Optimal water allocation of the Zayandeh-Roud Reservoir in Iran based on inflow projection under climate change scenarios

Water resources in developing countries are experiencing significant pressures due to the growing population, increasing demands, non-sustainable water consumption patterns, and poor management practices (Vorosmarty et al. 2000; Marin et al. 2020). The changes expected to occur in the coming decades will challenge or even jeopardize already constructed water schemes' performance. They will increase the pressure on water resources. The hydrologic changes resulting from climate change will affect the planning, design, and operation of water resource systems. Developing water schemes based on present conditions without considering possible future changes could increase water resource pressure in the coming period. Therefore, it is of utmost importance to consider these changes in the future design and management of water supply systems (Ashofteh et al. 2015). Most climate change studies reveal a significant impact on available water resources on global and regional scales (Vorosmarty et al. 2000; Abbaspour et al. 2009; Faramarzi et al. 2013; Luo et al. 2019; Shahvari et al. 2019).

The Soil & Water Assessment Tool (SWAT) (Arnold et al. 1998) is popularly used for large-scale hydrologic modeling. It simulates various hydrological cycle components under climate change scenarios (Chen et al. 2019; Ercan et al. 2020). The data derived from general circulation models (GCMs) under different greenhouse gas forcing scenarios are often used to analyze climate change impacts on the water resource availability and water supply systems (Krysanova & Hattermann 2017).

Vaghefi et al. (2015) coupled the SWAT with a water allocation model (MODSIM) to study climate change effects on cropping patterns in Iran's Karkheh River Basin. They concluded that the coupled SWAT–MODSIM program was more suitable for identifying adaptation options required by farmers and decision-makers to respond to changing water availability. Carvalho-Santos et al. (2017) used a SWAT model to evaluate the suitability of a single- or two-reservoir system for water supply under climate change conditions in the Alto Sabor watershed in Portugal. They found that the reservoir water storage will decrease under climate change conditions. Also, a two-reservoir system will have less reliability to supply water in the future. Fereidoon & Koch (2018) used a coupled simulation–optimization tool, SWAT–LINGO–MODSIM–PSO (SLMP), to determine optimal crop patterns for the Karkheh River Basin in Iran under climate change. They concluded that the SLMP model could serve as a practical tool to analyze the agricultural benefits under climate change impacts.

Kangrang et al. (2019) investigated the optimal reservoir rule curve for future climate change in the Ubolrat Basin in Thailand. They used a methodology that used a SWAT model, a coupled genetic algorithm, and a reservoir mathematical model. They found that the new rule curve's performance was better than the old rule in the climate change condition. Ahmadianfar & Zamani (2020) assessed the performance of the Jarreh reservoir in Khuzestan province in Iran under climate change conditions. They adapted the reservoir operation policy based on reservoir inflow variation under climate change conditions. They concluded that the reservoir performance and water resource management would be more reliable based on their operation policy. According to the previous studies, reservoir performance evaluation to supply water is an essential issue under climate change.

The Zayandeh-Roud Basin is one of the most critical basins in Iran, located in a semi-arid area. Erratic precipitation patterns, uneven distribution of the water resources, repeated prolonged droughts, and human factors in the region have led to considerable challenges for water availability and supply across the river basin. The multipurpose Zayandeh-Roud Reservoir with a storage capacity of 1.5 MCM (SWRC 2009) regulates the river flow. It supplies water during drought or water shortage. The reservoir inflow has considerably reduced in recent years (Figure 1), posing significant water availability problems for the various regional uses. According to the variation of water demands in different sectors and impacts of climate change, the water supply condition may become critical. The majority of the previous studies assessed the effects of climate change on the meteorological and hydrological components of catchments and optimized the benefits of the agricultural sector in the future. However, no study optimized water allocations to other water users besides the agricultural sector under climate change, along with considering the possible changes in different water demands. Most climate change studies on this region also used the third and fourth reports of IPCC (Intergovernmental Panel on Climate Change) scenarios.

In the present study, the impacts of climate change on the reservoir inflow and meteorological components were investigated using ISI-MIP5 (Inter-Sectoral Impact Model Intercomparison Project) GCMs. In addition, the allocation of available water to agricultural, industrial, and domestic sectors was optimized regarding possible changes in the future water demands.

