Bilingual
En/Fa
Geographical Research is Published in both Persian and English Full-text.
Volume 37, Issue 1 (2022)
GeoRes 2022, 37(1): 1-13 |
Back to browse issues page
Article Type:
Subject:
History
Received: 2021/08/21 | Accepted: 2021/09/30 | Published: 2022/01/22
Received: 2021/08/21 | Accepted: 2021/09/30 | Published: 2022/01/22
How to cite this article
Sanaifard A, Amirahmadi A, Zanganeh Y. Flood Susceptibility Mapping Using SVM Model and the Impact of Sabzevar City Development on Increasing Flood Peak and its Damages. GeoRes 2022; 37 (1) :1-13
URL: http://georesearch.ir/article-1-1196-en.html
URL: http://georesearch.ir/article-1-1196-en.html
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Rights and permissions
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
1- Faculty of Geography and Environmental Sciences, Hakim Sabzevari University, Sabzevar, Iran
Full-Text (HTML) (126 Views)
Introduction
Today, with the rapid increasing trend of converting non-urban watersheds into urban areas and, consequently, changing land use from natural conditions (agricultural, rangeland, and forest) to urban uses (residential, industrial, commercial, recreational, roads, and streets), a complex phenomenon known as the urban watershed has emerged [Alcántara-Ayala, 2002]. This process has adversely affected the hydrology of urban areas, leading to intensified flooding, increased runoff pollution, higher runoff coefficients, and reduced groundwater recharge [Pfeifer & Bennett, 2011]. In other words, the hydrological changes caused by urbanization and land-use patterns in urban watersheds can be summarized as follows: (1) changes in total runoff volume, (2) changes in recharge derived from precipitation, (3) changes in peak flood discharge, and (4) changes in water quality.
Most Iranian cities, including Sabzevar, are located at watershed outlets. The expansion of impervious surfaces resulting from urban development and construction on permeable soils has reduced the areas capable of absorbing rainfall, thereby increasing the total runoff volume in urban environments. Paved surfaces, building roofs, streets, and parking areas act as barriers to rainwater infiltration into the soil and groundwater recharge, causing a larger portion of precipitation to be converted into surface runoff [Asghari Moghadam, 2005]. An examination of flood damage statistics in Iran and worldwide demonstrates the extensive impacts of flooding on natural, human, and economic resources.
The Support Vector Machine (SVM) model is one of the most widely used machine learning algorithms. This model is capable of solving nonlinear problems more effectively; however, its accuracy strongly depends on the quality and distribution of sample points. In this study, the SVM model was applied to flood analysis, partly because it avoids subjective judgments and because weighting is addressed separately within multi-criteria decision analysis (MCDA) methods [Li & Liu, 2013].
Numerous studies have investigated applied models for flood hazard assessment. Cheraghi Qalehsari et al. [2020], in a study aimed at identifying flood-prone areas using the SVM model in the Nekaroud watershed, have concluded that the model had acceptable accuracy in identifying flood-sensitive zones. Mujaddidi et al. [2020] have applied a combined Frequency Ratio–SVM model in the Kalat watershed and found that this hybrid model achieved higher accuracy and efficiency than the SVM model alone. Fathalizadeh et al. [2020], using the HEC-HMS hydrological model and fuzzy logic in the Zanuzchai watershed, have demonstrated that these models were highly effective in identifying flood-prone areas.
Additionally, geological formations and soils with low permeability, steep slopes, lack or weakness of vegetation cover, and short concentration and lag times are among the most important factors contributing to high runoff generation and rapid runoff routing in sub-watersheds. Masri Alamdari [2021], in a study of the Ghaleh-Chai watershed in Ajabshir using GIS and HEC-HMS, concludes that this approach could identify areas with higher runoff production, representing a key step toward flood mitigation and control at watershed headwaters. Zanganeh Asadi et al. [2021], in a study have conducted in Khorasan Razavi Province using VIKOR, L-THIA, and artificial neural network models, have found that the L-THIA model showed the highest correlation with input layers and demonstrated greater accuracy and efficiency in flood prediction than the other models.
Kadaverugu et al. [2021], using simulations of the biophysical effects of nature-based solutions (NBS), concluded that such models are useful for reducing runoff volume and quantifying tangible benefits related to water regulation and urban flood control. Quantifying ecosystem benefits in both biophysical and economic terms is essential for integrating natural capital into policy planning. Oliveri and Santoro [Oliveri & Santoro, 2000] estimated urban flood damages in Palermo using physical, economic, hydrological, and hydraulic data. Douglas et al. [2007] have emphasized that urban flood impacts are largely due to high population density and its consequences, as well as natural factors and human interventions such as land-use change, population growth, urban expansion, and road construction. Saberifar [2006], in a study of northern Tehran watersheds, shows that increased urbanization nearly doubled the runoff coefficient and runoff volume in the studied basins.
Based on previous studies, urban development in watersheds leads to increased peak discharge, reduced concentration time, and increased runoff volume. However, the magnitude of these changes and their quantitative assessment have not yet been adequately incorporated into urban management programs. Therefore, the objective of this study was to prepare a flood susceptibility map and propose a flood prediction method for the city of Sabzevar, emphasizing the relationship between hydrological model parameters, urban development, and the scenarios proposed in the comprehensive plan of 2005–2006.
