Shakib et al. (2026) FloodWatch: An Automatic Machine Learning Tool for Flood Forecasting and Segmentation using Weather Data and Images

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-02-16
• Authors: Tasnim Ullah Shakib, Tariq Hasan Rizu, Ellora Yasi, Nusrat Sharmin, Rubyeat Islam
• DOI: 10.1007/s11269-025-04444-8

Research Groups

• Department of Computer Science and Engineering, Military Institute of Science and Technology, Dhaka, Bangladesh

Short Summary

This study introduces "FloodWatch," an automatic machine learning (AutoML) tool designed for comprehensive flood forecasting and segmentation. It integrates various machine learning and deep learning algorithms, Explainable AI (XAI), Generative Adversarial Networks (GANs) for data augmentation, and GIS visualization into a user-friendly, no-code web application to enhance flood prediction accuracy and accessibility for disaster management.

Objective

• To develop a practical and effective automatic machine learning tool, FloodWatch, for flood forecasting and segmentation, aiming to lessen the negative effects of flooding events on communities and ecosystems.
• To create a novel and seamless user interface for flood analysis that integrates diverse analytical tools and methodologies, making advanced flood prediction accessible to both meteorology professionals and non-experts.

Study Configuration

• Spatial Scale: Primarily Bangladesh, with data collected from 33 stations across the country.
• Temporal Scale: Weather data from 1949 to 2013; synthetic data generated for 2014 to 2021; GIS visualization covers 65 years of Bangladesh's history.

Methodology and Data

• Models used:
  • Machine Learning (Classification): AdaBoost, KNN Classifier, Logistic Regression, Decision Tree, Random Forest, Voting Classifier, Gradient Boosting, Support Vector Machine (SVM), Naive Bayes Classifier.
  • Deep Learning (Image Classification): Ensemble of 8 transfer learning models (pre-trained Convolutional Neural Networks).
  • Deep Learning (Image Segmentation): U-net (Convolutional Neural Network).
  • Data Augmentation: Conditional Generative Adversarial Networks (CTGAN).
  • Explainable AI (XAI): SHapley Additive exPlanations (SHAP).
  • User Interface: Streamlit (Python framework).
• Data sources:
  • Tabular Weather Data: Bangladesh flood dataset from a GitHub source (Gauhar et al 2021), consisting of meteorological parameters.
  • Flood and Non-Flood Images: Collected from an IEEE paper source (Rahnemoonfar et al 2021) and a Kaggle source (Krish Sharma 2022).
  • Flood Images and Masks for Segmentation: Kaggle source (Krish Sharma 2022).

Main Results

• Developed "FloodWatch," a comprehensive, no-code AutoML web application for flood forecasting, image segmentation, image classification, synthetic data generation, XAI, and GIS visualization.
• The flood forecasting module incorporates 9 machine learning algorithms, with performance evaluated using accuracy, precision, recall, F1-score, and AUC.
• Explainable AI (SHAP) provides transparent feature relevance analysis for flood prediction through Beeswarm and Bar plots.
• CTGAN successfully generated realistic synthetic tabular flood data for 2014-2021, closely mirroring the distribution of original data (1949-2013) and outperforming random data generation.
• Deep learning models, specifically an ensemble of transfer learning architectures, demonstrated proficiency in classifying flood and non-flood images.
• A U-net model achieved an Intersection over Union (IoU) score of 73.90% for accurate flood area segmentation within images.
• An interactive GIS map was developed to visualize historical flood events and rainfall patterns across Bangladesh over 65 years, aiding in identifying flood-prone areas.

Contributions

• Development of "FloodWatch," a novel automatic machine learning tool for flood prediction from weather data, offering statistical analysis, flood forecasting with 9 ML algorithms, hyperparameter tuning, and no-code accessibility.
• Integration of Explainable AI (XAI) to enhance the interpretability and trustworthiness of flood prediction results.
• Incorporation of Generative Adversarial Networks (GANs), specifically CTGAN, for realistic tabular flood data augmentation, a unique feature in existing AutoML tools for flood prediction.
• Advanced capabilities in flood image segmentation using deep learning (U-net) for precise identification of flooded areas.
• Proficiency in flood image classification through deep learning with transfer learning, enabling accurate discernment of flooded regions.
• Seamless integration of Geographic Information System (GIS) functionality for monitoring and visualizing flood-prone areas, supporting disaster preparedness.
• Contribution to the Fourth Industrial Revolution (4IR) by applying artificial intelligence to climate domain applications and addressing climate change challenges.

Funding

None.

Citation

@article{Shakib2026FloodWatch,
  author = {Shakib, Tasnim Ullah and Rizu, Tariq Hasan and Yasi, Ellora and Sharmin, Nusrat and Islam, Rubyeat},
  title = {FloodWatch: An Automatic Machine Learning Tool for Flood Forecasting and Segmentation using Weather Data and Images},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-025-04444-8},
  url = {https://doi.org/10.1007/s11269-025-04444-8}
}