Tag Archives: ABM

Surprising Findings: Using Mobile Phones to Predict Population Displacement After Major Disasters

Rising concerns over the consequences of mass refugee flows during several crises in the late 1970’s is what prompted the United Nations (UN) to call for the establishment of early warning systems for the first time. “In 1978-79 for example, the United Nations and UNHCR were clearly overwhelmed by and unprepared for the mass influx of Indochinese refugees in South East Asia. The number of boat people washed onto the beaches there seriously challenged UNHCR’s capability to cope. One of the issues was the lack of advance information. The result was much human suffering, including many deaths. It took too long for emergency assistance by intergovernmental and non-governmental organizations to reach the sites” (Druke 2012 PDF).

Forty years later, my colleagues at Flowminder are using location data from mobile phones to nowcast and predict population displacement after major disasters. Focusing on the devastating 2010 Haiti earthquake, the team analyzed the movement of 1.9 million mobile users before and after the earthquake. Naturally, the Flowminder team expected that the mass exodus from Port-au-Prince would be rather challenging to predict. Surprisingly, however, the predictability of people’s movements remained high and even increased during the three-month period following the earthquake.

The team just released their findings in a peer-reviewed study entitled: “Predictability of population displacement after the 2010 Haiti earthquake” (PNAS 2012). As the analysis reveals, “the destinations of people who left the capital during the first three weeks after the earthquake was highly correlated with their mobility patterns during normal times, and specifically with the locations in which people had significant social bonds, as measured by where they spent Christmas and New Year holidays” (PNAS 2012).

For the people who left Port-au-Prince, the duration of their stay outside the city, as well as the time for their return, all followed a skewed, fat-tailed distribution. The findings suggest that population movements during disasters may be significantly more predictable than previously thought” (PNAS 2012). Intriguingly, the analysis also revealed the period of time that people in Port-au-Prince waited to leave the city (and then return) was “power-law distributed, both during normal days and after the earthquake, albeit with different exponents (PNAS 2012).” Clearly then, “[p]eople’s movements are highly influenced by their historic behavior and their social bonds, and this fact remained even after one of the most severe disasters in history” (PNAS 2012).


I wonder how this approach could be used in combination with crowdsourced satellite imagery analysis on the one hand and with Agent Based Models on the other. In terms of crowdsourcing, I have in mind the work carried out by the Standby Volunteer Task Force (SBTF) in partnership with UNHCR and Tomnod in Somalia last year. SBTF volunteers (“Mapsters”) tagged over a quarter million features that looked liked IDP shelters in under 120 hours, yielding a triangulated country of approximately 47,500 shelters.

In terms of Agent Based Models (ABMs), some colleagues and I  worked on “simulating population displacements following a crisis”  back in 2006 while at the Santa Fe Institute (SFI). We decided to use an Agent Based Model because the data on population movement was simply not within our reach. Moreover, we were particularly interested in modeling movements of ethnic populations after a political crisis and thus within the context of a politically charged environment.

So we included a preference for “safety in numbers” within the model. This parameter can easily be tweaked to reflect a preference for moving to locations that allow for the maintenance of social bonds as identified in the Flowminder study. The figure above lists all the parameters we used in our simple decision theoretic model.

The output below depicts the Agent Based Model in action. The multi-colored panels on the left depict the geographical location of ethnic groups at a certain period of time after the crisis escalates. The red panels on the right depict the underlying social networks and bonds that correspond to the geographic distribution just described. The main variable we played with was the size or magnitude of the sudden onset crisis to determine whether and how people might move differently around various ethnic enclaves. The study long with the results are available in this PDF.

In sum, it would be interesting to carry out Flowminder’s analysis in combination with crowdsourced satellite imagery analysis and live sensor data feeding into an Agent Base Model. Dissertation, anyone?

Promises and Pitfalls in the Spatial Prediction of Ethnic Violence

My colleague Nils Weidmann just published this co-authored piece with Harvard Professor Monica Toft. The paper deserves serious attention. Weidmann and Toft review this article on the spatial prediction of ethnic conflict that was authored by Lim, Metzler and Bar-Yam (LMB) and published in the prestigious journal Science.

