Abstract
This study aimed to assess flash-flood susceptibility using a new hybridization approach of Deep Neural Network (DNN), Analytical Hierarchy Process (AHP), and Frequency Ratio (FR). A catchment area in south-eastern Romania was selected for this proposed approach. In this regard, a geospatial database of the flood with 178 flood locations and with 10 flash-flood predictors was prepared and used for this proposed approach. AHP and FR were used for processing and coding the predictors into a numeric format, whereas DNN, which is a powerful and state-of-the-art probabilistic machine leaning, was employed to build an inference flash-flood model. The reliability of the models was verified with the help of Receiver Operating Characteristic (ROC) Curve, Area Under Curve (AUC), and several statistical measures. The result shows that the two proposed ensemble models, DNN-AHP and DNN-FR, are capable of predicting future flash-flood areas with accuracy higher than 92%; therefore, they are a new tool for flash-flood studies.
Highlights
As a result of the climatic changes that occurred during in recent years as well as of the massive changes in land use, a significant growth in the number of extreme meteorological and hydrological phenomena can be observed [1]
The selection literature, the selection and the inclusion into the analysis of another samplewhere containing the locations and the inclusion into the analysis of another sample containing the locations the phenomenon is where are the mandatory phenomenon is absent are mandatory in order of tothe increase the performance of the another models absent in order to increase the performance models
The highest predictive capability was attributed to slope angle (0.92), followed by profile curvature (0.83), hydrological soil group (0.71), lithology (0.64), plan curvature (0.59), Topographic Wetness Index (TWI) (0.54), convergence index (0.43), Topographic Position Index (TPI) (0.41) and aspect (0.32)
Summary
As a result of the climatic changes that occurred during in recent years as well as of the massive changes in land use, a significant growth in the number of extreme meteorological and hydrological phenomena can be observed [1]. Popular floods and flash-floods are considered among the most devastating hazards [2,3] Worldwide, these natural hazards cause multiple economic losses annually, affecting over 200 million people [4,5]. Unlike simple neural networks whose architecture contains a single hidden layer of neurons, the DNN has a feed-forward structure, which, along with input and output layers, contains two or more hidden layers [33]. Because the presence of multiple hidden layers is intended to solve complex classification problems, DNN models are considered to be more powerful and more efficient than simple neural networks [34]. The input layer will contain information regarding the flash-flood predictors, which will be forwarded to hidden layers where this information will be analyzed and processed. In terms of the DNN, this algorithm calculates the gradient of the loss function with respect to each weight by the chain rule and avoids the redundant computations within the intermediate factors of this chain rule [36]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
