DATA FOR CHANGE
Can Human Mobility Disruptions in Cities Be Seen in Near-Real-Time?
This work has been done entirely using open data, and was co-authored with Kai Kaiser. All errors and omissions are those of the author(s).
More than ever, 2020 highlighted that the incidence and aftermath of geographic and public health crises can result in widespread disruptions to human movement. With emerging sources of big Data comes the promise of informing response, recovery, and ultimate resilience to these risks in near-real-time. Using location data derived from smartphones, we provide a comparative cross-border visualization of human movement in the face of such challenges in two Pacific countries: the Philippines and Vietnam.
On November 9, 2020, Typhoon VAMCO, a single weather front formed in the Pacific, spurring hundreds of thousands to evacuate, stranding passengers, suspending government work, closing schools, and grounding flights. Beginning in the eastern Philippines, Typhoon VAMCO lashed laterally through to Central Vietnam (trajectory illustrated in Figure 1). In total in 2020, even as policymakers grappled with continued questions of lockdowns, mobility restrictions and other non-pharmaceutical interventions to curb the spread of COVID-19, the Pacific saw 10 typhoons and 23 tropical storms. Metro Manila saw the worst flooding in decades, as the economic shocks from the pandemic compounded with destruction of housing and infrastructure, and urban islands where food could not reach for days.
Policy directives (lockdowns or evacuations), force majeure events (destruction of housing or infrastructure), and acts of self-preservation (choosing to stay home) sent shockwaves of disruption to ‘normal’ patterns of human movement and socio-economic activity this typhoon season, against the backdrop of the pandemic.
When people are ‘sheltering-in-place’ (such as during a pandemic or extreme weather), or trapped somewhere (such as during a flash flood or road network disruption), we expect mobility to decrease. If lockdowns or evacuations are ordered, we might see surges in movement as people relocate. Following directives, such as those typically seen on national bulletins, governments and policymakers might ask: Which areas were the or least responsive? Were some regions not responsive at all?. Was mobility affected in the areas we expected, or was movement in other areas affected too? And what data sources can help us monitor changes to human movement in real time, from the oncoming of a disaster through to its dissipation?
Smartphone mobility data: Near-real time sensing of human movement
As smartphones become ubiquitous across the world, insights derived from GPS-location data are providing unprecedented insight into how people move during a disaster. As part of a collective humanitarian effort—accelerated by COVID-19—technology companies such as Facebook and Google are increasingly making datasets available that enable us to measure how changes to mobility unfold on the ground as people brace for, and then tackle an emergency. In addition to near-real time, highly granular local insights for policymakers, the shared methods used to collect and generate this data also enable consistent and comprehensive comparisons of mobility change across borders.
Any crisis-related application of mobility data in time and space will ideally meet two criteria for policy practitioners. First, that results can be presented in a way that is useful to decision-makers. Second, that at any given level of spatial or temporal granularity, they are representative of underlying local people, geographies, and contexts. A companion blog shows how the representativity of mobility data can be assessed using set of open-source rapid appraisal methods that we have developed.
Here, by focusing on Typhoon VAMCO (Figure 1), we present a new use case for the Facebook Movement Range Maps dataset (originally intended for monitoring the mobility impact of COVID-19): measuring shocks caused by natural disasters. Because this data has been made available since the beginning of the pandemic, it enables us to track impacts on local mobility, even when compounded by manifestations of other risks, such as natural disasters.
Consistent measurement from first-waves to flooding in Vietnam and the Philippines
Facebook Movement Range Maps provide two metrics for tracking mobility: (i) change in movement (what was the range or radius of movement in a district on a given day?) and (ii) ‘staying put’ (what percentage of users stayed in a single location all day?). In Figure 2, we track the former, from the first waves of COVID-19 through to the 2020 Pacific typhoon season and the New Year. In the Philippines, the initial lockdown showed a 60% reduction in mobility compared to baseline, and remained at about a 30% reduction throughout the pandemic – a lockdown described as having ‘no end in sight’. In Vietnam on the other hand, the average reduction in the first wave was only 40% nationally, and lasted only a month (April-May). Mobility had all but returned to baseline (in some cases even overshooting due to bursts of activity), until the second wave of infections in Da Nang in August, where more localized lockdowns led to a smaller drop in mobility of 20%.
Considering this context, mobility was already below ‘normal’ in the Philippines when Super Typhoon GONI and Typhoon VAMCO made landfall, both of which led to short, rapid drops in mobility, mirroring the types of low mobility typically seen on low-movement days such as Christmas Day and New Year’s Eve.
In Figure 3, we evaluate the percentage of users ‘staying put’ during the pandemic and typhoon season. We see long, sustained peaks in periods where lockdowns were introduced (April and May), and short, temporary peaks during extreme weather events.
Tracking Typhoon VAMCO from formation to dissipation
Typhoon VAMCO first made landfall in Catanduanes in the Philippines on 11 November, 2020––of the most destructive typhoons of the season. In Figure 4a, we identify and track areas that were successively impacted by flooding, beginning with Catanduanes, followed by Metropolitan Manila, Pampanga Bulacan, Tarlac and Rizal on 12 November, and Tugueragao City on 13 November.
After moving laterally through the Philippines, Typhoon VAMCO moved to Vietnam, where it made landfall on 15 November. Figure 4b illustrates how VAMCO moved through the country, decreasing mobility in Ha Tinh, Thua Thien Hue, Da Nang, Quang Binh, and Quang Ngai within the span of two days. In both figures, greater drops in mobility are indicated by darker shades of red.
Movement Range Maps in the Open-Source: Continuous Mobility Monitoring for Compound Risk
The Movement Range Maps data used for the analysis above is accessible publicly on the Humanitarian Data Exchange, available globally, and updated daily. This enables almost immediate availability for policymakers, robust comparability across borders, and continuous monitoring of impact and degree of recovery down to district levels. Whilst Movement Range Maps were originally released for measuring mobility changes following the COVID-19 pandemic, we illustrate that the data is a unique compound metric that intrinsically layers the effects of multiple shocks to mobility. In contrast to some other datasets, the dataset does not need to be triggered or activated in the face of an event.
This information can be critical for policymakers and local governments at each stage of the disaster response cycle, from identifying areas which have been most impacted as an emergency unfolds (did the areas expected to be impacted differ from those that were actually impacted?), to tracking rates of local recovery (which city or town recovered the fastest?), identifying long-term damage, and eventually, ‘building back better’. In order to scale the application and adoption of this data and scale capacity in policy-making contexts, we have worked with the Facebook Data for Good team to release a Python-based tutorial on disaster analytics.
References
[1]. W. Maloney and T. Taskin, Voluntary vs mandated social distancing and economic activity during COVID-19, VoxEU.
[2] K. Kaiser and T. C. Peixoto, How governments can use data to fight the pandemic and the accompanying infodemic, World Bank Blogs.
[3]. M. Khan and K. Kaiser. How representative is mobility data? Rapid validation of Facebook’s Disaster Maps data.
[4]. M. Khan and K. Kaiser, How do people move during a disaster? A tutorial on a new use case for Facebook’s Movement Range Maps.
Acknowledgements
This analysis was conducted by the World Bank in partnership with Facebook Data for Good. Special thanks go to the Australian Government Department of Foreign Affairs and Trade (DFAT) for their support of this work through the Vietnam Big Data Observatory for COVID-19 Socio-Economic Response, Recovery, and Resilience as part of the Australia World Bank Strategic Partnership in Vietnam, Phase 2. All errors and omissions are those of the author.