## TL;DR
When disaster strikes, AI helps humanitarian organizations see the impact, coordinate the response, and deliver aid more efficiently. Satellite imagery AI detects damaged buildings within hours of an earthquake. NLP models scan social media for real-time crisis maps. Reinforcement learning optimizes routing of relief trucks through damaged infrastructure. As climate change intensifies disasters, AI is becoming a critical capability for the humanitarian sector.

## Core Explanation
AI for humanitarian aid operates across the disaster management cycle: (1) Mitigation/Preparedness -- ML models predict where disasters are most likely (flood risk mapping, famine early warning systems like FEWS NET, epidemic outbreak forecasting), integrating satellite data, weather forecasts, demographic data, and conflict monitoring; (2) Response -- satellite imagery AI compares pre- and post-disaster imagery to identify damaged buildings, blocked roads, flooded areas, and displaced populations. Social media NLP ingests millions of posts to create real-time crisis maps identifying where people need help and what they need; (3) Recovery -- AI assists in damage claims processing, reconstruction prioritization, and long-term resilience planning.

## Detailed Analysis
Key technologies: (1) Satellite damage assessment -- Deep learning models (U-Net, DeepLab, Swin Transformer with change detection heads) process satellite radar (SAR, which can see through clouds and at night) and optical imagery to identify damaged structures. The xView2 benchmark (DARPA/DIU) provides 850,000+ building annotations across 15 disaster types for training. Google's Skai framework and UN PulseSatellite have operationalized this technology; (2) Crisis mapping from social media -- AIDR (Qatar Computing Research Institute) pioneered ML-based tweet classification for disaster response. Modern systems use LLMs for nuanced classification and entity extraction from multilingual crisis communication; (3) Humanitarian logistics optimization -- route optimization under damaged infrastructure, facility location for aid distribution centers, and multi-agent coordination for drone delivery of medical supplies. The IEEE 2025 multi-satellite data fusion framework demonstrates combining Sentinel-1 (SAR), PlanetScope (optical), and UAV imagery for comprehensive real-time disaster assessment. Ethical considerations: AI must not perpetuate biases, data privacy in crisis situations, and the risk of automated decision-making without human oversight in life-or-death contexts.

## Further Reading
- xView2 Challenge: xview2.org (DARPA building damage assessment benchmark)
- AIDR: aidr.qcri.org (Qatar Computing Research Institute)
- UN Global Pulse / Google Skai: AI for crisis response