Improving Runoff Prediction Accuracy in a Mountainous Watershed Using a Remote Sensing-Based Approach

Abstract

Due to the limited number and sparse distribution of meteorological and hydrometric stations in most watersheds, the runoff estimation based on these stations may not be accurate. However, the accurate determination of the Antecedent Soil Moisture (ASM) in watersheds can improve the accuracy of runoff forecasting. The objective of this study is to utilize the ASM derived from satellite imagery to enhance the accuracy of runoff estimation in a mountainous watershed. In this study, a range of Remote Sensing (RS) data, including surface biophysical and topographic features, climate data, hydrometric station flow data, and a ground-based measured SM database for the Balikhli-Chay watershed in Iran, were utilized. The Soil Conservation Service Curve Number (SCS-CN) method was employed to estimate runoff. Two approaches were used for estimating the ASM: (1) using the precipitation data recorded in ground stations, and (2) using the SM data obtained from satellite images. The accuracy of runoff estimation was then calculated for these two scenarios and compared. The mean Nash–Sutcliffe statistic was found to be 0.63 in the first scenario and 0.74 in the second scenario. The inclusion of ASM derived from the satellite imagery in the precipitation–runoff model resulted in a 51% increase in the accuracy of runoff estimation compared to using precipitation data recorded in ground stations. These findings have significant implications for improving the accuracy of ASM and runoff modeling in various applications.

1. Introduction

Runoff is a crucial factor in the hydrological cycle, and changes in runoff have significant effects on the ecology, water, and soil storage resources in a watershed, particularly downstream [1,2,3]. The management of runoff in many watersheds is of utmost importance globally. The process of runoff production at the watershed scale is complex due to various factors such as precipitation patterns and surface characteristics [4,5]. Therefore, it is useful to determine and recognize runoff distribution patterns in order to develop appropriate control methods [6].

One important feature of watersheds that has been increasingly considered in runoff estimation systems is the Antecedent Soil Moisture (ASM) conditions prior to flooding [7,8,9,10]. As the ASM increases and the stability of aggregates decreases, there is a sharp increase in runoff and sediment production [11]. Consequently, the underestimation and overestimation of runoff discharge peaks in hydrological modeling, especially in runoff operational forecasts, is a persistent problem due to errors in estimating the ASM [12]. Incorporating observational ASM data into hydrological models is crucial for reducing errors in runoff estimation [13,14]. Therefore, the accurate measurement of this variable across watersheds will improve the performance of runoff forecasting, as rainfall–runoff models are sensitive to ASM changes [6,15].

The Soil Conservation Service Curve Number (SCS-CN) method is a common method used for runoff simulation [16,17,18], and it is supported by the Hydrologic Modeling System (HEC-HMS) runoff model. In the SCS-CN method, the ASM factor, along with land use and the hydrological group, is used to determine the Curve Number (CN) values. These variables are primarily calibrated using local data. Also, incorporating satellite data into hydrological models may help to overcome the limitations and uncertainties of hydrological modeling [19,20].

In several previous studies, the impact of ASM on the accuracy of runoff estimation was evaluated. Zhang et al. [8] assessed the effects of ASM on runoff generation in a semi-arid environment. The results of their study demonstrate the provision of concepts for runoff modeling based on the process. The study results show that a 1% change in the ASM resulted in a 0.05 mm change in the runoff. Shi et al. [21] presented a physical formula for estimating the Antecedent Soil Moisture. The study results show that their proposed method increased the model’s efficiency by 88% in both calibration and validation, surpassing the conventional method based on the recorded precipitation data at ground stations. Shi and Wang [17] evaluated the influence of three variables, the slope, ASM, and storm duration, on the SCS-CN method to improve the accuracy of runoff estimation. Their study results show that the impact of the ASM variable on improving the runoff estimation accuracy was higher than the two variables of storm duration and slope. Mishra et al. [22] improved the accuracy of runoff estimation in 234 watersheds in the United States by incorporating the ASM data into the SCS-CN method. Tramblay et al. [9] evaluated the performance of several soil moisture indicators in runoff estimation, including the local Time Domain Reflectometry (TDR) measurements of the SM, and modeled the SM through the Interaction between Soil, Biosphere, and Atmosphere (ISBA) component of the SIM model (Mto-France), the antecedent precipitation, and the base flow. Their study results show that the TDR measurements had a higher accuracy in runoff estimation, while the antecedent-precipitation-based method had less efficiency. Brocca et al. [23] evaluated the performance of two scenarios for estimating antecedent conditions, including the Antecedent Precipitation Index (API) and the Base Flow Index (BFI), in runoff estimation. The results indicate that considering the ASM variable in the rainfall–runoff model helped to predict both the volume and peak discharge well, with a Nash–Sutcliffe efficiency of over 90%. The BFI-based scenario showed a higher accuracy in the runoff estimation compared to the API-based scenario in the rainfall–runoff model.

