Security threats, humanitarian needs
Operations research, geospatial analysis boost public safety and humanitarian relief efforts.
By Colleen McCue
Recent advances in knowledge management and processing capabilities, coupled with operations research, have made it possible to reflect the subtle interplay of human activity and location in anticipating and influencing behavior. These advances have substantial uses in operational public safety and humanitarian relief.
Similar to marketing analytics, operational security analytics is based on the premise that human behavior – even extremely violent, predatory, or otherwise aberrant behavior – tends to be repetitive, predictable and location-dependent. The ability to model behavior is increasingly important as public safety agencies are moving from “doing more with less” to “doing almost everything with nothing” given residual budget shortfalls associated with the recent recession, as well as significant mission creep arising from new counterterrorism requirements .
Likewise, the ability to do more with less is a constant challenge facing humanitarian groups and others responding to natural disasters and other challenges, including the current Ebola crisis in West Africa where the ability to effectively allocate scarce healthcare resources can truly mean the difference between life and death. By leveraging analytic concepts, methods and techniques developed for use in the commercial setting, public safety and humanitarian professionals can optimize scarce or otherwise limited resources, while concomitantly improving outcomes.
As commercial retailers know, location matters. The physical and human attributes of space enable, inhibit or otherwise constrain activity, and human decisions are influenced either consciously or subconsciously by these spatial factors. Geospatial preference models (GPM) are a relatively new method to combine statistics and geospatial mapping to characterize and predict human behavior or place preferences.
Given the recent advances mentioned above, it is now possible to compile and more effectively exploit a tremendous breadth and depth of geospatial factor data in support of truly meaningful and detailed analysis of the specific attributes associated with a particular location. For example, Geospatial Big Data™ (GBD), an initiative of DigitalGlobe, facilitates the development of multivariate models that better reflect the complexity of location and its relationship to human activity by leveraging the close to 4 million square miles of satellite imagery collected daily by its constellation. This effectively translates to more than 50 terabytes of new data each day, which are combined with other sources of geospatial content to enable more comprehensive, information-based approaches to geospatial predictive analysis.
Feature space preference (FSP)  is a GPM that is based on a supervised learning algorithm that statistically characterizes known events; identifying and statistically characterizing geospatial similarities between the known or training events, and identifying and statistically characterizing geospatial differences between the training events and “nonevent” locations. Leveraging an ensemble approach, the analyst can use each specific factor to construct a submodel, with the final model representing a weighted sum of these submodels, where the submodel factor weighting is directly associated with the influence that each specific factor has on the final output. The results are expressed as an assessment score that represents an estimate of the relative likelihood that a future incident will occur at each location within the identified area of interest. The analyst can then project the assessment score back into the area of interest as a choropleth (color-shaded representation of intensity of numerical counts) map, which enables the visualization of these calculated likelihoods in a format that is relatively easy to interpret and use.
Unlike other geospatial techniques that focus primarily on the identification of adjacent spaces, the results of FSP can be extended to novel space in support of identifying noncontiguous locations that are statistically similar to the known events. This enables the identification not only of future locations, but also areas at increased likelihood for displacement, as well as “hidden” locations associated with incidents or activity that have not been reported or detected previously. These hidden locations can be particularly important in cases where reporting is difficult either due to communications challenges, fear of reprisal, or even apathy associated with locations or situations where crime is prevalent and law enforcement response is limited. Similarly, it is not unusual for crime to “jump” to a new location in response to heavy deployment or other enforcement efforts. The ability to effectively model and identify these noncontiguous locations can move the public safety community from chasing crime to being able to anticipate and proactively influence it.
Finally, geospatial tools and capabilities facilitate our ability to effectively convey complex statistical relationships in a relatively intuitive, operationally relevant and actionable analytic environment. At a minimum, the “go here now and expect this” attributes of geospatial output help the nontechnical end user to interpret the results and use them to support action. Ideally, the end user will be able to incorporate their tacit knowledge and domain expertise to extend from the results in support of novel insight and meaningful solutions to some of our hardest problems, including operational public safety and humanitarian efforts.
