Sunday, March 28, 2021

Spy agencies, COVID-19, and parking lots

In Canada and many other countries around the world, most government agencies reacted to COVID-19 by directing the bulk of their employees to work from home. But this option was not available for the majority of those working for intelligence agencies because most of their work is too highly classified to be done outside special high-security offices known as secure compartmented information facilities (SCIFs). So I was curious how Canadian agencies such as CSE and CSIS, and CSE's Five Eyes counterparts, addressed this problem. Did they keep a large part of their workforce at home at various points during the pandemic? Did they move people to off-hour times such as weekends and nights? How long did these changes go on for?

When you ask Canadian agencies questions like these the OPSEC klaxons sound, public affairs officials cry out in terror and are suddenly silenced, and a great and impenetrable darkness falls over the land. It can be pretty awkward. But it occurred to me that publicly available satellite photos might provide at least partial answers to some of these mysteries. Specifically, satellite photos of agency parking lots. As it turns out, you can learn a fair bit about how these agencies responded to COVID-19 by looking at their parking lots.

For this blog post I analyzed satellite photos of the parking lots at CSE headquarters, CSIS headquarters, Canadian Forces Station Leitrim, NSA Fort Meade, and GCHQ Cheltenham. With the exception of CSE (which uses a parking garage for most of its parking), roughly the same pattern can be seen at all of these sites: a sharp reduction in parking lot use around late March 2020 as the first wave of the pandemic struck, greater but still reduced occupancy in May and June 2020, and a return to full lots by the end of the summer of 2020. There is very little evidence of reduced parking lot use during the winter 2020/2021 wave of the pandemic.

PARKINT complications

Before we get into all that, though, we need to consider the connection between parking lot occupancy and building occupancy.

The first thing to recognize is that very few buildings have enough parking spaces for everyone who works in the building. Most of these agencies maintain at least a small 24/7 operations capability, which means not everyone is in the building at the same time. And even on the main Monday to Friday day shift, some percentage of the workforce is typically expected to take public transit, walk, ride a bike, carpool, or otherwise get to work without taking up a space in the parking lot. In some cases the parking available on site is insufficient even for that lower level of demand, and some of the workforce ends up parking on neighbourhood streets, sometimes leading to local tensions.

A second complication is that there is no standard ratio between the number of people in a building and the number of parking spaces provided. Agencies whose sites are located far from most housing and are poorly served by public transit may provide parking for nearly everyone who works there. Those located in cities well served by transit, on the other hand, may insist that a large percentage of their workers leave the car at home. Even agencies located beside one another, like CSE and CSIS, may differ in the amount of parking they provide per employee.

Third, if a reduction in the number of people working in the office frees up parking spots, employees who ordinarily would not have driven may switch to their cars to take advantage of the availability of spots. This tendency is likely to have been especially strong during the pandemic, when many people will have wanted to avoid using public transit. As a result, the number of people occupying a building can probably drop quite significantly before the parking lot becomes less than completely occupied.

This also means, however, that if large vacancies do appear in the parking lot, it's a safe bet that a very substantial reduction has taken place in the number of people coming in to the office at that time.

What is more difficult to decide is whether those reductions reflect a switch to work at home or just a change in the specific hours of the day being spent at the office. Satellite photos are typically taken within a few hours of mid-day and it is rare to get more than one photo on any given day, so evidence of reassignment to other shifts is mostly indirect. The question can be answered in part, however, by checking whether significant changes have occurred in daytime attendance on weekends.

Finally, a significant part of the agency's parking may be provided by parking garages, which obviously pose a major problem for analyses based on satellite photos. As mentioned above, this was specifically a problem for assessing CSE.

Suitable imagery

Another problem is accessing suitable imagery. Satellite images like those available on Google Earth are typically very high in resolution, making individual vehicles easy to count, but such images are not updated nearly often enough. The latest Google Earth imagery for Ottawa, for example, dates from 2018. You can easily purchase more up to date imagery from commercial providers, but that option is not available to those who, like me, are working with a budget of zero.

