Sentinel

A predictive defense grid that detects pre-ignition wildfire heat 48 hours before smoke appears.

Comment 1

Sentinel

Predictive Wildfire Defense Grid

Python FastAPI Redis Status

Problem

In 2021, the town of Lytton, BC was destroyed by wildfire in under 20 minutes. Current detection methods rely heavily on satellite imagery, which has significant drawbacks:

  • High Latency: Satellites only pass over BC every 6-12 hours.
  • Line of Sight: Cameras require visibility and are often blocked by smoke rising.
  • Reactive: Detection mostly occurs after the fire has grown significantly.

Solution

Sentinel is a network of sensors placed directly on the forest floor. Instead of waiting for satellites to spot smoke, Sentinel detects sudden temperature spikes in the soil. This could provide a 48-hour warning before a fire actually ignites, allowing crews to stop the threat early.

Architecture: Systems for Scale

The architecture prioritizes write-throughput and fault tolerance. The system handles 100,000+ sensor writes per second via a non-blocking pipeline.

1. Ingestion Layer (FastAPI)

  • Function: High-concurrency gateway accepting raw JSON packets.
  • Performance: Uses asynchronous non-blocking I/O.
  • Configuration: Logging disabled in production to maximize throughput.

2. Buffer Layer (Redis Streams)

  • Function: Decouples ingestion from the visualization/analysis layer.
  • Benefit: Prevents backpressure on the ingestion server during the visualization rendering. Data is buffered in RAM for super fast access.

3. Command Layer (Rich CLI)

  • Function: Low-latency Mission Control interface.
  • Design: A Terminal User Interface (TUI) was selected over a web dashboard to ensure functionality in low-bandwidth environments and on low-power hardware typical of base camps.

Theme Integration

The system integrates historical wildfire data to weight risk probabilities.

  • Historical Tracking: GPS coordinates for the Lytton (2021) and Kelowna (2003) fires.
  • Weighted Logic: Sensitivity multiplier is applied to sensors within these historical zones.
  • Outcome: Thermal anomalies in scar zones escalate to "Pre-Ignition Events" faster than random sectors, utilizing past data points to enhance future protection.

Quick Start

Three separate terminal windows are required to simulate the distributed system components. It is recommended that you create a virtual environment first.

Prerequisites

pip install fastapi uvicorn redis rich requests pydantic
#make sure redis is running locally

How to run

#run each command on a different window in linux
uvicorn gateway:application
python setsimulator.py
python monitor2.py

Built With

Share this project:

Updates