August 5, 2018

New data tools for relief organizations: Network coverage, power and displacement

By: Paige Maas, Laura McGorman, Chaya Nayak, Wonhee Park, Andi Gros

Just over one year ago, Facebook launched its Disaster Maps product to help communities recover and rebuild from natural disasters. Since that time, a variety of humanitarian organizations have used our suite of data tools to respond to natural disasters around the world.

In the last year, the Red Cross used Disaster Maps to restore internet connectivity to parts of Puerto Rico in the aftermath of Hurricane Maria, Direct Relief distributed 400,000 respiratory masks during the Southern California Wildfires using insights from the Disaster Maps tool, and the World Food Programme recently used the data to restore connectivity to neighborhoods in the east of Dominica.

Over the last year, we have also worked with our partners to continue to improve our Disaster Maps offerings for partners. We are excited to share that we have developed three new maps that help partners understand patterns of CellularNetwork Coverage, Battery Charging and PopulationDisplacement. This post presents these new Disaster Maps, with descriptions of the data and methods used to create the maps, alongside examples with interpretations from past disasters.

Cellular Network Coverage

Cellular coverage is critical during times of crisis, but hurricanes and tornadoes can damage cell towers and disrupt critical connectivity infrastructure. Affected populations need to be able to communicate with friends and family, and response organizations need to be able to communicate with affected populations and staff. Our new Cellular Network Coverage maps provide information on where cell coverage is available for populations affected by natural disasters.

  • Key Questions: Where has a disaster disrupted cellular service? Where can people access a cellular network to reach out to their friends and family?
  • Data & Methodology: When people use the Facebook mobile application with Location Services turned on, we collect data on which towers are providing the connection and where the device is located. This location data does not connect the device to a person’s name or any other identifying information. For each cell tower in the disaster-affected region, we estimate the size of the coverage area based on the locations of devices that have connected with it. We then compare this estimate with an aggregation of three months of historical data to identify changes in the coverage area caused by the disaster [1].
  • Metrics: The cellular network coverage map is presented as a raster grid. For each cell in the grid, we provide:
    • An estimate of the number of cell towers providing coverage.
    • A z-score representing the difference in standard deviations between the number of connections originating from the grid cell during the baseline period compared to the number of connections during the crisis.
  • Data Considerations: In areas with low numbers of people on Facebook, we cannot determine whether cell sites are operational with any level of certainty. Similarly, it is difficult to establish the status of a cell tower in evacuated areas since the tower may be operational but not in use. The combination of the two metrics — the differences in counts per grid cell of pings to Facebook, together with the coverage data during the disaster period — allows us to determine with higher certainty that a cell tower is operational.

Coverage Maps in Practice: Volcanic Eruption in Guatemala

The figure below is pulled from Facebook’s Global Coverage Map, which shows the 2G network coverage in Alotenango, Guatemala on April 22, 2018 before the volcanic eruption [2]. From this map, we see that under normal circumstances, there is network coverage in the area.

The visualization below shows 2G network coverage on June 10, 2018, four days after the volcanic eruption. Based on the sharp increases in red and orange areas, as well as complete blank spots on the map, we can see a sharp drop in network use in the areas surrounding the site of the eruption.

The Cellular Network Coverage maps were first tested in 2014 during the Ebola Crisis in East Africa to help NetHope connect Emergency Treatment Units to the internet and then used by NetHope to prioritize efforts for re-establishing internet connectivity for communities in Puerto Rico. The UN Emergency Telecommunications Cluster also used the data to understand the state of internet connectivity in other Caribbean Islands after Hurricanes Irma and Maria.

Battery Charging Map

Access to electricity enables critical activities like heating or cooling one’s home, lighting a room, or charging a cell phone. When power infrastructure is down, response organizations need to know where people are without power. Facebook is able to produce maps that estimate power availability by measuring the number of phones connecting to or disconnecting from power. If the volume of phones charging after a disaster drops significantly compared to normal levels, we can assert that power may not be available in that area.

