Natural disasters can be devastating to the environment and natural resources. Flood inundation mapping and hydraulic modeling are essential to forecast critical flood information, including flood depth and water surface height. In this research, several factors that influence floods were studied. These factors include the intensity of the rainstorm, the depth of precipitation, soil types, geologic settings, and topographic features. Furthermore, the research carried out hydraulic modeling of storm flows for 50- and 100-Year return periods and estimated that the water depth in Wadi Al Wala could reach 15m at 50 years of storm and 25m at 100 return years of storms. A DNN model is developed with good accuracy to predict flood flow based on historical records from 1980 to 2018 meteorological data. The goal of this research is to improve flood prediction, and risk assessment with the use of DNN integrated with hydrological and hydraulic models.
Research Article
Deep Neural Networks Hydrologic and Hydraulic Modeling in Flood Hazard Analysis
https://doi.org/10.21203/rs.3.rs-4107156/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
Floods Prediction
HEC-HMS
HEC-RAS
GIS
WMS Deep Neural Networks
Deep Learning
Natural disasters such as floods can have a devastating impact on individuals and communities. Floods, especially in areas with limited access to water, can lead to death and economic damage. Drought can cause water shortages in specific locations, which only exacerbates the problem [1]. Floods are considered one of the most critical factors affecting water resources. They have inverse impacts that exceed their positives in terms of increasing the surface water level by causing pollution and damage to infrastructure. Floods result from a quick and sudden storm that carries heavy rain during a short interval, which does not exceed several hours. Therefore, the quantity of water flow level in the basin rises at a speed that exceeds its ability to drain [2]. The degree of a high density of rainfall or flood damage depends on topography, soil types, and rocks. Morphological factors of the watershed and the density and volume of flow in the catchment [3]. Jordan suffers from water scarcity with different topographic features and precipitation distributions. It is a prone region that flashes flood storms in different regions. Wadi Al Wala, Wadi Musa, Wadi Al-Youtum, and Petra are among the touristic areas in Jordan that are the most affected by devastating and deadly floods. Water resources are limited, and water demand has been increasing daily due to rapid economic development, a fast population growth rate, and refugees from the vicinity. Therefore, managing water resources is a must of high importance [4].
The degree of flood risk in the watershed areas is estimated by the characteristics of rainfall, topography, types of soil and rocks, morphological factors of the watershed, and the density and volume of flow in the catchment [3, 5]. The main reason for floods is the output of the hydrological cycle which is affected by the increasing number of populations leading to urbanization [6]. Hydrological models are used to make hypotheses and help in the decision-making process. Thus, the importance of analyzing flood events increases with time to understand the response of watersheds in cases of sudden precipitation.
The most used methods to evaluate floods are hydrological and hydraulic modeling. Water Modeling Systems (WMS), Hydrologic Engineering Centre-Hydrologic Modeling System (HEC-HMS), and Hydrologic Engineering Center’s River Analysis System (HEC-RAS) software integrated with Geographic Information Systems (GIS) and Remote Sensing (RS) techniques are flood modeling examples [7]. These models can be utilized in different areas, like urban areas and agricultural areas.
With the tremendous progress in the GIS, simulation models have been run based on rainwater runoff to improve the accuracy of flood analysis results. These models utilize new technologies and engineering approaches by studying and analyzing the thematic layers of flood conditions. Deep Neural Networks (DNN) models have been applied to thematic layers by implementing and using the training points of the flood inventory map in the neural net package, and R statistics [8]. Such thematic factors include distance from drainage, network density of drainage, earth elevation slope angle, and other factors. As a result, a prediction model and a resource management approach utilizing new technologies are needed.
This research proposes a rainfall-runoff model for estimating the runoff rate and depth over Wadi Al Wala. The developed model has been employed to build flood simulations, evaluate the flash flood hazard in the watershed, and assess the risk of flash floods by studying the behavior of rainfall flow. Rain flow behavior is represented in the form of hierarchical models. Rain measurement data was recorded at meteorological stations provided by the Ministry of Water and Irrigation in Jordan and was collected for the period 1980–2018.
The results of the morphometric analysis showed that the morphological characteristics, including rainstorm intensity, rainfall depth for different return periods, and stormwater discharge amount, were determined. The intensity, duration, and frequency (IDF) curves were utilized for estimating rainfall depth. The amount of stormwater discharge is determined by the Soil Conservation Service Curve Number (SCS-CN) using GIS and WMS. In addition, by estimating these factors, peak discharge and flood magnitude are calculated by creating hydrographic charts for different return periods using HEC-HMS of 50 and 100-year storm flows performed by HEC-RAS. The results showed that the water depth in Wadi Al Wala could reach 15 meters in 50 years of storm and 25 meters in 100 years of storm.
Deep neural networks are artificial neural networks (ANNs) composed of many hidden layers. These layers extract features from the input data and make predictions. The layers in a deep neural network are made up of neurons. Each neuron receives input from other neurons and applies a set of weights and biases to the inputs, which are then passed through an activation function to produce the output. The outputs from one layer are then passed as inputs to the next layer, and this process is repeated until the final output is produced. The ability of DNNs to learn complex relationships and make predictions with high accuracy has made them a popular choice for a wide range of applications[9] [10] [11].
Furthermore, DNNs are utilized for flood forecasting using meteorological data from different gauge stations over watersheds prone to flooding[12]. In this research, statistical methods, including accuracy and Mean Square Error (MSE), have been employed to prove the validity of DNN models. This study integrated DNNs with Hydrological and Hydraulic Models for flood prediction and risk assessment. The proposed DNN model has achieved an accuracy rate of 92%.
The remainder of this paper is organized as follows: The next section presents background and review of related literature. The third section discusses the methodology and data collection, and section four discusses the results. Finally, section five presents the conclusion and future work.