In previous studies, meteorological station information has commonly been used to determine the ASM. However, the accuracy of ASM determination is dependent on the number and distribution of stations in the region, as well as the accuracy of their recorded information. In many watersheds, particularly in mountainous and steep areas, there is a scarcity of meteorological stations, resulting in a limited access to sufficient data for precipitation and ASM determination [24,25,26]. Additionally, the spatial resolution of precipitation data for the entire area is low due to the limited number of rainfall recording stations. Typically, only one meteorological station provides rainfall and SM data for the entire area. However, the SM is a dynamic characteristic influenced by various factors such as surface biophysical and topographic factors, soil properties, and climate. At the watershed scale, SM exhibits heterogeneity [27]. Therefore, relying on a single value of SM for the entire area makes accurate runoff prediction challenging.

Previous studies have attempted to enhance the accuracy of runoff estimation by utilizing satellite data to calculate the SM [10,20,28,29]. However, the effectiveness of satellite data in improving the runoff estimation accuracy varies depending on the conditions. Consequently, it is necessary to evaluate the effectiveness of satellite data separately in each region. Estimating the runoff in mountainous and semi-arid areas presents greater challenges due to the heterogeneity of topographic and biophysical conditions.

The objective of this study is to utilize the ASM derived from satellite imagery to enhance the accuracy of runoff estimation in a mountainous watershed. To address these challenges and improve the runoff estimation accuracy, two approaches were employed for runoff modeling: (1) utilizing meteorological station precipitation data to determine ASM, and (2) utilizing satellite-imagery-derived SM. Finally, the results of the two approaches were compared to determine the most effective method.

2. Study Area

The Balikhli-Chay watershed is a sub-basin of the Qarahsu watershed, located in the northwest of Iran. This watershed is situated upstream of the Yam-chi Hydrometric Station. The highest elevation in the watershed is 4365 m, which is associated with the Sabalan Mountains in the northeast. The lowest elevation is 1550 m, located at the exit of the watershed on the northeast side. The average height of the watershed is 2109 m, and it covers an area of 567 km², with rainfed agriculture accounting for 34.3% of the total area.

Bozgoosh Mountain, located in the southern part, has an elevation of 2672 m above sea level. The Bagherloodagh Mountains can be found in the eastern and southeastern parts of the watershed. The output of the watershed is a flat plain known as the Ardabil Plain. This river is an important tributary of the Aras River in northwestern Iran, flowing in a south–north direction and collecting water from sub-tributaries in the east and south of the watershed. The Balikhli-Chay River is particularly significant in the region as it flows through the center of Ardabil, providing water for both drinking and agriculture. It is also surrounded by industrial and recreational areas.

The study area, characterized by surface condition heterogeneity such as slope, elevation, land use, and vegetation, is vulnerable to floods and erosion. Over a statistical period of 20 years (from 1999 to 2019), the average annual rainfall recorded by the Ardabil Meteorological Department is 360 mm. The high elevation of the study area greatly influences the type of rainfall, evapotranspiration, vegetation, and ultimately, the runoff. The average daily flow is 1.46 m³·s⁻¹, with a base flow of 0.62 m³·s⁻¹. The main drainage slope of the watershed is 3.4%. The slopes of both the main river and the watershed itself play crucial roles in the occurrence and severity of floods in the study area, as rivers and tributaries follow the slopes until they reach the main river. The average slope in the watershed is 17.2%, further contributing to the occurrence of floods. The Balikhli-Chay watershed (Figure 1) was chosen as the most suitable test watershed for this study based on the available recorded data from hydrometric and rain gauge stations.

