Future runoff assessment under climate change and land-cover alteration scenarios: a case study of the Zayandeh-Roud dam upstream watershed

Since the advent of the Industrial Age in the nineteenth century, direct and indirect human activities have placed tremendous pressure on water resources (Emami & Zarei 2021), not least of these destructive activity consequences have been land-cover alteration and increasing average global temperature. These two effective factors can potentially cause changes in either the hydrological process of watersheds or regional water resources management (Dibaba et al. 2020). Predicting management schemes with respect to climate and land-use change scenarios is likely to play a fundamental role in reducing water allocation pressure on planners in the coming period (Pandey et al. 2021).

Hydrological responses to climate and land-cover change are much more sensible in areas with dry climates (Zhou et al. 2013; Getachew et al. 2021; Serur & Adi 2022). The upstream of the Zayandeh-Roud dam watershed, where located in central Iran with arid and semi-arid climate conditions, not only has witnessed dramatic changes regarding land cover but also has experienced a dry climate status, given the fact that the average rates of precipitation have diminished significantly in the last decades (Sayedipour et al. 2015; Azadi et al. 2022). Moreover, Zayandeh-Roud plays a potentially essential role in providing adequate water in order to allocate for domestic, industrial, and agricultural purposes (Ohab-Yazdi & Ahmadi 2018). In this circumstance, analysing the impacts of climate change and land-cover alteration on the upstream of Zayandeh-Roud basin flow seems to be drastically important in order that policymakers would have access to necessary data by which they would be capable of making appropriate decisions regarding water supplies (Javadinejad et al. 2019).

Analysing the effects of climate change and, specifically, land-use change (Hatamkhani et al. 2022) on water resources is considered a complicated and time-consuming procedure through which the nature and extent of land resources and changes over time should be evaluated (Santillan et al. 2019). Additionally, in some regions, due to the lack of historical recorded climatic and hydrological data, evaluating the effects of these factors on water resources is likely to be much more complicated (Yaseen et al. 2019). However, through the expansion of Remote Sensing (RS) and Geographic Information System (GIS), these challenges have been overcome as though using these tools, spatial data can be analysed, and dynamic land-cover maps can be generated (Fida et al. 2020; Khawaldah et al. 2020; Zarei et al. 2021). Classification of different categories in any land-use map with acceptable results has been another challenge that seems researchers have been dealing with (Congalton 1991), but by extension of machine learning (ML) methods such as Artificial Neural Network (ANN) model Zeshan et al. 2021; Baig et al. 2022) and Support-Vector Machine (SVM) model (Singh & Pandey 2021; Thamaga et al. 2022), this difficulty has been overcome as well. There are numerous studies that have used RS and GIS effectively in order to generate land-use maps. Kundu et al. (2017) used Landsat and LISS-III data sets, land-cover maps for 1990, 2000, and 2011 were generated, and then projected land-use maps were provided for 2020, 2030, and 2040 using the Markov Chain model by which the impacts of land-cover alteration on runoff variation were evaluated; Gessesse et al. (2021) used Maximum Likelihood Classification (MLC) in a supervised classification method in order to make thematic features in 1985, 1995, 2005, and 2015 extracted. Their results demonstrated that despite the fact that the coverage of grass and shrubland areas between 1985 and 2015 decreased by 18.1 and 11.9%, the amount of land which was occupied by agriculture and urban areas increased by 29.6 and 0.53%, respectively.

Calibrating and validating a rainfall–runoff model for historical datasets is a cornerstone for assessing the impacts of climate change and land-cover alteration on water resources (Usman et al. 2021; Idrissou et al. 2022). Previous studies have effectively used different hydrological models such as the Soil and Water Assessment Tool (SWAT) (Rouholahnejad Freund et al. 2017; Jalali et al. 2021), Variable Infiltration Capacity (VIC) (Yan et al. 2020), and Hydrologic Engineering Center-Hydrological Modeling System (HEC-HMS) (Hamdan et al. 2021; Chiang et al. 2022) for surface flow assessment. Compared to other hydrological models, HEC-HMS has been much more prevalent among hydrologists due to the fact that it not only makes them provided with a simple, user-friendly model but also gives them the chance of achieving fairly accurate results in the case of calibrating a rainfall–runoff model (Zhang et al. 2013; Chakraborty & Biswas 2021). The HEC-HMS model has also been used widely in area with dry climates as well (Derdour et al. 2018; Shakarneh et al. 2022). Ndeketeya & Dundu (2021) applied the HEC-HMS model to provide a runoff-rainfall model for Crocodile (West) and Marico (Crocodile) and Upper Vaal catchments in South Africa. They calibrated inlet runoff into 21 stations based on the NS and R² coefficients and their results showed that the model was accurate enough in simulating runoff in a dry region.