Methodology
Location of the Study Area
Sabzevar County is located in the northwestern part of Khorasan Razavi Province at an elevation of approximately 970 meters above sea level. It is bordered by Joveyn and Khoshab counties to the north, Neyshabur County to the east, Kashmar County to the south, and Shahroud County of Semnan Province to the west. Considering the influence of vegetation cover on surface runoff, a general assessment of the region’s plant ecology was conducted. Due to its geographical and climatic conditions, proximity to the northern margin of the Dasht-e Kavir, the presence of relatively high mountain ranges (the Joghatai Mountains), saline rivers, and human interventions such as tamarisk plantation for stabilizing shifting sands, the region supports diverse vegetation types.
Research Method
To estimate runoff in the studied watersheds, the Justin and ICAR methods were applied [Hosseini, 2013; Negarsh & Shaheh Hosseini, 2011]. Based on the hydrological, climatic, and geographical characteristics of the study area, the Sabzevar synoptic station was selected as the representative station for runoff estimation. Annual runoff values were then calculated using the relevant parameters.
Flood magnitudes in the study area were estimated using several empirical methods, including Krieger, Dicken, Creager, and Fuller methods. In these methods, the effects of physical characteristics, watershed shape, land use, soil properties, and other influencing factors are incorporated into empirical coefficients [Negarsh & Shaheh Hosseini, 2011].
Support Vector Machine (SVM) Model
In the study area, 56 flood locations were identified and recorded based on recent flood events. Seventy percent of these points were randomly selected for model calibration, and the remaining thirty percent were used for model validation. Flood occurrence probability was then modeled using the SVM approach.
After rasterization and normalization of all maps, the standardized layers were weighted using the SVM model. Modeling was performed using binary values, where zero represented non-flood conditions and one indicated flood occurrence. After standardization, known flood points were evaluated using the standardized maps. All layers were then converted into ASCII format and imported into the R software to generate the flood hazard zoning map using the SVM model.
Model Validation
The results of the SVM model were evaluated using various criteria, including root mean square error (RMSE), the Kappa coefficient (KC), the receiver operating characteristic (ROC) curve, and the precision–recall curve (PRC), using the R software.
The area under the curve (AUC) represents the predictive performance of the system by describing its ability to correctly estimate flood occurrence and non-occurrence. AUC values range from 0.5 to 1, with values closer to 1 indicating higher accuracy [Wahabzadeh et al., 2017]. In this study, the ROC curve was used to evaluate the accuracy of flood hazard zoning maps.
Runoff Prediction
Another objective of this research was to predict surface runoff under different urban development scenarios using urban hydrological models and geographic information systems (GIS). Relationships between hydrological factors and urban development were determined using statistical data and field studies. Impervious surfaces were estimated using empirical and theoretical approaches, and parameters such as concentration time, runoff volume, and channel travel time were determined.
Due to differences between urban and rural watersheds, a digital terrain model approach proposed by Jensen and Doming [Jensen & Doming, 1988] was employed, in which the watershed was divided into square grid cells. Hydrological modeling and GIS were used to design hydrographs. Although estimating these parameters requires extensive calculations, previous studies have shown that they can be reliably estimated using tested algorithms [Amir Ahmadi et al., 2012].
Vegetation cover changes significantly influence flood generation in the watershed. The percentage of impervious surfaces was considered a key model parameter for estimating effective rainfall. On impervious surfaces, runoff flows as surface flow until reaching drainage channels [Mahmoodzadeh et al., 2015]. In this study, an infiltration model was applied to remove the portion of rainfall stored in the soil and to estimate groundwater infiltration under different rainfall conditions [Campana & Tucci, 2001].
Rainfall Distribution
The overall rainfall design was based on intensity–duration–frequency curves derived from the Sabzevar station. Spatial rainfall distribution across the watershed was obtained using a representative historical rainfall event. In the study area, rainfall typically begins in October, continues until May, and is followed by a dry period lasting four to five months, with minimum rainfall occurring in August and September. In this study, the SCS method was used to distribute total rainfall over time.
Finding
The estimation of surface runoff based on the Justin and ICAR methods indicates that, with a rainfall of 192 mm and a temperature of 13.7°C, the runoff depth estimated using the Justin method is 15.6 mm, while with a rainfall of 196 mm and a temperature of 14.8°C, the runoff depth estimated using the ICAR method is 12.3 mm. The Justin method produces higher runoff depth estimates. Due to its reliance on statistical inputs from hydrometric stations, the Justin method is recommended as the most suitable approach for runoff estimation.
In this study, flood susceptibility mapping was carried out using parameters including slope, elevation, rainfall, curve number (CN), the spatial distribution of flood points and flood-prone zones, the location of urban drainage channels, residential density, runoff maps, land use, and urban open spaces. Spatial layers corresponding to each of these factors were prepared within the ArcGIS environment.
Flood zoning based on the Support Vector Machine (SVM) model indicates that areas with the highest potential vulnerability are primarily located around urban drainage channels. In particular, channels that receive higher flow volumes and discharges from upstream areas expose adjacent urban zones to a greater risk of flooding. Based on the flood susceptibility zoning results, vulnerability levels were higher in Zones 1, 2, and 3 of Sabzevar due to high building density. The overlay of flood zoning and urban land-use maps showed that residential and transportation land uses occupy the largest proportion of the urban area. Owing to the age of existing constructions, the extent of newly developed areas with higher stability and permeability characteristics was relatively limited.
