Introduction

Runoff is one of the most important components of the hydrological cycle, and in any basin, one of the most necessary measures is to estimate runoff for accurate planning of water resources management and understanding of the precipitation–runoff process (Teshome et al. 2020; Gholami and Khaleghi 2021). The limitation of water resources and the increase in the need for water, which is caused by the increase in population, the development of cities, and the new management policies of human societies, as well as the indiscriminate and unprincipled use of these resources, have caused increasing problems and disputes regarding the management of water resources (Gholami et al. 2020, 2022). Precipitation and runoff modeling is an important part of global research in the field of surface water hydrology (Dianati Tilaki et al. 2020). Despite extensive efforts to collect hydrological data (Varvani et al. 2019; Gentilucci et al. 2022; Varga and Breuer 2024), there are still regions of the world, especially arid and semiarid regions that lack hydrological statistics, especially hydrometric station statistics (Marques et al. 2006).

The use of new technologies and hydrological models to simulate water resources and water components is one of the ways to reduce water stress, so proper and reliable forecasting can be used to manage very beneficial planning because of the accuracy and skill in the model. Flow forecasting has a direct impact on water resource management decisions (Arnold et al. 2012). Since it is not possible to measure all the data related to hydrological systems and processes for various reasons, including spatial and temporal heterogeneity, the measurement methods are expensive and time-consuming. Therefore, hydrological models provide us with the opportunity to have a better understanding of the interaction between them and the system by changing the variables and parameters of the system (Sokolowski and Banks 2011). In recent years, hydrological models have been used as a useful tool for water management in watersheds. Each hydrological model has its strengths and weaknesses. From the point of view of comprehensive simulation of watersheds, distributed and semi-distributed models are superior to integrated models (Abbaspour et al. 2015).

Simulation models are used to better understand the role of hydrological processes in watersheds. The mentioned models are divided into different categories, including experimental models against physical models, single event models against continuous models, and integrated models against distributed models (Hosseini and Khaleghi 2020). The basic hydrological models were necessarily used in an integrated manner due to the limitation in computing resources and the lack of spatial description of the physical characteristics of the watershed, as well as the limitation of parameter measurement. Today, although due to the increase in computer capacity and speed, the tendency to use distributed models has increased, but often due to factors such as the ability to simulate design variables (runoff, underground water, sediment, etc.), available data, and the studied spatial and temporal scale, the most used hydrological models are of the semi-distributive type (WMO 2022; Sahour et al. 2023).

The simulation of hydrologic response of a basin system using hydrologic models involves calibration and validation process. Executing these processes in fact needs an widespread knowledge on the parameters that affect the considered hydrologic process. Distributive and semi-distributive models have many parameters, and to use these models optimally, these parameters should be calibrated and validated. Validation and calibration are currently being done using the latest optimization methods, and new researches are conducted every year to investigate these methods. Distributed models such as the Soil and Water Assessment Tool (SWAT) can simulate all hydrological components of the watershed. Therefore, these models have higher accuracy in hydrological simulation. However, these models need a lot of data for hydrological simulation (Abbaspour et al. 2007b). SWAT is one of the semi-distributive watershed models that play the main role in analyzing the effect of land management changes on water in the watershed. In this model, the spatial changes of watershed characteristics are taken into account at the hydrological unit scale (land use, elevation, and soil). This model, which was developed by (Arnold et al. 1998), has been widely used. In general, the study of different researchers indicates the proper performance of this method (Abbaspour et al. 2007a, b; Birhanu et al. 2007; Li et al. 2010; Cibin et al. 2010; Kigobe et al. 2009; Memarian et al. 2014; Raneesh and Thampi Santosh 2011). In limited cases, the weak results can be attributed to factors such as inadequate spatial data, inaccuracy and weakness in data measurements, lack of model calibration, and limited calibration and validation periods (Gassman et al. 2007; Yang et al. 2016). In Iran and in recent years, this model has been widely used to simulate runoff and sediment, as well as uncertainty analysis and optimization of model parameters, and it has been found that compared to other models, it has better results in estimating runoff and sediment (Abbaspour et al. 2007a, 2007b, 2015; Hosseini and Khaleghi 2020; Memarian et al. 2013, 2014).