As far as evaluating climate change and land-cover alteration impacts on runoff is concerned, a large number of studies have used a calibrated HEC-HMS model by which global circulation model (GCM) and Intergovernmental Panel on Climate Change (IPCC) scenarios or dynamic land-use maps have been applied and interpreted. Santillan et al. (2019) using Landsat 5 TM and Landsat 8 OLI images generated land-use maps in 1995 and 2017 for the Agusan River Basin in the Philippines and HEC-HMS was used to estimate river flow based on generated land-use maps. The results illustrated that discharge, flood depth, and flood extent increased between 1995 and 2017. In another study, Bekele et al. (2021) calibrated HEC-HMS through the period 1971–2000 for the Arjo-Didessa catchment, and then based on Representative Concentration Pathway (RCP) climate scenarios, stream-flow was estimated for 2041–2070. Their results showed that the annual runoff is expected to reduce by 1–3% in the future.

Previous studies have mostly focused on the impacts of climate change or land-cover alteration on stream-flow independently despite the fact that the combination of these factors seems to play a vital role in determining watershed management policies. In this study, first of all, land-use maps from three different years are generated based on Landsat 5 TM and Landsat 8 OLI images and then are classified into specific categories using the SVM model; secondly, a calibrated HEC-HMS model is provided based on produced land-use maps and historical climatic data over 20 years period (1988–2018) by which flow behaviour is interpreted, and eventually, the impacts of not only climate change scenarios (RCP 8.5 and RCP 2.6) but also land-cover alteration are analysed on future runoff both individually and combinatory. The results of this study will be a great help for policymakers in order to make management decisions regarding water allocation due to the water shortage that the watershed is getting used to facing.

Datasets in this study are categorized into four groups: satellite images, elevation maps, climatic data, and hydrometrical information (Table 1). First, Landsat 5 TM and Landsat 8 OLI images with pixel resolutions of 30 m for the watershed in 1996, 2008, and 2018 were derived from the United States Geological Survey (USGS) website. Prior to classifying land-use map categories, geometric accuracy, radiometrically correction, and atmospheric correction of the images were assessed. Second, a digital elevation model (DEM) map with a cell size of 30 m was obtained from the US National Aeronautics and Space Agency (NASA) database. Third, precipitation and minimum and maximum temperature information over the period of 1988–2018 which contained four rain gauge stations, four climatology stations, and two evaporation gauge stations were taken from Isfahan Regional Water Authority (IRWA) and Iran Meteorological Organization (IMO). In order to predict the impacts of climate change on runoff, Inter-Sectoral Impact Model Intercomparison Project (ISI-MIP5) GCMs under RCP 2.6 and RCP 8.5 scenarios were used. Finally, hydrometrical data including Eskandari and Ghale-Shahrokh hydrometric stations and also the amount of Koohrang and Cheshmeh-Langan discharge were collected from IRWA.

Classification of land-use maps is considered a multi-step process that requires an ML model and evaluation of some statistics indicators. In this study, the SVM model was used for the classification of different land-use map categories. SVM was initiated as a non-parametric supervised ML model based on statistical learning theory and the structural risk minimization (SRM) principle (Guo et al. 2021). It has been proved that SVM is likely to have an appropriate performance in working with small sample data, solving nonlinear problems, and identifying high-dimensional patterns (Ayesha et al. 2020). Despite the fact that SVM was organized in the latter part of the 1970s, its application in RS came into the picture in the late 2000s for the first time (Mountrakis et al. 2011). The SVM model initially was developed for solving linear classification problems with two classes; however, working with nonlinear training samples demonstrated that these samples could not be separated accurately through the original sample space (Silva et al. 2021). In this situation, a higher dimensional feature space should be used to map the samples and then a partitioned hyper-plane with a ‘maximum interval’ should be introduced in this space (Cai et al. 2017).

