SWAT sensitivity analysis
The first step in calibrating the model is adjusting the input parameters so that the simulated results match the observed values (Kumar et al., 2017; Zeckoski et al., 2015). Sensitivity analysis is used to determine the parameters that can most effectively control the Shazand Plain River discharge. In Table 1, these parameters have been determined based on previous studies and hydrological knowledge of the region. These parameters were determined based on the highest value of t-stat and the lowest value of p-value (Abbaspour, 2007). The most influential parameter was recorded by the soil evaporation compensation factor (ESCO), followed by the threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN), initial SCS runoff curve number for moisture condition II (CN2), and snow melt base temperature (SMTMP). In contrast, base flow alpha factor (ALPHA BF) had the smallest effect. The parameters of the method for changing the parameter values listed in Table 1 are as follows:
v parameter value is replaced by given value or absolute change; r parameter value is multiplied by (1 + a given value) or relative change (Abbaspour, 2007).
Calibration and validation of SWAT
A calibration procedure is used to reduce the uncertainty of model predictions by parametrizing local conditions (Abbaspour, 2005; Abbaspour et al., 2004; Kumar et al., 2017). Figure 4 shows the simulated and observed average discharge from the Pole Doab hydrometry station (basin outlet) over the period 2010–2016. As shown in Table 2, the calibration performance indicators (objective function) include NSE and R2. According to Moriasi et al. ( 2007), the NSE objective function will have a satisfactory performance if it exceeds 0.5 in the monthly time step. Also, Van Liew et al. (2003) stated that R2 should exceed 0.5 for a satisfactory performance. According to Nash and Sutcliffe (1970), an NSE value greater than 0.75 represents a good simulation, while a value greater than 0.36 is satisfactory. Based on the results of this study, the NSE and R2 values were 0.73 and 0.77, respectively, and the R-factor and p-factor were 0.11 and 0.67, respectively, indicating uncertainty in the conceptual model, parameters, and input data and good model performance.
Using validation allows for predicting future conditions without resetting the calibration parameters (Zheng et al., 2012). This validation was performed with the algorithm used in the calibration for a period of three years (2017–2019). The performance of the validation period is shown in Table 2. Thus, the R-factor and p-factor were obtained to be 0.0 and 0.13, respectively. Meanwhile, the NSE and R2 were estimated to be 0.57 and 0.65, respectively, indicating the validity of the developed model.
Table 1
SWAT sensitive parameters in calibration, t-stat, p-value maximum, and minimum new value
Code
|
Description
|
Variation
|
t-stat
|
p-value
|
Fitted value
|
New- minimum
|
New-maximum
|
V__ESCO.hru
|
Soil evaporation compensation factor
|
V
|
13.79
|
0.00
|
0.77
|
0.62
|
0.79
|
V__GWQMN.gw
|
Threshold depth …. occur
|
V
|
-3.93
|
0.00
|
2.13
|
62.04
|
3.80
|
R__CN2.mgt
|
Curve number II
|
R
|
-3.35
|
0.01
|
-0.18
|
-0.19
|
-0.16
|
V__SMTMP.bsn
|
Snow melt base temperature (°C)
|
V
|
2.50
|
0.04
|
3.64
|
2.86
|
3.86
|
V__CH_K2.rte
|
Effective … alluvium (mm/hr)
|
V
|
2.46
|
0.04
|
0.52
|
0.00
|
1.00
|
R__SOL_K(1).sol
|
Saturated …of first (mm/hr)
|
R
|
1.27
|
0.24
|
-0.72
|
-0.80
|
-0.71
|
V__EPCO.bsn
|
Plant uptake compensation factor
|
V
|
1.27
|
0.38
|
0.31
|
0.00
|
0.50
|
V__GW_DELAY.gw
|
Groundwater delay
|
V
|
-0.68
|
0.51
|
66.65
|
62.04
|
67.63
|
V__PLAPS.sub
|
Precipitation lapse rate
|
V
|
-0.52
|
0.62
|
0.77
|
0.38
|
0.78
|
R__SOL_BD(1).sol
|
Moist bulk ….. (mg/m3)
|
R
|
0.40
|
0.70
|
-0.30
|
-0.50
|
-0.29
|
R__SOL_AWC(1).sol
|
Available…. layer (mm/mm)
|
R
|
-0.32
|
0.75
|
0.86
|
0.85
|
0.87
|
V__SFTMP.bsn
|
Snowfall temperature
|
V
|
-0.21