Sensitivity is measured as the response (reaction) of an output variable to a change in the input parameter, the greater the change in the output response, the greater the sensitivity. The parameters determined in the sensitivity analysis that affect the outputs are often used to calibrate the model (Van Griensven et al. 2006). Therefore, sensitivity analysis as a tool to evaluate input parameters according to their effect on model output is not only useful for model development but also useful for validation and uncertainty reduction. There are different methods for performing sensitivity analysis and expressing its results (Beven 2001). The objective function for model calibration includes a statistical test, such as minimum relative error, minimum average error, or NS (Santhi et al. 2001; Grizzetti et al. 2003). Validation methods are similar to calibration methods in which predicted and measured values are compared to determine the appropriate objective function. The SWAT model is one of the most popular hydrological models among researchers (Teshome et al. 2020; Eini et al. 2019; Zhang et al. 2020). In addition, SWAT-CUP software has been introduced by Abbaspour et al. (2007a, b). This software uses SUFI-2, MCMC, PSO, ParaSol, and GLUE for validation and verification (Abbaspour et al. 2007b). In recent years, PSO, SUFI-2, GLUE, MCMC, and many other algorithms have been used for the validation process of hydrological models (Abbaspour et al. 2007b, 2015; Narsimlu et al. 2015). However, the SUFI-2 method is more common for SWAT model validation due to its ease of use and fast computation (Kumarasamy and Belmont 2018). In this algorithm, the uncertainty of the parameters includes all sources of uncertainty of the inputs, the conceptual model, and the parameters in the modeling discussion. The 95% uncertainty estimate is calculated at the 2.5% and 97.5% levels of the cumulative distribution function of the output variable (Memarian et al. 2014). R2 and NS coefficients have also been used as objective functions in determining the goodness of fit (Memarian et al. 2013). Many researches have confirmed the effectiveness of the SWAT model. For example, (Borah et al. 2007), after comparing the SWAT model with several other models (at the watershed scale), found that this model gives better results in continuous simulation of watersheds with agricultural use. Kavian et al. (2014) chose the SWAT model to simulate runoff in the Kechik watershed in Golestan province. According to the presented results, the curve number parameters (CN), soil evaporation compensation (ESCO), available water capacity in the soil layer (SOL_AWC), and soil hydraulic conductivity in the saturated state (SOL_K) are among the most important flow control factors, respectively. Also, their results showed that the CN parameter is the most sensitive parameter and the model has simulated the time of peak discharge and the amount of peak discharge with high efficiency in the investigated stations. Himanshu et al. (2017) simulated the water balance in the Indian River catchment using the SWAT model. The results of the estimated balance have allocated 44.6% of the annual precipitation to evaporation–transpiration, and the share of runoff and infiltration into the deep aquifer was 34.7% and 19.5%, respectively. A comparison of the SUFI-2 and GLUE algorithms can be seen in the studies of Nkonge (2017) evaluated in the Tana Watershed region of Kenya. They declared that the SUFI-2 algorithm was superior to the GLUE algorithm. Ang and Oeurng (2018) simulated the runoff in a non-statistical basin of Cambodia with the SWAT model and reported acceptable results. Mengistu et al. (2019) have obtained satisfactory results in their study to calibrate and validate the SWAT model in watersheds lacking statistics in the semiarid region of South Africa using a regional approach with the physical similarity method. In a research, Rivas-Tabares (2019) used the SWAT model to investigate the water balance of the Cega-Eresma-Adaja watershed. The NS was found to be 16% for the calibration period and 67% for the validation period. The obtained results indicated the high performance of the model in simulating the hydrological conditions of large-scale basins, especially areas with cold semiarid conditions. The results of Yang et al.’s (2019) studies in Norway showed that the physical similarity method is the best method among other methods. The studies conducted to simulate watershed runoff without statistics using the SWAT model in different parts of the world and with the zoning approach show the acceptability of the results of this model. Leye et al. (2020) also presented good results in the study of the Kayanga River basin with the SWAT model for water resources management purposes.

