Using Geolocation Data to Understand Consumer Behavior During Severe Weather Events

Our mission at Orbital Insight is to understand what’s happening on and to Earth through geospatial analytics. So, when major weather events like Hurricane Florence occur, we take the opportunity to explore how geospatial analytics can be most beneficially applied to aid decision-making.

In 2017, we worked on analyzing satellite imagery to understand the full flood extent in Houston associated with Hurricane Harvey. Our efforts showed how geospatial analytics can be useful for first responders, policy makers, insurance companies, and those who help with the rebuilding effort. For that project, we used both optical and synthetic aperture radar (SAR) satellite imagery, because SAR imagery has the advantage of being able to “see” through clouds—important for the often-overcast aftermath of weather events.

For Hurricane Florence, however, we took a slightly different approach, leveraging another type of data: geolocation information. We analyzed this data in order to better understand how people shop at stores like Walmart or Home Depot to prepare in the days leading up to (and following) a major storm. Our hope is that analysis like this may be useful to better understand consumer behavior so that, for future weather events, stores and officials can know when to expect higher-than-normal traffic. This can help ensure provisions are sufficiently stocked for when people begin preparing for the storm.

What is geolocation data?
At a high level, the geolocation data we work with consists of ad_IDs, which are strings of letters and numbers associated with a single device, such as a mobile phone. An ad_ID contains no personally-identifiable information (PII), and we only combine the ad_IDs with location data and a timestamp. Essentially, this gives us an understanding of how many unique devices are in a given area at a given time.

Geolocation data provides a different type of information than satellite or drone imagery and has unique benefits. For instance, ad_IDs from a device typically ping multiple times per day, creating a more granular dataset than a satellite image, which might only be captured once per day or even less frequently. Where satellite imagery is limited by revisit rates, geolocation data reliably offers multiple “snapshots” per day of ground activity. Additionally, devices ping ad_IDs anywhere that they have service, meaning that device counts can be aggregated inside buildings or other structures. Satellite imagery, which relies on a visual connection to the object being monitored, can be limited in that regard. Finally, geolocation data is complementary to our efforts to build a macroscope for the planet because it can be both highly granular, such as being restricted to the boundaries of a specific store, or viewed at scale, taking an entire county into consideration. This allows our team to analyze both micro and macro trends to better understand what’s happening on the ground.

Consumer behavior around Hurricane Florence
For Hurricane Florence, we worked with geolocation data in order to better understand how the weather event was impacting local store traffic. We specifically studied New Hanover County in North Carolina. In this area, we receive ad_ID data from a certain percentage of overall devices, which means that our analysis is most useful for understanding trends over time, rather than estimating raw numbers.

The graph below shows how device counts were distributed between three categories of stores—grocery, home improvement, and Walmart—in the days leading up to landfall. In this store comparison analysis, the category of “home improvement” included two Home Depots and two Lowe’s, while “grocery” included one Whole Foods, six Harris Teeters, five Food Lions, and two Lowes Foods. For the “Walmart” category, two Walmart stores were included. As you can see, all three categories show fairly similar device counts for most of the days leading up to Hurricane Florence. However, around the time of the voluntary evacuation notice, there is a sharp spike in visits to home improvement stores relative to the other two categories. This likely reflects residents working to weather-proof their homes before the storm hit and shows that many people took early action based on forecasts.

We also used geolocation data to create a county-level heat map for devices in New Hanover. As previously mentioned, most devices ping their ad_IDs multiple times per day. In our heat maps, we have recorded all ad_ID pings we received on a given day; as a result, these maps indicate how devices were spread out throughout each day.

The first heat map depicts ad_ID pings from Aug. 22, an average day in the county that’s not impacted by severe weather. You can see a fair amount of device density in the Wilmington area and other urban centers. Also, you’ll notice a few pings represented along the coast line.

On the heat map for Sept. 11, a few days before landfall, you can see an increased concentration of devices in city areas—possibly indicative of residents shopping for preparation supplies. Also, there are fewer pings along the coast on this map.

On Sept. 14, the day of landfall, we see less activity clustered in urban areas, and the overall density of pings also appears to be decreased, suggesting that some residents had evacuated or were sheltering in place. Also, this heat map shows the fewest pings along the coast.

The future of geolocation data
We continue to explore the best ways to incorporate geolocation data like this into useful analyses, whether for future natural disaster response or other industry segments. Our analysis in this case was conducted post-event, and we hope that it will aid in understanding how populations behave in advance of severe weather, especially with a voluntary evacuation notice issued. We’re also currently conducting analysis on data from areas affected by Hurricane Michael, in order to add to our understanding of consumer behavior during weather events. As the availability of geolocation data continues to evolve, we’ll be constantly iterating on the best way to incorporate it into analytical work.

Orbital Insight is a multi-modal geospatial data analytics company. As illustrated, geolocation data has several unique benefits, but ultimately, we believe that the richest analytics are derived from combinations of multiple data sets, such as optical satellite imagery, SAR satellite imagery, and geolocation. These combinations are an area we continue to explore in order to make our products as useful, accurate, and timely as possible.

We’ll share updates as we continue to work with different modes of geospatial data. We’re excited to see how these diverse data sources can improve the quality of our already-strong analysis. If you’re interested in working on this challenge with us, visit

Back to Blog