1. QUESTIONS

Disasters are increasing in intensity, frequency, and scope due to the continuing effects of climate change (IPCC 2023). During these disasters, transportation systems are essential to ensuring that people can evacuate safely and recovery supplies are distributed efficiently. However, gaps remain in understanding how people travel before, during, and after disasters, especially flooding events (Hamilton et al. 2020). An assessment of travel behaviour can help officials improve their overall transportation response, meet the needs of travelers, and improve evacuation planning. Moreover, current methods to assess disaster behavior, such as surveys and interviews, often fail to generalize results beyond the survey participants (Wong 2020) due to relatively small sample sizes (Wong et al. 2022; Kuligowski et al. 2022). Mobile phone data can solve this issue (e.g., Yabe and Ukkusuri 2020; Zhao et al. 2022), but some concerns remain about privacy controls and anonymization of individual traces (De Montjoye et al. 2018). Anonymized traffic count data have also been used (e.g., Staes, Menon, and Bertini 2021; Parr et al. 2022), but data are usually limited to highways and major roadways without clear origins and destinations. Consequently, our question for this research was: what emergent travel behavior for a flooding event can be determined using de-identified, aggregated, network mobility data? We answer our question using the 2021 British Columbia (BC) Floods in Canada as a proof-of-concept.

2. METHODS

The data used for this study was gathered through the TELUS Data for Good program. TELUS is a large telecommunications company in Canada that provides wireless coverage across the country. The program offers access via an application programming interface (API) to de-identified, privacy-protected network mobility data for researchers focusing on projects that can produce societal benefits. Importantly, the API platform provides secure, de-identified, and aggregated data. This negates privacy issues, while still providing sufficient detail to understand large-scale urban travel behavior. Rather than displaying individual mobile phone traces, the data provides the total number of mobile phone devices that use any cell towers within and between user-defined geographic locations (e.g., cell tower zone, census tract, city, province, etc.). For example, the API can determine the number of devices entering a city from another city. For increased temporal detail, aggregated counts were collected for one-hour segments.

For our case example, we analyzed the number of people traveling between highly impacted cities before, during, and after the 2021 BC Floods using device counts as a proxy. Resulting from an atmospheric river event, the floods severely damaged infrastructure, closed multiple highways and rail lines, and led to evacuations of several areas (British Columbia 2023). We focus our attention on Hope, BC, a small municipality outside of Vancouver that was cut off from the rest of B.C. We first conducted a simple city-to-city analysis between Hope and Merritt, BC (toward the larger city of Kamloops) and between Hope and Chilliwack, BC (toward Vancouver). Using the API, we gathered a device count (i.e., a proxy for trips) starting from Hope and terminating in Merritt and Chilliwack over a two-week span during the 2021 floods. This was matched with 2019 data by weekday to account for day-by-day variations.

3. FINDINGS

Our analysis of city-to-city flow found a clear change in device counts during the 2021 BC Floods (compared to 2019 data), which indicates a significant deviation in travel behavior. While percentage change varies between years, likely due to seasonal effects (from winter storms) and small population changes, a spike occurs on November 14 (Table 1, Figure 1) between Hope and Merritt, likely corresponding to a small evacuation before the worst of the storm. A clear reduction occurs on the second day of the floods (November 15) and continues through our dataset (Table 1), arising from damaged infrastructure and severe flooding in Merritt. Travel appears to briefly recover at time points for several following days (Figure 1), likely due to temporary access for evacuation purposes. More consistent travel begins on November 20 due to alternative routes, though there is still a daily reduction of devices by over 65%. Continued reduction may be a result of the evacuation order in Merritt on November 15 (Global News 2021), parts of which were lifted on December 3 (Little 2021a), or the continued closure of HW5. This highway reopens later to commercial traffic on Dec. 20 and all traffic on Jan. 19 (B.C. Ministry of Transportation and Infrastructure 2023).

Figure 1.Hourly device counts from Hope to Merritt in 2019 (Nov 7 – Nov 21) and 2021 (Nov 11 – Nov 25). Data matched by weekday to account for day-to-day variation (dates in 2021).

Similar results were found for Hope to Chilliwack (Table 2), with a large drop on November 15 and 16 in 2021. The increase at the beginning of the storms may indicate a small evacuation of Hope away from potential flood risks or early Remembrance Day travel to avoid the storms. In Figure 2, device counts increased for the next weeks at a faster rate than for Hope-Merritt, likely due to lower infrastructure damage. Chilliwack is also connected to Hope through several additional routes, which would indicate flexibility in infrastructure for travelers.