3. Materials and Methods

3.1. Data

To evaluate the efficiency of the ASM in runoff modeling, the data used include in situ-measured soil properties, climatic and meteorological data, and a set of satellite images.

3.1.1. Terrestrial Data

In this study, the effect of soil texture on the accuracy of SM modeling was investigated [30]. A total of 225 sampling points were selected in the area (Figure 1c) at a depth of 0–30 cm, and samples were collected in June 2018. The minimum and maximum distances between sampling points were 450 and 660 m, respectively. All samples were taken to a laboratory, air dried, and passed through a 2 mm sieve for texture analysis.

To model the SM and evaluate the accuracy of the model, ground measurements of the SM were taken at 148 points out of the 225 points. These measurements were taken simultaneously with the satellite overpass time in June, July, August, and September 2018. The reduction in the number of points for SM ground measurement was due to time limitations in relation to the size of the study area during the satellite overpass time. The SM was measured using an SM meter model, T SM150, at a depth of 0–30 cm. This portable device provides quick and accurate measurements of volumetric SM with an accuracy of 0.03 m³.m⁻³.

For runoff modeling, climatic and environmental information including daily and hourly rainfall (including intensity and duration) was obtained from the Meteorological Organization of Ardabil Province. Flow and flood hydrograph data were obtained from the Ardabil Regional Water Company. Data on air temperature, rainfall, and air pressure were collected from meteorological stations. Eight rainfall–runoff events were used to simulate runoff, as shown in

Table 1. The dates of the selected events numbered 1 to 8.

Number of Events	1	2	3	4	5	6	7	8
Date of events	15 March 2004	8 November 2007	27 September 2011	1 January 2014	10 March 2016	30 March 2015	26 April 2016	14 April 2017

. The events are referred to as numbers 1 to 8 in the main text.

3.1.2. Remote Sensing (RS) Data

In this study, the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Digital Elevation Model (DEM) was used to extract various surface topographic features. Spectral indices were extracted from Landsat 8 satellite images, which can be downloaded from https://earthexplorer.usgs.gov (accessed on 18 September 2019). The dates of the images used are presented in

Table 2. RS data used in the study.

Name	Dates
ASTER-DEM	-
Landsat 8	15 June, 1 July, 18 August, and 3 September, 2018
MODIS	(14, 13, 12, 11, and 10 March 2004), (7, 6, 5, 4, and 3 November 2007), (26, 25, 24, 23, and 22 September 2011), (31, 30, 29, 28, and 27 December 2014), (29, 28, 27, 26, and 25, March 2015), (9, 8, 7, 6, and 5 March 2016), (25, 24, 23, 22, and 21 April 2016), (13, 12, 11, 10, and 9 April 2017)
MOD07	The same as Landsat image dates
MOD11A1	The same as MODIS image dates

. These data were used on four different dates to determine the optimal model for calculating the SM in the study area. Additionally, in this stage, the average water vapor in the study area was obtained from MOD07 with a spatial resolution of 5000 m. These data were used to estimate the Land Surface Temperature (LST) based on the Landsat 8 image thermal bands. To determine the ASM in runoff estimation using the best model obtained in the first part, daily Moderate Resolution Imaging Spectroradiometer (MODIS) images from five days before and eight rainfall–runoff events were used (Table 2). Furthermore, in the process of generating the SM map based on the MODIS data, the LST was used as one of the spectral features to calculate the ASM. The daily MODIS LST product (MOD11A1) with a spatial resolution of 1000 m was used to determine the LST in the ASM estimation.

