Sudden Onsets

Computational Approach to Evaluate the Impact of Natural Disasters

Young people reporting through their mobile phones during the ebola crisis
UNICEF_UNI174449_Jallanzo

Although countries and regions prone to natural disasters can be mapped, they are often not equipped to prevent or respond comprehensively. Exogenous shocks like earthquakes, floods, landslides, and other natural occurrences often occur without any early warnings. 

 

Such disasters often have tragic consequences, affect massive populations, cause loss of life and leave lasting damage that requires years to repair. After a disaster, the first hours of the response are critical for saving lives. Emergency responders need to act very quickly to reach affected populations, as such, it is vital that they receive accurate and up-to-date information to help focus their limited resources. Long-term effects are equally important, it is absolutely necessary to know how communities are doing weeks, months, and years after an emergency, and whether they require additional support in rebuilding.

Changes in human behavior.

At the time of an emergency, people change behavior, we change our mobility patterns, our consumption habits, our daily routines. Often, the instant reaction is to reach out to a safe zone, notify others who may be concerned or in danger, or turn to social media for information.  These changes can be observed directly throughout various social and technological platforms, from Twitter and Facebook to cell phone activity (see figure below) and satellite imagery. All these interactions leave digital breadcrumbs which researchers and organizations can utilize in order to:

  • Monitor in real-time affected populations: Giving first-responders the opportunity to, in real-time, understand where affected people are and their needs, in order to deliver assistance in a quick and efficient manner.
  • Understand vulnerability and resilience of populations: Allowing international and relief organizations to efficiently assess how different regions recover after a shock, and to better prepare for future events.

The work we do.

UNICEF combines different datasets donated by private sector partners such as satellite imagery, population density, population movements and basic infrastructure. This allows first-responders to estimate, in real-time, where the population that require most help are, permitting a dynamic adaptation of the response. We also build mathematical models that, analyze data over longer periods, allowing governments and UNICEF field offices to understand how different communities recover after a shock, as well as monitor how fast people return to a pre-event behavior – potentially identifying the most vulnerable communities.

Visual map of data from mobile
UNICEF Innovation
Figure: Collective response to the March 10th 2015 earthquake in Colombia. The top panel shows the hourly number of calls in the region closest to the epicenter, exhibiting a clear spike right after the earthquake. Bottom left panel compares cellular activity in the two closest regions (light green circles) to the overall country, a few hours after the earthquake. Circles are scaled according to call frequencies, the larger a circle the more calls take place. The bottom right panel is a map of the most affected regions a few hours after the earthquake.

A closer look. 

UNICEF Innovation is partnering with the private sector to obtain information that can play a critical role in responding to a disaster in a country or a region. In partnership with Telefonica and Facebook, we are using aggregated data from phone usage to identify in almost real-time communities affected by a drastic shock such as an earthquake, a flood or a landslide. Our research aims to answer the following questions:

  • Can we accurately determine, based on digital traces, number and location of people affected?
  • Can we monitor the recovery of affected communities and track how fast they fall back into normal daily patterns?
  • Potentially identifying communities of interest that can benefit from receiving additional help, or communities that can play an active role in supporting others?
  • Can we automatically assess the impact of a natural disaster in an area by using machine learning using satellite imagery pre- and post-disaster?

Next steps include improving and refining the methodologies and testing the usability of these insights and data in emergency prone areas.

A practical example: Colombia

An adolescent girl walks through the flooded yard in front of her home near the Sinú River, in the northern municipality of Cotorra in Córdoba Department. Beside her is a bicycle.

Colombia is beset by regular natural disasters as well as humanitarian challenges emerging from the long-term conflict and current post-accord context. Real-time information on people’s movement, access to schools and other services, as well as local capacities, is critical to planning an emergency response. Together with Telefonica and Facebook, we are testing the possibilities to integrate these new datasets and insights, as well as the school mapping project, into the existing disaster preparedness and response systems operating in Colombia.