ObsCrisis-Bench: A Multimodal Benchmark for Extreme Weather Event Analysis

ObsCrisis-Bench is a multimodal benchmark containing 4,599 visual question-answering samples for evaluating large vision-language models. The dataset covers 127 extreme weather events across 8 disaster categories, combining satellite multispectral imagery with optional weather station data. It was created by author YYQ898 and last updated on the Hugging Face platform in June 2026.

Use Cases

• Benchmarking model performance on risk assessment tasks based on satellite imagery and weather data.
• Training models for extreme weather event type classification based on multispectral sensor data.
• Evaluating model accuracy for timing prediction of disaster events based on the described multimodal samples.
• Assessing model capabilities for predicting the impact of extreme weather based on the combined data sources.

Strengths

• Contains 4,599 VQA samples, providing a substantial testbed for model evaluation.
• Covers 127 distinct extreme weather events across 8 disaster categories.
• Integrates satellite multispectral imagery from AMSU-A, HIRS, and MHS sensors with optional weather station data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face, author YYQ898.

Collection Method

Likely compiled from satellite imagery and weather station data for benchmark creation.

Freshness

Last updated 2026-06-10 10:03:05; freshness should be verified.