Macroscopic Traffic Flow Network Modeling For Wildfire Evacuation: A Game-Theoretic Junction Optimization Approach with Application to Lahaina Fire

1. Problem Statement

The paper addresses the critical failure of evacuation systems during wildfires, exemplified by the August 2023 Lahaina fire in Maui, Hawaii, which resulted in 101 deaths. The core problem is that standard traffic flow models, calibrated for routine conditions, fail to capture the unique dynamics of evacuation scenarios. Specifically:

• Bottleneck Dominance: In many wildland-urban interface (WUI) communities (like Lahaina), evacuation routes funnel into a single arterial road with limited exit capacity.
• Behavioral Shifts: Evacuation traffic exhibits reduced speeds and flow rates due to smoke, caution, and unfamiliarity with routes.
• Junction Inefficiency: Traditional junction models (e.g., First-In-First-Out or capacity-proportional rules) often leave exit capacity unused when one outgoing road saturates, even if alternative exits are free.
• Dynamic Topology: Roads may become blocked by fire or debris, requiring models that handle time-varying network structures.

The authors aim to develop a rigorous mathematical framework to optimize traffic distribution at junctions and determine infrastructure interventions (like lane reversals) to maximize evacuation throughput.

2. Methodology

The authors propose an end-to-end computational pipeline combining macroscopic traffic flow theory, game-theoretic optimization, and numerical simulation.

A. Mathematical Framework: LWR on Networks

The traffic flow is modeled as a system of scalar conservation laws (Lighthill-Whitham-Richards model) on a directed graph:
$$\frac{\partial \rho_i}{\partial t} + \frac{\partial f_i(\rho_i)}{\partial x} = 0$$
where $\rho$ is vehicle density and $f$ is the flux function. The network consists of $N$ roads and $M$ junctions.

B. Game-Theoretic Junction Conditions

Instead of fixed preference matrices (FIFO), the authors formulate junction dynamics as an optimization problem.

• Objective: Maximize total network flux (throughput) subject to entropy conditions.
• Mechanism: Drivers are modeled as agents who dynamically adjust route preferences to maximize flow when congestion occurs.
• Regimes: The solution identifies three regimes based on capacity ratios:
  1. Supply-Limited: Incoming demand is less than outgoing capacity; preferences are respected.
  2. Bottleneck (Incoming): Incoming capacity exceeds outgoing; flux is scaled proportionally to incoming capacities.
  3. Bottleneck (Outgoing): Outgoing capacity is the constraint; flux is scaled to match outgoing limits, effectively forcing drivers to abandon preset preferences to utilize all available exit lanes.
• Theorem 2.1 provides the unique solution for flux distribution at junctions, ensuring well-posedness.

C. Evacuation-Calibrated Flux Functions

The authors construct a piecewise linear-quadratic flux function $f(\rho)$ to reflect evacuation conditions:

• Linear Branch: Free-flow regime ($0 \le \rho < \sigma$).
• Quadratic Branch: Congested regime ($\sigma < \rho \le \rho_{max}$).
• Calibration: Parameters (free-flow speed $v_f$, max flow $f_c$) are adjusted based on empirical data from the Kincade and Glass fires (showing ~3.5 km/h speed reduction and ~2-5% flow reduction) and local Lahaina road data.

D. Numerical Scheme

• Solver: Godunov's method with an exact Riemann solver is used to evolve densities on road segments.
• Junction Coupling: Uses a Supply-Demand formulation (Lebacque/Daganzo) where boundary fluxes are determined by solving the entropy maximization problem at each time step.
• Optimization Algorithm: A Stochastic Block Coordinate Descent (SBCD) scheme is used to tune junction distribution parameters ($\alpha$) via a loss function. The loss function minimizes the time-integrated number of cars on the network, weighted by distance to the exit (prioritizing cars closer to the exit).

3. Key Contributions

1. Game-Theoretic Junction Resolution: A novel optimization-based approach that dynamically reallocates flux to under-utilized exits during congestion, outperforming static FIFO or proportional rules.
2. Phase Transition in Exit Capacity: Theoretical identification of a critical threshold ($n^*$) for exit lane capacity. Below this threshold, adding lanes linearly increases throughput; above it, the system becomes supply-limited, and further lane additions yield no benefit.
3. Evacuation-Specific Calibration: A flux function and initial condition setup specifically tuned for wildfire scenarios, incorporating reduced speeds and capacities observed in real-world fire evacuations.
4. End-to-End Lahaina Case Study: A full simulation of the August 8–9, 2023 Lahaina fire, incorporating dynamic road closures, time-varying source densities, and infrastructure interventions.

4. Key Results

A. Toy Network Analysis

• Phase Transition: In a simplified network, the authors proved that the critical number of exit lanes $n^*$ is given by:
  $$n^* = \frac{f_{c,2} + f_{c,4}}{f_{c,1}}$$
  where $f_c$ represents the flux capacities of incoming and outgoing roads.
• Finding: Adding lanes improves throughput linearly only until $n^*$ is reached. Beyond this point, the bottleneck shifts from the exit capacity to the incoming supply, rendering additional lanes useless for throughput.

B. Lahaina Fire Simulations

• Optimization Impact:
  • When the network is not saturated, optimizing driver preferences (routing) significantly improves throughput.
  • Once the network is gridlocked (bottleneck-dominated), route optimization yields diminishing returns; the binding constraint becomes physical road capacity, not driver behavior.
• Lane Reversal Strategy (Contraflow):
  • Simulating the reversal of southbound lanes on Honoapi'ilani Highway (creating 3 northbound exit lanes) captured nearly all achievable improvement.
  • Critical Finding: Reversing one lane (total 3 lanes) was sufficient to handle the maximum incoming flux. Reversing a second lane (total 4 lanes) provided negligible additional benefit for civilian evacuation.
  • Recommendation: The 4th lane should be reserved for emergency vehicles, as it does not significantly reduce civilian clearance time.
• Multiple Exits: Opening southbound exits in the afternoon was critical. The study suggests that future plans should prioritize multiple exit directions over optimizing a single route, as this prevents single-point failures.

5. Significance and Implications

• Mathematical Rigor: The paper provides a rigorous PDE-based foundation for evacuation planning, moving beyond heuristic simulations to analytically solvable models with proven existence and uniqueness of solutions.
• Policy Guidance: The identification of the "phase transition" offers a clear, computable metric for infrastructure planners. It suggests that in funnel-to-arterial networks (common in WUI communities like Topanga Canyon or Malibu), investing in exit capacity up to a specific threshold is far more effective than complex traffic management software alone.
• Actionable Recommendations: The study validates the efficacy of contraflow lane reversal for Lahaina, providing a data-driven basis for emergency management protocols. It highlights that infrastructure changes (adding/reversing lanes) dominate behavioral optimization once saturation occurs.
• Generalizability: The framework is applicable to other WUI communities with similar topologies and can be adapted for urban planning or event traffic management by adjusting the loss function weights.

In summary, the paper demonstrates that for bottleneck-dominated evacuation scenarios, infrastructure capacity is the primary lever for success, and mathematical optimization can precisely identify the point of diminishing returns for infrastructure investments.