During the modeling process, 30% of the dataset was used for validation purposes. The Kappa coefficient, ROC, and PRC metrics indicated a significant relationship between observed flood occurrences and the trained models. Evaluation results demonstrated that the RMSE values obtained during calibration and validation for the SVM model indicated relatively low prediction error.
Furthermore, validation of the SVM model using the area under the curve yielded a value of 0.9123, indicating acceptable predictive accuracy.
To estimate flood magnitudes, the Creager and Fuller methods were applied, while the envelope curve and Dicken methods were used due to their reliance on common reference stations for calibration. Flood estimates were calculated for different return periods.
According to urban land-use information considered in the 2005 comprehensive plan, 25% of the urban area was allocated to streets and alleys, while 75% was devoted to building land uses. Due to the lack of updated information for these and other land-use categories, the analysis relies on the data provided in the 2005 urban comprehensive plan [Zysta Consulting Engineers, 2005].
Hydrological analysis of the study watershed was conducted for the period from 1979 to 2019 using data from the Sabzevar synoptic station. As precise flow measurements were not available, data from similar regions were used. Model calibration was performed in two stages: (a) calibration of the hydrological model for sub-watersheds, during which three rainfall events were selected and their data were used to estimate and evaluate model parameters; and (b) calibration of the hydrological model for the entire watershed.
Boundary conditions for the southern surroundings of the city were defined using hydraulic scaling and computational criteria, while downstream boundary conditions were set based on water surface levels. The calibration process incorporated two main components. The three selected rainfall events for the sub-watersheds included hydrological parameters derived using the aforementioned factors. The impacts of urban development were evaluated by simulating watershed behavior under both existing and future conditions, and flood risk was assessed for each urban scenario over the past 30 years. These results were calculated using rainfall events with return periods of 25 and 50 years.
The potential for flood occurrence was estimated for four different scenarios in the study area based on heavy rainfall events with return periods of 25 and 50 years. The concentration time of the watershed within the city of Sabzevar was also estimated. This analysis considered three scenarios: (a) rural conditions prior to urban development, (b) current conditions, and (c) projected future conditions based on the comprehensive plan.
Since one of the objectives of this project was to identify suitable flood discharge points, flood pathways were delineated based on the slope characteristics of Sabzevar derived from regional topography. These flood pathways represent thalwegs along which runoff is conveyed. The identified discharge points correspond to the lowest elevations within the urban area. Among these, the point formed by the confluence of two flood channels was identified as the most suitable location for flood discharge.
Discussion
The aim of this study was to prepare a flood susceptibility map and to present a model for flood prediction, with an emphasis on the relationship between hydrological model parameters and urban development, as well as the scenarios proposed in the 2005–2006 Master Plan of Sabzevar city.
Based on the results, changes in smaller areal extents had a significant effect on the model output, namely the peak flood discharge. As the area of a region increases, its influence on peak flood discharge decreases. According to this study, return period parameters and, in particular, area and regional coefficients for estimating instantaneous peak discharge had a substantial impact on the estimation of instantaneous discharge in ungauged basins or at stations that, for any reason, lack sufficient data. The effect of changes in the regional coefficient on instantaneous discharge was very large in extensive areas, such that the impact of variations in this coefficient exceeded that of changes in the area factor. In contrast, the influence of the area factor on instantaneous discharge was more pronounced in smaller areas. Although area was not the sole factor affecting maximum flood discharge and other factors such as slope, elevation, climatic conditions, main channel slope, channel length, vegetation cover, and similar parameters also play a role, many of these factors were often dependent on basin area and are interrelated. Therefore, to control flooding in the study area, maximum instantaneous discharge data from hydrometric stations can be utilized. Accordingly, if flooding occurs in a given region, the available information can be used for rivers located within the same region to estimate flood magnitudes at any point of interest.
One of the physical and morphological problems of Sabzevar city was the inundation of public streets during rainfall events, because the surface runoff collection system had not been comprehensively considered as a major issue in urban design. As a result, substantial costs were incurred annually for repairing and replacing the surface water drainage network. All alleys and streets of Sabzevar city were considered as the statistical population. Using checklists, field surveys, and data analysis, the condition of urban passages was assessed. Due to the inadequacy of discharge outlets for the safe conveyance of urban floods, rainfall-generated runoff accumulates and becomes concentrated within the drainage network, particularly in its downstream sections. After the conveyance capacity of the channels was exceeded, water spreads over the urban surface in low-lying and depressed areas that were nominally covered by the urban drainage system. Consequently, due to the lack of adherence to engineering principles in urban development, the city was facing a phenomenon known as human-induced flooding. This phenomenon led to urban flooding in various parts of Sabzevar during the early spring rainfall of 2019.
Investigations indicated that the Master Plan of Sabzevar city [Zysta Consulting Engineers, 2005] has not been able to achieve its intended objectives in this regard. In fact, the influx of migrants into the city and the formation of informal settlements resulted in the current urban structure without considering the limitations of the drainage network [Saberifar, 2013]. As a result, the city currently exhibits the maximum extent of residential land use. Although the area of permeable land in each zone has been anticipated, these spaces were limited in size. Consequently, it appears that the plan has not adequately addressed the expansion of impervious surfaces, which has led to problems associated with the safe disposal of urban runoff.
