Screen Shot 2016-06-13 at 1.31.11 PM.png
Screen Shot 2016-06-13 at 2.11.02 PM.png
Screen Shot 2016-06-13 at 1.31.11 PM.png

The Forces Behind Fire

Assessing Fire Risk in Detroit Property


The Forces Behind Fire

Assessing Fire Risk in Detroit Property


A model for assessing fire risk

Detroit suffered nearly 3,000 structure damaging fires in 2015. The human, physical, and financial cost of these fires and their destabilizing effects on Detroit's neighborhoods was tremendous:

  • 1,373 structure damaging fires in occupied homes

  • 1,238 catastrophic or major fires that will likely require demolition of the remaining structure and perhaps as much as $17 million in additional demolition funds

With stakes and costs this high, the benefits of improved fire prevention techniques could be significant. But fire in Detroit is chaotic and unpredictable — arson abounds, and with more than 50,000 vacant structures across the city, there is fuel to burn. The problem is complex.

In July 2015, LOVELAND Technologies started working with Joshua Pelletier and Demarius Chrite of the software design firm Pretty Robots LLC in an effort to identify what characteristics put residential structures in Detroit at greater risk for fires, and which Detroit properties are at highest risk. Pretty Robots used data collected in 2014's citywide property survey, Motor City Mapping, along with governmental data from the City of Detroit and Wayne County to inform their research. 

The goals were two-fold:

  • With the data available, attempt to identify what characteristics put Detroit homes at greater risk for fire.

  • Identify which properties in Detroit bear those characteristics and to what degree.

What follows is a collection of maps, data, and analysis showing where LOVELAND and Pretty Robots found at-risk properties, how the model identifying those properties was constructed, and how this might be used to inform programs and policies that can prevent fires and the massively destabilizing effects they have on Detroit.



detroit fire risk

detroit fire risk

What Does “HIGH Risk” Mean?

What Risk Looks Like: Case Study at 14615 Eastwood

This property showcases almost every feature of a high risk property available in the data used: it is open to trespassing, vacant, and publicly owned. The interplay of all these factors (each factor is listed below the left image) produced a risk assessment score in the Pretty Robots model of 16.2%.

A fire risk assessment of 16.2%, as seen in the property below at left, does not mean there is a 16.2% the property will catch on fire — the model is not predictive, it is an assessment of risk. A 16.2% risk assessment means that the collection of features seen in this property put it at greater risk than a property with none of those features.

A catastrophic fire such as the one suffered at 14615 Eastwood on September 28th, 2015 (and surveyed by LOVELAND staff later) can exhaust the fuel for subsequent fires, but it’s of little benefit: In addition to the risk and cost incurred by the Detroit Fire Department in putting out the fire, the city will still need to spend thousands in demolition funds to remove the foundation of the collapsed structure.

Detroit Fire Risk Map

Zoom into the map and click on any colored parcel to see property-specific information and photography

The Site Control map above shows all of the Detroit houses assessed by Pretty Robots' model as high risk. High risk is defined by Pretty Robots as those properties with a risk level greater than 5% -- one standard deviation above the average residential property risk score and therefore considered a statistically significant risk level. Of the 168,773 residential properties analyzed for the model, 24,819 (~15%) had risk scores greater than 5%. (Note that this includes both occupied and vacant residential properties.)

Number of high risk properties by tier, citywide

Number of high risk properties by tier, citywide

There are several areas of the city where at-risk properties are clustered. Click the Focus Areas button in the map to see a dropdown list of high risk cluster areas. Click to zoom to, and explore, those areas.

The areas with the densest clustering of high risk properties are outlined and shaded below. These parts of Detroit are, historically, areas of high fire activity, thick with vacant, city-owned property, and frequent arson. Below is a slideshow with greater detail for each of the four areas outlined below.

High Risk Clustering

Areas of the city where at-risk properties are clustered.

Detail of High Risk areas

using the model

using the model

While LOVELAND and Pretty Robots see ample opportunity to deepen and refine the model, there are insights to be gleaned in its current form. For example, there may be no greater benefit to the people and property of Detroit when it comes to fire damage mitigation than to focus demolition efforts on properties that are at high risk and located directly adjacent to an occupied home. 