|
0.83
|
0.91
|
0.85
|
1.00
|
V__ALPHA_BF.gw
|
Base flow alpha factor
|
V
|
0.02
|
0.97
|
0.31
|
0.31
|
0.32
|
Table 2
SWAT model performance in validation and verification phases
Hydrometric station
|
period
|
p-factor
|
R-factor
|
R2
|
NSE
|
Pole doab
|
calibration
|
0.67
|
0.11
|
0.77
|
0.73
|
validation
|
0.13
|
0.0
|
0.65
|
0.57
|
MODFLOW Calibration and Validation
The aquifer’s recharge rate was calculated using the SWAT model, and these values were used as inputs to the MODFLOW 10.5 model in the GMS package for steady and unsteady states. Calibration was based on a monthly recorded water table. According to Chiang and Kinzelbach (1998), MODFLOW requires an initial water table to begin modeling. For this reason, we assumed a steady state for one month (October to November 2009) in which certain initial parameters were changed by trial and error and then constructed an unsteady state for the remaining months. Markazi Regional Water Authority records 1134 pumping wells in the Shazand Plain (Fig. 3). After determining K and zoning hydraulic conductivity (Fig. 5a), the groundwater model was calibrated in an unsteady state for 84 stress periods (2009–2016), and the specific yield value was estimated (Fig. 5b). Changes occurred mainly in the northern parts (outlet of the basin), which may be attributed to the diverse geological facies and non-uniform erosion resistance of the basin (Markazi Regional Water Authority 1998). As shown in Fig. 5c, the 10-year mean recharge values for the aquifer vary from 1 to 550 mm. The highest recharge value was recorded in March due to spring rains. In addition, Fig. 6 shows the hydrograph of some unsteady state observation wells. The hydrographs are accurate in most of the observation wells, although some simulated values differ from the observed values in some wells. It may be caused by unauthorized wells, a lack of attention to recording statistics, or a model error in estimating values. In addition, the model was validated for the years 2017–2019. In Figs. 7 and 8, simulated and observed water tables are shown in calibration and validation modes, respectively. Table 3 shows the performance criteria for the model, including RMSE and R2. During calibration and validation, the RMSE is 1.96 m and 2.7 m, respectively, and the R2 is 0.97 m and 0.95 m, respectively. Hernández et al. (2012) modeled with MODFLOW and reported R2 of 0.81 and 0.67 and RMSE values of 25.1 and 25.8 m, respectively. Additionally, they found a good match between their study’s observed and calculated data, indicating that the results are acceptable.
Table 3
Model performance for calibration and validation periods
period
|
R2
|
RMSE
|
calibration
|
0.97
|
1.96
|
validation
|
0.95
|
2.7
|
Management scenarios
First, the results of land use maps produced for 2009 and 2019 and their impact on infiltration calculated by the SWAT model were examined. Next, the effects of reducing infiltration and increasing groundwater extraction due to cultivated area growth were explored. Finally, the effects of different scenarios for improving the water table were considered.
1- Land use changes: Fig. 9 presents land use changes for the 2009–2019 period. There have been changes in all 5 classes of land use, with most of these changes occurring in urban, agricultural, and pasture areas. As shown on this map, urban and cultivated areas have increased by 10% and 18%, respectively, since 2009, while pastures have decreased by 27.1%. Based on SWAT modeling and a comparison of output results for these two years, it is concluded that the annual groundwater recharge has decreased by 1 cm on average. Gyamfi et al. (2017) reported that the annual aquifer recharge level has decreased by a total of 1.271 cm from 2000 to 2013. In this regard, the agricultural sector experienced the greatest change, with a 20.1% increase during the same period. These findings are consistent with the results of this section.