3.2. Methods

In this study, several steps were performed to evaluate the efficiency of the SM used in the runoff modeling (Figure 2). First, the SM of the study area was calculated based on Landsat 8 satellite images and DEM using different models. The accuracy of the SM maps, modeled based on real ground data, was evaluated to determine the best model for SM modeling. Then, based on the best model, the SM of the study area for five days before the events was calculated using MODIS satellite images and DEM. Secondly, the ASM of the study area was modeled based on two strategies using precipitation data and the SM obtained from satellite images. In the third step, the study area runoff values at different events were calculated based on two strategies using the ASM based on the rainfall data and the SM obtained from satellite images in the SCS-CN model. The efficiency of the two strategies in runoff modeling was evaluated based on the actual peak flow data and flood volume. Finally, the effect of the SM on the amount of runoff was investigated.

3.2.1. SM Estimation

Before calculating and extracting different spectral indices (Table 2) from satellite images, radiometric and atmospheric corrections of their different bands were necessary. The Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) model was used for the atmospheric correction of reflective bands [31]. According to this model, spectral values resulting from water vapor and aerosol retrieval, as well as the adjacency effect, are corrected based on the spectral bands of satellite images. This model utilizes the moderate resolution atmospheric transmission (MODTRAN) for atmospheric correction, taking into account ancillary information such as satellite overpass time, sensor altitude, geographical location of the area, solar radiation angle, and regional atmospheric model.

The Indices required for SM modeling were extracted from Landsat, MODIS, and DEM images. In the first stage, Landsat and ASTER data (with a spatial resolution of 30 m) were used to determine the optimal model for calculating moisture in the study area. In the second stage, the optimized model was applied to MODIS data (1000 m) and ASTER data (1000 m resampled using the Nearest Neighbor method), and moisture maps for the days prior to the occurrence of runoff were prepared. In the first stage, after preparing the Landsat and DEM images and extracting all the indices, the Random Forest (RF) model was used to model the SM. The following indices, including albedo, brightness, greenness, wetness, Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Normalized Difference Built-up Index (NDBI), Land Surface Temperature (LST), elevation, slope, and local incident angle, were used as predictive variables in the SM calculation process [32,33,34]. A supervised learning algorithm, RF, employing a regression tree structure [35], was employed for the prediction of the SM. RF comprises an ensemble of predictor trees that were grown without pruning [36], with the model’s stability relying on the diversity of these trees [37]. In this research, tree diversity was achieved through the utilization of a range of RS datasets, encompassing surface biophysical characteristics and DEM-derived surface terrain indices as predictive variables. Concurrently, in situ SM data measured during satellite overpass time were integrated into the model. These predictors played a pivotal role in the creation of decision trees within the nodes. The construction of trees was carried out in parallel, with these variables being randomly selected to facilitate the process, and no interplay existing between them [33]. Additionally, the significance of variables in SM modeling was evaluated through the extraction of Variable Importance (VI) factors from the RF model. Noteworthy advantages of RF encompass (a) its high accuracy, rendering it suitable for various types of datasets; (b) its capacity to handle large datasets without necessitating data removal; (c) its ability to generate an unbiased prediction of generalization error, akin to a progressing forest; and (d) its efficacy in accurately predicting missing data, even when a substantial amount of data is absent [34,35,38].

After confirming the efficiency of the model, the best model was used to estimate ASM based on MODIS images and DEM in order to estimate the runoff in the SCS-CN model.

3.2.2. Antecedent Soil Moisture (ASM)

In this study, in the first runoff estimation strategy, based on previous studies, the average rainfall information of five days before the event, obtained from the meteorological station, was used to consider the ASM conditions (

Table 3. ASM condition classification in watershed based on the first strategy [16].

The Amount of Precipitation of the Previous Five Days
ASM Class	Dormancy Season (mm)	Growing Season (mm)
I	<12.5	<35
II	12.5–27.5	35–52.5
III	>27.5	>52.5

). As a result, the accuracy of determining the ASM depends on the number and distribution of stations in the region, as well as the accuracy of their recorded information. In many areas and watersheds, particularly in mountainous and steep regions, there is a limited number of meteorological stations, resulting in limited access to sufficient data for determining precipitation and ASM.

In the second strategy, the ASM conditions were estimated using MODIS satellite images from the previous five days and the average of the five-day SM from

Table 4. Classification of SM conditions in the watershed based on the second strategy.

