Flood susceptibility mapping using remote sensing and geospatial data in West Africa

Abstract

Floods in West Africa represent a challenging risk scenario that requires comprehensive action from stakeholders and the local community. To this end, precise and up-to-date data, alongside periodic hazard, susceptibility, and risk mapping, are crucial to understand flood disasters for decision-making. In particular, flood susceptibility maps quantify the likelihood of floods occurring in a spatial location based on topographical, geological, hydrographical, land cover, socioeconomic, urban, environmental, meteorological, and climatic flood-influencing factors. These maps support urban planning and disaster risk management. Thereby, remote sensing and Geographic Information Systems (GIS) contribute to the susceptibility analysis by providing inputs derived from satellite images, geodata, and tools to produce accurate flood susceptibility maps. Hence, this master thesis aims to obtain flood susceptibility maps in West Africa, specifically Nigeria, through open data, remote sensing, GIS, machine learning, and statistical approaches. First, four different Digital Elevation Models (DEMs) from different sources and four hydrological methods (D8, D-inf, Fd8, and Rho8) were combined and compared, and used to get the most applied topographic and hydrographic flood-influencing factors. At the same time, other factors derived from land cover, groundwater, soil, lakes, and coastline datasets were included. In total, twenty-three static flood-influencing factors were calculated at a spatial resolution of 30 meters, representing topographical, hydrographical, land cover, urban, and environmental characteristics of Nigeria. Secondly, preprocessing and feature selection were applied to the flood-influencing factors. Then, three different models were implemented to compare their performance. The selected models were Random Forest (RF), Binary Logistic Regression (LG), and Linear Discriminant Analysis (LDA); in this way, R and WhiteboxTools were used for the data preparation and modeling. As a result, a total of forty-eight flood susceptibility maps were calculated. The best map from each model was provided by the Global DEM (GLO30) from Copernicus and the D8 and Fd8 hydrological methods. Therefore, the GLO30/D8 RF, GLO30/Fd8 LG, and GLO30/Fd8 LDA maps were evaluated with reference data (flood events between September and October 2022) provided by the Deutsches Zentrum für Luft und Raumfahrt (DLR) and the Global Flood Monitoring (GFM). Consequently, the evaluation demonstrated a good performance of the proposed models. Lastly, the exposed population in 2020 was quantified along with the projected population in 2025 and 2030, where Anambra, Bayelsa, Borno, Delta, Rivers, Lagos, Jigawa, Kogi, Kebbi, and Sokoto are the most exposed to floods.