This study aimed to quantify the expected impacts of climate change in the future. Furthermore, the study was carried out to optimize water allocations by minimizing water deficits in the agricultural, industrial, and domestic sectors. To achieve these objectives, a SWAT model of the study area was built in both historical and future periods. The simulated reservoir inflow under climate change scenarios was used to optimize water distribution in the future.

To model the impact of climate change, we used five ISI-MIP5 GCMs: GFDL-ESM2Mt, HadGEM2-ES, IPSL-CM5A-LR, MIROC, and NoerESM1-M under four Representative Concentration Pathways (RCPs): RCP2.6, RCP4.5, RCP6.0, and RCP8.5. A nonlinear optimization model was applied to minimize the water deficits while also appropriately responding to agricultural, industrial, and domestic demands. The optimization problem was solved in the LINGO software. The reservoir performance was evaluated based on two indices: reliability and vulnerability.

The study area is upstream of the Zayandeh-Roud Reservoir (49°50′–50°40′E and 32°20′–33°10′ N), located in the Zayandeh-Roud River Basin in central Iran. It covers an area of around 4,100 km² (Figure 2). The study area is climatically and hydrologically different from the rest of the basin because of the high elevation. It receives annual precipitation of around 1,400 mm. The Zayandeh-Roud River is the most crucial surface flow in the basin. It supplies water for agriculture, industry, domestic, and other household purposes in Isfahan and Chaharmahal and Bakhtiari provinces in central Iran. The existence of high-flow rivers in the neighboring basins, such as Chaharmahal and Bakhtiari, has encouraged inter-basin water transfer projects to supply part of the water shortage in the Zayandeh-Roud River Basin. There are three tunnels in the study area for this purpose.

The SWAT was used to build a hydrologic model of the study area. Calibration/validation, sensitivity analysis, and uncertainty analysis were performed using the SUFI-2 algorithm (Abbaspour et al. 2007). Data derived from five GCMs were used to compute the variations in precipitation and temperature as well as their effects on the reservoir inflow. Finally, optimal water allocations of the reservoir to different uses were determined for future periods using the LINGO software. A schematic plot of the study process is shown in Figure 3.

The SWAT is a process-based, semi-distributed, continuous-time program and has an efficient computational simulator (Arnold et al. 2012). The river basin is initially divided into the smaller subbasins based on a digital elevation model (DEM). These subbasins are further divided into the hydrologic response units (HRUs) consisting of homogeneous units of land use, soil type, and slope (for more information, see Neitsch et al. 2011). The study region was divided into 45 subbasins and 405 HRUs using a 10% threshold for soil, 5% for land use, and 5% for slope. The thresholds were defined to eliminate small inconsequential units and to reduce the model calculation time. Data required to build the SWAT model and their sources are summarized in Table 1. There are three tunnels in this region, which transfer water from the adjacent basins to the study area. Monthly tunnel data were added to the streamflow in the SWAT model. In this study, the SWAT model was built from 1987 to 2015. The observed discharges from two stations, Ghaleshahrokh and Eskandari, were used to calibrate (1991–2008) and validate (2009–2015) the model with three years of the warm-up period (1987–1990).

The runoff will only occur when ⁠. The steps of SWAT runoff simulation are summarized in Figure 4. The SWAT model's automatic irrigation option was used for agricultural management and dividing the agricultural land uses into different crops. The data obtained from the Isfahan Agricultural Jahad Office on the cropping patterns in the townships across the Zayandeh-Roud River Basin were used to assign the irrigated crops. Given that the cropping patterns in the basin were diverse, the area under each crop was initially used to determine the predominant crop(s) in each township. The land-use maps, township boundary maps, and boundary maps of the subbasins constructed in the ArcSWAT model were then used to assign predominant crops to each SWAT subbasin in the model by the Geographic Information System. Ultimately, wheat, barley, alfalfa, rice, potato, and corn were identified as the dominant crops.

Data from GFDL-ESM2Mt, HadGEM2-ES, IPSL-CM5A-LR, MIROC, and NoerESM1-M models, and four RCPs such as RCP2.6, RCP4.5, RCP6.0, and RCP8.5 were used to simulate the climate data for the years 2020–2045 at four climate stations: Badijan, Ghaleshahrokh, Damanehfereidan, and Singerd (Figure 2).

