Attention is all you need for an improved CNN-based flash flood susceptibility modeling. The case of the ungauged Rheraya watershed, Morocco

Akram Elghouat,Ahmed Algouti,Abdellah Algouti,Soukaina Baid
2024-08-04
Abstract:Effective flood hazard management requires evaluating and predicting flash flood susceptibility. Convolutional neural networks (CNNs) are commonly used for this task but face issues like gradient explosion and overfitting. This study explores the use of an attention mechanism, specifically the convolutional block attention module (CBAM), to enhance CNN models for flash flood susceptibility in the ungauged Rheraya watershed, a flood prone region. We used ResNet18, DenseNet121, and Xception as backbone architectures, integrating CBAM at different locations. Our dataset included 16 conditioning factors and 522 flash flood inventory points. Performance was evaluated using accuracy, precision, recall, F1-score, and the area under the curve (AUC) of the receiver operating characteristic (ROC). Results showed that CBAM significantly improved model performance, with DenseNet121 incorporating CBAM in each convolutional block achieving the best results (accuracy = 0.95, AUC = 0.98). Distance to river and drainage density were identified as key factors. These findings demonstrate the effectiveness of the attention mechanism in improving flash flood susceptibility modeling and offer valuable insights for disaster management.
Machine Learning,Artificial Intelligence
What problem does this paper attempt to address?
The paper aims to address the following main issues: ### Main Issues Addressed by the Paper 1. **Improving the Performance of Convolutional Neural Networks (CNN) in Flash Flood Susceptibility Modeling**: - Utilizing attention mechanisms, particularly the Convolutional Block Attention Module (CBAM), to enhance the performance of CNN models in flash flood susceptibility prediction. - Addressing issues such as gradient explosion and overfitting that CNNs face as depth increases. 2. **Evaluating the Impact of CBAM at Different Positions in CNN Architectures**: - Comparing the effects of integrating CBAM into different CNN architectures (such as ResNet18, DenseNet121, and Xception). - Investigating the impact of placing CBAM in different layers of the CNN on flash flood susceptibility prediction results. 3. **Establishing an Effective Flash Flood Susceptibility Prediction Model**: - Selecting the unobserved Rheraya basin in Morocco as a case study, which is a region prone to frequent flash floods. - Using a dataset containing 16 conditioning factors and 522 flash flood inventory points for model training and validation. - Measuring model performance using various evaluation metrics (accuracy, precision, recall, F1 score, and Area Under the ROC Curve [AUC]). ### Contributions of the Paper 1. **First Application of Attention Mechanism-Based CNN in Flash Flood Susceptibility Prediction**. 2. **First Evaluation of the Effects of CBAM at Different Positions in CNN Architectures**. 3. **Generated the First Deep Learning-Based Flash Flood Susceptibility Map for the Rheraya Basin**, providing crucial information for risk management and preventive measures in the region. ### Conclusion Through the above research, the paper not only demonstrates the effectiveness of attention mechanisms in improving the accuracy of flash flood susceptibility prediction but also provides valuable insights for natural disaster management. Additionally, this study offers new ideas and technical means for future research in related fields.