Back to Research
paper
Featured Research
mit

AI-Driven Flood Prediction in Urban Areas using Sentinel-1 SAR and Machine Learning

Dr. Kwame Osei, Amina Hassan
Published
January 20, 2024
Read Time
18 minutes
Citations
42
Field
disaster_risk_reduction

Abstract

Development of real-time flood prediction system combining Sentinel-1 SAR data with XGBoost and LSTM models, achieving 94.3% accuracy in predicting flood-prone areas 48 hours in advance.

Citation

apa
Dr. Kwame Osei, Amina Hassan (2024). AI-Driven Flood Prediction in Urban Areas using Sentinel-1 SAR and Machine Learning. Remote Sensing of Environment.
mla
Dr. Kwame Osei, Amina Hassan. "AI-Driven Flood Prediction in Urban Areas using Sentinel-1 SAR and Machine Learning." Remote Sensing of Environment, 2024.
chicago
Dr. Kwame Osei, Amina Hassan. "AI-Driven Flood Prediction in Urban Areas using Sentinel-1 SAR and Machine Learning." Remote Sensing of Environment (2024).
bibtex
@article{spationex,
  title={AI-Driven Flood Prediction in Urban Areas using Sentinel-1 SAR and Machine Learning},
  author={Dr. Kwame Osei, Amina Hassan},
  journal={Remote Sensing of Environment},
  year={2024},
  volume={301},
  pages={113-125},
  doi={null}
}

Urban flooding poses significant risks to infrastructure, economies, and human lives. This research presents an AI-driven framework for early flood prediction in Nairobi's informal settlements using Synthetic Aperture Radar (SAR) data from Sentinel-1 satellite.

Methodology

We combined Sentinel-1 C-band SAR data (VV and VH polarizations) with meteorological data from local stations. The framework employs:

1. XGBoost for feature importance analysis and initial classification

2. LSTM Networks for temporal sequence prediction

3. Random Forest for ensemble learning and uncertainty quantification

Results

The system achieved:

94.3% accuracy in predicting flood-prone areas 48 hours in advance

F1-Score of 0.91 for binary classification of flood/no-flood scenarios

• Reduction of false alarms by 37% compared to traditional hydrological models

The model successfully predicted the April 2023 floods in Mathare Valley with 96% accuracy, enabling early evacuation of 2,300 households.

Publication Details

Journal/Conference
Remote Sensing of Environment
Volume/Issue
301
Pages
113-125
Funding
African Development Bank

Keywords

flood prediction
machine learning
Sentinel-1
SAR
urban flooding
LSTM
XGBoost
early warning system

Share This Research

Continue Exploring

Discover more research in geospatial intelligence and climate resilience.