Figure 2.Hourly device counts from Hope to Chilliwack in 2019 (Nov 7 – Nov 21) and 2021 (Nov 11 – Nov 25). Data matched by weekday to account for day-to-day variation (dates in 2021).

Our takeaways focus on both methodology and empirical analysis. First, de-identified network mobility data can approximate hourly travel in disasters at a reasonable level of granularity. Second, our analysis largely matched with route closures across BC (British Columbia, 2021; Serebrin 2021), indicating the relatively strong ability of this proxy method to show general behavioural trends. Third, device counts indicate a rapid movement of people away from the Hope area to other destinations on the first day of major rainfall. Finally, travel patterns suggest a slow recovery and poor resilience of some B.C. highways, routes, and infrastructure. With repair costs expected to reach over $1 billion CAD (Garrett 2023), lessons from this study indicate a need for more adaptive infrastructure for maintaining travel during major flooding events. This research also points to the feasibility of collecting de-identified network mobility data to approximate travel patterns before, during, and after disasters. With its strict privacy protections, simple interface, and zero cost structure, the TELUS data and associated API open a new methodological avenue for timely and big datasets for transportation resilience.

ACKNOWLEDGMENTS

We thank the TELUS Data for Good team, especially Kendal Burtch and Jesslyn Dymond, for their commitment to the research and providing the data. The research was supported by the Dean’s Research Award at the University of Alberta. We also thank the anonymous reviewers for their insightful and helpful feedback.

SUPPLEMENTAL INFORMATION

TELUS Insight data incorporates approximately 9.8 million cellular devices across Canada. A new timestamped record is created when a device:

a) Makes or receives a phone call;

b) Sends or receives an SMS/MMS message;

c) Sends or receives data;

d) Is dormant for over 30 minutes; or

e) Moves/connects to a new cell tower.

At a minimum, a non-moving device that is not actively used will generate 48 timestamped records per day, and an average device generates 300 timestamped records per day. Since the data results in 3 billion records per day, TELUS Insight data (de-identified network mobility data) provides absolute results for all TELUS customers.

For this research, the TELUS Insight team confirmed that there were no anomalies in the number of records each day since 2019 (inception of the de-identified network mobility data). However, some changes may exist in the number of customers as people switch carriers. The inability to determine this change is a limitation of the study. Moreover, the number of TELUS customers in a specific area is not available externally due to TELUS Privacy by Design guidelines. However, research by the Government of Canada has determined that TELUS accounts for about 28% of the Canadian market (Canadian Radio-television and Telecommunication Commission 2022). Within British Columbia, TELUS accounted for 40.6% of subscribers in 2019 and 43.0% of subscribers in 2021 (CRTC 2023). The following population changes also occurred between 2016 and 2021 (when a census was conducted in the area (Statistics Canada 2024))

• Hope: 6,181 to 6,686

• Chilliwack: 83,788 to 93,203

• Merritt: 7,139 to 7,051

These population differences do not account for changes in the number of travellers passing through Hope, Chilliwack, and Merritt, who would also be counted in the methodology based on the requirements of the timestamped records above.

Regarding resolution, TELUS Insights data is based on the locations of the cellular towers, and their coverage areas. The spatial granularity is dependent on the cellular coverage in the area. Coverage accuracy is typically below 200 meters in urban areas, between 200 meters and one kilometer in suburban areas, and between one and many kilometers in rural areas (depending on remoteness). The resolution of data for Hope, Merritt, and Chilliwack in the majority of the area is generally within 200 meters to one kilometer, though certain areas might fall within a one to five-kilometer range. In our analysis, we used the entire municipal district (DM) of Hope and the entire cities (CY) for Merritt and Chilliwack. Regarding overall geography, using straight-line distances and measuring from city center-to-center, Hope is approximately 93 kilometers from Meritt and 44 kilometers from Chilliwack.

For timestamps, the device counts were collected in 60-minute buckets, meaning that each device was counted once and only once if it recorded a timestamp in that 60-minute bucket. The trip terminates in that specific bucket. For routes, since a pass-through location was not included, the device counts factor in all available routes to Chilliwack and Merritt from Hope.