ASM Class	The SM of the Previous Five Days (Volumetric Percentage)
I	<10
II	10–30
III	>30

. The best SM model was utilized to calculate the ASM based on MODIS images, as discussed in the previous sections. The average SM for each sub-watershed was calculated, and the corresponding SM classes for each sub-watershed were determined based on Table 4 (there were a total of 11 sub-watersheds in the study area (SW1 to SW11)).

3.2.3. Runoff Modeling

This study employed the HEC-HMS rainfall–runoff model and the SCS-CN method to convert the rainfall–runoff relationship in the Balkhli-Chay watershed. The calculations included determining the amount of precipitation losses, converting excess precipitation into runoff, and calculating the baseline flow. Further details on the calculation of each parameter can be found in [16].

Model Calibration and Validation

Calibration is a process that involves adjusting the values of the variables in the model to achieve results that align with the observational data using an objective function. The objective function measures the variation between the observed and simulated hydrographs. The HEC-HMS model automatically minimizes the objective function and optimizes the values of the variables using a search method. In this study, the Nash–Sutcliffe criterion (CNS) based on Equation (1) was used as the objective function to optimize the CN variable and lag time:

$${\mathrm{C}}_{\mathrm{N}\mathrm{S}}=1-\frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{Q}}_{\mathrm{s}\mathrm{i}}-{\mathrm{Q}}_{\mathrm{o}\mathrm{i}})}^2}{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{Q}}_{\mathrm{o}\mathrm{i}}-{\mathrm{Q}}_{\mathrm{o}})}^2}$$

(1)

where Q_si is the simulated discharge, Q_oi is the observed discharge, and Q_o is the mean observed discharge.

For calibration and validation, five and three events were randomly selected, respectively. After running the model with the corresponding rainfall–runoff events, the events were randomly divided into two categories. The variables for the five calibration events were optimized using the CNS objective function. The mean values of the optimized calibrated variables were then used for validation, and the model was run for the remaining three events. The results of the validation were compared with the observed values to test the accuracy of the data obtained from the calibration. To validate the model, the feature space was formed between the observed and estimated values. It was then analyzed based on the CNS index (Equation (1)), line 1:1, the absolute mean error (MAE) (Equation (2)), and the root mean square error (RMSE) (Equation (3)) between the observed and estimated values.

$$\mathrm{M}\mathrm{A}\mathrm{E}=\frac{1}{\mathrm{n}}\sum _{\mathrm{i}=1}^{\mathrm{n}}│{\mathrm{Q}}_{\mathrm{s}\mathrm{i}}-{\mathrm{Q}}_{\mathrm{o}\mathrm{i}}│$$

(2)

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{\frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{Q}}_{\mathrm{s}\mathrm{i}}-{\mathrm{Q}}_{\mathrm{o}\mathrm{i}}\right)}^2$$

(3)

where Q_si and Q_oi are the simulated and observed discharge, respectively.

3.3. The Impact of SM on Runoff

The amount of rainfall is the most important factor in determining the amount of runoff in a watershed. Increased rainfall can lead to higher peak discharge and flood volume in a basin. However, factors such as CN also play a role in determining the amount of runoff in a field. The temporal variation of runoff in a watershed depends on both rainfall and the ASM. To investigate the effect of the ASM on the runoff, the regression relationship between the discharge and measured flood volume in each event was obtained, along with the recorded rainfall at the meteorological station and the measured flow and flood volume estimated using RS. The relationship between the flow and measured flood volume was obtained simultaneously for both the station-measured rainfall and estimated SM.

4. Results and Discussion

4.1. The Effect of Soil Texture on the Accuracy of SM Modeling

The mean error of the modeled SM obtained from the RF model for the sampling points with fine, medium, and coarse textures was 1.6%, 2.1%, and 2.6%, respectively. Furthermore, the R² (RMSE) values between the measured and modeled SM for the sampling points with fine, medium, and coarse soil textures were 0.82 (1.5%), 0.73 (2.2%), and 0.60 (2.7%), respectively. These results indicate that the accuracy of SM prediction based on surface biophysical and topographic characteristics increases in fine-textured soils. In fine-textured soils, increased soil suction allows for more moisture retention. The higher water-holding capacity of fine-textured soils leads to a lower soil temperature and an improved SM content from an RS perspective. On the other hand, coarse-textured soils have a lower water-holding capacity and lose moisture faster than fine-textured soils. As the SM increases, the reflection in the red band increases, while the reflection in the infrared band decreases significantly, resulting in better quality information from RS images in fine-textured soil conditions [39]. Prakash et al. [40] also demonstrated the significant role of soil texture in SM retrieval.