A review of previous studies has shown that in each region of the world with different weather conditions, these methods have different results in hydrological simulations. Many commercial and open source models are available for basin hydrological simulation, which are presented in different frameworks, assumptions, and limitations, so care must be taken for its application in a specific watershed. The basin studied in this study has special mountainous conditions from the hydrological aspect and is the main source of the Zoshk–Shandiz River. Although there have been many studies, for the first time in this region runoff classification has been done on both a daily and monthly basis, and a comparison of uncertainty analysis and model estimation with the SUFI2 algorithm was done. Despite SWAT’s capability to simulate hydrological processes, the majority of previous SWAT studies have been utilizing daily and monthly inputs. Few studies have been conducted to assess the impacts of the SWAT and SWAT-CUP for hydrological simulation and uncertainty analysis of the hydrological processes simulation performance. Since the SWAT model as a physical hydrological model is capable of simulating many key hydrological processes at the basin scale, and so far frequently has been widely used to evaluate water resources management (more than 600 articles in the last two decades in journals reliable science), with reported success (Gassman et al. 2014), it was used in this research. It is a semi-distributed physically based model to predict the impact of various land uses on environment at the basin scale (Ndomba and Birhanu 2008). There is an imperative need for more researches to assess the uncertainty analysis of the arid and semiarid watersheds by SWAT model. The purpose of this research is to investigate the effectiveness of the SWAT model in estimating and simulating runoff and investigating the uncertainty of hydrological parameters in arid and semiarid regions. This research highlights some problems and prospects of using SWAT and SWAT-CUP for hydrological simulation and uncertainty analysis of the arid and semiarid watersheds (Case study: Zoshk Watershed, Shandiz, Iran). The objective of the research was to identify the sensitive parameters, calibrate, and validate the model for the study area for its subsequent application in assessing the impact of water conservation and management strategies. Checking the accuracy and performance of the optimization method mentioned in the hydrological simulations of the target basin is one of the necessities of this research. In order to achieve this, after preparing the data according to the desired format, the model was executed. After sensitivity analysis and determination of important parameters, model calibration and validation was done. Then, the results were analyzed quantitatively and qualitatively. In general, the application of this model provides a better understanding of the complex process of the Zoshk Watershed in order to optimally manage water resources.

Methods and materials

Study area

Zoshk–Abardeh basin with an area of about 9225.9 hectares is located in the area of Torghabeh-Shandiz Township. This region has a moderate climate with cool summers and very cold winters. The average annual rainfall is about 348 mm per year. The climate of the region is cold semi-humid based on the Amberje climate classification method and Mediterranean based on the Domarten method. The mean annual rainfall is about 348 mm. The runoff coefficient for the basin is 48.94%, and the volume of runoff is 18.24 (million m3) (Memarian et al. 2013). The study area is located in the Binalud zone in terms of the divisions of the geological structural states (Fig. 1). It has 3 rock units of the second and third periods (Triassic, Jurassic period) including sandstone and shale (St), shale, and phyllite and metamorphosed sandstone with numerous streaks of milky quartz (qs), shale, and phyllite, metamorphosed sandstone, quartzite (TR3j1) with Triassic and Quaternary alluvial deposits including old alluvial blocks (Qt1), young alluvial blocks (Qt2), and river deposits (Qal).

Fig. 1
figure 1

Location of SWAT model applications in Iran

Full size image

Study method

In this research, the SWAT model was used as a semi-distributed and continuous physical model in simulating the effects of different climate change scenarios. Figure 2 shows the flowchart and the study method with the model.