According to the studies of the 1384–1385 Master Plan, the average residential building density across the city is approximately 97%, while this index reaches 84% in the city’s core area. Regarding the spatial distribution of average building density in the core zones, the lowest densities were observed in zones 10, 8, 7, and 6, which were located along the southern margin of the city. In contrast, the highest average densities were found in zones 1, 5, and 2, situated in the central part of the city. The most vulnerable areas corresponded to west–east and east–west networks that disrupted the natural slope of watercourses and hinder the continuity of flow networks along the dominant north–south slope of the region. In areas dominated by deteriorated urban fabric, high-quality vulnerability covered a larger area, such that more than 65% of the deteriorated fabric was distributed within this zone. This issue should be taken into account in flood management and hazard planning, as runoff contribution in this part of the city is also high. The findings of this study are consistent with the results reported by Darfashi et al. [2020], Baghalani et al. [2019], and Rashidi and Hosseinzadeh [2019].
According to the results, for all return periods considered and under different scenarios, the existing channels in Sabzevar city lack sufficient capacity to convey runoff. The only exception is related to severe floods with a 25-year return period occurring under scenarios 2 and 3. In this situation, most streets experience inundation. These overflow conditions are mainly due to the presence of specific bridges that obstruct normal water conveyance routes. The following conclusions are evident:
Conclusion
The SVM model demonstrates adequate accuracy and, due to its compatibility with the conditions of Iran, can be applied in similar contexts. The accuracy of the hydrological model proposed in this study is relatively consistent with the recorded observed data. Urban development in Sabzevar began in the central part of the watershed and gradually expanded toward the upstream and downstream areas. This pattern of development has increased the risk of flood-related damages in residential areas. The 2005–2006 comprehensive urban plan for this city was prepared without sufficient consideration of the impacts of the drainage network and the planned transformations. Therefore, future studies should give particular attention to estimating permeable surfaces in proportion to impervious surfaces, while taking existing conditions into account.
Acknowledgments: The authors sincerely thank Ms. Naemi for her cooperation in the preparation of this article.
Ethical Permission: There are no ethical issues to report.
Conflict of Interest: This article is derived from the doctoral dissertation of Abolghasem Sanaei-Fard, supervised by Abolghasem Amir-Ahmadi and Yaghoub Zanganeh.
Author Contributions: Sanaifard A (First Author), Methodologist/Principal Researcher/Discussion Writer/Data Analyst (50%); Amirahmadi A (Second Author), Assistant Researcher/Discussion Writer/Data Analyst (30%); Zanganeh Y (Third Author), Introduction Writer/ Assistant Researcher/Discussion Writer/Data Analyst (20%).
Funding: None.
Today, with the rapid increasing trend of converting non-urban watersheds into urban areas and, consequently, changing land use from natural conditions (agricultural, rangeland, and forest) to urban uses (residential, industrial, commercial, recreational, roads, and streets), a complex phenomenon known as the urban watershed has emerged [Alcántara-Ayala, 2002]. This process has adversely affected the hydrology of urban areas, leading to intensified flooding, increased runoff pollution, higher runoff coefficients, and reduced groundwater recharge [Pfeifer & Bennett, 2011]. In other words, the hydrological changes caused by urbanization and land-use patterns in urban watersheds can be summarized as follows: (1) changes in total runoff volume, (2) changes in recharge derived from precipitation, (3) changes in peak flood discharge, and (4) changes in water quality.
Most Iranian cities, including Sabzevar, are located at watershed outlets. The expansion of impervious surfaces resulting from urban development and construction on permeable soils has reduced the areas capable of absorbing rainfall, thereby increasing the total runoff volume in urban environments. Paved surfaces, building roofs, streets, and parking areas act as barriers to rainwater infiltration into the soil and groundwater recharge, causing a larger portion of precipitation to be converted into surface runoff [Asghari Moghadam, 2005]. An examination of flood damage statistics in Iran and worldwide demonstrates the extensive impacts of flooding on natural, human, and economic resources.
The Support Vector Machine (SVM) model is one of the most widely used machine learning algorithms. This model is capable of solving nonlinear problems more effectively; however, its accuracy strongly depends on the quality and distribution of sample points. In this study, the SVM model was applied to flood analysis, partly because it avoids subjective judgments and because weighting is addressed separately within multi-criteria decision analysis (MCDA) methods [Li & Liu, 2013].
Numerous studies have investigated applied models for flood hazard assessment. Cheraghi Qalehsari et al. [2020], in a study aimed at identifying flood-prone areas using the SVM model in the Nekaroud watershed, have concluded that the model had acceptable accuracy in identifying flood-sensitive zones. Mujaddidi et al. [2020] have applied a combined Frequency Ratio–SVM model in the Kalat watershed and found that this hybrid model achieved higher accuracy and efficiency than the SVM model alone. Fathalizadeh et al. [2020], using the HEC-HMS hydrological model and fuzzy logic in the Zanuzchai watershed, have demonstrated that these models were highly effective in identifying flood-prone areas.
