Evaluating Cooling Center Coverage Using Persistent Homology of a Filtered Witness Complex

This paper proposes a novel approach using persistent homology on a filtered witness complex to identify geographic gaps in cooling center coverage across four US cities, demonstrating that combining these topological findings with traditional heat vulnerability indices provides a more holistic assessment of heat-related mortality risk.

The Gist

Imagine a city during a scorching summer heatwave. The air is thick, the sun is relentless, and for many people, especially the elderly or those without air conditioning, this isn't just uncomfortable—it's dangerous. To survive, people need to find "cooling centers" (like libraries or community centers) to escape the heat.

The problem is: Where are these centers, and are they actually helping the people who need them most?

This paper is like a detective story where the authors use two different sets of "glasses" to look at the city and find the blind spots where people are at risk.

The Two Sets of Glasses

1. The "Demographic Glasses" (The Heat Vulnerability Index)

Think of this as looking at a city map and coloring it based on who lives there.

• How it works: Researchers look at census data. They ask: "Are there lots of old people? Lots of young kids? Is it a poor neighborhood? Is there no tree shade?"
• The Result: They create a "Heat Vulnerability Index" (HVI). Areas with high scores are colored red because they have many people who are physically vulnerable to heat.
• The Limitation: This method assumes that if you have vulnerable people, you need a cooling center nearby. But it doesn't actually check where the cooling centers are. It's like saying, "This house has a sick person, so they need a doctor," without checking if the doctor's office is actually open or if there's a road to get there.

2. The "Topological Glasses" (Persistent Homology)

This is the paper's main innovation. Instead of looking at people, they look at the shape of the space between the cooling centers.

• The Analogy: Imagine you are trying to cover a floor with a few scattered rugs (the cooling centers). You want to know where the bare floor is.
  • The authors use a mathematical tool called Persistent Homology. Think of this as slowly inflating giant, invisible balloons around every cooling center.
  • As the balloons get bigger, they start to touch and merge.
  • The "Holes": If there is a gap in the coverage, the balloons won't touch for a long time. That gap is a "hole" in the coverage.
  • The "Death Time": The authors measure how big the balloons have to get before the hole finally closes. If a hole stays open until the balloons are huge, it means there is a massive gap in coverage. If the hole closes quickly, the coverage is decent.
• The Magic: This method doesn't care about demographics. It only cares about the geometry. It finds the "dead zones" where a person would have to walk a very long way to find any cooling center, regardless of who lives there.

The Big Discovery: They Don't Always Agree

The authors tested this in four cities: Boston, Austin, Portland, and Miami. Here is what they found:

• Sometimes they agree: In some neighborhoods, the "Demographic Glasses" say it's dangerous (lots of old people), and the "Topological Glasses" say it's dangerous (no cooling centers nearby). This is a double whammy—these are the most critical areas to fix immediately.
• Sometimes they disagree (The Surprise):
  • Case A: A neighborhood has a high "Vulnerability Score" (lots of old people), but the "Topological Glasses" show it's actually well-covered by cooling centers. Maybe the people are safe because the centers are right there.
  • Case B (The Hidden Danger): A neighborhood has a low "Vulnerability Score" (maybe it's a young, wealthy area with lots of trees), but the "Topological Glasses" show a massive gap in cooling center coverage.
    • Why does this matter? Even if the population is "low risk" on paper, if a heatwave hits and there are no places to go, everyone is at risk. The topological method found these "invisible" gaps that the demographic method missed.

The "Moon Island" Example

The authors found a funny example in Boston. There was a tiny island (Moon Island) that had a huge gap in cooling center coverage (a big "hole" in the math). The demographic method said, "Don't worry, no one lives there!" And they were right—it's a fire department training ground.

• Lesson: The math is smart, but it needs context. The "Topological Glasses" found a gap, but the "Demographic Glasses" told us why we might not need to panic about it.

The Takeaway

The paper argues that we shouldn't just use one method.

• The HVI tells us who is most likely to get sick.
• The Topological Method tells us where the physical gaps in safety are.

By using both, city planners can stop guessing. They can see exactly where to build a new cooling center: not just where the vulnerable people are, but also where the "holes" in the safety net are, ensuring that no one is left stranded in the heat.

In short: It's like checking both the passenger list (who needs help) and the lifeboat map (where the safety is) to make sure everyone gets on a boat before the ship sinks.

Technical Summary

Here is a detailed technical summary of the paper "Evaluating Cooling-Center Coverage Using Persistent Homology of a Filtered Witness Complex" by Erin O'Neil and Sarah Tymochko.

1. Problem Statement

Extreme heat is the leading cause of weather-related mortality in the United States, with heat waves projected to intensify and become more frequent due to climate change and the urban heat island effect. While access to air conditioning and public cooling centers (e.g., libraries, community centers) significantly reduces heat-related mortality, current distribution strategies are often suboptimal.

Existing methods for evaluating cooling center coverage have limitations:

• Catchment Area Methods: Rely on arbitrary fixed distance thresholds, failing to capture the continuous nature of coverage gaps.
• Heat Vulnerability Indices (HVI): Rely heavily on demographic data (e.g., age, income, tree canopy) but often treat geographic units (census blocks) as isolated islands, ignoring spatial connectivity and the proximity of cooling centers in neighboring jurisdictions.
• Data Limitations: Many optimization tools rely on proprietary software or require complex demographic data that may not be available at fine spatial resolutions.

The authors aim to develop a topological data analysis (TDA) tool to identify "holes" (gaps) in cooling center coverage without relying on arbitrary distance cutoffs or extensive demographic data, and to compare this topological approach against standard HVI methods.

