Assessing the environmental impacts of flooding in Brazil using the flood area segmentation network deep learning model

Abstract

Natural disasters are sudden and unpredictable events that result from natural processes and often have a profound impact on people, the environment and the economy. Search and rescue operations conducted after such events are critical to saving lives and ensuring the safety of rescue teams. These operations involve entering disaster-affected areas, assessing damaged structures, rescuing trapped people, providing medical assistance and attending to urgent needs. In a recent study, a novel semantic segmentation model called flood area segmentation network (FASegNet) was proposed to speed up search and rescue operations after natural disasters. In this study, the FASegNet model was used to identify areas affected by severe flooding caused by heavy rainfall in Brazil in May 2024. The FASegNet model detected the affected regions with high precision and was compared with various commonly used segmentation models. In tests conducted with the floodplain and water bodies datasets, FASegNet achieved a mean intersection over union accuracy of 84.3 and 84.5%, respectively, without prior training or data augmentation. The model's ability to achieve higher accuracy with fewer parameters underscores its potential to help search and rescue teams work faster and more effectively in disaster-affected areas.