Fig. 2
figure 2

Flowchart and research steps using the SWAT model

Full size image

Hydrological simulation in the SWAT model is divided into two main phases. The land phase controls the amount of water, sediment, nutrients, and chemical pollution entering the main channel in each sub-basin. The riverbed phase or the routing phase controls the movement of water, sediment, nutrients, and chemical elements within the main waterway network of the sub-basin to the outlet of each sub-basin. Each sub-basin in the model is divided into subgroups called hydrological response units (HRU), which are homogeneous units in terms of soil, land use, and slope parameters. In each unit, the hydrological response of the water balance is calculated from the following formula:

$${SW}_{t}= {SW}_{0}+ \sum_{i=1}^{t}({R}_{\text{day}}- {Q}_{\text{surf}}- {E}_{a}- {W}_{\text{seep}}- {Q}_{gw})$$
(1)

where SWt is the final soil water content (mm H2O), t: time (days), SWo is the initial soil water content (mm H2O), Rday is the amount of precipitation on the i-th day (mm H2O), Qsurf is the amount of surface runoff on the i-th day (mm H2O), Ea is the amount of evapotranspiration on the i-th day (mm H2O), Wseep is the amount of water entering the vadose zone from the soil profile (mm H2O), and Qgw is the amount of return flow on the i-th day (mm H2O). The required basic maps include a digital elevation model (DEM), a land-use map, and a soil map, all three of which were introduced to the model in a raster format. Other information is related to comprehensive meteorological data, water quality, factors affecting surface flow and canal, underground water, water harvesting, land management, comprehensive information related to water quality, reservoirs, and some other fields (Nitsche et al., 2005). In SWAT, there are two methods of estimating surface runoff: the Green and Ampt infiltration method (Mao et al. 2016), and the Soil Conservation Service Curve Number (SCS-CN) procedure (Ghoraba 2015). In this study, surface runoff was calculated using the SCS-CN method. The SCS-CN is a function of soil properties, land use, and hydrological conditions.

Statistical indicators of model evaluation

To evaluate the model, the coefficient of determination (R2), bR2 coefficient, and Nash–Sutcliffe coefficient (ENS) are used:

$${R}^{2}=\frac{{\left[\sum_{i=1}^{n}\left({\begin{array}{c}Simulat\\ ed\end{array}}_{i}-{\text{Simulated}}_{\text{avg}}\right)\left({\text{Measured}}_{i}-{\text{Measured}}_{\text{avg}}\right)\right]}^{2}}{\sum_{i=1}^{n}{\left({\text{Simulated}}_{i}-{\text{Simulated}}_{\text{avg}}\right)}^{2}\sum_{i=1}^{n}{\left({\text{Measured}}_{i}-{\text{Measured}}_{\text{avg}}\right)}^{2}}$$
(2)
$${bR}^{2}= \left\{\begin{array}{c}\left|b\right|{R}^{2} if \left|b\right|\le 1\\ {\left|b\right|}^{-1}{R}^{2} if \left|b\right|>1\end{array}\right.$$
(3)
$${\text{NS}} = 1 - \frac{{\mathop \sum \nolimits_{i = 1}^n {{\left( {{\text{Measure}}{{\text{d}}_i} - {\text{Simulate}}{{\text{d}}_i}} \right)}^2}}}{{\mathop \sum \nolimits_{i = 1}^n {{\left[ {{\text{Measure}}{{\text{d}}_i} - {\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 n}}\right.\kern-0pt}\!\lower0.7ex\hbox{$n$}}\mathop \sum \nolimits_{i = 1}^n {\text{Measure}}{{\text{d}}_i}} \right]}^2}}}$$
(4)

where Simulatedavg is the average of simulated values, and Measuredavg is the average of measured values.