  • Key Question: Where are people able to charge their devices in a disaster-affected area? What does this imply about power availability?
  • Data & Methodology: Combining information from location services and the Android platform, we aggregate and smooth information on the number of phones connecting or disconnecting from power [2,3]. In order to establish baseline measures, we estimate the average number of phones across the disaster impacted region connecting to power before the disaster. We then estimate changes in the number of phones connecting to power in the area during the disaster compared to the baseline [4].
  • Metrics: Similar to the cellular network coverage map, the battery charging map is presented as a raster grid. For each cell in the grid, we provide:
    • The aggregated number of phones connecting to and disconnecting from power before, during, and after the disaster.
    • A z-score representing the difference in standard deviations between the number of battery charges during the disaster and the number of charges before the disaster began.
  • Data Considerations: In addition to the data considerations raised above for the Network Coverage Maps, for Battery Charging Maps, we cannot distinguish between all different types of power sources. We do filter out instances where people may be charging their phones in cars, in order to more closely proxy the availability of power infrastructure during disasters.

Battery Charging Map in Practice: Storms in Ranchi, India

The figure below shows changes in number of phones connecting to power, expressed as the percentage of the baseline average, in areas near Ranchi, India on May 28, 2018, which suffered from power outages after lightning and thunder storms. While power availability was not affected in the city center of Ranchi where the area is still pale blue, there was a sharp decrease in the number of phones connecting to power in areas outside the city, highlighted in red.

Displacement Maps

In 2016, 24.2 million people were newly displaced from their homes by disasters [5]. In order to provide resources and support to individuals who have been displaced, humanitarian organizations such as the Red Cross need to know where people have migrated. To surface these insights, we developed a new Displacement Map.

  • Key Questions: How many people have been displaced from their home city after a disaster? Which cities are these people displaced from? Which cities are displaced people traveling to?
  • Data & Methodology: We aggregate information from people using Facebook on mobile with location services on, at the city level. We focus on the population of people who were in the disaster-affected area during the five weeks before the disaster, and who were present in their most commonly occurring city or their “home city” during the week before the disaster. For this group of people, we then identify their most common city each week — their “current city” — in the months after the disaster and aggregate that information at the city-level, only including cities with a sufficient number of people in the data set to be statistically meaningful.
  • Metrics: We provide home city information, including relevant week by week information on cities in the disaster area. We also provide current city information, reflecting trends where large groups of displaced people moved during each week following the disaster.
  • Data Considerations: We may not have current city information for people without connectivity, people without location services turned on, or those who do not use the Facebook mobile application. To help understand if these potential sources of missing data may be affecting the map, we include information on the number of people whose city is currently unknown to make clear where there are gaps in our data.

Displacement Map in Practice: Wildfires in Northern California

In October 2017, devastating wildfires destroyed an estimated 8,900 structures in the Napa Valley region, forcing 100,000 people to evacuate, many of whom remained displaced from their homes well after the fires were extinguished [6]. The figure below shows the net difference in population during this time, reporting the total change in the number of people in each city in the week of December 4, 2017, approximately two months after the fires started, compared to the week before the fires started [7].

The dots in purple, blue, and green show cities with a decrease in people, while the orange and red dots show cities with an increase in population. From this visualization, we can see that nearly all of the cities in the Napa Region show some decrease in population after the fires, including Santa Rosa, Rohnert Park, Napa, and Vallejo. We can also see the cities where displaced people went in large numbers, including major cities outside the danger zone, such as San Francisco and San Jose.

Looking Forward

Our Disaster Maps program seeks to help NGOs and communities more effectively prepare and respond to natural disasters. With these new maps, our partners can get up-to-date information on whether people have access to cellular coverage and power, and whether they have been displaced from their homes. Building on our partnerships with humanitarian organizations around the world, we hope that when disasters do strike, these tools continue to help communities recover and rebuild.

Footnotes

[1] The Global Coverage Map presents coverage polygons, while the Disaster Coverage maps present cell coverage information at a raster tile level.
[2] This information is not available for IOS devices.
[3] Learn more about how we apply spatial smoothing methods to protect privacy in our previous blog post.
[4] As part of the baseline analysis we examine cellular coverage in areas during particular days of a week, which controls for situations such as towers that are always operational but only get traffic during certain parts of the week, for example an office park area, where people would be nearby during weekdays; but not over the weekend.
[5] Source: Internal Displacement Monitoring Centre’s 2017 Global Report.
[6] Source: California Statewide Fire Summary.
[7] We visualize changes on a log scale to minimize the influence of outliers in the map.