A GIS-Based Risk Assessment of the Columbia-Richland (SC) Fire Department

By TRACY WHELEN

Historically, partnerships between fire departments and local universities and colleges have focused on educational opportunities for firefighters, “live-in” programs for fire science students, and other programs focused on recruitment and retention. In recent years, partnerships have been expanded to help local fire agencies better use and interpret the mountains of data they collect. For example, the Houston (TX) Fire Department highlighted its partnership with Rice University in the April 2020 issue of Fire Engineering, which provided the fire department with custom advanced data analytics to optimize fleet placement and maintenance to improve the longevity of its fire apparatus. The project benefited the students and university by providing meaningful, real-world problems the students could apply their newly learned skills to help solve.¹

• Advantages of Fire Department-University Partnerships
• Using Big Data Analytics to Enhance Fire Response Capability
• Big Data Driving Operational Performance and Fireground Safety
• Risk Assessment: It Is Personal

The Columbia-Richland (SC) Fire Department (CRFD) used a similar model to develop a novel geographic information system (GIS)-based risk assessment to support its efforts to pursue international accreditation status. Given the proximity to the University of South Carolina (UofSC), CRFD often has a few current students as volunteer firefighters as well as some other ad-hoc interactions over the years. In 2018, I started a master’s degree in geography at UofSC and joined CRFD as a volunteer firefighter, having previously been a call firefighter/emergency medical technician outside the region. The CRFD’s volunteer coordinator, Battalion Chief William Broscious, asked if I could use my GIS skills to develop a risk assessment for the fire department as it aligned with the organizational accreditation initiative; this risk assessment became my master’s thesis.² This article explains the techniques and findings of that risk assessment as an example of how partnerships with local universities and outside skills brought by volunteer firefighters can be used to benefit the fire service.

The Columbia-Richland Fire Department

The CRFD serves 388,000 people in Richland County, South Carolina, covering 683 square miles. The center of the county around the city of Columbia is urbanized, with population density decreasing toward the county boundaries through suburban and rural areas. Approximately 500 career firefighters and 120 volunteers staff 32 fire stations organized into five battalions.

The stations within the Columbia city limits are staffed entirely by career personnel, with the remaining stations staffed by a combination of career and volunteer firefighters. All but one station has a fire engine, with specialized units (i.e., ladder trucks, rescue trucks, tankers, and so on) scattered across the county. Ambulance service is provided by Richland County EMS, with CRFD responding to potentially time-critical emergency medical services (EMS) incidents as first responders.

History of Risk Assessments

The three core questions when assessing risk are the following:

1. What types of hazards are under consideration?
2. What factors increase risk for this type of hazard?
3. Who or what is threatened by this hazard?

Often in the fire service, fire risk is thought of as being based on physical risk factors such as the distance to the first-due apparatus or building construction components. Reports on fire fatalities often include demographic characteristics such as age or race, but this information doesn’t necessarily get included in formal risk assessments, even though demographic and socioeconomic risk factors are common topics in the public health and hazards fields. This risk assessment used a combination of physical, demographic, and socioeconomic risk factors.

Risk is inherently spatial because hazards, people, and items at risk change from location to location. There is a long tradition of using spatial data to assess fire risk, first through paper maps and, more recently, GIS and other computer-based spatial modeling tools. Fire insurance maps showing individual buildings labeled with building construction and occupancy information were common in the late 19th and early 20th centuries.³ These maps were expensive to make and rapidly fell out of date, especially in rapidly growing regions.

In the mid-20th century, computers began to allow for more complex mathematical and statistical modeling based on actuarial and road network data.⁴ These coverage assessments focused on minimizing first-due response times and defined risk primarily around distance to the nearest fire station, ignoring most other potential risk factors. The classic example is the Fire Department of New York’s work with the RAND Corporation in the 1970s.^5,6

Since the mid-20th century, the U.S. Fire Administration and public health organizations have also produced many reports on fire victim demographics and socioeconomic traits (age, race, gender, poverty, and so on).⁷ More recently, these reports have been turned into vulnerability assessments by combining them with census data and geocoded fire incidents to predict what neighborhoods might be at higher fire risk. Liverpool and Cardiff in the United Kingdom are two communities to have this kind of risk assessment.8,9

Using GIS to Identify Risk

GIS is a great tool to use for risk assessments because it enables robust analysis of many spatial data layers related to risk such as combining U.S. Census data (demographic and socioeconomic risk factors) with physical risk factors like building age and construction, railroad-grade crossings that can slow responding units, and fire station location. Risk factors alone or in combination can be compared with incident locations, helping to identify which risk factors are most important for a particular neighborhood.

Many commercial software packages allow spatial data to be visualized, and this kind of simple map is a useful starting point. However, only a handful of systems (e.g., ArcGIS, QGIS) allow for more complex spatial analysis, particularly spatial statistics. This is important because the core tenet of geography—“near” things are more closely related than “far” things—means that spatial data immediately breaks the assumption of independence that’s required for most nonspatial statistics. Spatial analysis in a robust risk assessment requires more than mapping individual risk variables side by side.

This risk assessment used four GIS and statistical techniques (hot spot analysis, factor analysis, correlation analysis, and cluster analysis) to do two things: first, assess the distribution of structure fires, emergency medical incidents, and carbon monoxide (CO) incidents, comprising, in total, about 10,000 incidents per year during the 2012-2018 analysis period; second, examine the relationship between risk factors and incident occurrence.