Training samples that are on the edge of the category distribution in the feature space play a key role in determining the location of the optimal hyper-plane (Li et al. 2021). These samples are called support vectors. Moreover, a function is needed for mapping samples from the original space to the feature space by which the hyper-planes are established accurately and the classification errors are minimized (Shah & Chandra 2021). This function is called kernel and has a number of types: Gaussian, polynomial linear, and sigmoid. Unless a kernel function is selected correctly, classification accuracy will be affected negatively. In this study, 300 samples were collected from Google Earth Images for each year and applied to satellite images using ENVI5.3 software. The SVM model was trained and validated using these samples in 1996, 2008, and 2018 in order to classify land-cover map categories in the mentioned years.

After classifying land-cover classes, some indicators are required for evaluating the accuracy of classification. A confusion or error matrix is used to determine accuracy. A confusion matrix includes details about a classification system's actual and expected classifications (Naghdizadegan Jahromi et al. 2021). In this study, the Overall Accuracy and Kappa Coefficient were evaluated in order to assess the classification accuracy.

The number of properly identified values is added together and divided by the total number of values to get the Overall Accuracy. The correctly categorized values are found on the confusion matrix's upper-left to lower-right diagonal. The total number of values in either the truth or predicted-value arrays equals the entire number of values (Calixto et al. 2022).

The CA-Markov chain model combines cellular automata, Markov chain, multi-criteria, and multi-objective land allocation to predict land-cover change over time. It adds to the Markov model not only spatial contiguity but also the probable spatial transitions occurring in a particular area over time (Mokarram et al. 2021).

Providing a calibrated hydrological model is undeniably essential for analysing the effects of either climate or land-cover change on surface water resources (KhazaiPoul et al. 2019; Akbari et al. 2022). Two mentioned studies have used the SWAT and the WEAP models to simulate runoff, while, in this study, HEC-HMS was used in order to simulate runoff by provided historical climate data, DEM map, land-use map, and soil map. Runoff is simulated using this model based on two major components: loss (i.e., the amount of precipitation that evaporates) and transform (i.e., direct runoff caused by excess rainfall at the basin outlet). These components and also the approaches used in this project for estimating them are described in more detail subsequently.

Sensitivity analysis is a technique for figuring out which hydrological model factors have the most effects on the outcome of the model. Model parameters are ranked according to how they affect the overall accuracy of the model's predictions. The most sensitive factors will cause the output response to shift the greatest, which is crucial information for model calibration. Sensitivity analysis might be local or global, according to Lane (2002). The importance of each input parameter is assessed independently in the local sensitivity analysis by holding other model parameters constant. All model inputs are permitted to change simultaneously across their ranges in the global sensitivity analysis.

Hence, in this study, a local sensitivity analysis was selected for assessing the event model. The final set of the parameters of the calibrated model was considered a baseline/nominal parameter set. Formerly, the model was run repeatedly with the starting point value for each parameter multiplied, in turn, by 0.2, 0.4, 0.6, 0.8, 1.2, 1.4, and 1.6 while keeping all other parameters constant at their nominal initial values. In addition to this, the hydrographs resulting from scenarios of adjusted model parameters were then compared with the baseline model hydrograph.

Prior to applying climate and land-cover change scenarios, the generated HEC-HMS model should be calibrated and validated according to observational data over the period of 1988–2010 for calibration and 2011–2018 for validation. The calibration process was a cyclical one: first, generated model was performed; second, simulated values were compared to observed ones, and results were evaluated using the coefficient of determination (R²) and the Nash–Sutcliffe model efficiency coefficient (E_NS) in order to recognize whether the simulation values are accurate; third; the parameter values were assessed to find out if their values are reasonable; fourth, if not, input parameters were adjusted according to guidance within reasonable ranges of a parameter value; and fifth, this process repeated until reaching appropriate simulated runoff value. In this study, the model was first performed with respect to model inputs (climatic data, land-use map, soil map, and DEM map) and then simulated inflow for the hydrometric stations (Eskandari and Ghale-Shahrokh) was analysed over the calibration period.

