Predictive Firefighting with Premergency
for Seattle Fire Department
We will show you a risk model in risk-oriented strategic planning (Predictive Firefighting) data of the rescue services of Seattle Fire Department / USA.
As part of the city of Seattle’s open data initiative, multi-year data from fire and rescue services are made available. We analyzed this data using Premergency for Predictive Firefighting. The results can be found below. They show how to use limited information to draw conclusions about the type and time of operations. The security level of a city or county in quality and economic efficiency can be compared to the expected marginal costs. The goal is a key figure-oriented organization.
In this comparison, in contrast to the November 2015 article on methods of strategic planning and Predictive Firefighting, there is only one input parameter, the deployment data. This results from the lack of preventive fire protection or building management data. US cities are pioneers in open data and therefore more and more data will become available enriching digital models. The previously mentioned article explains how to deal with multivariate risks. A comparison with the methods from A. Ridder’s dissertation (see below) is also worthwhile.
Motivation for Predictive Firefighting and a risk model in active defense as such, can be:
- Quantification of the predominant security level
(regarding the probability of incidents)
- Quality improvement and key figure orientation
- Expected utilization of sites and personnel or their resources
- Site optimization
- Simulation of short-term performance peaks(e.g. mass events)
- Simulation of regionally different scenarios
- Observance of parallel operations
For example, locally, a lower deployment volume leads to a lower probability of deployment. After weighing the investment volume (profitability), graduated security concepts can be created for these areas without lowering the security level for citizens. The possible hazard scenarios in such areas are often diverse. The aim here is a locally customized organization for rescue services and fire protection.
The Risk Model
In the present (univariate) case, the probability of use or incident risk can be well calculated via the geo-based binning of deployment data. In this context, binning refers to the assignment of deployment sites to geographical cells in order to calculate a conditional probability per field and a spatial probability distribution.
For this purpose, we imported the application data into our system. After that, we applied various filters to calculate the geo-based, conditional probabilities of occurrence for the local deployment of the rescue service. The following image shows the deployment probability for rescue services (keywords Aid Response, Aid Response Freeway) for a Monday in winter, here based on the total volume.
Along with the differentiation of spatial information, the temporal dimension can be confined further. This results in conditional probabilities, e.g. for the case of deployment on weekdays between 10 p.m. and 6 a.m., which indicate the focal points for night-time deployments:
Dabei ergibt sich die örtlich zu erwartende Einsatzwahrscheinlichkeit als Summe über die benachbarten Sechsecke bzw. Raster (Bins). Somit lassen sich Standorte von Rettungsdienst und Feuerwehr (Rettungswachen, Feuerwachen) anhand des historischen Einsatzaufkommens (und weiterer Risikofaktoren, vgl. unseren Fachartikel) optimieren. Zwecks Auslastung eines Standortes kann man die einzelnen Wachbereiche nicht nur räumlich sondern auch zeitlich betrachten. In diesem Beispiel wurde der Montag und die Monate Dezember – Februar ausgewählt.
Temporal view of the deployment probability
Regarding the temporal and spatial analysis, it is important to have a sufficient number of cases within the analysis range in order to guarantee a selectivity of hypotheses or statistical significance within the Bayesian approach. In the example of Seattle, data from approximately 240,000 deployments are available for analysis. This guarantees sufficient accuracy on a daily and monthly basis. The advantage of the Bayesian approach is the evaluation of hypotheses with limited case numbers or limited information in contrast to the classical frequency analysis.
A comparison between weekdays and weekends for the summer months shows clear hotspots for weekdays. These hotspots can be included in the assessment of the shift system and individual areas can be reinforced in certain periods.
A comparison of the times of day for working days shows the following result: During nights (10 p.m. – 6 p.m.) there are significantly more deployments in the area of the city center than in the mornings (6 a.m. – 2 p.m.). The expected number of operations around rescue service locations is thus calculated as the sum or integral of the individual areas. This means that the number of staff at each location can be varied per shift reinforcing hotspots or shortages:
Comparing summer and winter months
Another comparison: leisure activities in summer usually differ from those in winter. This is also reflected in the number of deployments and their probabilities. The following map shows a decrease in deployment probability in summer (green) and clear hotspots or significantly higher deployment probabilities in summer – these hotspots are located at the city’s leisure centers and on the water or beach. By the way, the total deployment volume in the city hardly differs between weekdays.
If you calculate the ratio of deployment probabilities summer to winter for working days, other hotspots appear in the city center. In the section shown, the number of deployments is only slightly increased (<50%) in summer. Empty fields mean no change.
Number of deployments for Seattle Rescue Services during weekdays in comparison between summer and winter months. Veränderung = variation / Stadtgrenze = city limit.
All in all, it becomes clear how important it is for authorities to keep a permanent record of operational key figures. Already, by using models with few input parameters (coordinates, address, type of deployment, time) organizational decisions can be structured. These risk models can also be used in long-term location planning to separate service areas from each other and to predict and control deployment volumes through the size of deployment areas.
After all, non-optimal locations can significantly worsen the utilization of resources. As a result, construction or rental costs may be saved, but the personnel budget is significantly increased. One possible consequence is increasing employee dissatisfaction: some are bored, others are overworked. As a preventive measure, rescue service providers are increasingly renting rescue stations. Thus, you can react if a relocation of hotspots is detected. Often, a more suitable site can be rented.
Risk-oriented or risk-related approaches also offer advantages for fire departments. These models can make it possible to calculate the variation of unit strengths in day-to-day business considering structural changes for a forecast of the deployment volume. Locally increased deployment volume due to mass events or similar circumstances, can be simulated and planned with corresponding additional capacities. Preventive measures can be combined with fire protection.
This way hypotheses on fire frequency can be investigated and incorporated into inspections in order to reduce the frequency of damage. In his dissertation, Adrian Ridder was able to show that there is a correlation between building height, building age and fire frequency (A. Ridder: „Risikologische Betrachtungen zur strategischen Planung von Feuerwehren (Wuppertaler Berichte zur Sicherheitstechnik und zum Brand- und Explosionsschutz Band 11)“, VdS-Verlag, Köln 2015).
The data shown is based on data supplied by the city of Seattle. Neither we nor the City of Seattle guarantee the completeness and correctness of this data. The rights to the map material belong to HERE Inc.