Preparation of Flood Hazard Potential Map using EBF Statistical Method: The Case Study of Azarshahr Chai Basin

ABSTRACT
Azarshahrchai basin, located on the western slope of Sahand Mountain, is one of the flood-prone basins that every year, with the onset of spring rains, water flows in the valleys and floods occur in this basin. Thus, investigating and identifying flood-prone areas is a fundamental step to managing and reducing flood damage in this basin. The main goal of this research is to determine the effectiveness of the two-variable statistical method EBF (witness-belief function) by using the effective variables in flood occurrence and geographic information systems (GIS). In order to achieve this goal, 82 floodgates were first prepared, of which 57 points were randomly used for model training and 25 points for validation. In the next step, 14 parameters effective in the occurrence of floods, including elevation, slope, aspect, slope curvature, distance to river, distance to road, river density, TWI (Topographic Wetness Index), lithology, soil type, Stream Power Index (SPI), rainfall, Land-use, and NDVI were selected to prepare flood risk map. The results of the analysis of the parameters showed that the height classes of 1289-1500 meters, areas with a slope of 0-15 degrees, flat slope direction, areas with slopes with concave curvature, and areas near waterways and roads have a high potential for flooding. The evaluation of the accuracy of the research model using the ROC curve and the area under the curve (AUC) showed that the EBF model, with a value of 0.973, had an excellent performance in preparing the flood risk potential map in the study area. Therefore, the prepared map can be a reference framework for planners to improve and reduce flood risks along with flood risk management activities in this basin
Extended abstract
Introduction
Floods are the most common natural hazards that occur all over the world and are the product of runoff production and confluence processes in watersheds. Its formation process is complex and mainly influenced by precipitation, vegetation, topography, and soil factors. It is difficult to define a flood. In general, a flood is a relatively large flow that exceeds the capacity of the river channel. While normal currents occur within the channel, periodic strong currents pass over the banks of the channel and spill over the surrounding flood plains. Considering that floods are natural phenomena, and although they cannot be completely controlled, their risks can be minimized. Flood risk reduction planning requires mapping, modeling and predicting flood events at different spatial and temporal scales. Many researchers have considered flood risk mapping as one of the most efficient tools for prevention and mitigation. A flood risk map visualizes the spatial distribution of flood risk, and flood vulnerability assessment is recognized as a preventive measure for risk management and planning and helps planners and decision-makers identify highly vulnerable areas or communities.
 
Materials and methods
In this research, among the 14 factors of elevation, slope, aspect, slope curvature, distance to river, distance to road, river density, TWI (Topographic Wetness Index), lithology, soil type, Stream Power Index (SPI), rainfall, Land-use, and NDVI has been used to investigate the watershed in terms of flood potential. Using LANDSAT8 satellite images of the OLI sensor, the position of 82 flood points in the Azarshahrchai catchment area was determined; 57 flood points were randomly used for training data and 25 points for validation data. Then, using a digital elevation model (DEM) with a spatial resolution of 12.5 meters and GIS and ENVI software, the required information layers were extracted and using the EBF model, the basin was zoned in terms of flood risk and final maps were extracted. The EBF model or witness-belief function, which is also called Dempster-Shafer theory, includes 4 functions degree of confidence (BEL), uncertainty (DIS), uncertainty (UNC) and degree of plausibility (PLS), and its values are between -1 0 and its greatest advantage is the flexibility resulting from accepting uncertainty and combining many sources of belief. ROC curve is used to evaluate the accuracy of the model. A popular method for visualizing the discriminative accuracy of binary classification models is the area under the curve (AUC), which is a common measure of its accuracy.
 