After calibrating the model, it was also validated based on calibrated parameters but a different set of climatic data (2011–2018) to see how precise the model is regarding generating the simulated hydrograph. The simulated inflow, then, was compared to observed figures and R² and E_NS were applied.

In this section, results are reported in three main parts: first of all, land-use map generation, classification, and the accuracy assessment procedure are reported in detail; second, HEC-HMS model calibration and validation results are presented, and then annual and monthly simulated stream-flow charts are provided; and eventually, the effects of applied climate change scenarios with respect to predicted land-use map on future stream-flow are summarized.

Classification accuracy results for generated land-use maps showed that all three maps were classified properly due to the fact that 0.8 has been determined as the minimum acceptable magnitude for both Overall Accuracy and Kappa coefficients, and for all three produced maps, these coefficients were greater than 0.8 (Table 2). Particularly, Overall Accuracy and Kappa coefficients for generated land-use maps in 2018 stood at 92.2 and 89.3%, respectively.

The allocated categories' area proportion of the total watershed area in produced land-use maps are demonstrated in Table 3. Looking at the first table data set for upstream of the Eskandari station zone in more detail, it is evident that the percentage of the watershed area which was covered by mountains was the highest at 54.55, 64.9, and 59.64% in 1996, 2008, and 2018 respectively, followed by barren areas at 35.49, 27.50, and 32.33%, respectively. In contrast, urban areas made up 1.30, 1.44, and 1.55% of the watershed area in 1996, 2008, and 2018, respectively, which was the lowest value in this data set. Finally, the percentage of land which was occupied for agricultural purposes stood at 8.64% in 1996, this number first reduced by 2.52–6.12% in 2008, and then increased slightly to 6.50%.

Turning to the second table data set for upstream of the Ghale-Shahrokh station zone, it can be seen that there was a similar trend in this zone as well. The mountainous and urban areas made up 54.80 and 1.30% of the watershed area in 1996, respectively, and in this data set, these numbers rose to 64.27 and 0.8%, respectively, in 2018. In contrast, the percentage of barren land and farmland in 1996 stood at 40.13 and 4.38%, respectively, and these numbers decreased to 32.03 and 2.89%, respectively, in 2018.

Land-use maps in 1996, 2008, and 2018 were used as CA-Markov chain model input in order to predict land-use maps in 2033. The transfer area matrix and transfer probability matrix were identified according to the land-use data from 1996 to 2018. The predicted land-use map in 2033 and its classification category area are depicted in Figure 3 and Table 3, respectively. According to the table data set for upstream of the Eskandari station zone, the percentages of farmland, barren area, and urban area are expected to grow by 2.39, 3.10, and 0.1%, respectively, in 2033 compared to 2018. There was also a similar trend in upstream of the Ghale-Shahrokh station zone, where the percentages of farmland, barren area, and urban area are expected to rise by 0.71, 2.51, and 0.05%, respectively, from 2018 to 2033.

Climatic data from 1988 to 2018, land use, soil, and DEM maps in 2018 were used to set up the HEC-HMS model. Sensitive parameters (Table 4) of the generated model then were identified to be calibrated (1988–2010) and validated (2011–2018) using hydrometrical data from Eskandari and Ghale-Shahrokh stations. Similar to previous studies (Belayneh et al. 2020; Ukumo et al. 2022), results showed that Muskingum K and X coefficients were the most sensitive parameters in both stations. K is the travel time of a flood wave passing through the reach, X is a measure of the degree of storage varied from 0 to 0.5, in which x = 0 means a level-pool reservoir or maximum storage, x = 0.5 means a pure transmission reach in which there are no storing effects. Muskingum K and X coefficients played a significant role in adapting simulated peak flow and base flow with observational data. As though, given Muskingum K = 62.3, and 42.5 with Muskingum X = 0.1, and 0.38 coefficients for Eskandari and Ghale-Shahrokh stations, respectively, the most accurate simulated inflow was obtained compared to previous tries.

