Data mining used to improve disaster response

Citizens of the rural town of Oso, Wash. were greeted by tragedy on Mar. 23 as a waterlogged hillside gave way and unleashed a tsunami-like wave of earth, destroying dozens of homes residing in the river valley below.

Chaos ensued, and now entire communities lie under layers of mud as rescue careers and community volunteers work desperately to free survivors. Up to this point in time, 18 bodies have been recovered and identified. Assuming there are no additional survivors, the presumed death toll is believed to be 28.

As search and rescue crews work fervently on the ground to recover what they can, crises data analysts are sifting through large amounts of heterogeneous data to identify who exactly is missing and who needs to be found. Initially, over 90 people were reported immediately after the incident, but that number has since dwindled down to 30. These initial overestimates are common, as available information in the aftermath of a tragedy — like the Oso mudslide — is limited.

However, to assemble a more realistic picture of the event analysts must pull from a variety of data sources. Only then can accurate predictions of disaster situations, such as estimated persons missing, be made.

In a recent publication of the International Journal of Emergency Management, experts such as Adam Zagorecki of the Centre for Simulation and Analytics at Cranfield University UK point to the increasing abilities of data mining in the wake of natural disasters. Data mining is the process of pulling information from a variety of structured and unstructured data sources.

Unstructured sources can include news reports, local announcements, situational reports, and satellite imagery. All of this is used to generate a model of the event so that those involved can more effectively deploy strategies of mitigating damage. Traditionally, disaster models could only be applied to well-ordered data sets from emergency services and official damage assessments reports. However this information can be slow to gather especially in the immediate aftermath of a disaster, a critical period for emergency response crews.

Recently, data mining has begun to expand into social media applets, as experts use the increased data stream that occurs in the wake of a natural disaster. This is especially useful, when those affected still have access to wireless communications and or the internet. While this is unfortunately not the case for many of the victims of the recent mudslide, such a strategy could be employed in a variety of disaster scenarios.

Additionally, social media can help analysts and responders more quickly identify, contact and track suspected victims of natural disasters. This represents the growing strength of data mining to use information not normally applied to disaster modeling, and use data points from that information to create models to provide useful and relevant information to those on the ground.

The new capabilities of data mining could eventually lead a new generation of response strategies deployed after all sorts of disasters. With these tools we can hope to better predict disasters and mitigate the damage to communities like those near Oso, Washington.