1 Introduction

World population growth and the higher demand for water, food, and fossil fuel are among the most critical drivers of landuse and climate change. Additionally, landuse and climate change are reported to significantly alter hydrological processes and water resources over short- and long-term periods. For example, landuse substantially modifies surface runoff patterns (Ahmadisharaf et al. 2020), hydrological regimes (Awotwi et al. 2019), nutrient cycles (Delkash et al. 2018), sediment transportation (Haque et al. 2021), evapotranspiration (ET) rates (Pinto Dias et al. 2015), and water quality indicators (Tahiru et al. 2020). Likewise, climate change could potentially have a large impact on extreme rainfall (Tirupathi et al. 2018), stream discharge (Bao et al. 2019), flood events (Supharatid et al. 2016), and water resources (Tigabu et al. 2020).

The integrated landuse and climate change may have a larger overall effect on the hydrological response than either of these drivers operating individually. When the impacts of landuse and climate change are in the same direction, the hydrologic response would be amplified under the corresponding combined impact. On the contrary, the hydrologic response would be alleviated when the effects of landuse and climate change are in the opposite direction. The lack of integrated analysis implies that the hydrological studies that assess the impact of either climate change or landuse change individually are likely to either overestimate or underestimate the potential effects. Ignoring landuse change in climate change studies and ignoring climate change in landuse change studies can potentially result in misleading outcomes and consequently lead to inefficient policy and decision making. An integrated assessment of landuse and climate change will provide more valuable information to planners, watershed managers, and policy-makers and lead to better consequences for mitigating the undesirable impacts of landuse and climate change.

Although numerous studies are conducted on the topic of landuse, and climate change impacts on watershed hydrology (e.g., Mahmoodi et al. 2021; Bhatta et al. 2019; Aladejana et al. 2018), few have addressed the future combined effects of them (e.g., Dibaba et al. 2020; Torabi haghighi et al. 2020). Previous studies on the combined effects of landuse and climate change often designed hypothetical landuse change scenarios or investigated the historic landuse change patterns (Tan et al. 2014; Yin et al. 2017; Napoli et al. 2017; Li et al. 2015). As changes in landuse and climate dynamics have a potentially large impact on the hydrological response, simultaneous future landuse and climate change must be analyzed together.

Due to the complex and dynamic interaction between landuse, climate, and hydrology, various coupled hydrological models are used to quantify landuse and climate change impacts on river basins’ hydrology (Farjad et al. 2017). Among the more recently developed models, distributed physically based models are efficient tools to assess the effects of land management practices, as they can account for spatial heterogeneity and, consequently, more accurately predicting hydrological components (Aghsaei et al. 2020). In recent years, SWAT has been among the most widely used models to evaluate watershed hydrology under changing natural processes such as climate and landuse. For instance, Chen and Chang (2020) employed an SWAT model to determine the combined impacts of landuse and climate change on the streamflow in the Clackamas River watershed, USA. The results indicated an increase in peak flow and a decrease in low flow in the future. Tirupathi and Shashidhar (2020) applied it to examine water balance components’ response to the combined landuse and climate change in a large catchment in India using future landuse maps simulated from the Cellular Automata (CA) Markov Chain model. They reported that water balance components might increase by 1.5 to 2 times under the combined influence of landuse and climate changes. Shrestha et al. (2018), employing an SWAT model combined with a dynamic landuse simulation model (Dyna-CLUE) in Songkhram River Basin in Thailand, concluded that integrated landuse and climate change would decrease future streamflow and nitrate by 16% and 24.5%, respectively.

Gorganroud watershed in the Caspian Sea’s southeast coast contains ecologically significant ecosystems such as the Golestan biosphere reserve and Hyrcanian mixed temperate forests. It is also one of Iran’s most productive agricultural regions, with a high potential for the expansion of farmlands, orchards, and rangelands that supply food products for the growing population in the area (Asadolahi et al. 2017). In this watershed, the increasing need for agricultural water has resulted in water shortage and water resources pressure during recent decades. Additionally, due to intensive agrarian expansion, aggressive timber extraction, and livestock overgrazing, Gorganroud has been subjected to intense deforestation and land degradation over the past 50 years (Bahrami et al. 2010) with potentially adverse effects on various water balance components in the area. In addition to these problems, the Gorganroud watershed, as one of the country’s most flood-prone watersheds (Safaripour et al. 2012), is expected to be influenced by the increasing surface air temperature, precipitation, and extreme events as the results of global climate change (Azari et al. 2015). Climate change may also add additional pressure on water resources in this basin by increasing water demand, mainly in the agricultural sector.

Some studies have been conducted on modeling the individual impacts of landuse and climate change in the Gorganroud watershed. For instance, Azari et al. (2015) predicted an increase in streamflow and sediment yield due to a future rainfall increase. Likewise, Moghadasi et al. (2017) simulated the impact of landuse change on streamflow and concluded that a 5% increase in CN2 (SCS curve number) resulting from forest removal could enhance peak flow by up to 60%. Abbaspour et al. (2009) used a hydrologic model in Iran to assess the country’s water resources. They projected an increase in water resources for the Groganroud watershed due to climate change and increased rainfall. By predicting and classifying landuse classes for 2030, Moradi and Mikaeili Tabrizi (2020) concluded that the total water yield and flood frequency would increase due to Hyrcanian deforestation in the Gorganroud watershed. Additionally, the results of a study by Norouzi Nazar et al. (2018) show that the blue and green water resources in one of the main tributaries of the Gorganroud River would experience a substantial increase under the future climate projections. In another attempt, Asadolahi and Norouzi Nazar (2020) quantified the control of soil erosion under climate change in the Gorganroud watershed and suggested that the delivery level of erosion control service would be considerably affected by climate change. Finally, Irannezhad et al. (2018) evaluated the impacts of integrated landuse and climate over the recent decades on the Gorganroud watershed. They reported that the watershed has become wetter and warmer and has seen a considerable increase in farmlands due to landuse change. They stated that the combined landuse and climate changes have significantly increased flood intensity and frequency in the region. According to the authors, no study has investigated this watershed’s hydrological response to the future combined change of landuse and climate conditions. Therefore, this study aims to employ several analytic tools such as Land Change Modeler (LCM), Ashton Research Station Weather Generator (LARS-WG), and SWAT model to analyze the integrated impacts of landuse and climate change on future water balance components of the Gorganroud watershed.