To evaluate the reservoir performance, the reliability and vulnerability indices were used, which are calculated based on the following equations (Hashimoto et al. 1982).

Due to the imbalance between water availability and water consumption, the current policy based on the State Water Master Plan (SWMP) report increases irrigation while keeping the current area of agricultural lands constant (SWMP 2012). Also, the domestic water supply was estimated based on population growth. The following water needs were estimated based on the report provided by the SWMP.

The gross agricultural water need in the future was calculated based on the cultivated areas during 2016–2045. The total cultivated area was considered constant and equal to the areas reported in the SWMP for the base period. The total future gross water demand was calculated based on the past net demand and the projected agricultural efficiency provided by the SWMP report for 2041. According to the SWMP report, 30% of the gross agricultural demand is supplied by the reservoir. The remaining is supplied by other sources such as groundwater.

According to the SWMP report, the industrial demand supplied by the surface waters in the base period is 112 MCM/year. In this study, the industrial demand was assumed to remain the same.

The population living in regions whose domestic demands depended on the Zayandeh-Roud Reservoir was estimated to be 4,736,628 by 2041. Based on the average daily per capita water consumption of 267 litres, the domestic sector's future annual demand was estimated to be around 490 MCM/year (SWMP 2012).

For calibration and validation of the model based on measured runoff, the SUFI-2 algorithm in the SWAT-CUP software was used. This program also calculates parameter sensitivity and model uncertainty. Model prediction uncertainty accounts for all modeling errors (e.g., model parameters, observation data, and conceptual model) (Abbaspour et al. 2007). Propagation of the uncertainties in the parameters leads to uncertainties in the model output variables, expressed as the 95% prediction uncertainty (95PPU), computed at 2.5 and 97.5% levels of the cumulative distribution of an output variable. The p-factor and r-factor are two criteria used to measure the goodness of fit and model uncertainty, respectively. The p-factor is the percentage of observed data bracketed by the 95PPU band. Simultaneously, the r-factor is the average thickness of the 95PPU band divided by the standard deviation of observation data. The ideal values of the p-factor and r-factor are 100 and 0%, respectively, indicating a perfect match between the observed and simulated values. The objective in this process is, therefore, to obtain a high p-factor with a small r-factor. Abbaspour et al. (2015) suggested that for flow, values of >0.7 for p-factor and <1.5 for r-factor constitute a good calibration result.

A t-test was then used to identify the relative significance of each parameter b_i. The sensitivities given above are estimates of the average changes in the objective function resulting from changes in each parameter while all other parameters change. This gives relative sensitivities based on linear approximations and only provides partial information about the objective function's sensitivity to model parameters. In this analysis, the larger, in absolute value, the value of t-stat, and the smaller the p-value, the more sensitive the parameter (Abbaspour et al. 2017a).

A large number of parameters were initially selected for the sensitivity analysis. In the end, seven hydrologic parameters were sensitive to discharge at the 95% confidence level. The most sensitive parameters were CN2, followed by GWQMN, GW_REVAP, and RCHRG_DP (Table 2). Initially, the five snow parameters were calibrated and removed from further calibration to avoid identifiability problems with the other hydrologic parameters (Abbaspour et al. 2017a). The snowmelt base temperature (°C) (SUB_SMTM) was the most sensitive snow parameter.

For the calibration period, a p-factor of 0.79 at the Ghaleshahrokh station was obtained (Figure 5(a)), which means the 95PPU bracketed 79% of the observed data. The r-factor was 1.01, indicating an acceptable uncertainty level for discharge. For the Eskandari station (Figure 5(b)), a p-factor of 0.70 and an r-factor of 0.60 were obtained. The model prediction was less uncertain at the Eskandari station; however, it bracketed less observed data, indicating a slightly larger simulation error (=1 − p-factor) for small flow conditions, as also observed before by Abbaspour et al. (2015). The NSE values for the ‘best’ parameter set, i.e., the parameter set with the highest objective function value, were 0.75 and 0.61 for the Ghaleshahrokh and Eskandari stations, respectively. However, as the hydrologic model results are inherently uncertain because of unavoidable errors in the input data, the p-factor and the r-factor better describe a model's calibration reliability. All post-processings using a calibrated model must propagate the calibrated parameter ranges to obtain model prediction uncertainty and avoid using the ‘best’ parameter set only (Abbaspour et al. 2017b).

