1 Introduction

Floods are the most common natural disasters that cause human and financial losses (Shah et al. 2018; Vafakhah et al. 2020). Floods in Iran are primarily caused by natural-equilibrium disturbances such as land use changes and human intervention (Modarres et al. 2016; Peyravi and Marzaleh 2019; Yousefi 2020). Inconsistencies in land development capability and land use selection, as well as non-standard usage such as deforestation and subsequent increase in agricultural lands and urban development, result in higher discharge peak volume, increased flood occurrence, increased sediment, and erosion, all of which have negative environmental, economic, and social consequences (Calsamiglia et al. 2018; Llena et al. 2019; Sepehri et al. 2021). As a result, a better assessment of the effects of land use changes on flooding can be used as a powerful tool to reduce flood risk, which is one of the major issues in catchment planning, management, and sustainable development (Ali et al. 2011; Mustafa and Szydłowski 2020; Zuo et al. 2016). Hydraulic/hydrology models have become increasingly popular in recent years for assessing the impact of land use change on flooding severity. Dadhwal et al. (2010), for example, used VIC hydrological model to simulate the effect of land use changes in Mahanadi basin on rate of stream flow. Miller et al. (2002) used the Hydrologic Engineering Center-Hydrologic Modeling System (HEC-HMS) model to investigate the effects of land use change and vegetation on the hydrological responses of two United States (US) watersheds. In this regard, they prepared land use maps for these two basins at various intervals and in the forest, agriculture, urban areas, and bare lands classes. They concluded that from 1973 to 1997, the average annual runoff increased due to reduced forest area, agricultural land expansion, and urbanization.

Sajikumar and Remya (2015) used the Soil and Water Assessment Tool (SWAT) hydrologic model to investigate the effects of land use change on runoff parameters in India's Kerala watershed basin. The results showed that forestland decreased by 32%, resulting in a 15% increase in the watershed basin's discharge peak.

Brun and Band (2000) used Geographic Information Systems (GIS) and the Hydrological Simulation Program—FORTRAN (HSPF) to investigate the effects of land use change and urban development on the hydrologic behavior of the Glenfalls basin in the USA. According to the findings of this study, the change in land use resulted in a 20% decrease in base discharge flow. They also concluded that the severe impact threshold of urban development on the increase in runoff coefficient would be determined when 20% of the basin area is impenetrable.

Sanyal et al. (2014) investigated the impact of land use change in sub-basins on flood peak discharge at basin outlets in Eastern India. They began by preparing the land use maps for the 1976–2004 period using Landsat satellite images and ENVI software. They then investigated the effects of different land uses on the flood discharge peak using the HEC-HMS hydrologic model and the Natural Resources Conservation Service curve number (NRCS-CN) method.

Gao et al. (2020) used the HEC-HMS model to evaluate the effects of land use change on the hydrological responses of the drainage basin and discovered that the flood peak and volume in 2028 increased by about 3.5–2.9%, respectively, when compared to the flood peak and volume in 2010 in Qinhuai River Basin in the China.

Using the HEC-HMS model, Koneti et al. (2018) investigated the effects of land use change and vegetation on the hydrological responses of the Godavari River basin in India. According to the study, deforestation at the expense of urbanization and cropland expansion reduces overall evapotranspiration (ET) and infiltration while increasing runoff.

According to the studies cited, the effect of land use changes on runoff characteristics occurs locally. As a result, it is critical to assess the impact of land use change on runoff characteristics in the study area. Thus, the purpose of this study was to investigate the effects of land use change on flood hydrographs in Ekbatan Dam catchment basin during the years 1985, 2000, and 2015, using the HEC-HMS model, which is one of the most widely used hydrological models. The model for design flood estimation used some transformation methods, the most well known of which are the Soil Conservation Service (SCS) unit hydrograph and the Snyder unit hydrograph. In the present study, the accuracy of flood estimation methods was evaluated, and the best method was used to assess the effect of land use changes on flooding degree.

2 Materials and methods

2.1 Description of the case study

The study area of the Ekbatan Dam basin is located in southeast of Hamadan province, Iran, at the “34′ 34°”, “34′ 45°” northern latitudes and “48′ 42°”, “48′ 28°” eastern longitudes (Fig. 1). This area is one of the Gharachai River's sub-basins, located near Alvand Mountain to the southwest. The elevation map shows elevations ranging from 1948 to 3442 m above sea level. Based on Ekbatan Dam climatology data, the average annual rainfall and temperature are 343.11 mm and + 10.75 °C, respectively.

