Fire Hazard Resource Allocation

Inspiration

Wildfires are increasingly frequent and destructive, yet early response often relies on fragmented, delayed, or manually processed information. We wanted to explore how real-time satellite data and autonomous agents could be combined to improve early situational awareness and support faster, more informed response planning.

What It Does

Fire Hazard Resource Allocation ingests near real-time satellite fire detections and other external data sources including wind and weather, aggregates them through an agent-based pipeline, and produces a continuously updated heatmap and risk assessment of potential fire incidents.

The system evaluates thermal anomaly patterns, estimates incident likelihood and escalation risk, and generates evidence-backed recommendations for monitoring and resource prioritization.

Rather than treating raw detections as confirmed incidents, the platform fuses multiple weak signals into structured, interpretable intelligence.

How We Built It

• Python for data ingestion, processing, and orchestration

• Agent-based architecture for modular data collection, analysis, and decision support

• NASA FIRMS API for satellite fire hotspot data

• FastAPI backend for real-time data distribution

• REST endpoints for visualization and downstream integration

• Lightweight scheduling for continuous data polling and updates

Each agent operates independently and communicates through standardized payloads, enabling flexible integration of new data sources.

Challenges We Ran Into

• Interpreting noisy satellite hotspot data without overreacting to false positives

• Validating inconsistent or incomplete external datasets

• Distinguishing true fire incidents from industrial heat sources and artifacts

• Designing confidence and severity scoring that remained stable under sparse data

• Coordinating multiple agents without introducing blocking or race conditions

Accomplishments That We’re Proud Of

• Built a fully functional end-to-end pipeline from satellite ingestion to live visualization

• Implemented severity and confidence normalization to prevent misleading outputs

• Designed an interpretable agent communication format rather than opaque predictions

• Successfully integrated continuous polling and real-time updates

• Delivered a working system under tight time constraints

What We Learned

• How satellite-based disaster monitoring systems actually work in practice

• The importance of treating raw sensor data as probabilistic evidence, not ground truth

• Designing agent networks that remain robust under noisy inputs

• Building reliable real-time data pipelines with minimal infrastructure

• Balancing automation with explicit uncertainty and human-in-the-loop principles

What’s Next for Fire Hazard Resource Allocation

• Integrate additional data sources like Twitter API since that provides real near time signals that are very valuable

• Improve incident clustering and escalation forecasting models

• Add historical pattern analysis to identify persistent false positives

• Incorporate official emergency alerts and response data

• Enhance decision support with resource availability and logistics constraints

Built With

• fastapi
• openai
• python