Fortunately, there is a class of regularly updated, lower-resolution, free imagery available that is suitable for our purposes — if barely. At 10 metres per pixel, individual vehicles cannot be seen in Sentinel-2 images, but it is usually possible to tell the difference between occupied and unoccupied parking lots, as can be seen in these images of the lots at NSA headquarters. (See map showing the lots here.)

Even better is the 3-metre imagery collected by the Planetscope Dove satellites, which Planet Labs makes available to university-affiliated researchers through its Education and Research Program.

In principle, publicly available synthetic aperture radar (SAR) imagery could also be used to assess parking lot occupancy, and because SAR images are not dependent on daylight and thus can be taken at different times of the day, such images might be helpful in determining whether a significant part of an agency's workforce had switched to working at night. However, a brief survey of available Sentinel-1 SAR imagery did not turn up any images useful for this project.

Assessing the data

OK, so let's get on with it.

Stretching outwards from the main NSA headquarters buildings at Fort Meade, Maryland, is a vast expanse of parking lots covering around 30 hectares and containing roughly 10,000 parking spots. On normal weekdays, those lots are filled to full capacity, as demonstrated by the Planet Labs image below, taken on Monday, 16 March 2020, just as the pandemic's first wave was beginning to strike with force but before the U.S. government had begun telling its employees to stay out of the office.

By contrast, by the time the Planet Labs photo below was taken on Thursday, April 2nd, parking lot occupancy at NSA had plummeted by perhaps 80 percent, where it remained until roughly the end of June. This suggests that at least 8,000 (and probably actually many more) NSA employees, military personnel, and contractors who normally would have been in the buildings were told to stay home during this period.

Or maybe not to stay home, but instead to move from their normal daytime hours of work to different hours when fewer people would be in the complex. Like the other Five Eyes agencies, some parts of NSA run 24 hours a day, seven days a week, and thus there are always some vehicles in the agency's parking lot, but the overall number of shift workers is small in comparison to the day workers.

Imagery taken over the last year confirms that weekend parking lot occupancy has remained at its normal low level throughout the pandemic, indicating that there was no significant shift of Monday to Friday work to the weekends at NSA. However, the unusual distribution of vehicles in the lots during weekday images such as the one taken on April 2nd suggests that more than one large daily shift may have been used from Mondays to Fridays during the first wave of the pandemic. When parking lots are mostly empty you expect to see the vehicles that are there clustered around the entrances to the buildings, but as can be seen in the April image many are a considerable distance from the doors. This probably means there were already a lot of vehicles in the lot when the drivers of the ones seen in the image arrived to start their shifts. This pattern was evident in all the weekday images taken in the April to June period. According to this report, some elements of the U.S. intelligence community did adopt a two-shift day during the early months of the pandemic. It looks like NSA may have been one of those agencies.

By contrast, images from July and August show much higher occupancy in the NSA lot, perhaps 80%, which is still significantly below the pre-pandemic level but suggests that the workforce was back to a single main shift by this time. Weekday use of the lot increased further in September, rising to essentially full occupancy by the end of that month. It has remained there ever since, showing no reduction even during the peak of the winter 2020-21 wave of the pandemic.

As noted above, the relationship between parking lot occupancy and building occupancy is not straightforward. Despite the lot being full, occupancy of the buildings may still have been quite a lot below normal during this later period. It is safe to say, however, that no fewer than 10,000 people were in the complex during normal weekday hours during this period, and the number was almost certainly much closer to normal occupancy than that.

(What is that normal occupancy? If I had to guess, I'd say probably around 15,000, give or take a few thousand. But that is just a guess.)

Evidently, by the time the second wave was taking place, NSA felt that physical distancing measures and modifications to work stations and/or work practices were sufficient to enable a large percentage of its workforce to return to the office safely.


A broadly similar pattern can be seen at GCHQ's headquarters building, commonly called the Doughnut, in Cheltenham, U.K. The Doughnut is surrounded by about 7.5 hectares of parking containing around 3,000 parking spots (see map). Prior to the pandemic, all of those spots would be filled on a normal workday, as shown in the Planet Labs image on the left from Friday, 6 March 2020.

By the time the Planet Labs image on the right was taken, on Thursday, 26 March, parking lot occupancy had fallen to about 50%, which probably corresponded to a drop of more than 50% in the workforce in the building at any time. In mid- to late June we see parking lot use start to climb again, rising to around 80% in mid-September and perhaps 95% at the beginning of October.