The bar chart of Eskandari station's inflow illustrates that the average observed inflow in April from 1988 to 2018 in this station was the highest at 10.52 m³/s, this number was expected to reduce to 9.56 and 8.94 m³/s under RCP 2.6 and RCP 8.5 scenarios, respectively. Furthermore, the greatest decrease in projected inflow in light of climate change scenarios compared to the observed one was in March, where inflow magnitude was estimated to diminish from 8.11 to 6.54 and 5.56 m³/s, respectively. In contrast, observed and predicted inflow under RCP 2.6 and RCP 8.5 scenarios fell in the range of 0.47–0.9 m³/s, 0.41–0.78 m³/s, and 0.34–0.79 m³/s, respectively, from June to September, which was the lowest inflow value in this station.

Turning to the stream-flow chart in the Ghale-Shahrokh station, it is evident that similar to the Eskandari station, observed and projected inflow based on RCP 2.6 and RCP 8.5 scenarios had the highest value in April at 111.38, 87.07, and 80.32 m³/s, respectively, which was the greatest difference between observed and predicted inflow compared to other months. Oppositely, these numbers stood at 15.55, 12.31, and 10.28 m³/s, respectively, in October, which was the lowest amount of observed and predicted flow in this station.

After evaluating the impact of climate change on future runoff, the effect of land-cover change was assessed as well. In this regard, the first calibrated model, predicted land-use map in 2033, and RCP 8.5 and RCP 2.6 scenarios were used in order to measure runoff decrease affected by both climate and land-use changes in the future. Given the effects of these factors combinatory and also the effect of climate change independently on runoff decrease, the impact of the land-use change could be calculated easily (Table 6).

According to Table 6, the annual temperature average was expected to increase from 10.38 and 9.26 °C to 11.35 and 10.64 °C in Eskandari and Ghale-Shahrokh stations, respectively; however, annual precipitation average was anticipated to reduce from 389.47 and 406.25 mm to 376.61 and 373.34 mm, respectively.

Looking at the sixth column of Table 6 regarding impacts of both climate and land-use change combinatory in detail, it is clear that Eskandari and Ghale-Shahrokh stations' inflow was expected to decrease by 62.92 Million Cubic Meters (MCM) and 403.47 (MCM) under the RCP 8.5 scenario and 57.35 (MCM) and 295.11 (MCM) under the RCP 2.6 scenario, respectively. Given the fact that climate change (seventh column of Table 6) was expected to decrease stream-flow by 45.39 (MCM) and 346.12 (MCM) under the RCP 8.5 scenario, and 39.16 (MCM) and 233.35 (MCM) under the RCP 2.6 scenario, respectively, by itself, land-use change impact (eighth column of Table 6) was estimated to decrease stream-flow by 17.23 (MCM) and 57.35 (MCM) under the RCP 8.5 scenario and 18.19 (MCM) and 61.76 (MCM) under the RCP 2.6 scenario, respectively.

According to generated land-use maps in 1996, 2008, and 2018 (Figure 3) and classified category areas (Table 3), it is clear that the area of some specific categories has changed over the study period which normally should not have changed. For instance, the percentage of the watershed area which was a mountainous and barren area changed from 54.55 and 35.44% in 1996 to 59.64 and 27.50%, respectively, in 2018, although these areas seem to be permanently constant. The main reason for these changes has to do with the classification accuracy. Unless classification accuracy indicator magnitudes (Table 2) were 100%, the area of classified land-use map categories would not be completely accurate, which would, in turn, lead to a change in the area of classified categories through all generated land-use maps.

For urban area and farmland categories, however, the story is different. Despite the fact that the accuracy of classification has had effects on changes in area magnitudes of these two categories over the study period, there were also other reasons for area variation. First, due to the explosion of the population over the last decades, increasing the amount of land which has been covered by houses would be something inevitable. That is why the percentage of the urban area grew from 1.30 and 0.7% in 1996 to 1.55 and 0.8% in 2018 in upstream of Eskandari and Ghale-Shahrokh stations, respectively. Second, due to water management policies regarding agriculture, the area under cultivation experienced various magnitudes over the period. For example, due to adequate water availability in 1996, farmers were encouraged to cultivate crops as much as possible, which was why the percentage of farmland stood at 8.64 and 4.38% in upstream of Eskandari and Ghale-Shahrokh stations, respectively. However, because of the droughts that the watershed has experienced over the last few decades, farmers had to limit the area under cultivation based on their available water resources. Consequently, the percentage of farmland decreased to 6.5 and 2.89% in 2018, respectively.