Calibration, validation, and uncertainty analysis of the SWAT model

In this research, the SUFI-2 algorithm (Abbaspour et al. 2007a, b) was used in the sensitivity analysis, calibration, and uncertainty analysis of the SWAT model. In this algorithm, the uncertainty of the parameters includes all sources of uncertainty of the inputs, the conceptual model, and the parameters in the modeling discussion. The degree of uncertainty is calculated by two factors called R-factor and P-factor. P-factor is the percentage of observation data that are in the 95% uncertainty estimation band. The 95% uncertainty estimate is calculated at the levels of 2.5% and 97.5% from the cumulative distribution function of the output variable obtained by the Latin Hypercube sampling method (Memarian et al. 2014). R-factor is the division of the mean band of the 95% uncertainty estimate by the standard deviation of the observed data. SUFI-2 seeks a range of parameters where most of the observed data fall within the 95% uncertainty estimate band (i.e., large P-factor, maximum 100%, with the smallest possible R-factor value, the lowest zero). Calculations in the SUFI-2 algorithm continue until most of the observed data are in the 95% uncertainty estimation band and the thickness of the band is as small as possible. (Memarian et al. 2013).

Results

Physical characteristics of the study area

In the first step, the prepared statistics and information were converted into special model formats and introduced to the SWAT model. By processing the input maps, to simulate the spatial details, this model divided the basin into several sub-basins and each of the sub-basins into several HRUs which are homogeneous in terms of land use and soil characteristics. Based on the results, the study area has 12 sub-basins and 294 HRUs with an area of 91.40 km2 (Table 1).

Table 1 Area and number of HRUs of the sub-basins of Zoshk–Abardeh basin as a result of the SWAT model
Full size table

Sensitivity analysis of parameters affecting runoff

SUFI-2 program was used for sensitivity analysis. At first, 22 parameters were found to be effective in watershed runoff and sediment production. Then, 22 parameters along with the permissible range of their changes were included in the model and 300 simulations were performed to optimize the model outputs. The SUFI-2 program provided a t-stat value for each of them, and then, the sensitive parameters were determined (Table 2). Parameters with higher t-stat values have higher relative sensitivity. It shows the values of t-stat and p value for different effective parameters in the runoff and sediment output of the basin. Examining and comparing the obtained t-stat values for each parameter shows that SOL_BD, CN2 USLE_P parameters have the highest relative sensitivity, and SLSUBBSN, GW_DELAY, and ESCO parameters have the lowest relative sensitivity.

Table 2 Parameters entered into the model and minimum, maximum values and results, t-stat, and p value after sensitivity analysis
Full size table

Note: v means replacing the existing values of the parameter with the given value, and r means multiplying the existing values of the parameter by (1 + the given value).

Calibration and validation of the model

To improve the discharge simulation results of the Zoshk–Abardeh Watershed, the SWAT model was calibrated and validated. Calibration was done using 7-year monthly discharge statistics (2000–2006) and its validation using 3-year monthly discharge statistics (2007–2010) (Table 3).

Table 3 Effective parameters in runoff and their optimal values, t-stat and p value after calibration
Full size table

Calibration and validation of runoff

The efficiency of the model was evaluated using R2, bR2, and NS coefficients. Figure 3 shows the correlation diagram and the value of the coefficient of determination between the simulated and observed values in the calibration stages.

Fig. 3
figure 3

Correlation diagram between simulated and observed data in the calibration stage

Full size image

Table 4 shows the values of model evaluation indices in the stages of recalibration and validation.

Table 4 Values of model evaluation indicators in the calibration and validation stages
Full size table

To evaluate the efficiency of the model in the simulation of base flow and peak flow, as well as to check their time compliance with real data, from the graphs related to the observed and simulated monthly flow values, during the calibration and validation periods, examining these graphs shows that the model has modeled the occurrence time of peak flow and base flow values well. However, it has estimated the peak flow values more than the actual values, which is confirmed by the simulated average monthly flow during the calibration and validation periods (Figs. 4 and 5).