Due to frequent cloud cover and low light levels, good imagery is somewhat sparse during the subsequent winter months, but the GCHQ parking lots appear to be 100% occupied no later than Thursday, 26 November, and they seem to have remained that way throughout the following months. Like NSA, there is no sign of a significant shift to weekend work at any point during the pandemic. Also like NSA, the fact that the GCHQ parking lots are back to full occupancy does not necessarily mean that the full workforce is back to normal work hours in the building. It is likely, however, that the great majority were back during most of this period.

Canadian sites

The headquarters of the Canadian Security Intelligence Service, located at the corner of Blair and Ogilvie roads in east Ottawa, has about 3 hectares of parking, but the odd shape of the lot limits its capacity to about 900 vehicles (see map). Planet Labs imagery from Monday, 16 March 2020 (left), shows the lot more or less fully occupied, but by Friday, 27 March (right), occupancy had fallen to roughly 60%, suggesting an even deeper reduction in the number of personnel in the building.

Occupancy of the lot remained at that lower level until the summer, when it began to rise again. By early July, up to 90% of the lot was typically filled, and since the fall it has been back to essentially 100% full, which may or may not mean that occupancy of the building returned to normal.

The response of the Integrated Terrorism Assessment Centre (ITAC), a multi-agency organization with offices inside the CSIS building, may give an idea of how the changes in parking lot occupancy corresponded to workforce attendance at the office. ITAC reduced the number of people working in its spaces by as much as 80% during the early days of the pandemic. By the summer of 2020, the number of people working in ITAC spaces was back to half its normal level, and by the fall, following renovations to improve the safety of the centre, three-quarters or more of the personnel were back. The reductions in the CSIS workforce may not have been quite as sharp as those of ITAC, but as the parking imagery confirms, it is likely that they followed a broadly similar trajectory.

[Update 29 March 2021: Stephanie Carvin confirms that CSIS headquarters was back to 80% of normal staffing by January 2021.]

The Edward Drake Building, the headquarters of the Communications Security Establishment (CSE), is located beside the CSIS headquarters, just to its west (see map). Most of the parking at CSE is provided by an 800-car parking garage, which of course largely eliminates the value of satellite imagery for analyzing parking at the agency.

Fortunately, not all is lost. CSE's garage is too small to accommodate all the people who normally want to drive their vehicles to work, so parking has tended to overflow into the residential neighbourhood to the west of the complex, sparking complaints by residents and enforcement actions by city bylaw officers. In an attempt to reduce this problem, CSE opened a 440-car overflow parking lot on Enigma Private (see map) just north of the CSE/CSIS complex in January 2020, about two months before the pandemic hit. As it is considerably further from the building, this lot is likely to fill up last — which opens the possibility of observing occupancy drops at CSE as well.

Complicating the issue, however, is that CSE was also in the course of moving most of the 800 staff members of the Canadian Centre for Cyber Security, CSE's cyber security arm, to a separate building at 1625 Vanier Parkway. Moreover, because many of those employees work on less classified, and sometimes even unclassified, projects, it has also been possible for a significant part of their work to be performed at home, freeing up space at the Vanier Parkway building for other CSE employees who do need office spaces to do classified work but do not necessarily need the highly secure SCIF spaces required by most of the SIGINT part of the agency.

Nonetheless, there are probably as many as 1,500 CSE employees or contractors who would seek to work inside the Drake building during normal Monday to Friday hours if they could. Unless a lot of those people are using public transit, that's a lot more than an 800-vehicle garage is likely to be able to accommodate. Thus, the use or non-use of the overflow lot may give some indication of limits on building occupancy during the pandemic.

And what do the pictures show? This Planet Labs image from 5 November 2020 is fairly typical: the overflow lot, visible at the top of the picture, does not appear to be in use (compare to the CSIS lot also visible). It is possible, however, that a few vehicles are present in the lot.

The bottom line is that there does not seem to have been extensive use of this lot by CSE during most of the pandemic. This suggests that CSE did manage to significantly reduce the number of people using the building during peak hours, although it doesn't tell us what combination of working from home, working in the Vanier Parkway building, or moving to different work hours was used to accomplish this, or how those measures may have varied over time.

