CONSTRUCTION OF A FLOOD MONITORING MODEL USING DEEP LEARNING BASED ON GEOSPATIAL DATA FROM MULTIPLE SOURCES

• Grant number: 24/20838-3
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: October 01, 2025
• End date: March 31, 2027
• Field of knowledge: Interdisciplinary Subjects
• Principal Investigator: Thales Sehn Körting
• Grantee: Brenda Oliveira Rocha
• Host Institution: Instituto Nacional de Pesquisas Espaciais (INPE). São José dos Campos , SP, Brazil
• Associated research grant: 23/09118-6 - Content discovery in remote sensing image catalogs, AP.R

Abstract

Major floods were the type of disaster with the highest number of records in 2022,accounting for 45% of all disasters worldwide. Given the need for a comprehensive analysisof the affected areas, due to the extent and intensity of the damage, the use of RemoteSensing (RS) data is essential for flood management, considering the complexity of theevents. However, defining a method for the temporal monitoring of flood-prone areas ischallenging, given the complexity of RS data processing, the topographic characteristics ofthe environment, changes in land use and cover, and the analysis of large precipitationevents. From a temporal perspective, understanding these aspects can be essential to analyzeflood trends and patterns over time, aiding in strategic urban planning. Models that useRecurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) can offeran ideal solution for monitoring flood-prone areas, due to their powerful ability tosimultaneously extract temporal, spatial and spectral features from images at different scalesand levels of abstraction. The association of dynamic data (such as optical images, SyntheticAperture Radar, and precipitation data) and static data (such as attributes extracted from aDigital Elevation Model), structured in a DL network that considers sequential data, has thepotential to improve the identification of flood-prone areas.