Fig. 4
figure 4

Comparison of observed and simulated monthly runoff values after calibration

Full size image
Fig. 5
figure 5

Comparison of observed and simulated monthly runoff values after validation

Full size image

Uncertainty in water flow simulation

In this research, after calibration, the probability of data accuracy was determined between 92.5% and 97.5% (Fig. 6). The value of P-factor, which is one of the comparison criteria for better matching of simulated data with observational data (value equal to 0.65), indicates 65% presence of observational data in the uncertainty band (95PPU), and therefore, its value is acceptable but not desirable. One of the reasons for this is the lack of accuracy and precision in the existing discharge statistics, because there is no correct information about the amount of water collected in the gardens upstream of the area and also other different land uses. Also, there are several springs on the ridge of the area, the exact statistics of their water yield in different seasons of the year are not available, and only an approximate number of discharges are available from those springs (Mamarian, 2013). Also, the questions and answers of experts with experience in the regional waters confirm the existence of errors in the observational data.

Fig. 6
figure 6

Uncertainty band of Zoshk–Abardeh runoff after calibration

Full size image

Discussion

The goal of this study was capability assessment of the SWAT model and SWAT-CUP software in hydrological simulation and evaluation of the uncertainty of the SWAT model in estimating runoff in arid and semiarid watersheds. SUFI2 algorithm was implemented to perform model calibration and uncertainty analysis. In the stage of calibration and validation of water flow, the SWAT performance was evaluated using R2, bR2, and NS coefficients between observed and simulated records. Based on the results, the coefficients R2, bR2, and NS were estimated to be 0.75, 0.59, and 0.67, respectively, in the stage of calibration and those are 0.46, 0.24, and 0.42, respectively, in the stage of validation. The results of the model showed the model performance is weak in the stage of calibration. Also, the results of the model are acceptable, but the model performance is not significantly high. In general, more repetitions are necessary to obtain more accurate results. The results obtained from the first implementation of the SWAT model and the evaluation of the simulation accuracy indicators of this model show that the SWAT model, in the first implementation and with the default values of the parameters, has been able to correctly model the time of the occurrence of peak discharges. Obtaining low values of evaluation indices does not have acceptable accuracy for simulating the runoff flow, and recalibration and uncertainty analysis of the parameters of this model can help to improve the results and increase the accuracy of its simulation. Therefore, after this stage, the model was recalibrated to improve the simulation accuracy of the runoff discharge. Examining the indicators and graphs obtained in the calibration and validation phase of the model for simulating the monthly runoff discharge shows that the model has performed poorly in the validation phase. In general, the obtained results show acceptable ability and accuracy. Based on the results, the SWAT model is not optimal in simulating the monthly runoff discharge, and this is due to incorrect information on the amount of water harvested in the upstream gardens of the basin, as well as the lack of accurate statistics on the amount of spring water in different seasons of the year. The findings of this research confirm the results of Birhanu et al. (2007), Li et al. (2010), Jha et al. (2007), Jiang et al. (2008), Neitsch et al. (2009), Tobin and Bennett (2009), Ndomba and Birhanu (2008), Saleh and Du (2004), Setegn et al. (2008), Spruill et al. (2000), and (Tibebe and Bewket 2011), who generally confirmed the SWAT model’s ability to simulate the river flow in the studied basins as satisfactory. Rasoolzadeh darzi et al. (2022) implemented the SWAT Model in the analysis of uncertainty analysis in Iran. They found that in the monthly period, the accuracy of SUFI2 was acceptable in calibration and validation. Also, applying statistical indicators R2, bR2, MSE, RMSE, and efficiency coefficient ENS, the SWAT model presented satisfactory results.