2 Materials and methods

2.1 Research area

The study area is a part of the Gorganroud watershed on the southeast coast of the Caspian Sea in Northern Iran, which lies between 36.65–37.78°N and 55.01–56.45°E with a total drainage area of 6954 km2 (Fig. 1). The basin is located in Golestan Province, where intensive landuse changes have taken place during the last decades. Recent droughts and floods have severely affected the groundwater and surface water resources of the watershed. According to the world reference base for soil resources (FAO 1998) (https://www.fao.org), the study area’s soils are dominated by two soil groups: Leptosol (61%) and Calcisol (34.6%). Other soil groups include Cambisol (2.9%) and Regosol (1.5%). Based on the landuse map of the year 2013 classified in this study, the entire watershed consists of rangelands (42.7%), agricultural fields (35.3%), Hyrcanian forests (21.3%), urban areas (0.44%), and water bodies (0.26%). Urban developments and agricultural lands are more concentrated in the lower watershed, while forested areas and rangelands dominate the middle and upper parts. The elevation changes are rapid and range between 17 m above the sea level at the outlet and 2874 m above the sea level at the highest point of the watershed.

Fig. 1
figure 1

Geographic location of the Gorganroud watershed and gauging stations in Golestan Province, Northeastern Iran

Full size image

The climate of the study area is characterized by a Mediterranean climate with mild, wet winters and warm to hot, dry summers. Based on climate data (1976–2015) acquired from the Iranian Meteorological Organization (IMO), the basin’s annual average rainfall is 535 mm, more than 75% of which falls from November to May (wet months). The average air temperature is 4.6 °C in the coldest months (January) and 25.8 °C in the warmest month (July). The watershed’s major river originates from Eastern Alborz Mountains, flowing through Hyrcanian mixed forests and farmlands from east to west into the Caspian Sea. The continuous monitoring of discharge (1976–2015) by Gloestan Regional Water Authority (GRWA) indicates that annual streamflow at the watershed outlet (Gazaghli station) is 12.6 m3/s, varying from 3.9 m3/s in August to 31.9 m3/s in April.

2.2 Landuse change modeling

Spatial models of landuse change require information on the rate of change and where the change will occur. Land Change Modeler (LCM) makes this problem easy for the user through the implementation of the Cellular Automata–Markov Chain (CA-MC) model (Eastman 2012). LCM is now entering a phase with new possibilities for development that could help address a large range of decisions that influence human–environment systems. It has also been commonly used in different parts of Iran to report very good results (Shooshtari et al. 2020; Asadolahi et al. 2018, Shooshtari and Gholamalifard 2015). Thus, it was used in this study as a complete package for landuse change modeling to predict the landuse change in a stepwise trend from (1) change analysis, (2) transition potential modeling, to (3) change prediction (Eastman 2012).

2.2.1 Change analysis

For this study, Landsat satellite images collected by Multi-Spectral Scanner (MSS) in May 1986, Enhanced Thematic Mapper (ETM +) in July 2000, and Operational Land Imager (OLI) in August 2013 sensors, respectively, were pre-processed in ERDAS Imagine software for mosaicking, subsetting, and atmospheric corrections. These node years were selected based on data availability, spatial and radiometric compatibility between sensors, and the possibility of evaluating ground truth for classification accuracy. Also, more importantly, this time profile was selected because it represented a significant amount of landuse change in our study area, which can feed model calibration and landuse change prediction with sufficient information. These Landsat sensors have a spatial resolution of 30 m and were downloaded from the United States Geological Survey (USGS) website (http://earthexplorer.usgs.gov). The maximum likelihood method was employed for the supervised classification of the images to produce temporal landuse maps in the Gorganroud watershed. The watershed was classified into five major categories: forest, agriculture, range, urban, and water. A set of 200 control points from a field survey was used as ground truth data to verify the classification accuracy using the Kappa coefficient of agreement (Landis and Koch 1977).

After preparing landuse maps of 1986, 2000, and 2013, the LCM change analysis tab was run, and the historical landuse conversions between 1986 and 2000 were detected. LCM change modeling is based on implementing the Cellular Automata–Markov Chain (CA-MC) model (Eastman 2012). By utilizing the 1986 and 2000 landuse maps, the Markov Chain (MC) sub-model determined how much land would be anticipated to transition from the later year (Eastman 2012). Based on these analyses, a transition areas matrix and a Markov transition probabilities matrix were produced.

