Application of hybrid machine learning model for flood hazard zoning assessments

Developing flood hazard risk assessments is vital in early warning systems to mitigate damage resulting from floods. However, assessing flood risk zones is difficult because of complex physical processes. In this study, a two-step flood hazard zoning model based on the random forest (RF) and self-organizing map (SOM) is proposed to yield the flood hazard zones map. In the first step (flood susceptibility analysis), the flood conditioning factors are used to obtain the flood susceptibility values. In the second step (flood hazard zoning), the proposed model not only considers the flood susceptibility value of the self-pixel as the module input, but also simultaneously considers flood susceptibility values of surrounding pixels to yield the flood hazard zoning map. The proposed model was applied to the Lanyang Plain in Yilan County, Taiwan, to demonstrate its advantages. The results indicated that the proposed model with the flood susceptibility values of the self-pixel and surrounding pixels does improve the assessment performance, significantly improving percentage of the effective ratio (ER) from results for the very high and high-risk level zones is 6 and 46%, respectively. The ER of the proposed model also improved by 9.6 and 30% for the very high and high-risk level zones than the conventional model, and it could provide optimal flood hazard zoning maps. In conclusion, the proposed model is expected to be useful in supporting the formulation of adequate disaster mitigation strategies.