Hydrological models are used to generate hypotheses and assist in decisionmaking. Thus, it is critical to examine flood occurrence over time to understand how watersheds react to sudden precipitation[6]. Natural and man-made factors are the main causes of urban floods, and climate change typically contributes to natural causes of floods, whereas man-made causes are worse by urbanization. Figure 1 depicts the primary factors and mechanisms contributing to urban flooding.
2.1 Understanding Urban Flooding: Causes, Models, and Predictive Tools
The main factors of urban floods can be direct or indirect. Some factors may lead to severe, devastating floods, whereas others can be seen as natural or human. Climate change plays a crucial role in floods; warming may increase rainfall. Also, soil type significantly impacts floods, which can be reduced by increasing plants. The increase in population in some areas may change the soil’s nature and the climate, increasing the flood hazards. The classification scheme shown in Fig. 2 demonstrates the approaches and critical presumptions in developing urban flood models.
The intensity-duration-frequency (IDF) curve is “a graphical representation of the probability that a given average rainfall intensity will occur within a given period” [14]. It represents the relationship between rainfall intensity, duration, and recurrence period (or the probability of exceeding its inverse). IDF curves are commonly used in hydrological, hydraulic, and water resource systems design [15]. The U.S. Department of Agriculture developed the Soil Conservation Service Curve Number (SCS-CN). It is the most commonly used model to produce realistic results for direct surface runoff. This model was implemented by addressing the spatial distribution of soils and the land use of the watershed [16]. It reflects the variation between rainfall and runoff to find the curve number. The CN value is equal to or less than 100. If CN equals 100, this implies an impermeable land surface with no retention (all rainfall moves as runoff).
Calculating the peak discharge and the volume of direct runoff described graphically as hydrographs depends not only on the CN value of a watershed but also on other parameters such as Lag Time (TL) and the Time of Concentration (TC)[17]. Watershed Models (WMS) are effective hydrologic models that comprehensively simulate hydrologic processes. These models primarily concentrate on individual or multiple processes at a relatively small field scale without fully incorporating a watershed area. Watershed models are essential in understanding water pollution since they investigate the surface and subsurface of water movement [18]. WMS models can be classified into different categories based on the modeling approach. These models can be empirical, conceptual, or physically based [19].
2.2 Overview of Flood Modeling Techniques
HEC-HMS model is a widely used open-source hydrologic model for urban watersheds, and flood modeling that impacts land use changes [20]. This mathematical model is designed to simulate the complete hydrologic processes of dendritic catchment [21]. The HEC-HMS model represents a branching or catchment network and focuses on the following components: precipitation, losses, base flow, runoff diversion, and routing. The HEC-HMS model suffers from data size limitations as it is used to simulate the size of small watersheds. The topographic surface of the Earth is represented digitally in a digital elevation model (DEM), where the elevation is shown as a collection of discrete, regularly spaced points. These points are needed to build an accurate three-dimensional picture of the Earth’s surface and are often derived using lidar data or stereo satellite photography. The development of topographic maps, land use and land cover analysis, and water flow and erosion simulation are just a few of the uses for DEMs. They can also be used to visualize landscapes and assess the environmental effects of potential development initiatives. HEC-RAS is a computer program developed by the U.S. Army Corps of Engineers Hydrologic Engineering Center [22]. Researchers utilize the HECRAS for hydraulic modeling and surface water systems flood studies. It requires mainly the river center line, floodplain geometry, and Manning’s roughness values [23]. It is a widely used modeling tool in many applications for solving binary equations [24]. Different schemes govern the HEC-RAS codes, mesh representations, capabilities, and input data requirements [25]. The finite implicit size solves the two-dimensional calculations and propagation wave equations. The choice of the equations depends on several parameters, such as dam penetration, wave propagation analysis, and the presence of multiple hydraulic structures within the area [26]. GIS, hydrologic models (e.g., WMS, HECHMS), and hydraulic models (e.g., HEC-RAS) are used to build simulation models due to potential flood risks, which aid in understanding the relationship between rainfall depth and flood risk in different return periods. The precipitation simulation model contains the surface runoff amount and its relation to precipitation for the catchment area. In response to inputting rainfall data for the study area, the HEC-HMS model will estimate the amount of runoff in the canal or river. The Unit Hydrograph (UH) is the first result of hydrological modeling. It is a graphical representation that simulates the behavior of uniformly excess rainfall over a catchment area during a rainstorm. The UH relies on the precipitation depth derived from IDF curves, CN, TL, and TC [24]. It can be used to define the peak discharge using HEC-HMS and SCS methods. There is a need to support environmental planning options to obtain a physical description of natural phenomena such as floods in terms of the time of their occurrence and the areas at risk of flooding and link them with deep learning prediction models. Deep learning models have appeared as one of the most reliable approaches for predicting time series and have shown excellent results for wide areas and large time series data [27] [28, 29]. As shown in Fig. 3, machine learning (ML) is a branch of artificial intelligence (AI) that allows systems to learn automatically from data without explicit programming. In ML, learning is the ability of a machine to learn from data and the ability of the ML algorithm to train a model, evaluate its performance or accuracy, and then make predictions. An Artificial Neural Network (ANN) is a network of interconnected entities known as nodes where each node is responsible for a simple computational operation. When ANNs have more than one hidden layer in their architecture, they are called DNN [30, 31].
As illustrated in Fig. 4, DNN contains several hidden layers between their input and output. These networks can process both structured and unstructured data. Like the human brain, they feature a hierarchical architecture of neurons. The neurons relay the signal to other neurons depending on the information they receive. Numerous neurons make up a layer, and each neuron performs an activation function. They are a gateway for the signal to travel to the following linked neuron. The weight affects the input to the next neuron’s output and the final output layer. Initial weights are assigned randomly, but when the network is iteratively trained, the weights are optimized to ensure the network predicts correctly [32].