2.2.2 Potential transition modeling

The second step in the change prediction process is the transition potential modeling, where the potential of a landuse to transition is identified. The Cramer’s coefficients of correlation (i.e., Cramer’s V) between transformed areas and driving forces were calculated to determine the effective drivers of landuse change in the study area (Eastman 2012). A set of driver variables was selected: elevation, slope, aspect, distance from roads, distance from the forest edge, distance from residential areas, distance from agriculture, distance from rangelands, and distance from water sources. Similar studies also report that these variables are important for landuse change modeling in Northern Iran (Shooshtari et al. 2020; Asadolahi et al. 2018; Shooshtari and Gholamalifard 2015). Values less than 0.15 for Cramer’s V indicate a given variable’s weak contribution to landuse change detection (Eastman 2012). In the case of influential driving forces (Table 1), the CA sub-model conceptualized a collection of potential transition maps. After executing the MC-CA analysis by landuse maps of 1986 and 2000, a collection of conditional probability maps, a transition areas matrix, and a Markov transition probabilities matrix were generated.

Table 1 Cramer’s coefficients of correlation (Cramer’s V) between the location of landuse changes occurring for the period of 1987–2000 and driving forces of landuse change
Full size table

2.2.3 Change prediction

The third and final step in change analysis is change prediction. In this step, future landuse was predicted using the artificial neural network model named the multi-layer perception (MLP). The MLP was first calibrated using the historical rate of changes between 1986 and 2000 and predicting the landuse map of 2013. Next, the 2013 predicted map was cross-tabulated with the same year’s actual map, which was produced using satellite images in the first step (reference map). The model’s error and prediction accuracy was determined by using the null successes, hits, false alarms, and misses. Null success and hits indicate an accuracy of prediction. At the same time, false alarms and misses suggest errors due to inconsistency between predicted and reference maps (Chen and Pontius 2010). Finally, having a properly calibrated model, the landuse change was predicted up to the year 2050.

2.3 Climate change modeling

General circulation models (GCMs) are among the most widely used models to simulate future climate change scenarios in water resources and climate change studies. To select the most suitable GCM for the study area, the monthly precipitation and temperature of five GCMs (EC-EARTH, GFDL-CM3, HadGEM2-ES, MIROC5, and MPI-ESM-MR) derived from the website of the government of Canada (www.Canada.ca) and were compared with the monthly observed data for the baseline period of 1971–2000 using the correlation coefficient (R), root mean square error (RMSE), and mean absolute deviation (MAD) indices. To compare the gauging station’s data with GCMs gridded data, as all 7 gauging stations fell within a gridded network, the average data of all gauging stations were compared with the average data of GCMs gridded points. Among the considered GCMs, MPI-ESM-MR was ranked as the most suitable model for generating precipitation and temperature with higher R and lower RMSE values than other GCMs (Table 4). Therefore, it was selected to simulate future climate change scenarios in this study. This is in agreement with the results of a study by Ghahreman et al. (2015), who investigated the uncertainty of 37 GCMs output over the country and concluded that MPI-ESM-MR was the best climate model in estimating historical weather data for the period of 1960–2000 in Iran.

LARS-WG is among the two most commonly used weather generators around the world, with a high ability to capture the essential statistical characteristics of observed precipitation and minimum and maximum air temperatures (Mehan et al. 2017; Semenov et al. 1998). Thus, it was used to statistically downscale the MPI-ESM-MR data under the two most commonly used Representative Concentration Pathways (RCPs) 4.5 and 8.5, as the intermediate and worst-case greenhouse emission scenarios, respectively. In the first step, recorded daily precipitation and temperature from the 7 gauging stations (1971–2015) were used to calculate site-specific parameters of the LARS-WG stochastic generator (model calibration). To determine if LARS-WG was able to simulate the climate at each station, daily synthetic data for 300 years were generated based on site-specific parameters derived from the calibration process. To better deal with the uncertainty of stochastic weather generation, LARS-WG enables the user to generate synthetic data for any number of years (default is 300) with any random seed number (a start point when a computer generates a random number sequence). The 300-year synthetic data is broken into 10 possible 30-year time series (300/30) for the period of 1971–2000, and then the average daily value of the resulting 10 series is calculated. In the next step, to validate the performance of LARS-WG, some statistical tests were performed to compare the observed data’s statistical characteristics with those of synthetic data. The statistical tests used in this study included Kolmogorov–Smirnov (K-S) test to compare the seasonal distribution of wet/dry series and the distributions of daily precipitation and minimum and maximum temperature, Student’s t-test to compare the means of monthly precipitation and minimum and maximum temperature, and F-test to compare the standard deviation of monthly precipitation. Each of these tests computes a particular weather statistic (K-S value, T-value, or F-value) and a P-value, indicating the possibility that the observed and generated datasets may have come from the same distribution. A very low P-value means that the observed and simulated climate are not the same, and therefore, it should be rejected. While the large P-value shows that the difference between observed and generated climate is too small, and it is accepted. Semenov and Barrow (2002) suggested that a P-value of 0.01 can be taken as an acceptable significance limit for validation results. After verifying the performance of LARS-WG, the parameters were used to generate daily weather data corresponding to the climate change scenarios for the two future periods of 2021–2040 and 2041–2060.

2.4 Hydrological modeling

The process-based, time-continuous, and semi-distributed SWAT model (Arnold et al. 1998) was used to model the Gorganrouds Watershed’s hydrology. SWAT can deal with varying climate and land management activities. Table 2 provides a detailed description of the various data types used in this study. SWAT first subdivides a watershed into multiple sub-watersheds based on DEM properties, stream drainage network, and the user-defined drainage area threshold. Then, each sub-basin is discretized into multiple hydrological response units (HRUs) by overlaying landuse, soil, and slope class. This study divided the entire Gorganroud watershed into 99 sub-basins and 421 HRUs based on a threshold drainage area of 4000 ha. The surface runoff and ET were estimated using the Soil Conservation Service (SCS) Curve Number approach (SCS 1972) and the Hargreaves method (Hargreaves and Samani 1985), respectively. Also, the variable storage method (Williams 1969) was used for flow routing. Detailed information about the SWAT model can be found in Arnold et al. (2012).

