ObsCrisis-Bench

A multimodal benchmark for evaluating large vision-language models on extreme weather event analysis tasks.

Dataset Description

ObsCrisis-Bench contains 4,599 VQA samples across 127 extreme weather events in 8 disaster categories. Each sample combines satellite multispectral imagery (AMSU-A, HIRS, MHS sensors) with optional weather station data, and requires models to perform risk assessment, type classification, timing prediction, and impact analysis.

Disaster Categories

Category	Events	VQA Samples
cold-wave	5	154
earthquake	10	426
flood	17	580
heat-wave	5	184
mass movement (wet)	6	198
storm	55	2,066
volcanic activity	15	478
wildfire	14	513
Total	127	4,599

Task Structure

Each VQA sample belongs to one of three task categories:

Task Category	Samples	Percentage
Early Warning	1,524	33.14%
Impact Assessment	2,862	62.23%
Recovery Assessment	213	4.63%

Task Types

Early Warning (t1, t8, t12):

• Risk Detection: Assess the risk of an impending disaster
• Type Classification: Identify the specific type of potential disaster
• Arrival Time Prediction: Predict when the disaster will arrive
• Duration Prediction: Predict how long the disaster will last

Impact Assessment (tmax):

• CPI (Crisis Prevention Index): Estimate the crisis prevention index
• Total Affected: Predict total affected population
• Total Deaths: Predict total deaths
• No. Affected: Predict number of affected people

Recovery Assessment (t15):

• Recovery Time Prediction: Predict recovery time after the event

Timestep Convention

• t1 = 14 days before event start
• t8 = 7 days before event start
• t12 = 3 days before event start
• t15 = event start day
• tmax = maximum impact time
• t_number = (image_date - event_start_date).days + 15

Dataset Structure

ObsCrisis-Bench/
├── cold-wave/                          # Satellite imagery
│   └── cold-wave1/<event_id>/<sensor>/<band>/t1.png ... tN.png
├── cold-wave_json/
│   └── All.json                        # VQA records
├── heat-wave/
├── heat-wave_json/All.json
├── earthquake/
├── earthquake_json/All.json
├── flood/
├── flood_json/All.json
├── mass movement (wet)/
├── mass movement (wet)_json/All.json
├── storm/
├── storm_json/All.json
├── volcanic activity/
├── volcanic activity_json/All.json
├── wildfire/
└── wildfire_json/All.json

Satellite Sensors

The dataset includes multispectral imagery from three satellite sensors:

Sensor	Bands	Description
AMSU-A	15 bands	Advanced Microwave Sounding Unit-A
HIRS	20 bands	High-resolution Infrared Radiation Sounder
MHS	5 bands	Microwave Humidity Sounder

VQA Sample Fields

Field	Type	Description
Question_id	string	Unique identifier
Task	string	Task category (Early Warning / Impact Assessment / Recovery Assessment)
Subtask	string	Subtask with timestep, e.g. "Risk Detection (t1)"
Text	string	Question text
Image	string	Comma-separated relative image paths
Stations	object	Weather station sensor data (optional)
Ground truth	string	Standard answer
Event ID	string	Unique event identifier

Weather Station Data

Optional weather station data includes:

• METAR: Meteorological Aerodrome Report
• HadISD: Historical Integrated Surface Database
• GDAS: Global Data Assimilation System

Station data fields:

• Latitude, Longitude, Altitude
• Air Pressure, Temperature, Relative Humidity
• U-Speed, V-Speed (wind components)
• Dew Point Temperature

Evaluation Code

The evaluation framework is available at: https://github.com/YYQ898/ObsCrisis-Bench

Usage

Download Dataset

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="YYQ898/ObsCrisis-Bench",
    repo_type="dataset",
    local_dir="./ObsCrisis-Bench"
)

Load VQA Samples

import json

# Load a specific disaster type
with open("ObsCrisis-Bench/storm_json/All.json", "r") as f:
    samples = json.load(f)

print(f"Total samples: {len(samples)}")
for sample in samples[:3]:
    print(f"Question: {sample['Text']}")
    print(f"Answer: {sample['Ground truth']}")
    print(f"Images: {len(sample['Image'].split(','))} images")

Citation

If you use this dataset, please cite:

@misc{obscrisis-bench,
  title={ObsCrisis-Bench: A Multimodal Benchmark for Extreme Weather Event Analysis},
  author={YYQ898},
  year={2024},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/YYQ898/ObsCrisis-Bench}
}

License

This dataset is released under the MIT License.

Contact

For questions or issues, please open an issue on the GitHub repository.