Introduction and Goal Floods cause numerous financial losses and human casualties every year and have a devastating impact on the sustainable development of the country. However, generally, actual flood events are not used in spatial analysis and modeling, and flood susceptibility maps are prepared solely based on expert opinions and multi-criteria decision-making methods. On the other hand, management plans and actions of executive agencies are usually developed without considering the different flood susceptibility zones of watersheds. This study was conducted with the aim of utilizing observational data of flood events and determining the efficiency of the Naive Bayes model in the field of spatial prediction of flood susceptibility in the Zarineh-Rud watershed.
Materials and Methods First, a database for flood events was prepared based on flood event information in the Zarineh-Rud watershed recorded by the Provincial Disaster Management Office and the Regional Water Company. Given that various environmental factors play a role in the formation of floods and inundation of lands adjacent to rivers, flood modeling is not possible without considering them. Therefore, thirteen environmental factors affecting flooding were selected, including land elevation, slope direction, drainage density, land use, lithology, surface curvature, cross-sectional curvature, average annual rainfall, slope percentage, soil texture, flow power index, distance from the watercourse, and topographic moisture index. The multicollinearity of environmental factors was examined using the tolerance factor statistic. Raster layers of environmental factors were introduced as independent variables into the Naive Bayes model. The flood event locations were divided into two training and validation groups based on the spatial random method with a ratio of 30|70. After running the model, a flood susceptibility map of the Zarineh-Rud watershed was generated in such a way that each cell represents the probability of flooding. The accuracy of the flood susceptibility map was evaluated using independent and threshold-dependent statistics as well as data from the validation group.
Results and Discussion Based on the results of this study, the independent variables in question lacked multicollinearity and could be used as predictive factors in the modeling process. The validation results based on the area under the receiver operating characteristic curve statistic showed that the flood susceptibility map has an accuracy of 93.6%. According to the threshold-dependent statistics, it can be seen that the performance of the Naive Bayes model was obtained based on the Accuracy statistic of 85.7%, the Precision statistic of 82.6%, and the Recall statistic of 90.4%.
Conclusion and Suggestions The Naive Bayes machine learning model has shown good performance for spatial prediction of flood susceptibility at the watershed scale and can consider various variables for spatial analysis. The flood susceptibility map should be considered as the basis for planning river regulation operations (such as building coastal river walls and bed stripping), flood management (such as respecting river boundaries), and watershed management (such as building watershed management structures in the upstream of flood-prone areas) in the Zarineh-Rud watershed. It is suggested that this model be used to prepare flood susceptibility maps based on historical flood data in other basins of the country.
Extended Abstract Introduction and Goal Floods cause numerous financial losses and human casualties every year and have a devastating impact on the sustainable development of the country. However, generally, actual flood events are not used in spatial analysis and modeling, and flood susceptibility maps are prepared solely based on expert opinions and multi-criteria decision-making methods. On the other hand, management plans and actions of executive agencies are usually developed without considering the different flood susceptibility zones of watersheds. This study was conducted with the aim of utilizing observational data of flood events and determining the efficiency of the Naive Bayes model in the field of spatial prediction of flood susceptibility in the Zarineh-Rud watershed. Materials and Methods First, a database for flood events was prepared based on flood event information in the Zarineh-Rud watershed recorded by the Provincial Disaster Management Office and the Regional Water Company. Given that various environmental factors play a role in the formation of floods and inundation of lands adjacent to rivers, flood modeling is not possible without considering them. Therefore, thirteen environmental factors affecting flooding were selected, including land elevation, slope direction, drainage density, land use, lithology, surface curvature, cross-sectional curvature, average annual rainfall, slope percentage, soil texture, flow power index, distance from the watercourse, and topographic moisture index. The multicollinearity of environmental factors was examined using the tolerance factor statistic. Raster layers of environmental factors were introduced as independent variables into the Naive Bayes model. The flood event locations were divided into two training and validation groups based on the spatial random method with a ratio of 30|70. After running the model, a flood susceptibility map of the Zarineh-Rud watershed was generated in such a way that each cell represents the probability of flooding. The accuracy of the flood susceptibility map was evaluated using independent and threshold-dependent statistics as well as data from the validation group. |