📝 Related papers/other publications
[1] Han, S. Y., Lee, Y., Yoo, J., Kang, J. Y., Park, J., Myint, S. W., Cho, E., Gu, X., & Kim, J. S.* (2025). Spatial Disparities in Fire Shelter Accessibility: Capacity Challenges in the Palisades and Eaton Fires, arxiv. preprint: https://doi.org/10.48550/arXiv.2506.06803
[2] Lee, Y., Han, S. Y.*, Cho, E., Kim, J. S., & Myint, S. W.* (2026). Spatial Disparities in Fire Shelter Accessibility: Capacity Challenges in the Palisades and Eaton Fires. [Poster]
* corresponding author
On January 12, 2025, as the Palisades Fire and Eaton Fire burned across Los Angeles County, thousands of residents faced an urgent question: Where can I go to be safe?
Wildfires are not only destructive because of flames. They are also a major source of hazardous PM2.5, posing severe respiratory risks. Dense smoke reduces visibility to near zero, slows traffic dramatically, and often leads to internal road closures. In such conditions, evacuation becomes more than a logistical challenge—it becomes a matter of survival.
But what happens when shelters exist on paper, yet people cannot realistically reach them?
This question motivated our study on spatial disparities in emergency shelter accessibility under wildfire smoke exposure and traffic congestion.
The Overlooked Problem: Accessibility Under Real-World Constraints
Many previous studies on evacuation vulnerability have focused on infrastructure capacity or geographic distance. However, they often assume normal road conditions. In reality, wildfire emergencies bring:
Internal road closures
Severe congestion (we modeled smoke-induced speeds at 10 km/h)
Limited visibility and delayed response times
Shelter capacity shortages
Media reports during the 2025 fires indicated that official shelters were insufficient. Many evacuees had to rely on relatives, hotels, or even sleep in cars.
This gap between theoretical access and real-world mobility is where vulnerability becomes invisible.
Study Area and Data
Fig 1. Population and open shelter distribution on Jan 12 2025
Fig 2. Road network in study area
We focused on the fire zones on January 12, 2025 (data from CAL FIRE), integrating:
Population (Demand, Figure 1)
High-resolution residential density from LandScan Global (90m resolution)
Shelters (Supply, Figure 1)
Official shelters reported by Cal OES (with some capacity estimates based on floor area data)
Additional Candidate Sites
Facilities from the National Shelter System for optimization scenarios
Road Network (Figure 2)
OpenStreetMap data, incorporating:
Road closures within Evacuation Order and Warning Zones (EOWZ)
Congestion effects (maximum 10km/h)
Method: Measuring Accessibility Under Crisis
We applied the Enhanced Two-Step Floating Catchment Area (E2SFCA) model, a spatial accessibility framework that integrates:
Supply (shelter capacity)
Demand (population)
Distance decay (travel impedance)
We simulated two scenarios:
Road Closures + Traffic Congestion
Road Closures Only
What We Found: Accessibility Collapse
Figure 3. Shelter accessibility under road closures and
traffic congestion
Figure 4. Shelter accessibility under road closures
1️⃣ Critical “Dead Zones” in High-Elevation Areas
When both road closures and congestion were considered (Figure 3), accessibility collapsed near Evacuation Order and Warning Zones.
Residents in high-elevation terrain became effectively stranded due to:
Internal road blockages
Smoke-induced delays
These areas became what we call mobility dead zones.
2️⃣ Severe Capacity Mismatch
On January 12 (peak intensity), eight operational shelters could accommodate only 5,224 people.
Meanwhile, more than 80,000 residents within Evacuation Order and Warning Zones had no available capacity.
Even when congestion was removed from the simulation, accessibility improved slightly—but the capacity gap remained massive.
This reveals a structural mismatch between demand and emergency infrastructure.
Rethinking Shelter Planning: Two Optimization Strategies
Figure 5. Capacity-Driven Optimization
Figure 6. Proximity-Driven Optimization
To address this gap, we tested two shelter expansion strategies using facilities from the National Shelter System.
Scenario A: Capacity-Driven Optimization (Greedy Algorithm, Figure 5)
Prioritized shelters with the largest capacities.
Significantly increased average accessibility scores.
However, it created geographic blind spots.
Larger facilities were often farther away, increasing travel distances.
This approach is efficient—but not equitable.
Scenario B: Proximity-Driven Optimization (Incremental Addition, Figure 6)
Added shelters based on geographic closeness to residents.
Minimized individual travel distances.
Reduced stranded populations.
Improved equity in access.
This strategy ensured that even isolated residents had nearby safe options. From a social equity perspective, this approach performed best.
Policy Implications: Decentralization is Key
Our findings suggest that wildfire emergency planning must move beyond centralized, high-capacity shelters.
Instead, planners should prioritize:
A decentralized shelter network
Proximity-based safety guarantees
Real-time mobility modeling
Integration of smoke-induced traffic constraints
Without accounting for congestion and internal road closures, planners risk overestimating safety and underestimating vulnerability.