This research aims to establish the rainfall-runoff relationship for estimating runoff rate and depth to evaluate flash flood hazards in the watershed over Wadi Al Wala using Hydrological and Hydraulic models. Furthermore, this study uses Deep Learning to build flood simulation models for flood prediction.
This section presents the methods used to perform the hydraulic and hydrological analysis. The three models used in this research are WMS, HEC-HMS, and HEC-RAS. The WMS and HEC-HMS models were used for hydrology analysis, while the HEC-RAS model was used for hydraulic analysis. The study began by obtaining data from different resources. Three methods were applied to the data: curve Number (CN), Concentration Time (TC), and Unit Hydrograph (UH) to calculate the runoff depth from precipitation. The intensity and duration-frequency curves (IDF) were generated to estimate the maximum discharge for different return periods of 50 and 100 years. Arcmap and WMS interfaces were used to define the watersheds and find their morphological parameters, then to model the rainfall-runoff relationship using HEC-HMS and generate a triangular irregular network (TIN) map to use it for the extraction of the cross-sections back into the WMS interface. Finally, the resulting data is imported into the HEC-RAS interface to determine the network topology using Manning roughness coefficients (n) and to perform a static flow analysis that includes complete discharge information, boundary condition ranges, and flow regimes to calculate water surface height along the channel geometry and then generate intervals of flood height maps. The methodology is shown in Fig. 5.
3.1 Study Area and Basin Characteristics
Wadi Al Wala is the study area in this research. It is located in the central part of Jordan, between longitudes 35◦ ’38◦ and 36◦ ’12◦ east and latitudes 31◦ ’20◦ and 31◦ ’50◦ north. The maximum length of the basin is 71 km from east to west, while its extension from north to south reaches 33.6 km. The area of the basin is about 2050 km2, and it meets Wadi Al Wala with Wadi Al-Hidan to form one stream that meets with Wadi Mujib, which flows into the Dead [33] as shown in Fig. 6.
Wadi Al Wala forms part of the Dead Sea basin in Jordan and is one of the most exposed areas to frequent floods. DEM data is spatial data that provides the characteristics of the watershed. It was available on the Shuttle Radar Topographic Mission (SRTM) with a spatial resolution of 12.5m × 12.5m, and Arc-Map created them. Flow directions, drainage networks, slopes, and DEM-derived morphometric parameters of Wadi Al Wala as shown in Fig. 7.
The process of measuring the shape is a mathematical analysis of the arrangement of the Earth’s surface and the dimensions of its terrain. One of the main features of the description of the basin is the quantitative representation of the drainage system, which is determined by measuring the shape of the basin [34]. Based on the basin’s topography data, obtained from the geographical center and using both WMS and GIS software, the basin was demarcated, and the morphological standards were reached for Wadi Al Wala. It has an area of 2054.08 square kilometres and a circumference of 310296.92 meters, and the average height of the basin from upstream to the outlet is 744.67 m, as seen in Fig. 8.
3.2 Data Collection
The data used in this research was collected from four resources. Data related to soil and rock groups, the digital elevation model (DEM), Land Use Land Cover data (LULC), meteorological data containing precipitation and surface runoff, and finally, climatic data such as temperature, wind speed, and relative humidity. The rainfall data needed to analyze the process of rain runoff was collected from 13 different stations. The data types used in this research and their sources are detailed in Table 1.
| Data | Source |
|---|---|
| Precipitation Data | Ministry of Water and Irrigation (MWI) |
| Climate Data | Jordanian Meteorology Department (JMD) |
| Soil Data | Ministry of Agriculture |
| Geology Data | Natural Resources Authority |
3.2.1 Precipitation Data
The rainfall data needed to analyze the process of rain runoff was collected from 13 different stations. Climatic data consists of the average monthly and annual minimum and maximum temperatures, average wind speed, and average relative humidity of the four stations: Ez-Zeithna, Mushaqar, Amman Airport, and Qasr. The rainfall data recorded by the metrological stations distributed in the watersheds of Wadi Al Walaa were analyzed to determine the rainfall amounts and intensity. The reason behind selecting Wadi Al Wala as the area of study is that it is one of the areas in Jordan that are the most affected by devastating and deadly floods. Daily monthly precipitation data were collected to produce each year’s total monthly precipitation data. The average monthly precipitation data was then calculated for all years included in the period, which produces more accurate assessments of flood variability and water resources. Three parameters were extracted: curve number (CN), time of concentration (TC), and lag time (TL) to calculate the runoff depth from precipitation.
3.2.2 Climate Data
The climate is dry in summer and cold and humid in winter. The temperature ranges from 40◦ in summer to − 2◦ in winter. The average annual relative humidity ranges from (64.67% − 58.67%). The average annual precipitation exceeds (66 mm/hr) [35]. Natural recharging occurs in winter, as rainfall is usually concentrated between December and March when the evaporation rate is low. The reason behind selecting Wadi Al Wala as the area of study is that it is one of the areas in Jordan that are the most affected by devastating and deadly floods.
3.2.3 Soil and Geology Data
According to the results of soil sampling tests by the Jordanian Ministry of Agriculture, different soil types cover the study area. The GIS software was used to obtain the qualitative distribution of soil along the study area. The results revealed different soil types, predominantly loam, sandy loam, and silty clay loam. The covered soil over the area is a crucial factor in determining the increase or decrease in the hydraulic conductivity of groundwater [36]. The study area consists of a succession of limestone, dolomitic limestone, and fossil limestone that lie under the marl nodules, limestone, sand, and chert nodules belonging to Ajloun and Balqa groups [37]. Several major and minor faults cross the Wala Valley, and basalt covers a small portion of the valley [35] as shown in Fig. 9.