Fig. 1
figure 1

Geographical location of the study area

Full size image

Thus far, studies show that land use change in Ekbatan Dam's catchment area has influenced important hydrological parameters in the basin in recent decades. These parameters are as follows: 1. An increase in deposited sediments in the case study's outlet (Ekbatan Dam) as a result of land plowing, 2. Reduction in extractable water resources as a result of flow coefficient reduction, resulting in: A. decrease in the volume of water behind the Ekbatan Dam, B. A case study of irrigated land drying, not utilizing the majority of the dam lake's volume capacity, 3. Increase in spillway design flood as a result of increased flood potential caused by land use changes.

2.2 Required data

Different land uses were extracted from remotely sensed data to determine the hydrologic response to land use changes. Satellite images from the Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM +), and Operational Land Imager (OLI) sensors, as well as the Landsat satellite, were used to assess land use patterns and their changes over time in 1985, 2000, and 2015. The images were obtained from the website of the United States Geological Survey (USGS) (https: /earthexplorer.usgs.gov/) (Fig. 2). It was attempted to select images with high quality, that were cloud-free, and had a limited rate. Furthermore, in order to have a similar seasonal variation in land uses, the authors selected images that were taken during the same season, from July and September. In addition, all images processed with ENVI software were georeferenced and stacked to ensure clear images during the classification and accuracy assessment stages.

Fig. 2
figure 2

Real images of each year by a TM1985 sensor, b ETM + 2000 sensor, and c OLI 2015 sensor

Full size image

The land use classes of the desired years were merged, and six (6) new classes were created, namely rock outcrops, agriculture, residential, grass lands, water-covered, and dry farming. ENVI 4.5 software evaluated the accuracy of classified images from 1985, 2000, and 2015 using four (4) most popular factors, including total accuracy, kappa coefficient, manufacturer accuracy, and user precision, which were based on a comparison of the classified image and reference data, which accurately reflected the true land cover (Dissanayake et al. 2019; Lu and Weng 2007; Ranagalage et al. 2019). Meteorological, soil, and watershed characteristics were also required to simulate the rainfall-runoff process. The daily rainfall in Hamadan Province was obtained from the general meteorological office for 1985–2015. The observed discharge of events was obtained from the General Department of Natural Resources of Hamadan Province in order to calibrate and validate the data. In addition, the curve number (CN) as a soil infiltration agent was obtained from the General Department of Natural Resources of Hamadan Province. A digital elevation model (DEM) (scale 1: 25,000) was obtained from the Shuttle Radar Topographic Mission (SRTM) dataset to calculate watershed characteristics for HEC-HMS model inputs (using ArcGIS10.7 software) (such as lag time of the case study).

2.3 Detection of land use change

After developing land use maps using satellite imagery, the trend of different usage variations was studied over a thirty-year period. This analysis is extremely useful for management decisions in identifying and concluding that a class has been changed to other classes over a specified period of time (Chang and Franczyk 2008; Mustafa and Szydłowski 2020; Rahman 2016).

2.4 HEC-HMS model

The US Army Corps of Engineers developed the HEC-HMS model, a semi-distributed and conceptually based model, to simulate the rainfall-runoff process in a watershed system. This model has been widely applied to the study of water resources, urban drainage systems, flood frequency, flood warning system planning, reservoir overflow design, and other topics (Gao et al. 2020; Sepehri et al. 2018, 2019). This model is run by a system of interconnected components, each of which simulates a different aspect of the rainfall-runoff process in a sub-basin or sub-area. In general, this model consists of three major components: the basin model, meteorological data analysis, and control specifications (Ali et al. 2011). A basin model is a graphical user interface that displays components of a hydrological system (such as stream length, elevation, and slope) and their interactive relationships. The SCS-CN method was used for runoff estimation due to its simplicity, flexibility, reliance on only CN, and acceptable accuracy (Halwatura and Najim 2013; Meresa 2019; Sepehri et al. 2018, 2019). The CN parameter is determined by the main characteristics of runoff generation, which include soil hydrological groups, basin land use (agriculture, forest, and urban), hydrologic status, and soil antecedent-moisture status (Mishra and Singh 2013). In this study, the constant monthly discharge method was used. The separator line of the base flow and direct runoff in this method begins at a point of minimum discharge from the onset of the flood runoff and continues in a straight line paralleled with the time axis until it cuts the falling limb (Rajkumar et al. 2021; Smakhtin 2001; Szilagyi and Parlange 1998).