Table 2 Sources of model input data and description for Gorganroud watershed
Full size table

To update the dynamic landuse and climate for the SWAT model, we used parameter ranges obtained from the baseline model’s calibration without any further changes. The hydrological response change was derived from the changes in landuse, climate, or both simultaneously. This study focused on the monthly and annual changes in ET, lateral flow, baseflow, surface runoff, total water yield, and streamflow at the watershed scale for the two future periods of 2021–2040 (2030s) and 2041–2060 (2050s). Each period involves one landuse and two climate scenarios which result in two combined scenarios. The schematic diagram of the overall processes performed in this study is illustrated in Fig. 2. For ease of expression, different scenarios implemented in this study are referenced as C1, C2 (climate scenarios for the 2030s), L1 (landuse scenario for the 2030s), C1-L1, C2-L1 (combined scenarios for the 2030s), C3, C4 (climate scenarios for the 2050s), L2 (landuse scenario for the 2050s), and C3-L2, C4-L2 (combined scenarios for the 2050s) (Table 3).

Fig. 2
figure 2

Schematic diagram of landuse change modeling, climate change modeling, and hydrological modeling in this study

Full size image
Table 3 Landuse, climate, and combined scenarios implemented in this study
Full size table

2.5 Hydrological model calibration and validation

After setup and delineation of the baseline model in ArcSWAT, calibration, validation, and uncertainty analysis were performed using the Sequential Uncertainty Fitting (SUFI-2) algorithm (Abbaspour et al. 2007) in the SWAT-CUP (Calibration and Uncertainty Program). In SUFI-2, all the sources of uncertainties are assigned to the parameter ranges. The overall uncertainty in the output is quantified by the 95% prediction uncertainty (95PPU) calculated at the 2.5% and 97.5% levels of the cumulative distribution of an output variable obtained through Latin hypercube sampling (Abbaspour 2022). Two statistics quantify the performance of a combined calibration-uncertainty analysis: the P-factor, which is the percentage of observed data bracketed by the 95PPU band, and the R-factor, which is the average thickness of the 95PPU band divided by the standard deviation of the measured data. Thus, SUFI-2 seeks to bracket most of the measured data with the smallest possible uncertainty band (Abbaspour et al. 2007). The ideal values of the P-factor and R-factor are 1 and 0, respectively. We followed the protocol for the calibration of SWAT models as provided by Abbaspour et al. (2017) to calibrate the model. Nash–Sutcliffe Efficiency (NSE; Nash and Sutcliffe, 1970), percent bias (PBIAS; Gupta et al. 1999), and the ratio of the root mean square to the standard deviation of measured data (RSR; Singh et al. 2005) were used to assess the model performance. Using monthly observed discharge at Gazaghli station, the model was calibrated from 1976 to 1990 and then validated from 1991 to 1995.

3 Results and discussion

3.1 Landuse change analysis

3.1.1 Classification accuracy assessment

The overall classification accuracy of 91.2%, 89.1%, and 87.3% was obtained for the classified imageries of 1986, 2000, and 2013. The kappa coefficient was 0.9 in 1986, 0.88 in 2000, and 0.86 in 2013. The results of LCM validation are shown in Fig. 3 as the map of correctness and error. The comparison between observed and predicted 2013 maps yielded null success = 88.04%, hits = 1.01%, misses = 6.94%, and false alarms = 4.00%. These values are acceptable, based on Chen and Pontius (2010).

Fig. 3
figure 3

Map of correctness and error according to the predicted and actual map of 2013. Null successes are correct due to observed persistence predicted as persistence. Hits are correct due to observed change predicted as change. Misses are errors due to observed change predicted as persistence. False alarms are errors due to observed persistence predicted as a change

Full size image

3.1.2 Spatio-temporal landuse change

Analysis of landuse change in the region indicates that forest was the dominant landuse in 1986 with 40.5% of the total study area, followed by rangelands (35.6%) and agriculture (23.7%). The spatial distribution of landuse classes in 1986, 2000, and 2013 and the predicted ones in 2030 and 2050 are illustrated in Fig. 4. Based on the comparison of different landuse data during the simulation period, the most apparent landuse conversion in the region is the transformation of forest to agriculture, followed by forest conversion to rangeland (Fig. 5). Major causes of deforestation in the area are the growing population’s needs for supplying food, high-potential cropping lands, livestock overgrazing, ineffective regulations of regional landuse planning, and inadequate resources to enforce protection laws. As shown in Fig. 4, the western edges of forest cover mainly experience landuse transformation from forest to agriculture, while eastern borders see landuse conversion from forest to rangeland.

Fig. 4
figure 4

Temporal changes in landuse layers of 1986, 2000, 2013, 2030, and 2050

Full size image
Fig. 5
figure 5

The change percent of each landuse category in 1986–2030 and 1986–2050

Full size image

Additionally, most of the forest covers in the north transformed into rangeland. In contrast, most of the forests in the south were converted to agricultural fields. From 1986 to 2030, agriculture and rangeland are predicted to increase by 13.6% and 9.5% of the total watershed area, respectively, at the cost of 23.8% in forest cover (Fig. 4). Moreover, based on the analysis of landuse change between 1986 and 2050, forest area is predicted to decrease by 30.2% of the total study area, while agricultural fields and rangelands are projected to increase by 18.6% and 10.7%, respectively.

