Machine learning and global traffic data for smarter planning
By: Scott Farley
The 2018 hurricane season was among the most destructive and deadly on record. Seventeen named storms battered the Caribbean and American Gulf Coasts, killing thousands and causing hundreds of billions of dollars in damages. Since working on this visualization of the impact of Hurricane Maria on regional mobility in Puerto Rico, I’ve been curious about the spatiotemporal patterns of mass evacuations.
How long before landfall do people evacuate? Which routes do they take when they go? What are the similarities among evacuation events for different storms?
At this week’s #SmartDev2018 conference, hosted by the World Bank and the Center for Effective Global Action, I explored these questions using three of last year’s major storms as case studies: Hurricane Harvey, Hurricane Maria, and Hurricane Irma. Anonymized data from our global traffic sensor network provides valuable insight into regional patterns of mobility and connectivity in over forty countries. Using this location data, we can model how and when people evacuate from hurricanes and storm events around the world.
Initially, I was curious about the timing of when people are on the road during an evacuation. To investigate, I added up the total volume of traffic on all the roads within a 3° latitude by 3° longitude box. For Puerto Rico, there is a major spike in total volume the evening before Hurricane Maria makes landfall.
I repeated this analysis for Southeastern Texas during Hurricane Harvey and South Florida during Hurricane Irma. Each of these time series shows a precipitous volume spike in the evening before the storm hits. The timing of these spikes is consistent with a cohort of people waiting for the conclusion of the work day before evacuating.
Next, I looked at the routes chosen during pre-storm evacuations. It is common to see images of miles-long traffic jams of people evacuating from impending storms. I found that this congestion is common not just on interstate highways but on all roads that can be used for long-distance travel, such as state and county routes. To illustrate, I compared a typical Friday rush hour in Southeastern Texas with the Friday night before Hurricane Harvey made landfall.
On a normal Friday night, the majority of the intercity travel occurs along the Interstate 37 corridor between Corpus Christi and San Antonio. However, during the evacuation, traffic is widespread on a host of alternate routes between these two cities, as well as on routes that connect Corpus Christi with Victoria and Houston.
In this case, primary roads (state highways) increased in volume by approximately 150%. Secondary and tertiary roads (county highways and local routes) also experienced 50–75% increases during evacuation when compared to the baseline.
In other words, roads designed to carry low volumes of vehicles at low speeds are essentially functioning as freeways during evacuation events. This can be especially dangerous for drivers choosing to take these roads in the inclement weather that accompanies an impending storm.
These patterns found in Southeastern Texas are indicative of trends seen during other evacuation events as well. For instance, secondary, tertiary, and residential roads in Southern Florida had 50–100% more traffic during the Irma evacuation than during a typical rush hour. While the exploration has so far focused on Atlantic hurricanes, our global traffic coverage means we can analyze the effects of storms in diverse regions across the globe.
Cyclones all over the world are projected to increase in frequency and severity over the coming century. As we observe future evacuations, we could build up a model capable of predicting the volume of any given road segment during an evacuation. State and local governments could use this model to assess the readiness of their infrastructure and to properly position emergency resources prior to a hurricane.
At the #SmartDev2018 conference, I spoke with a mix of academic researchers, machine learning engineers, and development practitioners about how we could use our mapping tools and real-time traffic data to bring this model to life and keep people safer during evacuations.
Learn more about using our platform for humanitarian causes. If you’re using maps and location data for positive social and environmental change, reach out to our community team: community@mapbox.com. Follow the Community tag on Medium to read more about humanitarian projects built with Mapbox.
Exploring real-time hurricane evacuation patterns was originally published in Points of interest on Medium, where people are continuing the conversation by highlighting and responding to this story.