Similar to Chen et al. (2009) and Malekinezhad et al. (2017), evapotranspiration (ET) losses were ignored due to the intensity of the storm event and the assumption that the ET volume is negligible in comparison with the runoff volume. For transformation, the SCS unit hydrograph and Snyder unit hydrograph methods were investigated, and the Muskingum–Cunge method was applied to routine river flow to basin outlet. HEC-HMS Technical Manual can be referred for more details on the model’s description, as well as other options and parameters (USACE 2003).

2.4.1 Model calibration and validation

The sensitivity of the model to parameter change was investigated prior to calibration and validation of the used model. To this end, the results of the model curve variations were plotted for each of the input HEC-HMS model parameters. As the slope of the line rose, a small change in the desired parameter caused a large variation in the model's response, so the model is referred to as sensitive (Cunderlik and Simonovic 2004; Ouédraogo et al. 2018). In this study, three parameters were examined: lag time, CN, and initial abstraction (an empirical parameter for all losses prior to runoff calculated using CN). Following sensitivity analysis, a rainfall-runoff model was calibrated by selecting several flood events and changing the values of more sensitivity parameters within a reasonable range to match the simulated hydrograph to the observed hydrograph with the highest possible accuracy.

Following the completion of the calibration process, the model was validated with at least one more rainfall-runoff event. The model would be acceptable if it rebuilds the observed hydrographs of the hydrometric stations with acceptable accuracy and without changing the calibrated parameters (Katwal et al. 2021; Teng et al. 2018). Validation would continue until the best fit is found. Otherwise, because of the poor matching in the validation process, the calibration process is repeated by changing the calibration parameters (Mandal et al. 2016; Roy et al. 2013). Three criteria were used to evaluate model performance accuracy: root-mean-square error (RMSE), coefficient of efficiency (CE), also known as the Nash–Sutcliff coefficient, and R-square (R2) (Eqs. 1–3).

$${\text{RMSE}} = \sqrt {\frac{{\sum {\left( {Q_{{\text{o}}} { - }Q_{{\text{e}}} } \right)} }}{n}}^{2}$$
(1)
$${\text{CE}} = 1 - \frac{{\sum {\left( {Q_{{\text{o}}} - \, Q_{{\text{e}}} } \right)^{2} } }}{{\sum {\left( {Q_{{\text{o}}} - \, \overline{{Q_{{\text{o}}} }} } \right)^{2} } }}$$
(2)
$$R^{2} = \, \frac{{\left[ {\sum {\left( {Q_{{\text{o}}} - \, \overline{{Q_{{\text{o}}} }} } \right)\left( {Q_{{\text{e}}} - \, \overline{{Q_{{\text{e}}} }} } \right)} } \right]^{2} }}{{\sum {\left( {Q_{{\text{o}}} - \overline{{ \, Q_{{\text{o}}} }} } \right)^{2} \sum {\left( {Q_{{\text{e}}} - \overline{{ \, Q_{{\text{e}}} }} } \right)^{2} } } }}$$
(3)

where Qe and Qo are the estimated and observed hydrograph at the moment t. The larger value of RMSE, CE, and R2 shows the higher model capability for simulations (Ali et al. 2011; Chen et al. 2009).

3 Results and discussion

Because of the effects on water resources, the hydrological response to human activity-induced land-cover changes in the Ekbatan Dam basin has received a lot of attention in recent years.

3.1 Land use map of Ekbatan Dam basin

The maximum likelihood method was used in the present study to calculate the classification accuracy of the produced land use change maps. The rate of accuracy was calculated by dividing the number of correctly classified reference points in a given class by the total number of reference points in that category. This ratio can be used as an omission error indicator (Mustafa and Szydłowski 2020; Story and Congalton 1986). The commission error was measured by the user's accuracy, which is the ratio of the number of correctly classified reference points of a specific class to the total number of points classified as that category (Fung and LeDrew 1988). The accuracy assessment for three images (TM (1985), ETM + (2000), and OLI (2015)) revealed an overall classification accuracy of more than 85%. Furthermore, the kappa coefficient computes the classification in comparison with a completely random classification. In this case, more accuracy is obtained than with an image that is classified completely randomly (Pontius et al. 2001; Story and Congalton 1986; Wu et al. 2019). The kappa coefficient has an advantage over overall accuracy in that it calculates accuracy using marginal (non-diagonal) elements of the error matrix (Mustafa and Szydłowski 2020). In this study, the calculated kappa coefficient using the maximum likelihood method was greater than 0.8. Different satellite images were used in the current study, and each image was analyzed and assessed separately (Table 1).

