Flood hazard zones prediction using machine-learning-based geospatial approach in lower Niger River basin, Nigeria

Abstract

Flooding has had devastating impacts on lives and properties over the years, caused as a result of climate change, rapid population growth, urbanization, and poor urban planning. The recurring events of this hazard necessitate the development of accurate flood hazard maps to better inform disaster preparedness and mitigation strategies. Therefore, this study aims to integrate Machine Learning Models (MLM) with Geographic Information Systems (GIS) techniques to predict flood hazard zones in the lower Niger River basin in Nigeria. The Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN) machine learning models were employed to assess flood-prone areas based on twenty (20) influencing factors, categorized into topographic, hydrologic, environmental/anthropogenic, and climatic factors. Based on historical flood events from different sources for the period of 1998–2023 within the study area, data from 1164 flooded and non-flooded points were utilized to train and test the models. Following the evaluation by statistical metrics such as precision, recall, f1-score, overall accuracy, and Receiver Operating Characteristics Area Under the Curve (ROC-AUC), XGBoost was found to have the best performance with an overall accuracy of 91% and ROC-AUC score of 0.89 compared to SVM and ANN with overall accuracy 88% and 85% respectively, and ROC-AUC scores 0.82 and 0.86 respectively. The flood hazard maps showed that areas near the river, particularly in the central and southern part of the basin, including the river confluence areas, are most prone to flooding which is likely to affect critical elements such as croplands, settlements, population centers, and infrastructures. This study provides a foundation to prioritize efforts and resources toward mitigating flood impacts in highly vulnerable areas.