To know where our increasingly limited emergency resources will be needed next, we need to understand where future requests for service will originate. If we knew exactly where the next call would come from, we could proactively dispatch a resource there even before it is requested (watch the movie “Minority Report” for an idea of how that might work.) Unfortunately, the nature of emergency response is not nearly that easy, but that is not to say it is impossible to recognize useful patterns across both time and space. While the 2002 Spielberg movie was set 50 years into the future, it correctly predicted the use of several new technologies that have become reality in less than twenty years. And although we don’t use “precogs” in forecasting demand, the ability of data to show future patterns that effectively influence deployment is also now well established within some agencies.
No one can tell you who will be that very next person to dial 9-1-1; however, it is imperative for the effectiveness of deployment that we concede that people and events often follow certain predictable patterns. Let me explain how this works in just a few steps. First, consideration of the repeatable nature of the temporal distribution of calls has been used for years in making shift schedules. The following chart represents the daily call volume from a specific study, but without a scale along the vertical axis, it could easily be representative of almost any agency regarding their relative hourly volumes.
The daily behavioral routine of individuals perpetuates the collective pattern for the larger community. These daily patterns not only replicate over the years, but across various types of political jurisdictions according to a 2019 Scandinavian study on the “Use of pre-hospital emergency medical services in urban and rural municipalities over a 10 year period: an observational study based on routinely collected dispatch data.” The following graphs from that study represent the relative call volumes of rural, small and large towns, as well as medium and large cities over a decade showing the reproducibility of call volume forecasts by hour of the day.
If we segregate the total call data by weekday, we can capture variations by the hour-of-the-day within each day-of-the-week. The chart of call volumes by day over a twenty-week timeframe, shown below, displays the commonly repeated variation throughout each week. It is the reproducibility of these volumes that allows us to schedule adequate crews to cover these anticipated call volumes.
The next step is to adequately distribute those available resources spatially to address the variation over the geographic area by time which requires an even deeper understanding of the call patterns. The fact that we, as social creatures, often live or work in communities that share similar and predictable risk factors allows us to generalize assumptions of individual activities over larger community groups. Corporations have used targeted demographic profiles to understand local populations for many years. Community profiling has even been recognized by the World Health Organization as an essential skill for all health professionals to help understand the specific and detailed needs of focused populations. (See “Community Profiling. A Valuable Tool for Health Professionals” published in Australia during 2014.) Beyond predictable human variables that focus primarily on medical emergencies are the physical characteristics of our built environment that determine the repeatability of traumatic accidents. A 2009 publication by the Association for the Advancement of Automotive Medicine looked specifically at “Identifying Critical Road Geometry Parameters Affecting Crash Rate and Crash Type” to aide road safety engineers with the challenge of addressing safety issues related to the shape of motorways. The existence of identifiable causes explains the ability to properly forecast the vicinity of calls in addition to their timing.
The following animation demonstrates several spatial demand forecasts in quick succession that are normally separated in the real world by hours. Your existing historical CAD records contain the necessary information to build such dynamic views in real-time.
The demonstrated reliability of demand forecasts, both spatially and temporally, is well known to MARVLIS users and proven to provide the critical information necessary to make decisions in prepositioning resources to reduce the time of emergency responses and limit the distances travelled in emergency mode to enhance the protection of crews and citizens. Furthermore, the Demand Monitor has the capability of grading demand hotspot calculations specific to your service by comparing actual call locations – as they are being recorded – with the forecast probability surface to highlight both the accuracy and precision of our demand forecasts over time that is specific to your agency data and query parameters. The following screenshot shows comparisons of various forecast models.
The percentage of calls that correspond with each shaded area over the selected timeframe quantifies the query accuracy while the hotspot size denotes the relative precision. Accuracy could be increased easily by enlarging the hotspots, but this would be at the cost of precision. A well-balanced query should result in a relatively small-sized hotspot that properly captures a significant portion of actual calls.
Still, knowing when and where to anticipate calls is not enough in itself to determine resource deployment. Some number of outlier calls will likely occur outside of the forecast hotspots, so it is critical to also develop a strategy for managing the risk of covering demand versus geography as weighted factors in any deployment decision. Where we need to be next is well beyond the simple strategies we typically employ now and must fully leverage the depth of our data for deeper understanding and action.