shows that the maximum and minimum CWU are 124 mm and 30 mm during the months of May and December, respectively. This variation in CWU, especially during the Rabi and Kharif seasons, is due to changes in those climatic parameters which directly influence the CWU.

The correlation of CWU with rainfall and air temperature for the base period showed that the CWU is significantly controlled by rainfall during the Rabi season, whereas temperature has a significant impact on CWU during the Kharif season. Moreover, CWU varies in space with the lowest and highest values of 400 mm year^–1 and 1,100 mm year^–1 at the tail and head end reaches of the LCC, respectively. Results of the SWAT modelling for a climate change scenario showed that the annual water consumption in the basin would increase overall by approximately 7% and 11% at the end of 2020 under RCP 4.5 and RCP 8.5 climate change scenarios, respectively. The monthly variations would be –40% to 60% and –17% to 80% in both scenarios. The maximum increase (>60%) in CWU would occur during October and November whereas the highest decrease of –40% and –17% in CWU would occur during February and April for RCP 4.5 and RCP 8.5, respectively. The demonstrated results and methodology could be of great value for the policy makers in the region for optimum management of surface and groundwater resources under changing climate.

The International Water Management Institute (IWMI) is in receipt of financial support from the Embassy of the Kingdom of Netherlands, Islamabad, Pakistan through Grant #22294 and the CGIAR Research Program on Water, Land and Ecosystems (WLE) which were used in part to support this study. The first author was supported through a grant for PhD studies by the Higher Education Commission (HEC), Pakistan. The authors are also grateful to the anonymous reviewers for their valuable suggestions to improve the quality of our manuscript.