3.2.4 Hydrological Data
Hydrological data include the hydrologic cycle and movement of water between the surface of the Earth and the atmosphere, as well as the quantity and quality of available water resources. The data collected from various sources; such as meteorological station data, satellite images, and ground-based measurements of precipitation, evaporation, and soil moisture; is used to perform a hydrological analysis. The output resulting from the hydrological analysis is used to develop models that can explain how water moves across the surface of the Earth and locate and evaluate possible water sources. Rainfall shows a very high spatial and temporal variability which can be divided into non-zero precipitation variance and intermittent rainfall, so it is rare for the rain to fall evenly and synchronously on the basin. Rain gauges in the basin are precisely designed to capture the average precipitation over different areas[38]. Rainfall storms fluctuate throughout the year and merely happen in the winter. In the study area, the most rainfall occurs in November, December, January, February, and March. The average annual rainfall varies from one rain gauge station to another in the catchment area and is related to the terrain variable of the area. Table 2 shows the daily rainfall data which was collected for the period (1980–2018) (data source: MWI) for 13 precipitation measurement stations distributed over different areas (Khaniz-Zebeeb, Jdeideh, Melih, Umm Al Risas, Nursery Duba, Widyan, Wadi Al Wala, Jiza, El-Muwaqar, Sahab, Ma’in, Madaba, Na’ur). After collecting the daily rainfall data from the various monitoring stations, the annual rainfall average for the study period was also calculated, as shown in Fig. 10. The highest rainfall in the study area was in Naour station (381.15 mm), and the lowest was in Duba nursery station (104.68 mm). Both Naour and Madaba show variations of precipitation in the watershed.
Table 2: Statistics of the annual rainfall data for the rain gauge stations
The Thiessen polygon is a standard method applied by hydrologists to calculate the spatial distribution of precipitation over a catchment area by multiplying each area of the rain-gauging stations with the value of the precipitation recorded at the stations during the year [38]. The sum of each day’s value was then divided by the basin’s total area to obtain the rainfall depth over the study area in the desired year, as illustrated in Fig. 11.
where p: weighted average depth of precipitation (areal mean precipitation), piis the precipitation recorded in I station inside or outside the watershed and ai is the area of the polygon surrounding i station.
Each gauge and the estimated areal rainfall measure the average annual precipitation depth. According to the results shown in Table 3, the highest weighted precipitation is recorded in Dhaba’ Nursery and Khanez-Zabeeb. The total areal mean precipitation for the Wadi Al Wala watershed is 159.10 mm, which is the closest to the mean rainfall recorded in El-Muwaqqar and Um El-Risas gauge stations.
Table 3 Weighted Rainfall using Theissen Polygon Method
3.2.5 Flood Data
Intensity Duration-Frequency (IDF) curves are used in hydrology and civil engineering to describe the relationship between the intensity of precipitation (usually measured in millimeters per hour), the duration of the precipitation event (measured in hours), and the frequency of the event (measured in years). The IDF curve can be used to estimate the flood event of a given intensity and duration occurring in a particular area and is used in the design of flood control and drainage systems [39]. Rainfall design depth is found from the IDF curve as an essential parameter for building hydrographs and estimating the flow discharge and runoff volume [40]. In this study, the raw data required for constructing the curves is a historical record of daily rainfall obtained from the MWI (1980–2018) to be statistically analyzed. The durations used when constructing the curves are 5-min, 10-min, 20-min, 30-min, 1-hr, 2-hr, 3-hr, 6hr, 12-hr, and 24-hr. IDF curves were prepared for all rain gauge stations for different return periods. The CD0016 station was taken as an example and is illustrated in Fig. 12.
3.3 SCS-CN Method
The Curve Number (CN) method is standard for estimating the flood or runoff from a rainfall storm. It was developed in 1964 by the American Soil Conservation Service (SCS), recently known as the Natural Resources Conservation Service (NRCS). The CN value of a catchment depends on the soil type and the area’s land use, reflecting the variation between rainfall and runoff. The CN value is equal to or less than 100. When CN equals 100, this implies an impermeable land surface with no retention (all rainfall moves as runoff)[41].
Calculating the peak discharge and the volume of direct runoff described graphically as hydrographs depends not only on the CN value of a watershed but also on other parameters (i.e., TL, TC). In this study, the computed CN is 86 as shown in Fig. 12, which means that most rainfall in any rainstorm converts to surface flood.
3.4 Building the DNN model
This study proposes a deep learning approach of artificial neural networks to develop a flood prediction model based on rainfall and climate data, as shown in Figs. 13 and 14. Regression using ANN is chosen for its performance and high accuracy in prediction in supervised learning. A deep neural network model is developed and trained on the rainfall and climate data training subset. Upon completing the training, the trained model is tested on the unseen test subset of the data for flood prediction. The predicted data is then compared with the actual data to improve the model’s accuracy.
The code in Fig. 13 defines a neural network model in Keras by adding layers. The first line adds an input layer to the model. This layer has 4 input nodes and uses the “ReLu” activation function. This layer is also defined as the input shape of the network. It is important to note that this layer defines the number of input features and sets the input shape for the rest of the model, making it the first layer and should be added only once. The following 3 lines add 3 hidden layers to the network. These layers are fully connected and called Dense in Keras. Each hidden layer has 10, 20 and 30 units, respectively, using “ReLu” activation function. These layers are meant to learn and extract features from the input data, using the number of units to determine the capacity of the layer. The “ReLu” activation function applies an element-wise rectified linear activation function to the output. In summary, this script creates a neural network with 4 input nodes, 3 hidden layers with 10, 20, and 30 neurons, and “ReLu” activation function that learns to extract features from the data.
The neural network model is designed to predict Runoff Depth based on the input features: Relative Humidity, Daily Rainfall, Wind Speed, and Wet Bulb Temperature, as shown in Fig. 16. The figure plots the architecture of the defined Neural Network model. It displays the layers that have been added to the model with the information provided. The input layer with the name ”dense input” with 4 inputs and an output of 10. Then it will show the next two layers with 20 and 30 outputs each. Each layer is represented by a box with the layer’s name, the number of units in the layer, and the input and output shape. All the layers are fully connected by arrows that point from input to output.