Table 1 Accuracy of classification coefficients using the maximum probability method
Full size table

The rate of accuracy is determined by the quality of remote sensing data, which includes the resolution, available bands, and image quality (Alexakis et al. 2012; Iino et al. 2018; Jin-Song et al. 2009). There was a limitation in this study to examine the accuracy of classified maps due to the nature of land use changes, which vary depending on regional conditions, and also due to the lack of availability of ground values.

After developing land use maps using satellite imagery, the trend of different usage variations was studied over a thirty-year period. Figure 3 depicts each land use type and its changes over the specified time period.

Fig. 3
figure 3

Land use maps of the case study in a 1985, b 2000, and c 2015

Full size image

The following is an explanation of land use change in the basin, based on Fig. 3 and Table 2: Residential use occupied 1.14 km2 in 1985, accounting for 0.51% of the basin’s total area; this would increase to 1.37 km2 in 2000 (0.61%) and 1.71 km2 in 2015 (0.76%). In 1985, grassland covered 157.89 km2, accounting for 70.74% of the basin’s total area. In 2000, the area of this use was 156.34 km2 (70.03%), and it would be 154.51 km2 in 2015. (69.2% of the total area of the basin). In 1985, agricultural lands and gardens covered 24.61 km2, accounting for 11.02% of the basin’s total area. This area would grow to 21.17 km2 (9/49%) in 2000 and 20 km2 (8.8% of the basin’s total area) in 2015. In 1985, dry land farming covered 31.64 km2, accounting for 14.17% of the basin’s total area. It would reach 36.4 km2 (16.31%) in 2000 and 39.07 km2 (17.68% of the basin’s total area) in 2015. In 1985, 2000, and 2015, the area of the rock outcrop without vegetation was 7.06 km2, accounting for 3.16% of the basin’s total area. Water-covered lands had an area of 0.91 km2 in 1985, 2000, and 2015, accounting for 0.4% of the basin’s total area.

Table 2 The area of total land use change in the target years
Full size table

3.2 Model sensitivity, calibration, and validation

In the case study, the values of each parameter were changed, and the results were analyzed at the basin outlet. The values of these three parameters were changed from − 15% to + 15% with a 5% interval, and the effects on flood peak discharge were studied. As shown in Tables 3, 4, 5, the model’s sensitivity to CN variations was high, so calibration processes were performed based on this parameter for four different flood events.

Table 3 Results of model sensitivity to lag time
Full size table
Table 4 Results of model sensitivity relative in proportion to initial abstraction
Full size table
Table 5 Results of model sensitivity to curve number (CN)
Full size table

In the flood event of March 23, 2012, the observed and calculated peak discharges before calibration were 11.1 and 13.4 m3/s, respectively, indicating a − 20.7% error, and after calibration, the calculated peak discharge would be 10.9 m3/s (1.8% error), whereas in the Snyder method, the calculated peak discharge values before and after calibration were 8.2 and 9.4 m3/s, respectively, with an error of 26.12 and 15.31% compared to the observed peak discharge (Fig. 4a–d).

Fig. 4
figure 4

Observed and calculated hydrograph for considered flood events

Full size image

In the case of April 17, 2001, the output of the SCS and Snyder methods before calibration was 6.8 and 10.1 m3/s, respectively, which, when compared to the observed peak discharge (8.4 m3/s), showed errors of 19.04 and − 20.23%. The error of calculated discharges for SCS and Snyder methods would then be 4.76 and − 8.33%, respectively, after the calibration process (Fig. 4e–h).

The observed and calculated peak discharge before calibration for February 12, 2002 event using the SCS method were 7.4 and 9.2 m3/s (− 24.32% error), respectively, and 7.4 and 5 m3/s (32.43% error) using the Snyder method. After calibration, the calculated peak discharge in the SCS and Snyder methods was 7 and 6.3 m3/s, respectively (Fig. 4i–l). Furthermore, in the event of November 2, 1997, the performance of the SCS method in calculating the peak discharge before and after calibration was 32.45% and 1.9%, respectively, whereas these values for the Snyder method were − 22.51% and 11.9%, respectively (Fig. 4m–p).

The calibration results for SCS and Snyder showed that both methods had acceptable accuracy in modeling the rainfall-runoff process, with the exception that the SCS unit hydrograph had more accuracy. On the other hand, the flood event of January 30, 2009 was used for validation, and the results showed that the SCS method outperformed the Snyder method (Fig. 5 and Table 6). As a result, the future analysis was based on the SCS unit hydrograph method.

