Simantiris et al. (2026) AIFloodSense: A Global Aerial Imagery Dataset for Semantic Segmentation and Understanding of Flooded Environments

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-03-19
• Authors: Georgios Simantiris, Konstantinos Bacharidis, Apostolos Papanikolaou, Petros Giannakakis, Costas Panagiotakis
• DOI: 10.3390/rs18060938

Research Groups

[Information not available in the provided text.]

Short Summary

This paper introduces AIFloodSense, a comprehensive and globally diverse evaluation benchmark designed to advance domain-generalized Artificial Intelligence for climate resilience and flood detection. It demonstrates that rigorous dataset diversity, rather than sheer scale, is more effective for training robust flood detection models, leading to superior generalization capabilities.

Objective

• To address the scarcity of diverse and granular annotated datasets that hinder the development of robust, domain-generalized AI models for accurate flood detection and disaster response.

Study Configuration

• Spatial Scale: Global, covering 64 countries and six continents.
• Temporal Scale: 2022–2024.

Methodology and Data

• Models used: State-of-the-art architectures were used to provide baseline benchmarks for all tasks.
• Data sources: AIFloodSense dataset, comprising 470 high-resolution aerial images capturing 230 distinct flood events.

Main Results

• AIFloodSense is a comprehensive dataset of 470 high-resolution aerial images, capturing 230 distinct flood events across 64 countries and six continents, ensuring global diversity and temporal relevance (2022–2024).
• The dataset supports three complementary tasks: (i) Image Classification (including novel sub-tasks for environment type, camera angle, and continent recognition); (ii) Semantic Segmentation (providing precise pixel-level masks for flood, sky, buildings, and background); and (iii) Visual Question Answering (VQA) for natural language reasoning.
• Baseline benchmarks using state-of-the-art architectures demonstrate the dataset’s complexity and utility for robust AI tools in environmental monitoring.
• AIFloodSense, despite its compact size, enables better generalization on external test sets compared to much larger alternative datasets, validating that rigorous diversity is more effective than scale for training robust flood detection models.

Contributions

• Introduction of AIFloodSense, a novel, comprehensive, and globally diverse evaluation benchmark for advancing domain-generalized AI in flood detection and climate resilience.
• Provides a dataset with exceptional global diversity and temporal relevance (2022–2024), overcoming limitations of previous benchmarks.
• Supports a multi-task approach including Image Classification (with novel sub-tasks), Semantic Segmentation (pixel-level), and Visual Question Answering.
• Empirically demonstrates that rigorous diversity in a dataset is more critical than sheer scale for achieving robust generalization in flood detection models.
• The dataset is made publicly available to accelerate further research in the field.

Funding

[Information not available in the provided text.]

Citation

@article{Simantiris2026AIFloodSense,
  author = {Simantiris, Georgios and Bacharidis, Konstantinos and Papanikolaou, Apostolos and Giannakakis, Petros and Panagiotakis, Costas},
  title = {AIFloodSense: A Global Aerial Imagery Dataset for Semantic Segmentation and Understanding of Flooded Environments},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18060938},
  url = {https://doi.org/10.3390/rs18060938}
}