The neural network model is designed to predict Runoff Depth based on the input features: Relative Humidity, Daily Rainfall, Wind Speed, and Wet Bulb Temperature, as shown in Fig. 14. The figure plots the architecture of the defined Neural Network model and displays the layers that have been added to the model with the information provided.
The input layer with the name ”dense input” has 4 inputs and an output of 10. Then it will show the next two layers with 20 and 30 outputs each. Each layer is represented by a box with the layer’s name, the number of units in the layer, and the input and output shape. All the layers are fully connected by arrows that point from input to output.
Several parameters were studied in Wadi Al Wala, including the rainstorm’s intensity and the depth of precipitation through IDF curves created for different return periods and the rate of infiltration and runoff by SCS-CN using GIS and WMS. The discharge peak and flood depth were also calculated by constructing hydrographs using HEC-HMS. HEC-RAS is used to build hydraulic models to determine water surface profiles, including depth and velocity, and floodplain mapping. Furthermore, the proposed Deep Learning model is trained on the training subset of rainfall and climate data for flood prediction. The following introduces the results obtained from Hydrologic, Hydraulic, and DNN models.
4.1 HEC-HMS Hydrologic Model Results
Unit Hydrograph (UH) is a graphical representation that simulates the behavior of uniformly excess rainfall over a catchment area during a rainstorm [42]. UH relies on precipitation depth derived from IDF curves, CN, TL, and TC [43]. It can be used to define the peak discharge using HEC-HMS and SCS methods. The HEC-HMS model was carried out for a well-defined part of Wadi Wala (about 21 Km long). First, the basin data were prepared using WMS, where the basin was subdivided into sub-basins. Then, the hydrologic HECHMS model was defined using the curve number and lag-type routing, which requires lag time (TL = 0.6*TC). Precipitation data for HEC-HMS modeling was obtained from the 13 rainfall gauge stations as illustrated in Fig. 15.
Figure 16 shows the resulting hydrographs at Wadi Al Wala outlet (21C sub-basin) for 50 and 100 years return period storms from the HEC-HMS model. The peak discharge value ranges from 44.71 MM at 50 years storm to 53.35 MM at 100 years storm.
4.2 HEC-RAS Hydraulic Model Results
Researchers commonly utilize the Hydraulic model (HEC-RAS) for hydraulic modeling and flood studies in surface water systems. It requires mainly the river center line, floodplain geometry, and Manning’s roughness values [44]. The Manning’s roughness coefficient (n) used for bare soil is approximately 0.10, 0.03, and 0.08 for the agricultural, river, and urbanized areas, respectively. The (n) coefficient for the study area (i.e. Wadi Al Wala) was 0.04. The water surface profile designed by the HEC-RAS can be presented as flood inundation maps. This study created 1D steady flow, 2D unsteady water depth, velocity maps, and flood inundation maps. DEM and hydrographs are required to build the hydraulic model as data inputs are imported from WMS and HEC-HMS, and RAS Mapper is used to obtain the required geometric data. The unsteady-flow simulations revealed that large settlement areas are prone to flood hazards. Along the twenty-one-kilometer profile, two stations were selected for HEC-RAS modeling using the HEC-HMS hydrograph data. Flow velocities and water depths were computed at 50 and 100 return periods along the profile and presented graphically in Fig. 17.
1D profiles were created for the selected locations along the Wadi al Wala stream network (100ST and 21000 ST) to determine the maximum water level for 50 and 100-year return periods. The highest values for max surface water level and the velocity of water flow are recorded at 100 ST (i.e. location A: the outlet of the reach). The Maximum water level for 100 year storm is 257.15 meters below sea level at 100ST, and 312.42 m above sea level at 21000 ST, while for 50 years, it is − 258 m and 311.69m at 100ST and 21000ST, respectively, as shown in Fig. 18.
One aspect of Hydrologic processes, characterized by a constant movement of water leaving the Earth’s surface and eventually returning in the form of precipitation, is represented by flood inundation modeling. Flood inundation modeling is crucial in giving the correct information about potential impending floods, which is vital to issuing flood warnings. Engineers, planners, and governmental organizations use flood inundation mapping as a crucial tool for several studies, such as planning municipal and urban growth, emergency response plans, and flood insurance rates [45]. Precise topographic information about rivers and floodplains is required to simulate flood propagation throughout the floodplain accurately, and inundation modeling [46]. To extract the cross sections of the river and the geometry of floodplains, DEM was transformed into a continuous surface TIN. The Triangulated Irregular Network was processed further in the HEC-GeoRAS tool in GIS. After that, all the parameters were exported for use in the HEC-RAS model. Manning’s roughness coefficient (n) values were applied to each river cross-section. The hydrographs for the 50 and 100 return periods were used as boundary conditions to assign the flow to cells covering the complete river width. HEC-RAS interpolates the surface water elevation data results and delineates the flood inundation, as shown in Fig. 19. It is revealed that upstream and downtown areas (marked by low altitudes) are prone to flooding. Downstream, the water depth exceeds 5m for a 50-year flood and 6 m for a 100-year flood. The width of the flooded areas reached during the 50-Year flood and 100-Year flood was 74 m and 77.74 m, respectively. The flooded water will exceed Wadi Al Wala bank for the 50 and 100-Year return period, and flood inundation will occur.
Figure 20 illustrates how the surface water flow velocity fluctuates. The flow direction and velocity are dependently changing over Wadi al Wala based on the characteristics of the watershed.