The infamous bank robber Willie Sutton, when asked why he robbed banks, reportedly answered: “Because that’s where the money is.” Criminals tend to be very interested in two attributes of the environment: access to victims and/or targets and an environment where they believe they can successfully perpetrate their act, which includes a subtle balance between attractors or other enabling features and avoidance of impediments. Like other behavior, these factors tend to be subtle, nuanced and comingled. They also tend to be both offense- and offender-specific. Therefore, there is no “one-size-fits-all” or easy button solution, even for a single offender involved in a series. Moreover, like the “wicked” problems described by Rittel and Weber , intervention changes the situation, sometimes profoundly. Therefore, one occupational hazard in operational security analytics is the requirement for an increased frequency of model updates as the criminal adapts to law enforcement intervention.
Personnel resources frequently represent the single biggest expenditure in most public safety agency budgets. Identifying ways to optimize this very expensive and often limited resource without compromising public safety outcomes presents a significant challenge to most police managers and command staff. Not only is a “cop on every corner” terribly inefficient, but it also has the tendency to be perceived as intrusive. Therefore, the ideal is to preposition police officers when and where they are likely to be needed as a visible deterrent to crime. Second best would be to preposition these resources in support of timely response and apprehension. Operationalizing these goals, concepts like “risk-based” deployment  are essentially “just-in-time” policing. Similarly, informed segmentation and characterization of the public safety problem set supports anticipation of specific public safety requirements and outcomes, and associated optimization and targeting of resources.
Example: Serial Shooter
A team including the author conducted a practical example of this approach during the northern Virginia military shooting series . In the fall of 2010 shots rang out in northern Virginia in a series of shootings apparently targeting facilities of interest to the military. The first four incidents occurred in an area of approximately 750 square miles with literally thousands of locations and facilities with a direct link to or of interest to the military. Similar to other O.R. problems, some hard decisions had to made regarding when and where to allocate a fixed set of resources. In addition, one early concern associated with this series was the potential for escalation. Would the suspect move from shooting at things to shooting people? Given the season, there were a number of high-profile events including Veteran’s Day, the U.S. Marine Corps anniversary celebrations and the Marine Corps Marathon.
Three days after the team completed and briefed the initial FSP assessment, a fifth shooting was reported at a U.S. Coast Guard recruiting station. Unlike the first four incidents, there was no direct link to the U.S. Marine Corps. However, the location fell directly into the high likelihood area identified by the model, which suggested that it fit the offender’s place preference.
While accurately “predicting” future events may represent solid validation of advanced analytics modeling techniques, there is very limited satisfaction in being “right” about most operational security analytics work given that new incidents frequently are associated with significant consequences or some other poor outcome. Unfortunately, similar to many other O.R. challenges, novel insight and prescient models do not necessarily translate directly to improved outcomes. In other words, the ability to effectively characterize and model behavior in support of informed anticipation does not always translate directly into action. In this particular case, however, operational validation of the model with the new incident supported the use of FSP as an effective method for characterizing the offender’s place preferences.
Like other rule induction techniques, the factors identified by FSP are correlational, not causal, although they can provide some high-level insight. In this particular example, factors associated with easy or rapid egress (e.g., proximity to highway on-ramps), emerged early and were not surprising. Other factors that generated greater concern, however, included proximity to motels and cemeteries. While the relative privacy created by a cemetery at night may have provided attractive concealment, this particular factor caused additional concern given the proximity to Veteran’s Day and the large number of memorial events scheduled for these locations. Similarly, many of the Marine Corps celebrations were scheduled to take place at the various motels along the I-95 corridor, which fell in the area of interest. So although what we had was mostly just correlational, law enforcement agencies focused special attention on these particular locations given their sensitivity and relevance.
After the fifth incident, there were no additional reports of shots fired, and the series appeared to stop. While it is unclear whether the break in the series was associated with the heavy deployment of resources to the locations identified as being consistent with the offender’s place preference or some other unknown factor, several months later a suspect was apprehended in the Arlington National Cemetery – a location identified previously as being at high likelihood for a future incident. This apprehension and subsequent prosecution of the suspect provided additional validation of the FSP model, and resulted in a permanent end to the series.
‘H2.0’ Helps Quench Humanitarian Thirst for Resources
The humanitarian community also has adopted the concept of “doing more with less” as a way of life. Again, the requirement to efficiently allocate and optimize scarce or otherwise limited resources is a common O.R. challenge. Adopting information-based approaches – Humanitarian 2.0 (H2.0) – provides novel approaches to effectively model and account for the subtle interplay between human and physical terrain necessary to making truly informed resource allocation and optimization decisions, particularly in a resource-constrained environment.