4.2. Determination of ASM

The ASM status of events 1 to 8 were III, I, I, I, II, I, I, and II, respectively. In this study, in addition to the usual method (referred to as the first strategy), another strategy was used for the first time. The second strategy utilized satellite imagery and the best model for SM prediction to estimate the ASM. This allowed for the calculation of the average SM per sub-watershed instead of an overall estimate for the entire watershed. The ASM information for each sub-watershed was then used to determine the CN value of the watershed and its sub-watersheds. There was a total of 11 sub-watersheds in the area (SW1 to SW11).

In the first strategy, the ASM for the entire watershed was determined based on the precipitation information from a single meteorological station. Since there was only one data point, the precipitation value was the same for the entire area, resulting in the same ASM value for the whole area. This affected the spatial variation of the calculated CN for the sub-watersheds, which was influenced by the hydrological groups of soil and land cover. On the other hand, the second strategy took into account the amount of ASM in addition to the aforementioned factors, resulting in different CN values for different parts of the watershed. This is important for accurate runoff forecasting, as the CN plays a crucial role in runoff production. The proposed method in this study can be particularly useful in watersheds that lack sufficient and appropriate data, allowing for a more accurate determination of the CN. The ASM, being a hydrological variable, has a non-linear relationship with the runoff, and its values significantly impact runoff production. The spatial and temporal dynamics of the ASM are commonly used to describe the internal state of a watershed, indicating the sensitivity of the watershed to the production of surface runoff [7]. The spatial distribution of the ASM conditions for different events at the sub-watershed scale is shown in Figure 3.

4.3. Runoff Modeling with HEC-HMS Model

After preparing the model input variables and physically modeling the Balikhli-Chay watershed in the HEC-HMS model, the model was implemented for eight corresponding rainfall–runoff events based on two different strategies. The observational and modeled flow hydrographs based on the model implementation using two different strategies are presented in Figure 4.

A visual examination of the hydrographs shows that the hydrographs modeled with the second strategy, in which RS is used to estimate the ASM, are closer to the (real) observational values. This result demonstrates the efficiency of the RS method as a more accurate estimation of the SM at the watershed, and it is a better result than the one obtained when using the meteorological station data.

To compare the efficiency of the runoff prediction strategies, the peak discharge and volume of the hydrograph flood in both strategies were calculated and compared to the observational mode (

Table 5. Peak flow and flood volume of observed and modeled hydrographs in 8 corresponding rainfall–runoff events.

Event	Peak Discharge (m3·s−1)			Flood Value (1000 m3)
	Observed	Model 1	Model 2	Observed	Model 1	Model 2
1	2.5	4.9	3.8	34.8	70.4	48.5
2	2.2	4.4	3.5	28.6	58.3	40.7
3	2.8	4.7	3.7	38.3	63.4	47.5
4	5.6	9.8	8.1	152.1	276.8	196.4
5	3.9	7.5	5.0	60.4	280.2	160.6
6	2.2	1.5	1.7	39.6	27.5	33.1
7	0.65	1.5	1.1	8.7	17.4	13.6
8	7.4	11.5	9.1	199.8	203.6	191.4
Mean	3.40	5.72	4.50	74.63	124.70	91.47

). An examination of these values showed that the results of the runoff modeling with the second strategy (using RS to estimate the ASM) were closer to the observational values than the first strategy (using meteorological station data to estimate the ASM). The mean values of the observed and modeled peak discharge for the eight events, using the first strategy and the second strategy, were 3.40, 5.72, and 4.50 m³·s⁻¹, respectively, and the mean flood values for these conditions were 74.63, 124.70, and 91.47 thousand m³, respectively. Based on these values, the average differences between the modeled and observed peak flows using the first and second strategies were 2.32 and 1.10 m³·s⁻¹, respectively, and the differences between the flood volume were 50.07 and 16.84 thousand m³, respectively. In both cases, the difference between the modeled and observed values using the RS method (second strategy) was greater than the meteorological station data (first strategy).

