ENGINEERING REPORT: LESSONS FROM BIG SUR AND APPLICATIONS TO LOS ANGELES FIRE RESPONSE SYSTEMS

Abstract

The Big Sur fires of 2020 were a pivotal moment for many, including myself, a female engineer passionate about using technology to mitigate wildfire damage. Displaced to San Francisco under an apocalyptic red sky, I witnessed firsthand the limitations of existing emergency response systems. This experience inspired the development of the EMBER project and its corresponding codebase. This report provides an in-depth analysis of each component of the uploaded code, detailing its technical implementation and relevance to enhancing wildfire response strategies, particularly in the context of Los Angeles.

1. Background and Problem Statement

The Big Sur wildfires exposed significant deficiencies in current evacuation systems, including delayed risk assessments, lack of personalized evacuation guidance, and inadequate real-time route optimization. These issues, compounded by poor data integration and decision-making tools, led to unnecessary chaos and loss.

In Los Angeles, where wildfires are more frequent and urban density further complicates evacuation logistics, there is a pressing need for scalable, adaptive systems capable of handling real-time complexities. The EMBER system—leveraging deep learning, reinforcement learning, and retrieval-augmented generation (RAG)—was designed to address these challenges. This report evaluates each component of the uploaded code and its engineering contributions in detail.

2. Detailed Code Analysis

The uploaded codebase represents a comprehensive framework aimed at enhancing wildfire response capabilities. Below is an exhaustive breakdown of each module:

2.1 Risk Assessment (risk_assessment.py)

• Objective: Evaluate and prioritize evacuation zones based on real-time data.
• Technical Workflow:
  • Data Ingestion:
    • GIS Data: Geopandas is utilized to read and process GIS data outlining evacuation zones in formats such as Shapefile (.shp) and GeoJSON (.geojson).
    • Wildfire Perimeter Data: Fetches real-time wildfire perimeter data from government APIs or fire monitoring services, typically in GeoJSON or KML formats.
  • Data Processing:
    • Feature Extraction: Calculates features such as distance of each zone from the fire perimeter, population density, historical fire data, and current weather conditions.
    • Data Normalization: Normalizes features to ensure uniform scaling for model input.
  • Risk Prediction:
    • Model Selection: Employs a pre-trained deep learning model, possibly a convolutional neural network (CNN) or recurrent neural network (RNN), to predict risk levels for each zone.
    • Inference: Feeds processed features into the model to obtain risk scores.
  • Prioritization:
    • Priority Queue Implementation: Implements a priority queue using Python’s heapq library to rank zones based on risk scores, ensuring high-risk areas are addressed first.
• Engineering Challenges:
  • Data Latency: Ensuring timely data retrieval and processing to maintain real-time assessment.
  • Model Accuracy: Selecting and fine-tuning models to balance accuracy with computational efficiency.
  • Scalability: Designing the system to handle large datasets typical of urban environments like Los Angeles.
• Relevance to LA: Rapid and accurate risk assessments are crucial in a densely populated city like Los Angeles to facilitate timely evacuations and resource allocation.

2.2 Route Optimization (route_optimization.py)

• Objective: Generate the safest and most efficient evacuation routes under dynamic conditions.
• Technical Workflow:
  • Data Integration:
    • Real-Time Traffic Data: Integrates traffic data from APIs such as Google Maps or Waze to assess current road conditions.
    • Hazard Mapping: Overlays hazard data, including fire perimeters and blocked roads, onto the traffic data.
  • Route Calculation:
    • Algorithm Selection: Utilizes graph-based algorithms like A* or Dijkstra to compute initial routes.
    • Dynamic Cost Function: Develops a cost function that accounts for factors such as traffic congestion, proximity to hazards, and road capacity.
  • Reinforcement Learning (RL) Agent:
    • Simulation Environment: Creates a simulated environment to train the RL agent using frameworks like OpenAI Gym.
    • Policy Optimization: Applies algorithms such as Proximal Policy Optimization (PPO) to train the agent to select optimal routes based on the dynamic cost function.
  • Route Adjustment:
    • Real-Time Updates: Continuously updates routes in response to changing conditions, leveraging the trained RL agent for decision-making.
• Engineering Challenges:
  • Computational Load: Managing the computational demands of real-time route calculations and RL training.
  • Data Accuracy: Ensuring the reliability of real-time data inputs.
  • Scalability: Extending the system to handle large-scale evacuations involving thousands of individuals.
• Relevance to LA: Los Angeles' complex traffic patterns and infrastructure necessitate advanced route optimization to ensure efficient evacuations.