The calibration and validation periods ranged between 6, 1, and 4 years, respectively. Data for calibration were split into two portions with nearly 70% for calibration and 30% for verification. The results of the model showed the model performance is weak in the stage of calibration. Based on the results of the sensitivity analysis of 22 parameters affecting runoff in the SWAT model, three parameters are: the USLE_P soil protection factor, wet soil density (SOL_BD), and CN among the most important parameters in determining the amount of output runoff. Among these factors, SCS-CN was recognized as the most sensitive parameter. Based on the results of the research, the SWAT model can simulate the runoff flow of the Zoshk–Abardeh basin with acceptable accuracy, and CN is the most sensitive factor affecting the runoff flow. The results of Kaibin et al.’s (2010) studies also show the different sensitivity of the modeled river flow to different parameters in different climatic conditions. Nasiri et al. (2020) simulated the stream flow in Samalqan Watershed Using SWAT model. To do this, they implemented SUFI2 algorithm to perform model calibration and validation for 1995–2012 and 2012–2014, respectively. The sensitivity analysis showed that the parameters RCHRG_DP (value of penetration into the deep aquifer), GWQMN (amount of water in the shallow aquifer to produce the base stream), ALPHA_BF (groundwater reaction coefficient), SOL_AWC (soil available water capacity), and CN (SCS curve number) had the most effect. Also, for calibration and validation periods, the Nash–Sutcliffe index and R2 coefficient were 0.65–0.80 and 0.40–0.65, respectively.

Conclusion

The quantification of the uncertainties of the parameters of hydrological models plays an essential role in the management of water resources. This process is a challenging thing that due to the large number of parameters and the lack of proper physical understanding of them, these models face problems in the calibration stage. In general, in this research, according to the analysis of the obtained indices and graphs, the model simulated the hydrological processes of the basin with relatively good accuracy. Of course, on the condition the input data are used with appropriate accuracy in the modeling, and sufficient accuracy and attention are paid in the calibration of the model so that the model can represent the real conditions of the basin as much as possible. In the calibration and validation stage, as in most studies in this field, the simulation during the calibration period has been done more accurately than the validation period, as well as the monthly time base compared to the daily one. In general, several factors are involved in the accuracy of modeling results. Some of these factors were related to the climatic and geological conditions of the basin and the collected information, including the inaccuracy and homogeneity of observational discharge data. In this study, the results of the model are acceptable, but the model performance is not significantly high. This is due to the lack of accuracy and precision in the statistics available in the region, the lack of statistics on the amount of water collected from the gardens upstream of the area, as well as the lack of statistics on the existing springs. Also, it is related to the complex nature of the dominant hydrological processes in the region, as well as human activities and interventions in the hydrological cycle in the basin. In general, by necessity, more repetitions are needed to obtain more accurate results. The model is therefore recommended for applications in catchments within Iran with similar data availability situations. The overall evaluation indicated that the SWAT model satisfactorily simulates river flows in the study catchments with limited data availability and where global spatial data are appropriate. As a final conclusion, it can be stated that the uncertainty analysis shows the high efficiency of the scenario used in this study to estimate the uncertainty of the parameters in the Zoshk Watershed. Therefore, the range of parameters optimized by this scenario can be used to conduct more research in this basin.

Implications and applications

The model is recommended for applications in catchments within Iran with similar data availability situations. The use of this model is very efficient due to the reduction of the cost of field operations to measure the required components and especially due to the reduction of the time required to analyze the issues, improve the level of water resources management, and preserve the environment. In addition, by using this tool, it becomes possible to evaluate various management programs that cannot be implemented in the short term and with reasonable costs and to make the best decision by analyzing the results. It suggested that poor catchment representation of important hydrological features such as low discharge may lead to poor performance of the model. SWAT model is a tool for water and soil evaluation that is recommended for improving watershed management. This model can be used for subsequent analyses of the basin and related sub-basins and to investigate various components of the hydrological cycle. The results of this research can be used to predict the effects of climate change and applicable management measures in the region, which are presented to the model in the form of scenarios. Due to the limited availability of hydrological data in Iran, this study has not assessed and compared the uncertainty related to the SWAT model of future runoff.