4.3 DNN Flood Prediction Model Results
The flood prediction model is based on rainfall and climate data analysis. The model utilizes information on precipitation and weather patterns to predict the probability of flooding in a specific location. Data on factors such as total rainfall, precipitation intensity, and historical weather patterns are incorporated into the model to make predictions. Rainfall data is used to estimate the potential runoff and forecast floods accordingly. It is important to note that this model may not consider all possible factors affecting flooding, such as land use, topography, and human activity. As such, it is crucial to consider the model’s limitations and use it in conjunction with other data and methods for more accurate predictions. The model is trained using historical data and then tested using new data to evaluate its accuracy and reliability. The model’s accuracy for the runoff depth is around 92%, as shown in Figs. 21a and 21b.
This work aims to assess the risk of flash floods in Wadi Al Wala watershed by studying the behavior of rain flow over the area. Wadi Al Wala was chosen as it is one of the most exposed areas to frequent floods. The study utilized Hydrological and Hydraulic Models with Deep Neural Networks. The rainfall-runoff relationship was studied by analyzing the daily rainfall data and representing it in hydrological models for thirty-eight years (1980 to 2018). Morphometric parameters of the watershed were computed using GIS and WMS. The watershed is characterized by steep slopes, different topographic features, and clayey loam, sandy, and Silty loam soil textures that prohibit fast water percolation into the ground leading to a high runoff rate. These features cover a substantial portion of the study area and play a significant role in water flow velocity. The preliminary results of the morphometric analysis showed that the morphological characteristics of the watershed of Wadi Al Wala contribute to the flow of floods and their high velocity. Hydrograph models were run using HEC-HMS based on CN, TC, and TL parameters. Hydraulic modeling was carried out using HEC-RAS to assess the flood risk based on the hydrographs. The DNN model developed in this research for flood prediction uses four parameters: relative humidity, daily rainfall, wind speed, and wet bulb temperature. The accuracy of the proposed model is around 92%, which is reliable and deemed high. Based on the results obtained from the Hydrological, Hydraulic models, and Deep Neural Networks, the study revealed the probability of flood hazards and infrastructure damage in Wadi Al Wala. The results showed that the water depth in Wadi Al Wala could reach 15 meters at 50 years of storm and 25 meters at 100 years of storm. A deep learning model can be used for future work to predict floods based on big data by analyzing relevant information such as historical floods, weather patterns, topography, and land use.
Author Contributions:
Alaa Hawamdeh: Conceptualization, Methodology, Results, Writing- Original draft preparation. Anwar Tarawneh: Methodology, Writing- Original draft preparation. Yousef Sharrab: Formal Analysis, Writing- Original draft preparation, Investigation. Dimah Al-Fraihat: Methodology, Formal Analysis, Writing- Reviewing and Editing, Administration
Funding: Not applicable
Conflicts of Interest: The authors declare no competing interests.
Ethical Approval: Not applicable
Consent to participate: Not applicable
Consent for publication: Not applicable
Data Availability Statement: Please see the following:
Software and Data Availability:
- DNN Model: The code used for measuring the DNN performance using Python language is based on the Keras library in the Google Colab environment. The code and data can be found in GitHub: https://github.com/YousefSharrab/Flood-Prediction
- Name of the software: ArcGIS10.7
- Availability: https://www.esri.com/en-us/home
- Operating System: Windows, 64-bit
- Requirements: The minimum RAM requirement for ArcGIS GIS Server, ArcGIS GeoEvent Server, ArcGIS Image Server, or ArcGIS Business Analyst for Server is 8 GB per unique license role.
- Developer: ArcGIS10.7 was developed by ESRI (Environmental Systems Research Institute), founded in 1969 for creating and managing geographic information. The software includes a wide range of data analysis, mapping, and visualization tools. This work utilized software for mapping the drainage basin, terrain, soil, and geologic data.
- Name of the software: WMS.11.0.8
- Availability: https://www.aquaveo.com/software/ wms-watershed-modeling-system-introduction
- Program Size: 1.1GB
- Operating System: Windows 10, 64-bit
- Description: The Watershed Modeling System (WMS 11.0.8) is a very familiar and among the most utilized software in hydrological studies, developed by Aquaveo, an engineering services company with many years of experience developing watershed modeling solutions. It provides many models, such as HEC-1. This study utilized software to extract the drainage network from DEM data and compute CN and morphometric parameters for building hydrograph models.
- Name of the software: HEC-HMS 4.10
- Availability: https://www.hec.usace.army.mil
- Program Size: 202 MB
- Operating System: Windows, macOS, Linux
- Description: HEC-HMS (Hydrologic Engineering Center’s Hydrologic Modeling System) has been developed for the U.S. Army Corps of Engineers at the
- Hydrologic Engineering Center. HEC-HMS is utilized to simulate the hydrologic and hydraulic processes of a watershed. In addition, it is used
- for designing and analyzing flood control and drainage systems and assessing the impacts of land use changes on water resources.
- Name of the software: HEC-RAS 6.3.1
- Availability: https://www.hec.usace.army.mil
- Program Size: 205 MB
- Operating System: Windows, Linux
- Developer: The Hydrologic Engineering Center - River Analysis System (HEC-RAS) is a software program developed by the Hydrologic Engineering Center (HEC) of the U.S. Army Corps of Engineers. It is hydraulic river analysis software that can simulate steady and unsteady flow in natural and man-made channels and culverts. It can be used for several purposes, like floodplain mapping.
- Name of the software: ArcGIS10.7
- Availability: https://www.esri.com/en-us/home
- Operating System: Windows, 64-bit
- Requirements: The minimum RAM requirement for ArcGIS GIS Server, ArcGIS GeoEvent Server, ArcGIS Image Server, or ArcGIS Business Analyst for Server is 8 GB per unique license role.
- Developer: ArcGIS10.7 is developed by ESRI (Environmental Systems Research Institute), founded in 1969, for creating and managing geographic information. The software includes a wide range of tools for data analysis, mapping, and visualization. This work utilized software for mapping the drainage basin, terrain, soil, and geologic data.