Fig. 5
figure 5

Observed and calculated hydrograph in validation process of flood event of January 30, 2009 by a SCS and b Snyder methods

Full size image
Table 6 Results of HEC-HMS model calibration and validation
Full size table

3.3 Influence of land use change on flood hazard using HEC-HMS model

Flood hydrographs depicted the hydrological response to changes in land use, so land use maps for the case study were created using satellite images for the years 1985, 2000, and 2016 (Fig. 2). The extracted land uses were divided into six groups: grasslands, agricultural lands, dry land farming, residential lands, rock outcrops, and water-covered lands.

To calculate runoff, the SCS-CN loss method in the HEC-HMS model was investigated.

In this regard, the rainfall/runoff process was simulated using the weighted CN, which was calculated based on the type, area, and percent of various land uses (Table 2). The HEC-HMS model output for the first period of study (1985) revealed that the runoff height and peak discharge of runoff are 283.72 mm and 64.3 m3/s, respectively (Fig. 6). The most common land uses at the time were grasslands, dry land farming, and agricultural lands. In next years, due to increased population and reduced water storages (Noori and Ilderomi 2013), the grasslands mainly were transferred to dry land farming and partly to residential areas, so that dry land farming and residential areas increased from 31.64 and 1.14 km2 to 39.7 and 1.71 km2, respectively. It is obvious that these transfers would increase the values of CN, which is a land use-related parameter, and thus the values of runoff volume and peak discharge. In this regard, the values of runoff height and peak discharge for the second period, i.e., 2000, were 453.19 mm and 67.3m3/s, respectively, indicating an increase of approximately 169.47 mm and 3 m3/s compared to the base period (i.e., 1985) (Fig. 7). Similarly, surveying the following and final period (2015) land uses revealed that this transferring continued. As a result, the runoff volume and height were higher in previous periods, at 494.71 mm and 70.5 m3/s, respectively (Fig. 8).

Fig. 6
figure 6

Simulated daily discharge at the basin outlet in 1985

Full size image
Fig. 7
figure 7

Simulated daily discharge at the basin outlet in 2000

Full size image
Fig. 8
figure 8

Simulated daily discharge at the basin outlet in 2015

Full size image

In a nutshell, decreasing the area of grasslands over time resulted in a significant increase in surface runoff. The increase in surface runoff was a result of a decrease in infiltration rate and, as a result, the process of groundwater recharge. Increased dry land farming and build-up areas as a result of population growth have resulted in an increase in land use changes. Increased dry land farming and development areas resulted in soil loss and, as a result, soil erosion. Soil erosion reduced either infiltration rate or terrain roughness, resulting in increased surface runoff. It should be noted that the occurrence of such changes is related to a small portion of the overall case study. The HEC-HMS model can be used as a suitable means to simulate the role of land use scenarios and related concepts on hydrologic response at regional and global scales over the selected time frames.

4 Conclusion

Human life on Earth has resulted in enormous changes to the Earth’s surface, such as the destruction of vast forests and pastures and the creation of dry agricultural lands; these factors have a significant impact on flood proneness. Land use changes and noncompliance with land ability are two of the most significant effects of human activity on flood occurrence (for cultivation). The effects of land use changes on basin hydrology include changes in peak discharge characteristics, total volume of runoff, water quality, and hydrologic equilibrium. Using flood hydrographs, the present study investigated the effects of land use change on the Ekbatan Dam basin in 1985, 2000, and 2015. The results revealed that there is a link between land use change and CN. The interaction of basin physical parameters in the HEC-HMS hydrologic model influenced their relationship with flood peak discharge in rainfall-runoff simulations. The effects of land use change are easily understood by examining CN, runoff, and flood peak discharge values. The simulation results for the target years (1985, 2000, and 2015) revealed that land use change during these intervals increased flood peak discharge, as well as runoff coefficient. As land use changes continue, grasslands and agricultural lands would decline, while dry lands would increase in the target basin. As a result, it is expected that the basin will be more prone to flooding in the future, with increased runoff coefficient and peak discharge. The non-structural method can reduce peak discharge and flood volume in order to manage and reduce floods in the study area. As a result, in order to prevent flooding in the region, extending the dry lands by increasing gardens and agriculture, as well as managing the catchment basin’s vegetation cover in the form of rangeland rehabilitation, can reduce peak discharge and runoff volume.