Previous studies have used different hydrological models to provide a rainfall–runoff model in order to estimate the impact of climate change or land-use change on future water resources. Jalali et al. (2021) and Saedi et al. (2021) applied the SWAT model for runoff simulation of the Zayandeh-Roud dam upstream watershed. Their calibration results showed that was for the Eskandari station and for the Ghale-Shahrokh station. In this study, the HEC-HMS model was applied as the hydrological model of the watershed. Compared to previous studies, the calibration results of this study were accurate enough (⁠ for the Eskandari station and for the Ghale-Shahrokh station). Considering the results of this study, it can be concluded that despite the complexities that relatively distributed models such as the SWAT model have during calibration and validation, selecting a less complex model such as the HEC-HMS model could be an appropriate alternative, particularly, in arid and semi-arid regions which most of them are engaged with lack of sufficient data.

It also should be noted that relatively distributed models provide some features (modelling dynamic land-use maps) to apply management scenarios. These features are not accessible using the HEC-HMS model. If applying management scenarios is required in a specific study, the HEC-HMS model would not be an alternative. Given the fact that this study has concentrated on management scenarios, the HEC-HMS model was used practically to simulate runoff.

According to Figure 6, it is clear that runoff volume and its distribution in upstream of Eskandari and Ghale-Shahrokh stations are different. In the Eskandari station, for instance, the highest amount of stream-flow has been reported in April and March and was expected to be the highest in future as well while the lowest amount of reported and predicted inflow was in June, July, September, and October. In contrast, the differences in reported and predicted monthly inflow in the Ghale-Shahrokh station were not that significant compared to the Eskandari station. The main reason for this occurrence has to do with not just water transfer tunnels but also the geographical condition of the Ghale-Shahrokh station upstream. These tunnels discharge a significant flow into the stream-flow of the station permanently by which the stream-flow is balanced over a year. Moreover, the upstream of Ghale-Shahrokh is covered by mountains with heavy snow in winter. Converting the snow into flow in spring discharges the inflow of the station considerably, this was why the stream-flow magnitude in April and May was remarkably more than in other months.

Another point which should be taken into consideration is regarding the striking reduction of predicted inflow in the Ghale-Shahrokh station in light of climate change scenarios from March to July. Given the fact that the global temperature increase would reduce the amount of snowfall, the produced runoff caused by snowmelt in spring would decrease inevitably. As a result, Ghale-Shahrokh station's inflow was expected to reduce significantly from March to July over the period of 2018–2033.

In contrast with climate change effects on stream-flow, where the Ghale-Shahrokh station was more sensitive compared to the Eskandari station due to its topography and snowfall, impacts of land-use change on runoff reduction were more considerable in the Eskandari station. The main justification for this has to do with the more fertilized soil for cultivation in upstream of this station compared to upstream of the Ghale-Shahrokh station, which has led to higher runoff sensitivity regarding the change in land cover and specific areas under cultivation (farmland).

Jalali et al. (2021) separated human activities and climate change contributions on flow reduction in the upstream of Zayandeh-Roud dam watershed over the period of 1980–2016. According to their results, human activities had played a detrimental role in reducing runoff at (68–84)% in Eskandari station; while, climate change had contributed (16–32)%. For the Ghale-Shahrokh station, however, human activities and climate change impacts on flow reduction were almost the same at 41–49% and 51–59%, respectively. The results of this study illustrated that impact of climate change on surface flow reduction in 2033 was expected to grow compared to the historical period in both stations. Climate change was anticipated to decrease runoff by 68–72% and 79–86% in Eskandari and Ghale-Shahrokh stations, respectively.