Additionally, geological formations and soils with low permeability, steep slopes, lack or weakness of vegetation cover, and short concentration and lag times are among the most important factors contributing to high runoff generation and rapid runoff routing in sub-watersheds. Masri Alamdari [2021], in a study of the Ghaleh-Chai watershed in Ajabshir using GIS and HEC-HMS, concludes that this approach could identify areas with higher runoff production, representing a key step toward flood mitigation and control at watershed headwaters. Zanganeh Asadi et al. [2021], in a study have conducted in Khorasan Razavi Province using VIKOR, L-THIA, and artificial neural network models, have found that the L-THIA model showed the highest correlation with input layers and demonstrated greater accuracy and efficiency in flood prediction than the other models.
Kadaverugu et al. [2021], using simulations of the biophysical effects of nature-based solutions (NBS), concluded that such models are useful for reducing runoff volume and quantifying tangible benefits related to water regulation and urban flood control. Quantifying ecosystem benefits in both biophysical and economic terms is essential for integrating natural capital into policy planning. Oliveri and Santoro [Oliveri & Santoro, 2000] estimated urban flood damages in Palermo using physical, economic, hydrological, and hydraulic data. Douglas et al. [2007] have emphasized that urban flood impacts are largely due to high population density and its consequences, as well as natural factors and human interventions such as land-use change, population growth, urban expansion, and road construction. Saberifar [2006], in a study of northern Tehran watersheds, shows that increased urbanization nearly doubled the runoff coefficient and runoff volume in the studied basins.
Based on previous studies, urban development in watersheds leads to increased peak discharge, reduced concentration time, and increased runoff volume. However, the magnitude of these changes and their quantitative assessment have not yet been adequately incorporated into urban management programs. Therefore, the objective of this study was to prepare a flood susceptibility map and propose a flood prediction method for the city of Sabzevar, emphasizing the relationship between hydrological model parameters, urban development, and the scenarios proposed in the comprehensive plan of 2005–2006.
Methodology
Location of the Study Area
Sabzevar County is located in the northwestern part of Khorasan Razavi Province at an elevation of approximately 970 meters above sea level. It is bordered by Joveyn and Khoshab counties to the north, Neyshabur County to the east, Kashmar County to the south, and Shahroud County of Semnan Province to the west. Considering the influence of vegetation cover on surface runoff, a general assessment of the region’s plant ecology was conducted. Due to its geographical and climatic conditions, proximity to the northern margin of the Dasht-e Kavir, the presence of relatively high mountain ranges (the Joghatai Mountains), saline rivers, and human interventions such as tamarisk plantation for stabilizing shifting sands, the region supports diverse vegetation types.
Research Method
To estimate runoff in the studied watersheds, the Justin and ICAR methods were applied [Hosseini, 2013; Negarsh & Shaheh Hosseini, 2011]. Based on the hydrological, climatic, and geographical characteristics of the study area, the Sabzevar synoptic station was selected as the representative station for runoff estimation. Annual runoff values were then calculated using the relevant parameters.
Flood magnitudes in the study area were estimated using several empirical methods, including Krieger, Dicken, Creager, and Fuller methods. In these methods, the effects of physical characteristics, watershed shape, land use, soil properties, and other influencing factors are incorporated into empirical coefficients [Negarsh & Shaheh Hosseini, 2011].
Support Vector Machine (SVM) Model
In the study area, 56 flood locations were identified and recorded based on recent flood events. Seventy percent of these points were randomly selected for model calibration, and the remaining thirty percent were used for model validation. Flood occurrence probability was then modeled using the SVM approach.
After rasterization and normalization of all maps, the standardized layers were weighted using the SVM model. Modeling was performed using binary values, where zero represented non-flood conditions and one indicated flood occurrence. After standardization, known flood points were evaluated using the standardized maps. All layers were then converted into ASCII format and imported into the R software to generate the flood hazard zoning map using the SVM model.
Model Validation
The results of the SVM model were evaluated using various criteria, including root mean square error (RMSE), the Kappa coefficient (KC), the receiver operating characteristic (ROC) curve, and the precision–recall curve (PRC), using the R software.
The area under the curve (AUC) represents the predictive performance of the system by describing its ability to correctly estimate flood occurrence and non-occurrence. AUC values range from 0.5 to 1, with values closer to 1 indicating higher accuracy [Wahabzadeh et al., 2017]. In this study, the ROC curve was used to evaluate the accuracy of flood hazard zoning maps.
Runoff Prediction
Another objective of this research was to predict surface runoff under different urban development scenarios using urban hydrological models and geographic information systems (GIS). Relationships between hydrological factors and urban development were determined using statistical data and field studies. Impervious surfaces were estimated using empirical and theoretical approaches, and parameters such as concentration time, runoff volume, and channel travel time were determined.
Due to differences between urban and rural watersheds, a digital terrain model approach proposed by Jensen and Doming [Jensen & Doming, 1988] was employed, in which the watershed was divided into square grid cells. Hydrological modeling and GIS were used to design hydrographs. Although estimating these parameters requires extensive calculations, previous studies have shown that they can be reliably estimated using tested algorithms [Amir Ahmadi et al., 2012].
Vegetation cover changes significantly influence flood generation in the watershed. The percentage of impervious surfaces was considered a key model parameter for estimating effective rainfall. On impervious surfaces, runoff flows as surface flow until reaching drainage channels [Mahmoodzadeh et al., 2015]. In this study, an infiltration model was applied to remove the portion of rainfall stored in the soil and to estimate groundwater infiltration under different rainfall conditions [Campana & Tucci, 2001].