3.2 Climate change analysis

3.2.1 Selection of suitable GCM and the LARS-WG validation results

The comparison results between observed monthly variables and those of GCMs data (Table 4) indicate that MPI-ESM-MR is the most suitable GCM for the study area with the highest correlation coefficient of 0.99 and 0.91 and the lowest RMSE of 3.7 °C and 0.9 mm for precipitation and temperature, respectively. Table 5 depicts the proper performance of LARS-WG in downscaling MPI-ESM-MR and generating synthetic climate variables with the same statistical characteristics as those of observed data at each station for the baseline period. It can be seen that all P-values in Table 5 are greater than 0.01 indicating acceptable assessment results (Semenov and Barrow 2002).

Table 4 Comparison results of observed monthly precipitation and temperature with monthly projections of five GCMs
Full size table
Table 5 K-S test for daily precipitation and temperature, T-test for the means of monthly precipitation and temperature, and F-test for the standard deviation of monthly precipitation
Full size table

In addition to statistical comparison, a graphical comparison was made between the monthly mean and standard deviation (SD) of observed and generated variables for the two example stations of Arazakouseh and Lezoreh (Fig. 6). The correlation coefficients ranged from 0.66 to 0.93 for all 7 rain stations and ranged from 0.98 to 1 for all 7 temperature stations indicating satisfactory results. As depicted in Fig. 6, the agreement for rain was, however, less satisfactory. This could probably be attributed to the higher uncertainties in precipitation projections than temperature projections due to larger spatial variability and seasonal dependence of precipitation (Orlowsky and Seneviratne 2012).

Fig. 6
figure 6

The mean and standard deviation of downscaled and observed monthly precipitation and temperature for the two example stations: Arazkouseh station and Nodeh station

Full size image

3.2.2 Future climate change

Gorganroud watershed will likely experience a warmer and wetter climate in the future (Fig. 7). Average annual precipitation could increase by 19.3% (RCP4.5) and 31.7% (RCP8.5) for the period of 2021–2040 and by 16% (RCP4.5) and 27.1% (RCP8.5) for the period of 2041–2060 relative to the 1975–1995 baseline of 541.7 mm. Compared to the baseline period, precipitation increase will likely be higher in the first than in the second period (Fig. 7). Moreover, during both future periods, the projection from RCP8.5 shows a higher increase in precipitation compared to RCP4.5. Precipitation increase in the Gorganroud watershed is in agreement with the global water cycle change due to climate change which implies that wet regions are getting wetter and dry regions are getting dryer in a warming climate (Held and Soden 2006; Skliris et al. 2020). Gorganroud watershed’s annual precipitation increase has been reported in previous climate change studies by Abbaspour et al. (2009) and Abbasi et al. (2012). Additionally, Ashraf Vaghefi et al. (2019) predicted a rise up to 100 mm year-1 in the Caspian zone of Iran based on the ensemble of 5 GCMs outputs.

Fig. 7
figure 7

Monthly precipitation and temperature for RCP4.5 and RC8.5 future emission scenarios compared to the baseline period

Full size image

The findings of our study show that average monthly precipitation would increase in all months, with the most significant increases being in dry months (June to October) for all climate scenarios. July shows the highest increases of 41.1% (RCP4.5) and 52.4% (RCP8.5) in 2021–2040 and 33.6% (RCP4.5) and 55.5% (RCP8.5) in 2041–2060, followed by June and September, implying a considerable potential for flooding in dry months (Fig. 7). March and May under all climate scenarios and August–October under C3 scenario are projected to see the smallest increases comparing to other months. Significant precipitation increases in all the months, especially in dry months, are consistent with the climate change trends in the region during the past decades. In this regard, Modaresi et al. (2010) used statistical and regional assessments to identify climate change trends from 1977 to 2006 in the Gorganroud watershed. They found an increasing trend for all seasons, especially summer and autumn. Future projections of precipitation by Tavangar et al. (2020) also show an increase in all months of the year in the region. As well as the results of a study by Rahimi et al. (2019) indicated a considerable increase in precipitation in dry months as a result of future climate change in Eastern Caspian sea regions. The temperature in the Gorganroud watershed will likely be higher in the future. Increases in expected average temperature are 1.2 °C and 1.3 °C in the 2021–2040 and 1.8 °C and 2.5 °C in 2041–2060 for RCP4.5 and RCP8.5, respectively. The average monthly temperature increase is not uniformly distributed throughout the year for both future periods and RCPs (Fig. 7). Monthly temperature variations indicate increases in the dry months, with the most substantial increase happening in August by 2.3 °C (RCP4.5) and 2.4 °C (RCP8.5) for 2021–2040 and 3.3 °C (RCP4.5) and 3.8 °C (RCP8.5) for 2041–2060. These results imply that the effect of climate change will be more intense during the dry months than during the wet months. Our results of temperature increases corroborate the previous findings of Norouzi Nazar et al. (2018), Azari et al. (2015), and Vaghefi et al. (2019) for the study area.

3.3 SWAT calibration and validation analysis

Before the streamflow calibration in SWAT-CUP, a global sensitivity analysis was performed by taking 25 hydrological parameters affecting streamflow. Fourteen parameters with the highest relative sensitivity values were selected for optimization at the watershed outlet (Table 6). Among the parameters, CN2 was identified as the most sensitive parameter, consistent with several other studies (Norouzi Nazar et al. 2020; Aghsaei et al. 2020).