- Name of the software: WMS.11.0.8
- Availability: https://www.aquaveo.com/software/ wms-watershed-modeling-system-introduction
- Program Size: 1.1GB
- Operating System: Windows 10, 64-bit
- Description: The Watershed Modeling System (WMS 11.0.8) is a very familiar and among the most utilized software in hydrological studies developed by Aquaveo; an engineering services company with many years of experience developing watershed modeling solutions. It provides many models such as HEC-1. The software was utilized in this study to extract the drainage network from DEM data, compute CN and the morphometric parameters to be used for building hydrograph models.
- Name of the software: HEC-HMS 4.10
- Availability: https://www.hec.usace.army.mil/
- Program Size: 202 MB
- Operating System: Windows, macOS, Linux
- Description: HEC-HMS (Hydrologic Engineering Center’s Hydrologic Modeling System) has been developed for the U.S. Army Corps of Engineers at the Hydrologic Engineering Center. HEC-HMS is utilized to simulate the hydrologic and hydraulic processes of a watershed. In addition, it is used for the design and analysis of flood control and drainage systems, and for assessing the impacts of land use changes on water resources.
- Name of the software: HEC-RAS 6.3.1
- Availability: https://www.hec.usace.army.mil/
- Program Size: 205 MB
- Operating System: Windows, Linux
Developer: The Hydrologic Engineering Center - River Analysis System (HEC-RAS) is a software program developed by the Hydrologic Engineering Center (HEC) of the U.S. Army Corps of Engineers. It is hydraulic river analysis software that can simulate steady and unsteady flow in natural and man-made channels and culverts and can be used for several purposes including floodplain mapping.
- Zamani AA, Abbasi HR, Alipour V, Peyravi M, Shojaei P, Goli A, Mohammadinia L (2022) Alternative water resources selection to supply drinking water in flood disasters by multicriteria decision-making techniques (danp and vikor). Journal of Environmental and Public Health (2022)
- Jacinto R, Grosso N, Reis E, Dias L, Santos F, Garrett P (2015) Continental portuguese territory flood susceptibility index–contribution to a vulnerability index. Nat Hazards Earth Syst Sci 15(8):1907–1919
- Ahmadisharaf E, Tajrishy M, Alamdari N (2016) Integrating flood hazard into site selection of detention basins using spatial multi-criteria decisionmaking. J Environ Planning Manage 59(8):1397–1417
- Adib-Moghaddam A (2010) A Metahistory of the Clash of Civilisations: Us and Them Beyond Orientalism. Columbia University, ???
- Ahmadisharaf E, Kalyanapu AJ, Chung E-S (2016) Spatial probabilistic multi-criteria decision making for assessment of flood management alternatives. J Hydrol 533:365–378
- Chourushi S, Lodha P, Prakash I (2019) A critical review of hydrological modeling practices for flood management. Pramana Res J 9:352–362
- Ochoa-Rodriguez S, Wang L-P, Gires A, Pina RD, ReinosoRondinel R, Bruni G, Ichiba A, Gaitan S, Cristiano E, van Assel J et al (2015) Impact of spatial and temporal resolution of rainfall inputs on urban hydrodynamic modelling outputs: A multi-catchment investigation. J Hydrol 531:389–407
- Falah F, Rahmati O, Rostami M, Ahmadisharaf E, Daliakopoulos IN, Pourghasemi HR (2019) Artificial neural networks for flood susceptibility mapping in data-scarce urban areas. Spatial Modeling in GIS and R for Earth and Environmental Sciences. Elsevier, ???, pp 323–336
- Li H, Zhang C, Chen M, Shen D, Niu Y (2023) Data-driven surrogate modeling: Introducing spatial lag to consider spatial autocorrelation of flooding within urban drainage systems. Environmental Modelling & Software, p 105623
- Amaranto A, Mazzoleni M (2023) B-ama: A python-coded protocol to enhance the application of data-driven models in hydrology. Environ Model Softw 160:105609
- Liang Y, Li X, Tsai B, Chen Q, Jafari N (2023) V-floodnet: A video segmentation system for urban flood detection and quantification. Environ Model Softw 160:105586
- Castangia M, Grajales LMM, Aliberti A, Rossi C, Macii A, Macii E, Patti E (2023) Transformer neural networks for interpretable flood forecasting. Environ Model Softw 160:105581
- Qi W, Ma C, Xu H, Chen Z, Zhao K, Han H (2021) A review on applications of urban flood models in flood mitigation strategies. Nat Hazards 108(1):31–62
- Sun Y, Wendi D, Kim DE, Liong S-Y (2019) Deriving intensity–duration– frequency (idf) curves using downscaled in situ rainfall assimilated with remote sensing data. Geoscience Lett 6(1):1–12
- Escobar-Gonz´alez D, Singan˜a-Chasi MS, Gonz´alez-Vergara J, Erazo B, Zambrano M, Acosta D, Villac´ıs M, Guallpa M, Lahuatte B, Peluffo-Ord´on˜ez DH (2022) Intensity-duration-frequency curve for extreme rainfall event characterization, in the high tropical andes. Water 14(19):2998
- Rawat KS, Singh SK (2017) Estimation of surface runoff from semi-arid ungauged agricultural watershed using SCS-CN method and earth observation data sets. Water Conserv Sci Eng 1:233–247
- Deng Z, Wang Z, Wu X, Lai C, Zeng Z (2022) Strengthened tropical cyclones and higher flood risk under compound effect of climate change and urbanization across china’s greater bay area. Urban Clim 44:101224
- Daniel EB, Camp JV, LeBoeuf EJ, Penrod JR, Dobbins JP, Abkowitz MD (2011) Watershed modeling and its applications: A state-of-the art review. Open Hydrology J 5(1)