Calibration and Validation of HEC-HMS Model

The average optimal values of the CN and lag time were calculated based on the calibration of the HEC-HMS model in rainfall–runoff events. These values were determined under three ASM conditions and are presented in

Table 6. Average optimized values of CN variable and lag time in three ASM conditions: I, II, and III.

Sub-Watershed	CN			Lag Time (h)
	I	II	III	I	II	III
SW1	62.10	79.60	89.97	213.45	131.70	91.40
SW2	58.57	77.10	88.56	304.79	185.47	126.23
SW3	56.94	75.90	87.86	112.35	67.97	45.87
SW4	61.09	78.90	89.58	156.78	96.34	66.48
SW5	55.09	74.50	87.04	211.69	127.27	85.08
SW6	60.67	78.60	89.41	117.81	72.27	49.75
SW7	63.12	80.30	90.36	187.13	115.96	80.95
SW8	63.27	80.40	90.41	110.15	68.30	47.72
SW9	59.68	77.90	89.01	106.93	65.34	44.74
SW10	68.48	83.80	92.24	122.40	77.81	56.15
SW11	56.54	75.60	87.69	64.31	38.85	26.16
Whole watershed	60.67	78.60	89.41	224.77	137.88	94.91

.

During the validation stage, the calibrated variables were entered into the model, and it was run for three other events. The results of the model implementation in the validation stage are shown as observed, and the modeled flow hydrographs are shown in Figure 5. A visual evaluation of the observed and simulated hydrographs in both the implementation and validation stages demonstrated the accuracy of the model in simulating the flow hydrograph. Based on these results, it can be concluded that the model performance is good. Additionally, a visual comparison of the results of the two strategies in the hydrograph simulation revealed that the second strategy produced more accurate results. This means that the hydrographs related to the second strategy in all three events were closer to the observed hydrographs. Therefore, using the RS method to determine the ASM in the watershed leads to more accurate results in predicting the runoff.

The results of the flood volume and peak discharge of the modeled and observed hydrographs in the model validation stage for the three corresponding rainfall–runoff events are presented in

Table 7. Peak flow rates and flood volume of observational and modeled hydrographs using two different strategies in three corresponding rainfall–runoff events.

Event	Peak Discharge (m3·s−1)			Flood Volume (1000 m3)
	Observed	First Strategy	Second Strategy	Observed	First Strategy	Second Strategy
1	2.5	3.1	2.6	34.8	47.6	38.6
6	2.2	2.6	2.3	39.6	49.8	45.5
7	0.65	0.85	0.75	8.7	9.6	9.1

. The peak flow rate and the flood volume of the hydrograph simulated with the second strategy were closer to the actual values. For example, in event 1, the differences between the modeled peak discharge values using the first and second strategies and the observed values were 0.6 and 0.1 m³·s⁻¹, respectively. Similarly, the differences between the modeled flood volume using the first and second strategies and the observational values were 12.8 and 3.8 thousand m³, respectively. These results indicate a better performance using the second strategy, which was based on RS.

In the next step, the accuracy of the HEC-HMS model in simulating the flow hydrograph was evaluated using the CNS, MAE, and RMSE evaluation indices (

Table 8. Values of runoff accuracy assessment indices modeled based on different strategies.