2. Methodology

The paper proposes a novel application of Persistent Homology (PH) using a Filtered Witness Complex to model the spatial coverage of cooling centers.

A. Topological Approach (Persistent Homology)

1. Data Representation:

   • Landmarks ($L$): The centroids of census blocks (the smallest geographic unit) serve as the vertices of the simplicial complex.
   • Witnesses ($W$): The geographic locations of cooling centers serve as the "witnesses."
   • Filtration: A scale parameter $\alpha$ is increased. A simplex (edge, triangle, etc.) is formed between landmarks if they are all within distance $\alpha$ of the same cooling center. This creates a Witness Complex $W(L, W)_\alpha$.
   • Distance Metric: Geodesic distance (straight-line distance along the Earth's surface) is used to avoid projection distortions inherent in map projections.

2. Persistent Homology Analysis:

   • As $\alpha$ increases, the complex evolves. Connected components merge (0-dimensional homology), and holes (cycles) form and eventually fill in (1-dimensional homology).
   • Birth and Death Times: A feature is "born" when it first appears and "dies" when it merges with another component or fills in.
   • Death Simplices: The specific geometric structures (edges for 0D, triangles for 1D) that cause a feature to die are identified.
     • 0D Death Simplices (Edges): Represent the distance required to connect two previously disconnected regions. Large death times indicate large gaps in coverage between neighborhoods.
     • 1D Death Simplices (Triangles): Represent the "epicenter" of a hole in coverage (a region surrounded by cooling centers but lacking one in the middle). Large death times indicate significant, persistent gaps.

B. Comparative Approach (Heat Vulnerability Index - HVI)

To benchmark the TDA results, the authors constructed a standard HVI:

• Variables: Four standardized variables were used: afternoon temperature, percentage of area without tree canopy, percentage of population under 5, and percentage of population over 65.
• Calculation: An unweighted z-score composite was calculated for each census tract.
• Limitation: HVI relies on demographic data available only at the census tract level (coarser than the block level used in TDA) and treats areas as isolated based on internal demographics rather than external resource access.

3. Key Contributions

1. Novel Application of Witness Complexes: The authors adapt the witness complex construction (typically used for dimensionality reduction) to resource coverage problems, allowing for the integration of two distinct point sets (neighborhoods and resources) efficiently.
2. Parameter-Free Gap Detection: Unlike catchment area methods, PH identifies coverage gaps across a continuous range of scales, eliminating the need to arbitrarily select a "cut-off" distance.
3. Spatial Connectivity: The method naturally incorporates spatial relationships, including access to cooling centers in neighboring cities, which is critical for residents on city boundaries.
4. Holistic Vulnerability Assessment: By comparing TDA results with HVI, the paper demonstrates that demographic vulnerability (HVI) and resource accessibility (TDA) identify different, yet complementary, high-risk areas.

4. Results

The methodology was tested on four US cities: Boston, MA; Austin, TX; Portland, OR; and Miami, FL.

• Persistence Diagrams:

  • Boston exhibited the most significant 1-dimensional holes (large cycles), indicating substantial gaps in coverage surrounded by cooling centers.
  • Austin showed high 0-dimensional death times (large gaps between components), particularly in the southern and eastern regions.
  • Miami and Portland showed moderate gaps, with specific localized vulnerabilities.

• Death Simplex Visualization:

  • 0D Analysis: Identified specific edges connecting disconnected census blocks. In Austin, a large gap connected the southernmost block to the city center. In Boston, the southeastern area showed poor connectivity.
  • 1D Analysis: Identified "holes" (triangles) where cooling centers were absent. Boston had the largest 1D death times, suggesting the most critical unmet needs in specific neighborhoods.

• Comparison with HVI:

  • Low Correlation: There was generally low statistical correlation between the TDA death times and HVI scores (only Boston showed a weak significant correlation).
  • Complementary Insights:
    • Agreement: In some areas (e.g., Northeast Austin), high HVI overlapped with large topological gaps, confirming severe vulnerability.
    • Disagreement (TDA > HVI): Some areas with low HVI (low vulnerable population, high tree cover) had large topological gaps. This suggests that while the population is less vulnerable demographically, the lack of access to cooling centers remains a critical infrastructure issue.
    • Disagreement (HVI > TDA): Some areas with high HVI had good topological coverage, suggesting the risk is driven by demographics rather than lack of nearby centers.

5. Significance and Implications

• Infrastructure Planning: The TDA approach provides city planners with a tool to identify where cooling centers are missing based on spatial geometry, independent of demographic assumptions. This is crucial for optimizing resource placement.
• Multi-Perspective Risk Assessment: The study concludes that relying solely on HVI or solely on spatial coverage is insufficient. A holistic strategy requires combining demographic vulnerability (HVI) with topological coverage gaps (PH).
• Scalability and Adaptability: The method is adaptable to other resources (e.g., food banks, AEDs, EV charging stations) and does not require complex demographic data, making it useful for preliminary assessments in data-scarce environments.
• Limitations: The authors acknowledge limitations, including reliance on OpenStreetMap data quality, the use of geodesic distance (ignoring traffic/road networks), and the instability of "death simplices" (small data perturbations can shift the location of identified gaps).

In summary, this paper successfully demonstrates that Persistent Homology offers a robust, mathematically rigorous framework for visualizing and quantifying resource coverage gaps, providing a vital complement to traditional demographic vulnerability indices in the fight against heat-related mortality.