Rainfall Distribution
The overall rainfall design was based on intensity–duration–frequency curves derived from the Sabzevar station. Spatial rainfall distribution across the watershed was obtained using a representative historical rainfall event. In the study area, rainfall typically begins in October, continues until May, and is followed by a dry period lasting four to five months, with minimum rainfall occurring in August and September. In this study, the SCS method was used to distribute total rainfall over time.
Finding
The estimation of surface runoff based on the Justin and ICAR methods indicates that, with a rainfall of 192 mm and a temperature of 13.7°C, the runoff depth estimated using the Justin method is 15.6 mm, while with a rainfall of 196 mm and a temperature of 14.8°C, the runoff depth estimated using the ICAR method is 12.3 mm. The Justin method produces higher runoff depth estimates. Due to its reliance on statistical inputs from hydrometric stations, the Justin method is recommended as the most suitable approach for runoff estimation.
In this study, flood susceptibility mapping was carried out using parameters including slope, elevation, rainfall, curve number (CN), the spatial distribution of flood points and flood-prone zones, the location of urban drainage channels, residential density, runoff maps, land use, and urban open spaces. Spatial layers corresponding to each of these factors were prepared within the ArcGIS environment.
Flood zoning based on the Support Vector Machine (SVM) model indicates that areas with the highest potential vulnerability are primarily located around urban drainage channels. In particular, channels that receive higher flow volumes and discharges from upstream areas expose adjacent urban zones to a greater risk of flooding. Based on the flood susceptibility zoning results, vulnerability levels were higher in Zones 1, 2, and 3 of Sabzevar due to high building density. The overlay of flood zoning and urban land-use maps showed that residential and transportation land uses occupy the largest proportion of the urban area. Owing to the age of existing constructions, the extent of newly developed areas with higher stability and permeability characteristics was relatively limited.
During the modeling process, 30% of the dataset was used for validation purposes. The Kappa coefficient, ROC, and PRC metrics indicated a significant relationship between observed flood occurrences and the trained models. Evaluation results demonstrated that the RMSE values obtained during calibration and validation for the SVM model indicated relatively low prediction error.
Furthermore, validation of the SVM model using the area under the curve yielded a value of 0.9123, indicating acceptable predictive accuracy.
To estimate flood magnitudes, the Creager and Fuller methods were applied, while the envelope curve and Dicken methods were used due to their reliance on common reference stations for calibration. Flood estimates were calculated for different return periods.
According to urban land-use information considered in the 2005 comprehensive plan, 25% of the urban area was allocated to streets and alleys, while 75% was devoted to building land uses. Due to the lack of updated information for these and other land-use categories, the analysis relies on the data provided in the 2005 urban comprehensive plan [Zysta Consulting Engineers, 2005].
Hydrological analysis of the study watershed was conducted for the period from 1979 to 2019 using data from the Sabzevar synoptic station. As precise flow measurements were not available, data from similar regions were used. Model calibration was performed in two stages: (a) calibration of the hydrological model for sub-watersheds, during which three rainfall events were selected and their data were used to estimate and evaluate model parameters; and (b) calibration of the hydrological model for the entire watershed.
Boundary conditions for the southern surroundings of the city were defined using hydraulic scaling and computational criteria, while downstream boundary conditions were set based on water surface levels. The calibration process incorporated two main components. The three selected rainfall events for the sub-watersheds included hydrological parameters derived using the aforementioned factors. The impacts of urban development were evaluated by simulating watershed behavior under both existing and future conditions, and flood risk was assessed for each urban scenario over the past 30 years. These results were calculated using rainfall events with return periods of 25 and 50 years.
The potential for flood occurrence was estimated for four different scenarios in the study area based on heavy rainfall events with return periods of 25 and 50 years. The concentration time of the watershed within the city of Sabzevar was also estimated. This analysis considered three scenarios: (a) rural conditions prior to urban development, (b) current conditions, and (c) projected future conditions based on the comprehensive plan.
Since one of the objectives of this project was to identify suitable flood discharge points, flood pathways were delineated based on the slope characteristics of Sabzevar derived from regional topography. These flood pathways represent thalwegs along which runoff is conveyed. The identified discharge points correspond to the lowest elevations within the urban area. Among these, the point formed by the confluence of two flood channels was identified as the most suitable location for flood discharge.
Discussion
The aim of this study was to prepare a flood susceptibility map and to present a model for flood prediction, with an emphasis on the relationship between hydrological model parameters and urban development, as well as the scenarios proposed in the 2005–2006 Master Plan of Sabzevar city.
Based on the results, changes in smaller areal extents had a significant effect on the model output, namely the peak flood discharge. As the area of a region increases, its influence on peak flood discharge decreases. According to this study, return period parameters and, in particular, area and regional coefficients for estimating instantaneous peak discharge had a substantial impact on the estimation of instantaneous discharge in ungauged basins or at stations that, for any reason, lack sufficient data. The effect of changes in the regional coefficient on instantaneous discharge was very large in extensive areas, such that the impact of variations in this coefficient exceeded that of changes in the area factor. In contrast, the influence of the area factor on instantaneous discharge was more pronounced in smaller areas. Although area was not the sole factor affecting maximum flood discharge and other factors such as slope, elevation, climatic conditions, main channel slope, channel length, vegetation cover, and similar parameters also play a role, many of these factors were often dependent on basin area and are interrelated. Therefore, to control flooding in the study area, maximum instantaneous discharge data from hydrometric stations can be utilized. Accordingly, if flooding occurs in a given region, the available information can be used for rivers located within the same region to estimate flood magnitudes at any point of interest.
