Research map / Large language models / Language, safety and interpretability
Flood Insights: Integrating Remote and Social Sensing Data for Flood Exposure, Damage, and Urgent Needs Mapping
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents Flood Insights, an end-to-end system fusing satellite, geospatial, and Twitter text/image data to map flood extent, exposure, damage, and urgent needs, deployed during the 2022 Pakistan floods.
Problem
Humanitarian organizations lack timely situational awareness (flood extent, population exposure, damage, and urgent needs) during disasters, and traditional ground assessments are slow and hard to reach.
Approach
Combines remote sensing (Sentinel-1 SAR flood segmentation with nighttime-light urban/rural split and WorldPop demographics) with social sensing (zero-shot tweet classification into damage, affected-people, and needs classes; ResNet-based image relevance and flood/damage classification) and validates district-level impact scores against official government data via geographically weighted regressi
Key results
- Zero-shot tweet classifier reaches F1 77.3% at top-1/2 (vs 78.1% HumAID baseline) and 84.1% at top-3.
- Combined remote and social sensing impact scores correlate with official ground truth at R2 = 0.7824 (social alone 0.7744, remote alone 0.3483) across 53 districts.
- About 18.95 million people estimated exposed to 2022 Pakistan floods; ~11.48% of country flooded over one month.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Combine remote sensing impact scores with social sensing impact scores | Flood Insights impact scoring · Pakistan floods, 53 districts | GWR R²: 0.7744 -> 0.7824 | Helped |
| Use social sensing impact scores instead of remote sensing impact scores | Flood Insights impact scoring · Pakistan floods, 53 districts | GWR R²: 0.3483 -> 0.7744 | Helped |
| Use top-3 rather than top 1–2 predictions | BART-large zero-shot classifier · HumAID | F1-score: 77.3% -> 84.1% | Helped |
| Use the BART-large zero-shot classifier rather than Alam et al. (2021) | HumAID classification · HumAID | F1-score (top 1–2 predictions): 78.1% -> 77.3% | No clear effect |
Surprising findings
- Social sensing alone correlated more strongly with ground truth (R2 0.7744) than remote sensing alone (R2 0.3483).
Limitations
- Remote sensing impact score correlates weakly with ground truth; rural and low-connectivity districts show poor social-sensing fit.
- Ground-truth validation available for only 53 of 160 districts.
- Dashboard granularity is fixed at activation; changing AOI requires costly recomputation.
Experimental setup
Datasets: Cloud to Street Microsoft Flood Dataset JRC global surface water HumAID WorldPop harmonized nighttime lights Places365 ImageNet Pakistan official flood ground-truth (PDMA) Microsoft Building Footprints Sentinel-1 Twitter/X streams
Benchmarks: HumAID Pakistan 2022 official flood ground truth
Metrics: F1-score R2 (geographically weighted regression) throughput latency
Baselines: HumAID zero-shot baseline (Alam et al. 2021) UNOSAT flood segmentation model (Nemni et al. 2020)
Builds on: U-Net BART BERT toponym extractor (Suwaileh et al. 2022) Incidents model (Weber et al. 2020) ResNet-50
Closest papers by meaning
- DisasterM3: A Remote Sensing Vision-Language Dataset for Disaster Damage Assessment and Response (NeurIPS 2025 · similarity 0.86)
- MONITRS: Multimodal Observations of Natural Incidents Through Remote Sensing (NeurIPS 2025 · similarity 0.86)
- RSCC: A Large-Scale Remote Sensing Change Caption Dataset for Disaster Events (NeurIPS 2025 · similarity 0.85)
- Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency Operations (NeurIPS 2026 · similarity 0.83)
- Kuro Siwo: 33 billion $m^2$ under the water. A global multi-temporal satellite dataset for rapid flood mapping (NeurIPS 2024 · similarity 0.82)
- Off to new Shores: A Dataset & Benchmark for (near-)coastal Flood Inundation Forecasting (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.