- Kauffeldt A, Wetterhall F, Pappenberger F, Salamon P, Thielen J (2016) Technical review of large-scale hydrological models for implementation in operational flood forecasting schemes on continental level. Environ Model Softw 75:68–76
- Ikhwali MF, Pawattana C, Nur S, Azhari B, Ikhsan M, Aida N, Silvia CS (2022) Reviews, challenges, and prospects of the application of hydrologic engineering center-hydrologic modelling system (hec-hms) model in indonesia. Eng Appl Sci Res 49(5):669–680
- Soulis KX (2021) Soil conservation service curve number (SCS-CN) Method: Current applications, remaining challenges, and future perspectives. MDPI
- Namara WG, Damisse TA, Tufa FG (2022) Application of hec-ras and hecgeoras model for flood inundation mapping, the case of awash bello flood plain, upper awash river basin, oromiya regional state, ethiopia. Model Earth Syst Environ 8(2):1449–1460
- Pathan AI, Agnihotri PG, Patel D, Prieto C (2021) Identifying the efficacy of tidal waves on flood assessment study—a case of coastal urban flooding. Arab J Geosci 14(20):1–21
- Liu Z, Merwade V (2018) Accounting for model structure, parameter and input forcing uncertainty in flood inundation modeling using bayesian model averaging. J Hydrol 565:138–149
- Shustikova I, Domeneghetti A, Neal JC, Bates P, Castellarin A (2019) Comparing 2d capabilities of hec-ras and lisflood-fp on complex topography. Hydrol Sci J 64(14):1769–1782
- Samadi A, Azizian A (2020) Influence of the concept of subgrid variability and computational mesh dimensions on the performance of hec-ras 2d model in simulating river floodplains (case study: Sarbaz river). J Water Soil Resour Conserv 9(3):1–12
- Zhou Y, Cui Z, Lin K, Sheng S, Chen H, Guo S, Xu C-Y (2022) Short-term flood probability density forecasting using a conceptual hydrological model with machine learning techniques. J Hydrol 604:127255
- El-Shebli M, Sharrab Y, Al-Fraihat D (2023) Prediction and modeling of water quality using deep neural networks. Environ Dev Sustain, 1–34
- Tom RO, George KO, Joanes AO, Haron A (2022) Review of flood modelling and models in developing cities and informal settlements: A case of nairobi city. J Hydrology: Reg Stud 43:101188
- Ali K, Alzaidi M, Al-Fraihat D, Elamir AM (2023) Artificial Intelligence: benefits, application, ethical issues, and organizational responses. Intelligent Sustainable Systems: selected Papers of WorldS4 2022, vol 1. Springer Nature Singapore, Singapore, pp 685–702
- Al-Fraihat D, Sharrab Y, Alzyoud F, Qahmash A, Tarawneh M, Maaita A (2024) Speech Recognition Utilizing Deep Learning: A Systematic Review of the Latest Developments. Human-Centric Computing and Information Sciences, p 14
- Sharrab YO, Alsmira M, Dwekat Z, Alsmadi I, Al-Khasawneh A et al (2021) : Performance comparison of several deep learning-based object detection algorithms utilizing thermal images. In: 2021 Second International Conference on Intelligent Data Science Technologies and Applications (IDSTA), pp. 16–22 IEEE
- Cordova CE (2008) Floodplain degradation and settlement history in wadi al-wala and wadi ash-shallalah, jordan. Geomorphology 101(3):443–457
- Gunjan P, Mishra SK, Lohani AK, Chandniha SK (2020) The study of morphological characteristics for best management practices over the rampur watershed of mahanadi river basin using prioritization. J Indian Soc Remote Sens 48(1):35–45
- Jawad T, Yahya A-B (2013) Y, A.-J.: Application of gis and remote sensing to groundwater exploration in al-wala basin in jordan. Journal of Water Resource and Protection (2013)
- Baker MB, Elektorowicz M, Hanna A (2018) Electrokinetic nondestructive in-situ technique for rehabilitation of liners damaged by fuels. J Hazard Mater 359:510–515
- Alkhaddar RM, Sheehy WJ, Al-Ansari N (2005) Jordan’s water resources: Supply and future demand. Water Int 30(3):294–303
- Yoo C, Jung K, Lee J (2008) Evaluation of rain gauge network using entropy theory: Comparison of mixed and continuous distribution function applications. J Hydrol Eng 13(4):226–235
- Al-Wagdany AS (2021) Construction of idf curves based on nrcs synthetic rainfall hyetographs and daily rainfall records in arid regions. Arab J Geosci 14(6):1–20
- Shrestha R, Di L, Yu G, Shao Y, Kang L, Zhang B (2013) Detection of flood and its impact on crops using ndvi-corn case. In: 2013 Second International Conference on Agro-geoinformatics (Agro-geoinformatics), pp. 200–204 IEEE
- Chow V, Maidment D, Mays L (1988) Applied hydrology, McGraw-hill. Inc., New York
- Acanal N (2021) Snyder-gamma synthetic unit hydrograph. Arab J Geosci 14(4):1–12
- Chen Y, Shi P, Ji X, Qu S, Zhao L, Dong F (2019) New method to calculate the dynamic factor–flow velocity in geomorphologic instantaneous unit hydrograph. Sci Rep 9(1):1–13
- Timbadiya P, Patel P, Porey P (2011) Hec-ras based hydrodynamic model in prediction of stages of lower tapi river. ISH J Hydraulic Eng 17(2):110–117
- Afzal MA, Ali S, Nazeer A, Khan MI, Waqas MM, Aslam RA, Cheema MJM, Nadeem M, Saddique N, Muzammil M et al (2022) Flood inundation modeling by integrating hec–ras and satellite imagery: A case study of the indus river basin. Water 14(19):2984
- Smirnov S, Werner W (2001) Critical exponents for two-dimensional percolation. arXiv preprint math 0109120