To my mind, the most intriguing phenomenon turns up during the winter of 2020/21. By December 2020, after the snow starts to arrive, it is clear that the overflow lot is being plowed. This suggests it was in use at least somewhat by that time or at least that CSE expected it to be imminently in use. The plowing continues in January but then abruptly stops, with the lot appearing completely snow-covered for the last two-thirds of the month. The same pattern appears in February: plowed for the first third of the month and then a snow-covered wasteland for the rest. It gets plowed again at the beginning of March, and from that point on appears to be in consistent use.

January 2021 was the worst month to date for new COVID-19 cases in Ottawa, so it may be that the agency implemented additional peak-hours reductions in occupancy of the building during that month, and perhaps February as well, and thus didn't need the lot during those months. That theory doesn't explain why the lot was cleared in early February, however. Maybe the agency's snow-clearing contract specified a minimum number of days of work per month and the contractor plowed the lot until those days were used up whether the lot was in use or not.

Complicating analysis of this question is the fact that the winter imagery was frequently difficult to interpret, due to lower light levels, fewer clear days, and less contrast between snow-covered vehicle roofs and parking lots that themselves might have some snow on them. The CSIS lot seemed less affected by this problem, possibly because it is more sheltered from blowing snow.

What about the CFIOG workforce at the intercept station at CFS Leitrim? Satellite imagery shows there are around 500-550 parking spots at Leitrim, of which 350-400 were typically in use on pre-pandemic weekdays. There is little affordable housing near the station and it doesn't have good transit connections, so unlike the other sites discussed here, the number of cars in its lots is probably pretty close to the number of people working at the station at that moment. Imagery from the pandemic period suggests that the CFIOG reduced peak-hours staffing at the station by as much as 40% from late March to May 2020, with occupancy returning to 80% or 90% of normal levels only in the fall. This was probably mostly accomplished by moving people to non-peak-hours shifts in the evening and overnight, a change that presumably was easier to implement with the predominantly military personnel at Leitrim than it would have been at other sites. As with CSE, the winter imagery was often too poor for clear interpretation.

Interestingly, in no case is there any evidence that a significant amount of work was moved to weekends at any of these sites. Spreading five days of work across seven would seem like an easy way to reduce the number of people in the buildings at any time, but no, weekends appear to be sacrosanct.


Analysis of satellite photos of the parking lots at CSE headquarters, CSIS headquarters, Canadian Forces Station Leitrim, NSA Fort Meade, and GCHQ Cheltenham showed clear evidence of staffing changes at most of these sites in response to the COVID-19 pandemic. With the exception of CSE, where use of a parking garage complicates the question, roughly the same pattern was seen at all of these sites: a sharp reduction in parking lot use (implying even deeper reductions in peak-hour building occupancy) around late March 2020 as the first wave of the pandemic struck; greater but still reduced parking lot occupancy in May and June 2020; and a return to full lots by the end of the summer of 2020. There was very little evidence of reduced parking lot use during the winter 2020/2021 wave of the pandemic. However, the winter imagery was more difficult to interpret, particularly for CSE and Leitrim, so this observation is necessarily more tentative.

There are undoubtedly easier ways for intelligence agencies — and even individuals who aren't working from home on a zero-dollar budget — to answer these questions. For example, a couple of days of surveillance sitting in a car in the shopping centre lot across from the CSE and CSIS buildings would get you a much more accurate estimate of the number of people working in those buildings and their various hours of work. Commercially available smartphone location and activity data would probably also reveal a great deal, and the smartphone data potentially available to intelligence agencies could be even more revealing. Access to higher-resolution satellite imagery would also be very helpful.

Still, as this blog post shows, even relatively low-resolution satellite imagery can provide some intriguing insight into the ways Canadian and partner intelligence agencies responded to COVID-19.

This research was undertaken as part of my research fellowship with the Citizen Lab, at the Munk School of Global Affairs and Public Policy, University of Toronto. Planet Labs imagery was accessed with the assistance of Citizen Lab director Ron Deibert. All Planet Labs Imagery © 2021 Planet Labs Inc.