Since many risk variables were from the U.S. Census, analysis was done using census block groups as proxies for neighborhoods. Results highlighted which risk factors are most important for specific neighborhoods with high incident rates.

Incident Hot Spots

Most experienced firefighters can tell you what neighborhoods in their territory are the busiest for incidents. However, do you know which neighborhoods in your region are busier than average for their population, rather than which ones are busy because they have a large, dense population?

Hot spot analysis can tease these questions apart. This technique compares each neighborhood with the neighborhoods around it and identifies which clusters of neighborhoods are hot spots—i.e., have higher incident rates than would be expected for that size population. Likewise, it identifies cold spots—the corresponding areas with lower-than-expected incident rates—as well as standalone high and low incident rate outliers.

Hot spot analysis allows us to gain two insights based on the data. First, we can confirm or deny anecdotal statements from line firefighters about which areas are the busiest. Second, we can identify unrecognized areas with higher-than-expected incident rates. Identification of key areas with high incident rates is important because these areas use fire service resources at an above-average rate and could benefit the most from prevention activities. Hot spot analysis works best on relatively common incidents (or all incidents combined), where every neighborhood has one or more instances of the incident type. It does not work well with rare incident types like fire fatalities, where many neighborhoods haven’t suffered this type of incident within the analysis time frame.

Think about the types of calls to which your station responds most frequently. Maybe it’s medical calls; hot spot analysis could help support the need for and best placement of a community paramedicine program to intervene with high-volume service users, repeatedly calling for non-emergent medical issues. If you spend all day responding to malfunctioning fire alarms, this type of data could pinpoint commercial districts needing alarm maintenance code enforcement or residential neighborhoods to target in a public education campaign about testing and replacing batteries in smoke detectors, or it could show that your response area is busy because of a large population rather than an above-average incident rate.

In the case of Columbia, hot spot analysis worked great when looking at structure fires and medical calls, identifying hot spots that aligned with department anecdotal evidence and were visible in thematic maps of incident rates. For structure fires, one big hot spot was identified on the northern edge of the city (Figures 1, 2). EMS had a similarly located large hot spot, with another smaller hot spot in the southeastern part of the county.

Grouping Risk Factors

Sometimes, risk factors come in groups, where two or more risk factors are similarly distributed throughout a community. For example, communities with lower education often have lower incomes; a statistical technique known as “factor analysis” can identify these groups. Each group or principle factor is built from risk factors that share similar patterns—“positive loadings”—whereas risk factors with opposing patterns are known as “negative loadings.” This technique helps identify high-risk communities more effectively than just looking at single risk factors. Marketing companies use similar methods when they segment customers into personas (e.g., Esri Tapestry Segmentation, Experian Mosaic).

The CRFD risk assessment identified nine principle factors (Table 1) derived from 31 original risk variables. We can see that, in some communities, risk is primarily related to demographic and socioeconomic risk factors (Factors 1, 5, 8), whereas in other communities, physical risk factors (Factor 3) or a combination of risk factors are more influential. One neighborhood’s risk can be influenced by multiple principle factors, just like it could be influenced by multiple individual risk variables. This shows us that targeted prevention activities may need a many-pronged approach.

We can look at two factors, Factor 1 (Systematic Hard Times) and Factor 3 (Rural), as case studies into tailoring fire service activities to best meet the needs of specific communities. With Factor 1, we see that fire safety education could be targeted to two key audiences: kids (and their caregivers) who might have limited supervision and adults with disabilities and thus potentially mobility issues. Likewise, in an area with lower high school graduation rates, it is important to ensure fire safety messaging uses easily understood images and language so that all members of the community can grasp the important takeaways. In contrast, Factor 3 shows mostly physical factors that could be addressed by a combination of fire safety education and department preplanning. Education activities could focus on safety for mobile home residents, whereas department preplanning could focus on best practices for response in an area with no hydrants and longer response times.

Correlation Between Risk Factors and Incidents

Sometimes, the broad, overall patterns between risk factors and incidents can be a useful starting place to validate one’s assumptions before diving deep into the data. Correlation analysis describes the relationship between two variables, such as a risk factor and incident type, in terms of the relations being positive or negative and the strength of the relationship. A positive correlation means that, as one variable increases, so does the other. For example, for an increase in the risk factor corresponding to an increase in incidents, correlation strength refers to how tightly related the two variables are. This type of analysis can be helpful to identify which risk factors are related to specific incident types.

One downside is that, as a nonspatial technique, it looks at overall relationships for your entire area and cannot distinguish between neighborhoods. This means that a risk factor that is only important in some, but not all, neighborhoods will not be properly understood.

In the CRFD risk assessment, there were many risk factors that had statistically significant correlation with one or more incident types, but the strength of these correlations was weak to moderate. This means that risk is complex, and no single risk variable (or group of risk variables—i.e., principle factor) explains all risk. Although this may be obvious to firefighters because no two incidents are identical, this isn’t always obvious to elected officials and community members. Correlation data showing many risk factors are equally important and can be useful when advocating for increases in fire department resources or as a counter statement if being pressed by local officials to name the one or two most important risk factors for an entire city or county.