Mapeamento de inundações e alagamentos em área urbana utilizando Aprendizagem de Máquina: abordagem baseada em relatos da imprensa e fatores naturais e antrópicos

Abstract

Floods and inundations are considered the most frequent extreme adverse events on the planet, and in urban areas, they are exacerbated by local and global environmental changes, bringing significant economic impacts and harming the lives of millions of people. This dissertation aims to investigate, in the first chapter, the state of the art of floods and inundations, within a final portfolio of 43 articles that highlights the most disseminated techniques and methodologies to promote the mapping of flood-prone and inundation-prone areas in cities, identifying methodological indicators, research gaps, and opportunities, and clarifying the current discussion on the topic in the academic field. The second part of this work discusses the use of Machine Learning methodologies to map potential flood and inundation zones in the urban perimeter. By applying the Gradient Boosting Classifier (GBC) method, we identified the possibility of floods and inundations with 433 occurrences per year in the municipal area of Campo Grande (MS), using locations reported by the press as ground truth and input factors that characterize the study area in natural and anthropogenic aspects.