I reviewed the article myself earlier this year and while I was highly suspicious of the findings—correlations of 0.9 (!) and above—I did not dig deeper. But Weidmann and Toft have done just this and their findings are worth reading.

The authors clearly show that the analysis by LMB “suffers from a biased selection of groups and regions, an inadequate null hypothesis and unit of analysis.” This really begs the following question: how did the LMB paper ever make it through the peer-review process?

The authors’ case selection is seriously biased as it “seems to adjust the group map as to better fit the model predictions,” for example. The isolationist policy recommendations that LMB put forward are thus founded on misleading methods and ought to be entirely dismissed.

Better yet, Science should retract the LMB paper or at least publish the commentary by Weidmann and Toft. Indeed, another question that follows from the conclusion reached by Weidmann and Toft is this: how many other below-par papers have been accepted and published by Science?

In sum, not only are the methods used by LMB questionable, but as Weidmann and Toft conclude, “the model provides little advance on prior research” in the field of crisis mapping.

On the plus side, the fact that there is push back on early articles in the field of crisis mapping is also a good sign and evidence that the field is becoming more formalized. In addition, the general approach taken by LMB still holds much promise for crisis mapping—it simply needs to be done with a lot more care and transparency. Indeed, combining agent-based models with real world empirical data and a sound understanding of ethnic conflict could become a winning strategy for crisis mapping analytics.

In closing, I look forward to following Nils Weidmann’s work at Princeton and have no doubt that he will continue to play an important role in the development of the field, and will do so with integrity and rigorous scholarship.

Patrick Philippe Meier

Crisis Mapping and Health Geographics

Crisis Mapping is by definition a cross-disciplinary field. Crises can be financial, ecological, humanitarian, etc., but these crises all happen in time and space, and necessarily interact with social networks. We may thus want to learn how different fields such as health, environment, biology, etc., visualize and analyze large complex sets of data to detect and amplify or dampen specific patterns.

We can’t all become specialists in each others’ areas of expertise but we can learn from each other, especially if we share a common language. Like the field of complexity science, Crisis Mapping can provide a common but malleable language, taxonomy and conceptual framework to facilitate the exchange of insights driven by innovative thinking in diverse fields.

This explains why I was excited to come across the International Journal of Health Geographics a few days ago. The Journal is an online and open-access resource. This means new ideas can be shared openly, which is conducive to innovation, just like arXiv.

Two of the Journal’s latest articles caught my interest:

1) An Agent-Based Approach for Modeling Dynamics of Contagious Disease Spread

This study developed a spatially explicit epidemiological model of infectious disease to better understand how contagious diseases spatially diffuse through a network of human contacts. To do this, the authors developed an agent-based model (ABM) that integrates geographic information systems (GIS) to simulate the spatial diffusion. (See my previous post on ABM and crisis mapping).

What is very neat about the authors’ approach is that they chose to draw on georeferenced land use data and census data. In other words, they combined the fomalistic rules of ABM with empirical GIS data. This means that the model can actually be tested and different scenarios can be played out by adding or changing some of the parameters. Could we use this model for conflict contageon?

2) Combining Google Earth and GIS Mapping Technologies in a Dengue Surveillance System

This study overlayed georeferenced epidemiological data on a town in Nicaragua with satellite imagery from Google Earth to enable dengue control specialists to prioritize specific neighborhoods for targetted interventions. The authors used ArcGIS to “accurately identify areas with high indices of mosquito infestation and interpret the spatial relationship of these areas with potential larval development sites such as garbage piles and large pools of standing water.”

It’s worth noting that the above Google Earth imagery was not particularly high resolution but the authors were still able to make full use of the imagery.

This approach to mapping for decision-support is particularly relevant for resource-limited settings since. As the authors note, the surveillance project “utilizes readily available technologies that do not rely on Internet access for daily use and can easily be implemented in many developing countries for very little cost.”