2- Fig. 10 depicts the hydrograph of the aquifer studied by piezometers in the region over seven years (2009–2016). The figure demonstrates a downward trend with a 5 m decline. According to the Markazi Regional Water Authority, the withdrawal from this aquifer in the agricultural sector is 231 million m3 per year. Meanwhile, the needs of this sector are met by only 5 million m3 of surface water per year. This phenomenon shows that this region is dependent on groundwater.
Evaluation of scenarios
Removal of irrigated crops
Wheat, barley, beans, and alfalfa, which had the largest area under cultivation, were eliminated by applying this scenario. According to the results, the water table increases by 42 cm every year. As a result, after 7 years, a 3-m increase in the water table is expected. In addition, aquifer extraction declines due to water demand and cultivation area of almost 71 million m3 annually. Considering that the Shazand Plain is dependent on groundwater, there was no significant change in surface water as expected. de Almeida et al. (2018) evaluated the effects of soil tillage and land cover on water infiltration into the soil. They concluded that soil water infiltration is more influenced by vegetation cover than by tillage. Our results also highlight the effect of vegetation on infiltration rate.
Modification of cropping patterns by reducing water consumption and considering the economic situation
In MATLAB, the area under cultivation was optimized based on each plant’s water consumption and income. The best result was obtained in the range of ± 50. Thus, water consumption was reduced, and the income level was decreased by the least extent. Table 4 presents the area under the initial and optimized cultivation. According to this scenario, the water table will grow 28 cm in 7 years, and extraction from the aquifer will decrease by 2 million m3 over that period. Despite the low increase in the water table, Dowlatabadi and Ali Zomorodian (2016); Kouchakzadeh and Saleh (2014) have shown that the recharge values obtained from the SWAT model used in MODFLOW have good certainty. The explanation for this result is that the total cultivated area is considered constant in this scenario. Ghaffari et al. ( 2010) studied the impact of land use change and concluded that pasture and agricultural land have decreased by 34.5% and 14.3%, respectively. Subsequently, these changes have led to a 22% decrease in groundwater recharge. These findings are consistent with the results of this study. Likewise, (Sun et al. ( 2018) identified that land use changes are often considered the main factor for soil infiltration. Also, they showed that soil infiltration rate decreased by 45.23% when pasture land was converted to farmland.
The scenario of removing 30%, 20%, and 10% of the cultivated area
Practical recommendations should be made on changing the cultivated area to be successful with the proposed models. It is therefore important to make these changes gently to be applicable (Alipour et al., 2019; Ghodspour et al., 2021). This section aimed to examine the effect of a 10–30% reduction in the cultivated area on the water table. According to Scenarios 30, 20, and 10%, the water table will probably rise by 1.5, 1, and 0.49 m, respectively. Every year, almost 31, 20.5, and 10.3 million m3 less will be extracted from the aquifer. Kibii et al. (2021) found that with the reduction of forest cover and change in land use to urban, aquifer recharge increased by 17% and surface flow decreased by 9%. Overall, it can be argued that the observed increase in the water table and no change in surface runoff are consistent with our findings. According to Lerner and Harris (2009), groundwater provides an excellent water supply for domestic use, and land use significantly affects groundwater resources through changes in recharge rate.
Table 4
Initial and optimized values of the dominant crop products of the region
Crop
|
Basic cultivated area
|
Optimization cultivated area
|
Wheat
|
8200
|
11317
|
barley
|
4000
|
2000
|
Beans
|
3000
|
1500
|
potato
|
400
|
200
|
tomato
|
35
|
52.5
|
Watermelon
|
200
|
100
|
Apple
|
770
|
1155
|
grape
|
1810
|
905
|
Almond
|
835
|
417.5
|
Walnut
|
749
|
397
|
Alfalfa
|
4000
|
6000
|