While there is evidence of this approach in city demolition patterns, a focused policy to remove fuel for fires next to occupied homes could have stabilizing effects for many Detroit properties. 

In advance of Angel’s Night in October 2015, LOVELAND published a blog post looking at high-risk areas of the city for arson in advance of the traditional uptick in fires around Halloween. Part of the analysis included the identification of occupied homes that are adjacent to vacant, publicly owned structures. At the time, LOVELAND identified approximately 11,700 occupied homes that have 8,800 vacant, publicly owned structures next to them.

The Site Control map below shows:

  • 9,500 Hardest Hit Fund demolitions carried out over the last two years (light blue)

  • 11,700 occupied homes (green) that are adjacent to 8,800 city-owned vacant structures (red)

  • Properties in black are those which suffered a suspicious fire in 2015 -- some of which occurred in vacant structures next door to occupied homes

With the fire risk model developed by Pretty Robots, city-owned vacant properties next door to occupied homes can be evaluated for which are, per the model, at highest risk for fire. This would amount to a demolition triage list of high-stakes properties where a fire can mean the loss of an occupied home. This is the kind of application of the fire risk model that LOVELAND and Pretty Robots envision.

Further citywide surveys could produce vastly higher resolution data for this model, which will allow the model to output more accurate and more insightful results. Further, the model exists now as a protocol — a series of steps and equations that produce fire risk assessments based on the data input. This can be turned into an application that is a component of the Motor City Mapping interface and that self-updates as new information comes into the system, allowing the fire department and others to see risk in real-time.

Screen Shot 2016-06-13 at 2.11.02 PM.png

Developing the Model

Developing the Model

Background on the Model

Pretty Robots utilized the records of 168,773 residential properties in Detroit collected during the 2014 Motor City Mapping campaign to test and train their model. The model only addresses residential properties, and only uses records where data was available for all fields. The independent variables that entered the model included: boarding, dumping, occupancy, public ownership, 2014 Tax Assessment, and 2014 Tax Foreclosure data. The end product is a percentage score that sets a fire risk level for all 168,773 property records used, based on the model created. The data used in the model was taken from the City of Detroit Assessor, the Wayne County Treasurer, as well as the Motor City Mapping campaign.

The Motor City Mapping survey collected categorical data about property use as well as condition, whereas the Assessor and Treasurer’s data is focused on ownership, value and tax status. 

Motor City Mapping surveyors reported on whether or not a parcel had fire exposure. Between December 5, 2013 and July 21, 2015 they had collected information on fire exposure for over 215,283 unique properties across Detroit. This simple piece of information, indicating whether or not a property had evidence of fire exposure or not, is invaluable to calculating odds, which is a useful notion for the likelihood of an event. The odds of an event is based on a ratio; it is the probability of an event occurring divided by the probability of the event not occurring.  

With odds as a baseline statistical concept and fire exposures as dependent variable, Pretty Robots was able to build contingency tables and logistic regression models in order to search for associations with other attributes within these datasets and measure the likelihood of an event.

Opportunities for Improvement

The fire risk model developed by Pretty Robots and LOVELAND can be improved through better and more frequent resurveys of Detroit property. Data from Motor City Mapping's 2013 / 2014 property survey is now more than two years old. Given the rapid rate of change in Detroit property this model would benefit greatly from updated condition information, as well as a second, citywide, snapshot of property conditions which would bring change over time into the equation. More nuanced questions of Detroit property, such as building materials, would be valuable in further development of the risk model.

The Nature of Fire in Detroit

The number of properties at high risk for fire is not static in Detroit. Fires that occur in occupied, good condition homes can lead to abandonment and leave behind a structure that is at greater risk for subsequent fires. 

This was the case at 15045 Faircrest — a property that was not included in Pretty Robots’ list of high risk properties because, at the time, the property was an occupied home with no risk factors. However a suspicious fire after a break-in led to a property that is now at greater risk for subsequent fires, and is adjacent to another occupied home that is now at greater risk, too.

Despite the real progress in Detroit over the last several years since bankruptcy, there are headwinds that remain strong. The more we can do to reduce their strength, the more successful our recovery and development efforts will be, and the more equitable their distribution. Fire sets back lives, blocks, and whole neighborhoods. Together, LOVELAND and Pretty Robots believe there is much we can do to fight fires before they start.