One of the physical and morphological problems of Sabzevar city was the inundation of public streets during rainfall events, because the surface runoff collection system had not been comprehensively considered as a major issue in urban design. As a result, substantial costs were incurred annually for repairing and replacing the surface water drainage network. All alleys and streets of Sabzevar city were considered as the statistical population. Using checklists, field surveys, and data analysis, the condition of urban passages was assessed. Due to the inadequacy of discharge outlets for the safe conveyance of urban floods, rainfall-generated runoff accumulates and becomes concentrated within the drainage network, particularly in its downstream sections. After the conveyance capacity of the channels was exceeded, water spreads over the urban surface in low-lying and depressed areas that were nominally covered by the urban drainage system. Consequently, due to the lack of adherence to engineering principles in urban development, the city was facing a phenomenon known as human-induced flooding. This phenomenon led to urban flooding in various parts of Sabzevar during the early spring rainfall of 2019.
Investigations indicated that the Master Plan of Sabzevar city [Zysta Consulting Engineers, 2005] has not been able to achieve its intended objectives in this regard. In fact, the influx of migrants into the city and the formation of informal settlements resulted in the current urban structure without considering the limitations of the drainage network [Saberifar, 2013]. As a result, the city currently exhibits the maximum extent of residential land use. Although the area of permeable land in each zone has been anticipated, these spaces were limited in size. Consequently, it appears that the plan has not adequately addressed the expansion of impervious surfaces, which has led to problems associated with the safe disposal of urban runoff.
According to the studies of the 1384–1385 Master Plan, the average residential building density across the city is approximately 97%, while this index reaches 84% in the city’s core area. Regarding the spatial distribution of average building density in the core zones, the lowest densities were observed in zones 10, 8, 7, and 6, which were located along the southern margin of the city. In contrast, the highest average densities were found in zones 1, 5, and 2, situated in the central part of the city. The most vulnerable areas corresponded to west–east and east–west networks that disrupted the natural slope of watercourses and hinder the continuity of flow networks along the dominant north–south slope of the region. In areas dominated by deteriorated urban fabric, high-quality vulnerability covered a larger area, such that more than 65% of the deteriorated fabric was distributed within this zone. This issue should be taken into account in flood management and hazard planning, as runoff contribution in this part of the city is also high. The findings of this study are consistent with the results reported by Darfashi et al. [2020], Baghalani et al. [2019], and Rashidi and Hosseinzadeh [2019].
According to the results, for all return periods considered and under different scenarios, the existing channels in Sabzevar city lack sufficient capacity to convey runoff. The only exception is related to severe floods with a 25-year return period occurring under scenarios 2 and 3. In this situation, most streets experience inundation. These overflow conditions are mainly due to the presence of specific bridges that obstruct normal water conveyance routes. The following conclusions are evident:
- In scenarios that consider planned urban development, runoff in downstream areas increases by approximately 20% to 50%.
- Implementing regulations to allocate open spaces in areas designated for construction can reduce this effect by up to 5%.
- Improving bridges and directing urban runoff toward surrounding areas can prevent the occurrence of floods with a 25-year return period. This approach can also be used to identify the impacts of other regulations designed for flood mitigation.
Conclusion
The SVM model demonstrates adequate accuracy and, due to its compatibility with the conditions of Iran, can be applied in similar contexts. The accuracy of the hydrological model proposed in this study is relatively consistent with the recorded observed data. Urban development in Sabzevar began in the central part of the watershed and gradually expanded toward the upstream and downstream areas. This pattern of development has increased the risk of flood-related damages in residential areas. The 2005–2006 comprehensive urban plan for this city was prepared without sufficient consideration of the impacts of the drainage network and the planned transformations. Therefore, future studies should give particular attention to estimating permeable surfaces in proportion to impervious surfaces, while taking existing conditions into account.
Acknowledgments: The authors sincerely thank Ms. Naemi for her cooperation in the preparation of this article.
Ethical Permission: There are no ethical issues to report.
Conflict of Interest: This article is derived from the doctoral dissertation of Abolghasem Sanaei-Fard, supervised by Abolghasem Amir-Ahmadi and Yaghoub Zanganeh.
Author Contributions: Sanaifard A (First Author), Methodologist/Principal Researcher/Discussion Writer/Data Analyst (50%); Amirahmadi A (Second Author), Assistant Researcher/Discussion Writer/Data Analyst (30%); Zanganeh Y (Third Author), Introduction Writer/ Assistant Researcher/Discussion Writer/Data Analyst (20%).
Funding: None.