Results and discussion
Elevation: The highest BEL or degree of confidence was recorded for the heights of 1289-1500, with a number of 0.736, which indicates the high probability of flooding in these areas. Altitudes of 1500-2000 and 2000-2500 have a lower probability of flooding than heights of less than 1500 meters, with confidence levels of 0.236 and 0.027, respectively. Confidence level 0 for heights higher than 2500 meters depicts the impossibility of flooding in these areas.
Slope: Areas with a slope of 0-15 degrees with a confidence level of 0.865 and slopes above 30 degrees with a confidence level of 0 have the highest and lowest probability of flooding and injury from this hazard, respectively.
Aspect: The highest degree of confidence was recorded for flat slopes with a value of 0.228, for east-facing slopes with a value of 0.132, for northwest slopes with a value of 0.131, and for a northeast-facing slope with a value of 0.115.
Lithology: The highest degree of confidence was recorded for the lake salt formation with a value of 0.388, deposits of high-level pediments and valley terraces with a value of 0.220, and semi-volcanic rocks from rhyolite to rhyodacite with a value of 0.186. These formations with a high degree of certainty have the possibility of aggravating the risk of flooding.
Soil type: The degree of reliability of Aridosol soil with a value of 0.768 is very high compared to the other two types of soil in the basin. Aridosols are followed by inceptisols and stone outcrops with confidence levels of 0.135 and 0.096, respectively.
Distance to the river: 200 meters from the river, one of the basin's most vulnerable areas, with a confidence level of 0.564. Distances of more than 800 meters from rivers with a confidence level of 0.205 have taken the second rank of vulnerable areas in this index.
Distance to road: the highest level of confidence with a number of 0.319 is assigned to distances of 0-200 meters from roads and the lowest level of confidence with a value of 0.069 is assigned to distances of more than 800 meters. Distances 200-400, 600-800 and 400-600 are next in terms of flood vulnerability with confidence levels of 0.296, 0.168 and 0.146, respectively.
River density: In the sections where the river density is more than 0.8 square kilometers per square kilometer, the highest degree of confidence has been recorded with a value of 0.321. After that, in the places where the density of the river is 0.4-0.2 square kilometers per square kilometer, the lowest level of confidence with the number 0.045 has been assigned to it.
Topographic wetness index: In this basin, TWI 7-10-94, with the highest score of 0.400, is the most vulnerable area against floods. The regions with TWI 3.7-2.5, with a confidence level of 0.032, are the least vulnerable to flood risk.
Precipitation: The areas with more than 500 mm of rainfall, with a confidence level of 0.652, and the areas with 273-330 mm of rainfall are the most vulnerable areas against floods.
The normalized index of vegetation difference (NDVI): the areas with high vegetation difference with a confidence level of 0.597 and the areas without vegetation with a confidence value of 0 have the highest and lowest vulnerability.
Land use: From the point of view of riverbed land use, with a confidence value of 0.680, garden agriculture with a confidence level of 0.145, water-agriculture with a confidence value of 0.095, residential areas with a confidence value of 0.042, agriculture-garden with a confidence level of 0.027, rainfed areas with a confidence level of 0.006, and pasture use with a confidence level of 0.001, these areas are highly vulnerable to flood risk.
SPI: The most vulnerable areas in this index are the river's power in distances of 0-1500 meters, with a confidence level of 0.437.
ROC curve was used for the overall evaluation of the model. which with a number of 0.973 indicates the good performance of this model in the basin in terms of flood vulnerability.
Conclusion
In this research, using the EBF model, flood-sensitive areas were zoned based on 14 factors affecting their occurrence, using 57 flood-prone and 25 flood-free points. Based on the results obtained from this research, 12.7% of the basin is in the high class, 20.6% is in the very high vulnerability class, and the height factors are 1289-1500 meters, areas with a slope of 0-15 degrees, flat slope direction, Areas with slopes with concave curvature, flat slopes, lake salt formation, Aridosol soil, distances of 200 meters from rivers and roads, river density in areas greater than 0.8 square kilometers per square kilometer, Topographic Wetness Index (TWI) 10.94 - 7 areas with Precipitation more than 500 mm, Normalized Difference Vegetation Index (NDVI), River Bed, Erosion and Stream Power Index (SPI) at distances of 1500 kilometers from the river with weights of 0.736, 0.865, 0.228, and 703. 0, 0.388, 0.768, 0.564, 0.319, 0.321, 0.400, 0.652, 0.597, 0.680, and 0.432 are the most vulnerable areas of the basin and need more attention. The evaluation of the obtained result using the ROC curve with a value of 0.973 depicts the good performance of this model for basin zoning. The prepared flood susceptibility map of the Azarshahrchai watershed provides important information on vulnerable areas and can be a reference framework for planners to improve and reduce flood risks along with flood risk management activities in this basin. Therefore, in order to reduce the damage caused by floods in the coming years, careful planning should be done for high-risk areas.
 
Funding
There is no funding support.
 
Authors’ Contribution
All of the authors approved the content of the manuscript and agreed on all aspects of the work.
 
Conflict of Interest
Authors declared no conflict of interest.
 
Acknowledgments
We are grateful to all the scientific consultants of this paper.