While the team had a free copy of ArcGIS thanks to the Global Fund, they plan to consider free and low-cost alternatives such as SaTScan, MapServer and Quantum GIS in the future. (See this post for additional alternatives like GeoCommons). I hope the authors also know about Walking Papers. I’ll email them just in case. Here’s to cross-disciplinary collaboration!

Patrick Philippe Meier

Crisis Mapping and Agent Based Models

The idea of combining crisis mapping and agent based modeling has been of great interest to me ever since I took my first seminar on complex systems back in 2006. There are few studies out there that ground agent based models (ABM) on conflict dynamics within a real-world geographical space. One of those few, entitled “Global Pattern Formation and Ethnic/Cultural Violence,” appeared in the journal Science in 2007.

Note that I take issue with a number of assumptions that underlie this study as well as the methodology used. That said, the study is a good illustration of how crisis mapping and ABM can be combined.


The authors suggest that global patterns of violence arise due to “the structure of boundaries between groups rather than the groups themselves.” In other words, the spatial boundaries between different populations create a propensity for conflict, “so that spatial heterogeneity itself is predictive of local violence.”

The authors argue that this pattern is “consistent with the natural dynamics of “type separation,” a specific pattern formation also observed in physical and chemical phase separation. The unit of analysis in this study’s ABM, however, is the local ethnic “patch size,” which represents the smallest unit of ethnic members that act collectively as one.

The Model

A simple model of type separation assumes that individuals (or ethnic units) prefer to move to areas where more individuals of the same time reside. Playing the ABM yields progressively larger patches or “islands” of each ethnic group over time. The relationship between patch size and time follows a power law distribution, “a universal behavior that does not depend on many of the details of the model […].”

In other words, the model depicts scale invariant behavior, which implies that “a number of individual agents of the model can be aggregated into a single agent if time is rescaled correspondingly without changing the behavior at the larger scales.”

To model violent conflict, the authors assume that both highly mixed regions and well-segregated groups do not engage in violence. The rationale regarding the former being that in highly mixed regions, “groups of the same type are not large enough to develop strong collective identities, or to identify public spaces as associated with one or another group. When groups are much bigger, “they typically form self-sufficient entities that enjoy local sovereignty.”

To this end, the authors argue that partial separation with poorly defined boundaries fosters conflict when groups are of a size that allows them to impose cultural norms on public spaces, “but where there are still intermittent violations of these rules due to the overlap of cultural domains.” In other words, conflict is a function of population distribution and not of the “specific mechanism by which the population achieves this structure, which may include internally or externally directed migrations.”

The model is therefore founded on the principle that the conditions under which violent conflict becomes likely can be determined by census.

The Analysis

The authors used 1991 census data of the former Yugoslavia and the Indian census data from 2001 and converted the data into map form (see figure below), which they used in an ABM simulation. “Mathematically, the expected violence was determined by detecting patches consisting of islands or peninsulas of one type surrounded by populations of other types.”


A wavelet filter that has a positive center and a negative surround (also called a Mexican hat filter) was used to detect and correlate the islands/peninsulas. scienceabm1

The red overlays depicted in Figure D above represents the maximum correlation over population types. The diameter of the positive region of the wavelet, i.e., “the size of the local population patches that are likely to experience violence,” is the main predictor of the model.


To test the predictive power of their model, the authors compared the locations of red overlays with actual incidents of violence as reported in books, newspapers and online sources (the yellow dots in the crisis map below).


Their statistical results indicate that the Yugoslavia crisis map model has a correlation of 0.89 with reports. Moreover, “the predicted results are highly robust to parameter variation [patch size], with essentially equivalent agreement obtained for filter diameters ranging from 18 to 60 km […].”

The statistical results for the India crisis map model indicate a correlation of 0.98. The range of the patch size overlapped that of the former Yugoslavia but is shifted to larger values, up to 100km. This suggests that “regions of width less than 10km or greater than 100km may provide sufficient mixing or isolation to reduce the chance of violence.”


While the authors recognize the importance of social and institutional drivers of violence, they argue that, “influencing the spatial structure might address the conditions that promote violence described [in this study].” In sum, they suggest that, “peaceful coexistence need not require complete integration.”

What do you think?

Patrick Philippe Meier