Integrating Harris Hawks optimization and TensorFlow deep learning for flash flood susceptibility mapping using geospatial data

Flash floods are recognized as some of the most devastating natural disasters globally, causing significant damage to socio-economic infrastructures, ecosystems, and human lives, thus highlighting the critical need for accurately identifying areas at risk. In order to address this challenge, our study introduces a novel approach by integrating Harris Hawks Optimization (HHO) with the TensorFlow Deep Neural Network (TFDNN), termed HHO-TFDNN, for assessing flash flood susceptibility. The innovation of HHO-TFDNN resides in its dual structure: TFDNN is employed to develop flash flood prediction models, while HHO is utilized to optimize their parameters. This methodology was applied to a region in northern Vietnam, frequently impacted by flash floods. A detailed flash flood database was assembled using various geospatial data sources for the model’s training and validation. The results underscore the model’s exceptional predictive accuracy, demonstrated by a high F-score of 0.913, a Kappa statistic of 0.825, and an overall accuracy of 91.2%. These findings establish HHO-TFDNN as a highly effective tool for predictive modeling in flash flood management.