In this study, the combined effects of climate and land-use changes were analysed by which policymakers have access to necessary information on surface flow reduction factors. This means that probable adaption strategies against surface flow reduction which will be made by planners would not be in light of only climate change impact (Javadinejad et al. 2021; Saedi et al. 2021) or land-use alteration effect (Ali et al. 2011), but the impacts of both factors and their role on runoff decrease will be considered on the management decisions. Based on the results of this study, inlet runoffs into Eskandari and Ghale-Shahrokh were anticipated to reduce by 68–72% and 79–86%, respectively, which means that planners should prioritize climate change in applying their management scenarios; however, the impact of land-cover alteration must not be underestimated. Particularly, inlet runoff into Eskandari was expected to decrease by 28–32% under land-use change conditions. Unless planners take these values into account in their predictions, water resources would face serious challenges.

Several factors contribute to the study's uncertainty. The input data, such as precipitation, temperature, runoff, land use, and soil, are the first sources of uncertainty. To make it clear, the way of recording the data, human errors, and the resolution of input maps could affect the output of the HEC-HMS and the CA-Markov models. Particularly, if the land-use maps included more resolution, the produced land-use map for the year 2033 using the CA-Markov would contain more classification classes.

The insufficient number of stations and their distributions at the basin level is a second source of uncertainty. The total number of stations that had adequate precipitation data were 10 stations. This lack of adequate precipitation stations has had a direct impact on runoff simulation accuracy, as though the more the number of precipitation stations existed, the more the accuracy of simulation would be.

Finally, because Zayandeh-Roud's upstream watershed includes water transfer tunnels and the records of these tunnels should be entered in the model manually, it is defined as a complex basin to model, and modelling complex watersheds increase the level of uncertainty (Jalali et al. 2021). However, the most critical case of uncertainty is regarding the predicted land-use map in 2033, where it has been assumed that management policies will not affect the area of classified land-use categories and they have been predicted based on historical land-use map trends.

Given the increasing prevalence of the global water crisis, identifying the impacts of climate change and land-cover alteration to variations in river flow would assist planners and policymakers in making better water resource management decisions. This study explored the impact of these factors on the future runoff into the Zayandeh-Roud dam. In this regard, first using the Landsat time-series dataset (TM and OLI) and the SVM classification approach, land-use maps with five classes in 1996, 2008, and 2018 were generated. Based on generated land-use maps and the CA-Markov chain model and to analyse the impact of land-cover variation on future runoff, a land-use map in 2033 was predicted as well. Although the percentage of urban area and farmland in 2033 compared to 2018 was anticipated to rise by 0.1 and 2.39% in upstream of the Eskandari station and 0.05 and 0.71% in upstream of the Ghale-Shahrokh station, respectively, the percentage of the barren area was expected to remain almost unchanged. In order to evaluate the impacts of climate change on future runoff, GCM outputs under RCP 8.5 and RCP 2.6 scenarios were downscaled and prepared for simulating runoff in the future. The HEC-HMS model was calibrated and validated over the period of 1988–2018 and the agreement between reported and simulated runoff values confirmed that the calibrated model could be used to simulate water balance responses to climate and land-use changes in this research region.

The effects of climate change under RCP 8.5 and RCP 2.6 scenarios and also land-use change with respect to the land-use map in 2033 were assessed in annual and monthly scales. The results showed that future runoff, affected by increasing temperature and decreasing rainfall, was expected to reduce by 68–72% in the Eskandari station and 79–86% in the Ghale-Shahrokh station. Additionally, changes in the land-use map in future were also expected to reduce the inflow by 28–32 and 14–21%, respectively, as well.

Future water resource planning should have a long-term view, adapt to new findings, and take into account these consequences. Understanding the contribution of each runoff reduction element is an important tool for policymakers in order to apply adopting management scenarios, especially in problematic and complicated watersheds. Particularly, the upstream of the Zayandeh-Roud dam watershed has been located between two different provinces, which have had always arguments regarding runoff reduction causes, as though one of the provinces refers runoff reduction to climate change impacts and the other one to land-cover change impacts. In this circumstance, climate and land-use change impacts on runoff variation should be identified in order to help managers to make great decisions.

In spite of the fact that the case study in this study was a semi-arid catchment in central Iran, the methodology which was used in this study can be applied to other watersheds, specifically those basins which has located in arid and semi-arid regions with water shortage challenges. Despite existing complexities, these approaches can be used for catchments and can be considered for larger case studies.

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

The authors declare there is no conflict.