Flood Mapping through Sentinel-1, Sentinel-2 Imagery and U-NET Deep Learning Model

Abstract

Natural disasters are inevitable phenomena that occur more frequently around the world. They often cause severe damage to populations, infrastructure and economic activities. Floods are the most common occurrences due to extreme meteorological events suchas intense rain fall, tropical storms and hurricanes. In Mexico, floods happen every year in different states ofthe country. The state of Tabasco is one of the most affected due to its numerous water bodies. In the state, floods impact agriculture, livestock and other economic sectors, causing significant damage to the population.This has led to efforts to develop strategies to reducethe impact on populations. In recent years, various studies have been conducted to detect floods. Most of these studies rely on the use of satellite images and deep learning algorithms with the purpose of mapping areas affected by floods. The combination of these technologies is becoming one of the most effective methods. This paper presents a methodology for flood mapping in the Rıos zone of Tabasco State using Sentinel-1 SAR, Sentinel-2, and U-Net deep learning architecture. The study period was from 2019 to 2023. The results obtained show that with more data and training periods, accuracy in detecting floods improves.