Similarly, acceptable results were obtained for the validation, which had similar climatic and hydrologic conditions as the calibration period. For the Ghalehshahrokh station (Figure 5(c)), the p-factor was 0.71, and the r-factor was 0.70. These values for the Eskandari station (Figure 5(d)) were 0.72 and 0.67, respectively.

The variations in mean annual temperature during the period of 2020–2045 exhibited an increasing trend from 19.9% (1.8 °C) in RCP2.6 to 29.4% (2.7 °C) in RCP8.5 compared to the base period (Table 3). In the future, mean annual precipitation is projected to decline under all the region scenarios, with the maximum reduction of 8.2% (32.8 mm) expected to occur under RCP8.5.

The SWAT model results indicated reductions of 41–51% inflow volumes relative to the average volumes in the base period under the different scenarios (Table 4). The highest percentage variation was projected to occur under RCP8.5, indicating the most severe radiative forcing in the future.

The upper and lower limits of the monthly reservoir storage volume were 1,400 and 120 MCM, respectively. In addition, the initial dam storage volume was assumed to be equal to 260 MCM, and the last dam storage was the same as the initial storage. The maximum and minimum of optimized water release were under RCP2.6 and RCP8.5 in all months, respectively (Figure 6). Based on the water allocation priority mentioned in the ‘Optimal water allocation of the reservoir’ section, the amount of water released under both scenarios completely satisfied the domestic and industrial water demands even in the dry months (March–August). At the same time, the agriculture sector did not receive its full demand even under the RCP2.6 with the highest water release over the projection period (Figure 7).

The water deficits will probably persist due to climate change under all the scenarios and throughout all months (Figure 8(a)). The highest water shortage is expected to occur in the dry months (March–August) under all scenarios. Also, the highest irrigation demand belongs to these months. Therefore, it is evident that the demand at its peak in these months could contribute to increasing water shortage and adversely affect crop yield. Moreover, mean annual water deficits in the future years will vary between 730 and 847 MCM, with the highest and the lowest estimated to occur under RCP8.5 and RCP2.6, respectively (Figure 8(b)).

The lowest reliability index of 0.6% and the highest vulnerability index of 59% are related to RCP8.5 compared to other scenarios (Table 5). The present results obtained for the reservoir performance under RCP8.5 agreed with those obtained in the previous sections showing the highest water deficit for this scenario. Overall, the agricultural sector's water shortages will affect the overall reservoir performance and make it unreliable and vulnerable in the future.

The study limitations are partly due to the assumptions that (i) the effect of climate change was evaluated based on four meteorological stations, (ii) the optimal water allocation was investigated based on the water demands that are supplied by the reservoir and the water demands which, supplied by groundwater, were not considered, and (iii) the water demands were calculated based on the projection of the SWMP report. Climate change affects the crop water demand, which should be considered in future studies. Other study limitations are the general errors of large-scale modeling, assumptions, and data limitation, especially on the agricultural management and management of the rivers and the reservoir.

Several sources of uncertainty in the reservoir operation include climate data, projected reservoir inflow, and catchment characteristics, which were considered in the hydrological modeling and climate change analysis. Hydrological models suffer from several sources of uncertainty, such as the conceptual model, input data, model parameters, and observation data. In this study, the SUFI-2 algorithm was used for the model calibration/validation, which accounts for all model uncertainties. The uncertainty of GCM outputs was considered by assigning weight to the GCMs based on their historical period performance. Other sources of uncertainty can be examined in future research.

In this study, we assessed the impacts of climate change and the future performance of the Zayandeh-Roud Reservoir in Iran. The results showed that climate change would likely affect regional precipitation and temperature and significantly reduce the reservoir inflow. Due to the water allocation priority of domestic and industrial sectors over agricultural, the latter sector will face significant water shortages under climate change conditions. The awareness of water availability in the future periods by considering both climate change impacts and variations in the demands in different sectors across the basin provides a better picture of water resource reliability. This realization should help water managers with the planning of future water structures.

All relevant data are included in the paper or its Supplementary Information.