Table 6 The most sensitive parameters to streamflow and their optimal values
Full size table

The values of 0.7 and 0.67 for NSE, 6.2 and 8.7 for PBIAS, and 0.56 and 0.59 for RSR for the calibration and validation, respectively, indicated “good” performance of the SWAT model in simulating monthly streamflow (Moriasi et al. 2007). Furthermore, based on the results of uncertainty analysis, the 95PPU band bracketed 73% of the data in calibration and 70% in validation (Fig. 8). The R-factor values of 1.12 for calibration and 1.13 for validation were around the suggested desired value by Abbaspour et al. (2007).

Fig. 8
figure 8

Comparison of simulated and observed monthly streamflow at the watershed outlet (Gazaghli station) during calibration and validation periods

Full size image

3.4 Hydrological response to landuse change

The change in annual water balance components shows ET, lateral flow, and baseflow will likely decrease in the future, whereas surface runoff, total water yield, and streamflow will probably increase (Figs. 9, 10, and 11). Generally, the changes under the L2 scenario are greater than those under the L1 scenario due to a more extensive transformation of forest to agriculture.

Fig. 9
figure 9

Monthly and annual ET for the baseline, landuse, climate, and combined scenarios during the 2030s and the 2050s

Full size image
Fig. 10
figure 10

Monthly and annual lateral flow, baseflow, and surface runoff for the baseline, landuse, climate, and combined scenarios during the 2030s and the 2050s

Full size image
Fig. 11
figure 11

Monthly and annual water yield and streamflow for the baseline, landuse, climate, and combined scenarios during the 2030s and the 2050s

Full size image

This study reveals that deforestation significantly affects ET under both L1 and L2 scenarios. As described in “Spatio-temporal landuse change,” L1 and L2 scenarios represent Hyrcanian deforestation by 23.8% (resulted from comparing the landuse of 2030 with the baseline landuse of 1986) and 30.2% (resulted from comparing landuse of 2030 with the baseline landuse of 1986) of the total watershed area. Simulation results indicate that these changes will reduce annual ET by 9.1% and 11.3% for L1 and L2 scenarios, respectively (Fig. 9 and Table 7). A possible reason for the decrease in ET is the conversion of forest to rangeland and non-irrigated agriculture. In the study area, forest removal is commonly attributed to the expansion of rangeland and rain-fed agricultural fields, which generally have a lower ET than forested areas. Therefore, deforestation in the study area typically increases ET. This is in agreement with Liu et al. (2008), who stated that deforestation to grassland and non-irrigated agricultural field would generally decrease ET, and deforestation to paddy land and irrigated agriculture would increase ET. ET reduction due to deforestation and the expansion of agriculture were concluded in previous research (Kundu et al. 2017; Awotwi et al. 2019). On the contrary, the ET increase due to afforestation activities is reported by Mwangi et al. (2016) and Woldesenbet et al. (2018).

Table 7 Annual change in hydrological components (%) during the 2030s and the 2050s between considered scenarios and the baseline scenario
Full size table

The contribution of lateral flow and baseflow to river discharge would probably decrease by 6.5% and 7% under the L1 scenario, respectively, while the surface runoff would increase by 12.7% compared to the baseline scenario. Similarly, for scenario L2, lateral flow and baseflow are predicted to decrease by 8.5% and 8.8%, respectively, and the surface runoff is expected to increase by 14.6% (Fig. 10 and Table 7). Surface runoff was the most affected water balance component by deforestation in the study area, owning the most significant relative change percent among the water balance components. These results are consistent with studies of Teklay et al. (2019) and Dibaba et al. (2020). They found that continuous expansion of cultivated lands due to forest withdrawal reduces lateral flow and baseflow while increasing surface runoff. Guzha et al. (2018) reported a significant increase of 45% in surface runoff due to forest cover losses in the East Africa region. The major cause of the decrease in lateral flow and baseflow and the increase in surface runoff is the reduction of infiltration and soil water capacity due to soil compaction resulted from forest removal, causing precipitation to largely turn into runoff. Based on our analyses, deforestation will probably induce an increase of 4.5% and 5% in total water yield (the summation of surface runoff, baseflow, and lateral flow) and an increase of 4.1% and 4.6% in streamflow under L1 and L2 scenarios, respectively (Fig. 11 and Table 7). Deforestation is widely known to be associated with an increase in water yield and streamflow, primarily due to the reduction of transpiration because forest trees have higher ET than other landuses. Furthermore, other landuse/landcovers from deforestation, such as rangelands and agricultural fields in the Gorhanroud Watershed, lead to a lower storage capacity of soil water due to smaller root penetration. This could result in a decrease in infiltration rate and groundwater recharge and an increase in surface runoff. In this study, the baseflow and lateral flow reductions were offset by the increase in surface runoff, which increased total water yield and streamflow. This implies the significance of surface runoff in water balance components and its contribution to streamflow. Increasing water yield and streamflow due to deforestation are in line with the findings of Guzha et al. (2018), Munotch and Goyal (2019), and Dibaba et al. (2020). They reported that forest removal and agriculture expansion would increase the total water yield and streamflow in East Africa, India, and Ethiopia, respectively.