In one particular example, approximately 6,000 people living in Phalombe District, Malawi, existed for years without access to a reliable source of water. They relied exclusively on three water wells, which were overused or in poor condition . Further complicating the situation, flooding during the rainy season often prevented or otherwise limited access to the existing wells. In this type of environment the Euclidian distance between points becomes increasingly less relevant as both physical and human terrain are considered.
For example, the local population is not uniformly distributed, and local customs, norms and relationships also influence movement and interaction in the region. This results in a subtle dance that reflects a nuanced relationship between the human terrain within the context and constraints of the physical environment; complexity that requires advanced statistical modeling techniques to truly understand. The seasonal flooding effectively bisects the space, thereby limiting access to possible well locations. Using the FSP model, which incorporated both physical and human terrain, a short list of locations that would effectively serve the water needs for the associated communities under different conditions, including the seasonal river flooding, was created and used to inform decisions regarding placement of the wells.
Going forward, geospatial content and associated analytic capabilities can provide unique insight in support of information-based approaches to resource allocation and response to other humanitarian challenges, including the 2014 Ebola outbreak and future epidemics. In particular, human geography content can provide additional context and related insight regarding the physical and human terrain associated with the outbreak to augment current disease surveillance and response efforts. Moreover, geospatial analytic capabilities and associated derived content regarding the distribution, movement and interaction of specific groups of people as it relates to culture, language, religion and other sociocultural variables within the constraints of the physical environment could bring important context to humanitarian decisions. Sociocultural mapping in particular not only provides valuable insight regarding putative transmission routes, but also could be used to inform anticipatory guidance regarding future spread.
This content and related capabilities could be used to further inform disease surveillance, as well the distribution of limited or otherwise scarce medical resources and healthcare personnel, particularly in remote, physically impenetrable and otherwise nonpermissive or denied areas. Again, by provisioning this content and related analytic capabilities in a geospatial environment, epidemiologists, infectious disease experts, healthcare providers and other professionals can effectively collaborate in the same analytic workspace, incorporating their specific tacit knowledge and domain expertise in support of novel insight and meaningful approaches to surveillance, prevention, mitigation and response.
As in other domains, data collection is necessary but not sufficient for insight and meaningful response. The use of good social science research, O.R. concepts and methods and commercial best practices can change outcomes in operational security analytics and humanitarian response. While we cannot mitigate the risk to zero, the ability to effectively characterize incidents enables information-based approaches to prevention, response and consequence management. So the next time that you are at your local grocery store and feel compelled to grab that box of cereal from the end cap, or see the strategically positioned adult beverages beckoning you, know that the same analytic tools and concepts are being used to keep you safe and secure in the parking lot and beyond.
Colleen McCue (Colleen.McCue@digitalglobe.com), Ph.D., is senior director, Social Science & Quantitative Methods, at DigitalGlobe, a leading provider of commercial satellite imagery.
1. Beck, C. & McCue, C., 2009, “Predictive Policing: What Can We Learn from Wal-Mart and Amazon about Fighting Crime in a Recession?” The Police Chief, November 2009. (http://www.policechiefmagazine.org/magazine/index.cfm?fuseaction=display_arch&article_id=1942&issue_id=112009).
2. Dalton, J.R. & M.D. Porter, 2009, “Geospatial Preference Models in Signature Analyst,” SPADAC, Inc. (#GSPWP20091006).
3. Rittel, H. and M. Webber, 1973, “Dilemmas in a General Theory of Planning,” Policy Sciences, Vol. 4, pp. 155-169.
4. McCue, C., 2007 (update), “Data Mining and Predictive Analysis: Intelligence Gathering and Crime Analysis,” Burlington, Mass.: Butterworth-Heinemann.
5. McCue, C., Miller, L. and Lambert, S., 2013, “The Northern Virginia military shooting series: Operational validation of geospatial predictive analytics,” The Police Chief, February 2013. (http://www.policechiefmagazine.org/magazine/index.cfm?fuseaction=display&article_id=2871&issue_id=22013).
6. “New Water Wells in Phalombe, Malawi, DigitalGlobe Case Study for Water Wells for Africa,” 2012. c.f. (http://www.digitalglobefoundation.org/sites/default/files/WWFA.pdf).