Event	Assessment Indices
	CNS		MAE (m3·s−1)		RMSE (m3·s−1)
	Strategy 1	Strategy 2	Strategy 1	Strategy 2	Strategy 1	Strategy 2
1	0.41	0.56	0.38	0.13	0.66	0.35
6	0.69	0.78	0.31	0.11	0.55	0.23
7	0.80	0.89	0.13	0.05	0.21	0.11
Mean	0.63	0.74	0.27	0.09	0.47	0.23

). Additionally, scatterplots and 1:1 lines are presented in Figure 6. The mean CNS, MAE, and RMSE statistical criteria for the three events in the validation stage indicate the ability of the HEC-HMS model to simulate the flow hydrograph in the Balikhli-Chay watershed. The mean values of the evaluation indices (CNS, MAE, and RMSE) for the three events (first strategy and second strategy) were (0.63, 0.74), (0.27, 0.09), and (0.47, 0.23), respectively. Based on these results, it can be concluded that the HEC-HMS model has the ability to simulate the flow hydrograph in the Balikhli-Chay watershed. Previous studies have also shown that this model can provide satisfactory performance in estimating the runoff. However, this model typically includes a significant number of conceptual variables that are difficult to measure directly. Uncertainty in runoff estimation is often related to the choice of hydro-meteorological input for the hydrological model. The ASM is one of the known sources of uncertainty in runoff estimation in rainfall–runoff models [5]. For this reason, many studies have performed rainfall–runoff simulations without including ASM data in the model. However, various studies have also demonstrated that utilizing ASM information obtained from physical methods, RS, etc., can significantly enhance the accuracy of runoff estimation [8,9,10,17,21,22].

4.4. The Impact of ASM on Runoff

The amount of runoff in a watershed is primarily influenced by the amount of rainfall. Increased rainfall in a basin can lead to a higher peak discharge and flood volume. However, factors such as the CN also affect the amount of runoff in a field, in addition to the amount of rainfall. The temporal variation in the runoff for a watershed depends on the amount of rainfall and the ASM. To investigate the effect of SM on the runoff, the regression relationship between the discharge and measured flood volume was first determined for each event, using the recorded rainfall at the meteorological station and the measured flow and flood volume, while estimating the SM using RS. Additionally, the relationship between the flow and measured flood volume was simultaneously obtained for both the station-measured rainfall and the estimated SM. The R² value between the flow rate and the modeled flood was calculated for each of the three mentioned cases for all eight events (Figure 7).

The results indicate that the ASM has a significant impact on the amount of runoff, with the runoff increasing as the ASM increases. The R² value between the observed and modeled peak flow rate (flood volume) was 0.56 (0.51) based on precipitation and 0.64 (0.58) based on the ASM. Considering both factors (precipitation and the ASM) resulted in an R² value of 0.81 (0.79), which represents an increase of approximately 0.25 (0.28) in the R² value compared to using precipitation information alone, and an increase of about 0.17 (0.21) compared to using ASM information alone. Zhang et al. [8] demonstrated that a 1% change in the ASM led to a 0.05 mm change in the runoff. Additionally, Brocca et al. [23] showed that incorporating the ASM variable in the rainfall–runoff model improved the accuracy of estimating both the volume and peak discharge of the runoff.

5. Conclusions

Effective water runoff management is crucial for the hydrological cycle, ecological conditions, water resource conservation, soil preservation, flood management, and other aspects of a watershed. Estimating the runoff in mountainous and semi-arid areas presents greater challenges due to the heterogeneity of topographic and biophysical conditions on the surface. In these areas, the ASM is one of the influential variables affecting the runoff. Most conventional methods for estimating the ASM rely on precipitation data collected at ground stations. However, watersheds located in mountainous regions often lack a sufficient and dense network of rain gauge stations. Therefore, the objective of this study is to utilize the ASM derived from satellite imagery to enhance the accuracy of runoff estimation in a mountainous watershed. In conclusion, this study demonstrates the advantages of using RS data and satellite imagery for estimating the runoff at the watershed scale. The following conclusions can be made based on the findings of this study:

–

There is a correlation between the SM and soil properties, such as soil texture.

–

The correlation between the ASM and runoff is higher than the correlation between precipitation and runoff. When considering both factors simultaneously, the correlation is even higher than their individual effects on the runoff.

–

The results demonstrate that the use of RS techniques, which provide better temporal and spatial coverage of an area, can offer more accurate and comprehensive information about the ASM.

–

The model’s efficiency in predicting the runoff using RS-derived SM is superior to using precipitation information from meteorological stations.

–

The results indicate a strong correlation between the SM in the watershed and the runoff.