On the monthly scale, landuse change’s impact of landuse change on water balance components is more discernible than the annual (Wang et al. 2018; Aghsaei et al. 2020). The monthly effects of landuse change show decreased ET, lateral flow, baseflow, and an increase in surface runoff in all months. Output results also indicate an increase in monthly water yield and streamflow except for the summer and early autumn months (Figs. 9, 10, and 11 and Table 7). The significant decreases explain the exceptions in water yield and streamflow in lateral flow and baseflow, which cannot be offset by the increase in surface runoff during the dry months, leading to a decline in water yield and streamflow. As a result, the ET, baseflow, and surface runoff changes would be greater in dry months than in wet months. The largest change is predicted to be under the L2 scenario by 24.9% (September), 10.8% (October), and 72.5% (August) for ET, baseflow, and surface runoff, respectively. On the contrary, the changes in lateral flow, water yield, and streamflow would be greater during the wet months than during the dry months. The largest change in lateral flow, water yield, and streamflow is projected to be under the L2 scenario by 13.3% (December), 9.2% (March), and 8.9% (March), respectively.

3.5 Hydrological response to climate change

The hydrological response of the watershed to climate change was found to be more intense than landuse change. All climate change scenarios indicate that the Gorganroud watershed will significantly increase all hydrological components compared to the baseline scenario due to an increase in precipitation (Figs. 9, 10, and 11 and Table 7). Changes in hydrological variables are greater during the first period than in the second one. Likewise, hydrological components will increase more under RCP 8.5 (C2 and C4 scenarios) than under RCP4.5 (C1 and C3 scenarios). As described in “Selection of suitable GCM and the LARS-WG validation results,” scenarios C2 and C4 represent precipitation increase by 31.7% and 27.1%, respectively, and temperature rise by 1.3 °C and 2.5 °C. Combining an increase in precipitation with the rise in temperature leads to increases of 25.3% and 25.6% in ET under C2 and C4 scenarios, respectively (Fig. 9 and Table 7). The equal amount of ET increase in both C2 and C4 scenarios is attributed to the smaller increase of rainfall (27.1%) combined with a larger increase in temperature (2.5 °C) in C4 in comparison to those under the C2 scenario (31.7% and 1.3 °C). The increase in ET as a result of climate change is mentioned by Desai et al. (2021), Yan et al. (2019), and Bajracharya et al. (2018).

Baseflow and lateral flow were the most and the least influenced water balance components by climate change, respectively (Table 7). Modeling outcomes indicate increases of 23.1% and 16.1% for lateral flow, 194.5% and 98.3% for baseflow, and 43.1% and 27.8% for surface runoff under C2 and C4 scenarios, respectively. After the baseflow, the total water yield is the most affected component by climate change. Average annual water yield and streamflow are projected to increase by 47.4% and 45% under C2 and by 33.4% and 29.3% under the C4 scenario, relative to the reference period (Fig. 11). The projected increase in streamflow is similar to the results of Bajracharya et al. (2018) in the Kaligandaki basin in Nepal. They reported a rise of 50% in streamflow due to an increase of 26% in precipitation. Azari et al. (2015) also mentioned an increase of 21.8% for streamflow under HADCM3 projections in the Gorganroud watershed in the same hydrometric station and the same future period (2040–2069).

There is an increasing trend in the monthly response of hydrological components under all climate scenarios (Figs. 9, 10, and 11). However, there are some exceptions for the surface runoff and lateral flow. For instance, the surface runoff will decrease in May for all scenarios, and lateral flow will decline in September and October for the C3 scenario. The possible reason for these exceptions is the slight precipitation increase and the significant temperature increase in these months. As a consequence, ET significantly increases, causing surface runoff and lateral flow to decrease. This is consistent with Gleick (1987), who stated that more ET resulting from higher temperatures would cause a corresponding reduction in surface runoff.

Generally, the largest increases in hydrological components due to climate change occur in dry months except for the baseflow, which increases almost uniformly throughout the year. The highest increase in ET, lateral flow, baseflow, surface runoff, water yield, and streamflow are predicted to occur in July, November, July, October, November, and November by 52.4%, 49.6%, 196.2%, 250%, 120%, and 105%, respectively, under the C2 scenario. Although the maximum precipitation increase is projected for July, the maximum increase of water balance components is not projected just for July, implying the significance of temperature increase and the interactions between water balance components and climate variables. Other studies also reported larger changes in water balance components during the dry months compared to wet months (Worqlul et al. 2018; Yang et al. 2019).

3.6 Hydrological response to the combined landuse and climate change

Hydrological response to the combined landuse and climate change provides a deeper and more accurate understanding of what will happen in the watershed in the future. In this study, the decrease in ET, lateral flow, and baseflow under landuse scenarios decline their increase under climate scenario, resulting in an alleviated response (the smaller increase) under the combined scenario (Figs. 9 and 10 and Table 7). Under combined scenarios, annual ET, lateral flow, and baseflow would increase by 13.9%, 14.6%, and 166.7% in the first period for the C2-L1 scenario and by 11.4%, 7%, and 77.2% in the second period for the C4-L2 scenario, respectively. In contrast to ET, lateral flow, and baseflow, the increase in surface runoff, water yield, and streamflow under both landuse and climate scenarios results in an amplified response (the greater increase) under the combined scenario (Figs. 10 and 11 and Table 7). Under combined scenarios, annual surface runoff, water yield, and streamflow would see an increase by 54.2%, 50.5%, and 48.6% in the first period for the C2-L1 scenario and by 41%, 35.9%, and 32.1% in the second period for the C4-L2 scenario, respectively.