Keywords:
References
1. Amir Ahmadi A, Behniafar A, Ebrahimi M (2012). Icrozonation of flood risk in Sabzevar suburb with the aim of sustainable urban development. Environmental Based Territorial Planning (Amayesh). 5(16):32-17. Persian [Link]
2. Alcántara-Ayala I (2002). Geomorphology, natural hazards, vulnerability and prevention of natural disasters in developing countries. Geomorphology. 47(2-4):107-124. [Link] [DOI:10.1016/S0169-555X(02)00083-1]
3. Asghari Moghadam M (2005). Urban water and habitat (Hydrological studies in urban planning) (Urban hydrology). 1st ed. Tehran: Sara Publications. [Persian] [Link]
4. Baghalani M, Rostami N, Tavakoli M (2019). Identification of effective factors on the occurrence of urban floods in Ilam watershed. 11(2):523-536. [Persian] [Link]
5. Campana NA, Tucci CEM (2001). Predicting floods from urban development scenarios: Case study of the Dilúvio Basin, Porto Alegre, Brazil. Urban Water. 3(1-2):113-124. [Link] [DOI:10.1016/S1462-0758(01)00004-8]
6. Cheraghi Qalehsari A, Habibnejad Roshan M, Roshan SH (2020). Flood susceptibility mapping using a support vector machine models (SVM) and geographic information system (GIS). Journal of Natural Environmental Hazards. 9(25):61-80. [Persian] [Link]
7. Douglas I, Garvin S, Lawson N, Richards J, Tippett J, White I (2010). Urban pluvial flooding: a qualitative case study of cause, effect and nonstructural mitigation. Journal of Flood Risk Management. 3(2):112-125. [Link] [DOI:10.1111/j.1753-318X.2010.01061.x]
8. Darfashi KB, Adeli SF, Malek Mohammadi B (2020). Present the pattern in vulnerability level zoning and analysis of urban areas at flood risk Case study: Regions 10 and 22 of Tehran. Journal of Emergency Management. 9(1):5-16. [Persian] [Link]
9. Fathalizadeh B, Abedini M, Rajabi M (2020). Investigation of flood causes and hazards in Zanuzchay catchment using HEC-HMS hydrological model and fuzzy logic. Quantitative Geomorphological Research. 9(1):134-155. [Persian] [Link]
10. Hosseini Y (2013). Selecting a Method for Calculating the Annual Estimation of a Drain Basin by Examining Various Methods for Estimating the Annual Flow (Jaroo Basin). First National Conference on Drainage in Sustainable Agriculture. 2013, February 27: Tarbiat Modares University. [Persian] [Link]
11. Jensen SK, Doming JO (1988). Extracting topographic structure from digital elevation data for geographic information system analysis. Photogrametric Engineerin and Remote Sensing. 54(11):1593-1600. [Link]
12. Kadaverugu A, Nageshwar Rao C, Viswanadh GK (2021). Quantification of flood mitigation services by urban green spaces using InVEST model: A case study of Hyderabad city, India. Modeling Earth Systems and Environment. 7:589-602. [Link] [DOI:10.1007/s40808-020-00937-0]
13. Li Q, Jiang X, Liu D (2013). Analysis and modelling of flood risk assessment using information diffusion and artificial neural network. Water SA. 39(5):634-648. [Link] [DOI:10.4314/wsa.v39i5.8]
14. Masri Alamdari P (2021). Spatial analysis of flood risk in Ajabshir tea castle catchment using GIS and HEC-HMS. Quantitative Geomorphological Research. 10:93-111. [Persian] [Link]
15. Mahmoodzadeh H, Emamykya V, Rasouli AA (2015). Micro zonation of flood risk in tabriz suburb with using analytical hierarchy process. Geographical Research Quarterly. 30(1):167-180. [Persian] [Link]
16. Mojaddidi Rizehi H, Habibnejad R , Pradhan B (2020). Efficiency of frequency model of combination ratio - support vector machine in identifying flood prone areas of Kalat watershed. Echo Hydrology. 7. [Persian] [Link]
17. Negarsh H, Shaheh Hosseini M (2011). Investigation of methods of estimating annual discharge and estimating floods in Tigh Ab watershed of Khash city. Geography Space. 11:255-282. [Persian] [Link]
18. Oliveri E, Santoro M (2000). Estimation of urban structural flood damages: the case study of Palermo. Urban Water. 2(3):223-234. [Link] [DOI:10.1016/S1462-0758(00)00062-5]
19. Pfeifer LR, Bennett EM (2011). Environmental and social predictors of phosphorus in urban streams on the Island of Montréal, Québec. Urban Ecosystems. 14:485-499. [Link] [DOI:10.1007/s11252-011-0160-0]
20. Rashidi M, Hosseinzadeh MM (2019). The role of sub-basins overlooking the city in the occurrence of urban floods in Izeh (Khuzestan). Geography and Environmental Hazards. 29:25-42. [Persian] [Link]
21. Wahabzadeh AH, Mirki Sh, Shirzadi A (2017). Landslide Sensitivity Zoning with GIS and Comparing the Efficiency of Logistic Regression Methods and Frequency Ratio (Case Study: Cheshidar Watershed, Kurdistan). Journal of Geographic Information System Application and Remote Sensing in Planning. 8:11 -21. [Persian] [Link]
22. Zanganeh Asadi M, Amir Ahmadi A, Naemi Tabar M (2021). Evaluation of the efficiency of Victor, L-THIA and artificial neural network models in regional flood analysis (Case study: Khorasan Razavi province). Echo Hydrology. 8:108-89. [Persian] [Link]
23. Zysta Consulting Engineers (2005). Detailed plan (improvement and renovation) of Sabzevar. Sabzevar: Sabzevar Municipality; [Unknown cited]. Available from: Not Found [Persian] [Link]