It is interesting to note that when the impacts of landuse and climate change are in the opposite direction (ET, lateral flow, and baseflow), the impact of landuse change is enhanced under combined scenarios (Table 7). For example, if landuse scenario L1 shows a 9.1% decrease for ET, and climate scenario C1 results in a 21.2% increase, then this leads to an increase of 10.2%, which is smaller than the summation of the two numbers (e.g., 21.1–9.1% = 12.1%). The combined scenario C1-L1 accounts for landuse change enhancement. In contrast, when the impacts of landuse and climate change impact are in the same direction (surface runoff, water yield, and streamflow), landuse change’s impact of landuse change is diminished under combined scenarios. For instance, the surface runoff increase of 12.7% under landuse scenario L1 and the surface runoff increase of 19.7% under climate scenario C1 result in an increase of 31%, which is smaller than the summation of both values (e.g., 12.7 + 19.7% = 32.4%). Therefore, the combined scenario C1-L1 implies a reduced landuse change impact. Different behaviors of landuse change are found in various studies when combined with climate change. For example, the results of a study by Tan et al. (2014) in the Johor River basin, Malaysia, indicated the reduced landuse change impact on streamflow under combined scenarios. Zuo et al. (2016) also reported that the impact of landuse change on water yield is not obvious when it is combined with climate change impact in the Huangfuchuan River basin, China.

On the monthly scale, hydrological components’ response to the combined scenarios differs slightly from the annual one. For example, for some hydrological components, the change under combined scenarios is in the same direction as landuse scenarios. It is in contrast to the annual response in which the changes under the combined scenario are all in the same direction as the climate scenarios (Table 7). Some examples are the change of ET under C3-L2 and C4-L2 scenarios during the dry months and lateral flow change under C3-L2 scenarios during almost all months. In these cases, the decrease under the landuse scenario cannot be offset by the increase under the climate scenario, resulting in a reduction under the combined scenario.

Like climate scenarios, the largest increases in hydrological components under integrated scenarios are seen for the dry months except for the baseflow, which almost increases uniformly throughout the year. Occurring the largest increases in runoff and streamflow in the dry months creates conditions favorable for more frequent, larger, and higher intensity floods in the area, as one of the most prone Iranian watersheds to flooding. Flooding is historically common in the Gorganroud watershed, especially during dry months. The integrated landuse and climate change scenarios represent that the Gorganroud watershed would experience more extensive and more intense flooding events in the future, especially in dry months in which both landuse and climate change have the highest impacts.

4 Conclusions

This study demonstrates the importance of dynamic quantification of individual and combined landuse and climate change impacts on the Gorganroud watershed’s hydrology. Both landuse and climate change are projected to have a prominent effect on water balance components. Comparatively, climate change indicates a more pronounced impact. However, landuse change significantly alters the hydrological regime, and its influence cannot be ignored. It is evident that improving landuse planning and management efforts can mitigate climate change’s undesirable impacts on the hydrological cycle. This study’s findings show that the future integrated landuse and climate change will force an increase in all water balance components. Concerning the projected increase in precipitation, surface runoff, total water yield, and streamflow, there does not seem to be a water scarcity problem in the Gorganroud watershed in the future. However, dynamic landuse and climate change are the two critical challenges in this watershed that would increase the frequency and intensity of flooding. As impacts of both landuse and climate change are predicted to be larger during dry months than wet months, the region’s historical potential of flooding in dry months is expected to increase. This study’s findings could provide valuable information to alleviate severe floods due to climate change through effective landuse planning and management.

While the results of this research are useful findings, there were some limitations that affected the study. Our study has lacked a high-resolution soil map and its physical and chemical properties. As it is known, soil plays a crucial role in the hydrological cycle and it can capture and store water, making it available for crop absorption and thus minimizing surface evaporation. Therefore, large-scale soil databases can be a significant source of uncertainty to hydrological studies under landuse and climate change. The low spatial and temporal resolution of rainfall and temperature networks was another limitation of our study. Moreover, spatially distributed hydrological models require a fine-resolution gauges network to simulate proper streamflow in watersheds, and unreliable weather gauges network can bring challenges to everyone who wants to calibrate the model. Another limitation was that the model did not provide a basis to address multicultural plant communities in the simulation process, which are common in agricultural fields, rangelands, and Hyrcanian forest ecosystems in the Gorganroud watershed as they were originally configured for monocultures. In addition, SWAT has limitations in the spatial representation of HRUs across each sub-watershed. It made the model simple and supported the application of the model to almost every watershed. In this regard, landuse, soil, and slope heterogeneity of the model are considered through sub-watersheds. This approach disregards flow routing among different HRUs. In terms of model calibration, prior knowledge is needed to optimize many parameters during the calibration process which discourages using the model. Disregarding and underestimating the significance of parameters could result in incorrect model outputs. However, the SWAT-CUP is an important improvement for the calibration procedure: Uncoupling of SWAT and SWAT-CUP decreases the efficiency of modeling.

As an intrinsic aspect of any modeling study, simulation results and findings of this study contain uncertainties. In this study, future landuse change is predicted based on historical landuse change, which can be a source of uncertainty. In addition, future landuse changes may be affected by future policies, strategies, and human activities that are difficult to predict. On the other hand, future climate change scenarios are generated based on the projections from only one GCM. Similarly, this can also be another source of uncertainty. Furthermore, the uncertainty of watershed hydrological modeling was not addressed in this study. Future works can improve the prediction of future landuse and climate change and address the uncertainty in their projections and the uncertainty in hydrology modeling. This can result in a better understanding and interpretation of the effects of landuse and climate change.

Our research findings could contribute to effective landuse management and water resources planning under the future climate change in the Gorganroud watershed. As land degradation and climate change continue in the watershed, integrated land and water resources planning is needed to deal with the potential changes in the hydrologic regime. This study’s results could provide a helpful guideline for local authorities and decision-makers to better understand the future hydrological response to the combined landuse and climate change to make future strategies, policies, and water resource management plans.