We ran out of time to live answer all the great questions posed by the audience at this month’s State of the Science presentation by Aaron Hipp and Katie Burford of North Carolina State University. They graciously agreed to give us responses by email.

How feasible do you think using Placer data would be for community organizations/local governments or other practitioners to justify funding and demonstrate impact of environmental changes?

We have spoken with several local governments/practitioners (e.g., Alamance County, NC, City of Raleigh-associated employees, KABOOM!) who are using or plan to use these data (mobile phone location-based) in their park-related efforts. For example, see this grey paper from KABOOM! about the use of location mobile device data to measure the impacts of playground renovation efforts. They intend to expand this to other communities. We learned from this report and speaking with KABOOM! that they use these findings within dissemination materials (e.g., on their website, presentations) to seek out new opportunities. The EPA and NRPA also have recent reports.

Parks and Recreation Departments, systems,  and agencies small to large are definitely using mobile phone location data. This gives them a sense of true use of their parks – especially comparing across parks in the same system. The data also provides information on seasonal and weekly variations. We have been told they are using this to request additional funding and make maintenance prioritization decisions.

Specific to environmental change, there have been a few reports and papers that look at the use of larger parks (e.g., national parks) and use of spaces, especially backcountry. This may get at the potential human impact on more natural spaces.

Do you think that there may be some possible use of cameras and machine learning to count people and possibly classify activity? For instance like trail cameras but maybe ‘park cameras’.

Yes! This method is being explored by some of our academic collaborators at the World Playground Research Institute. The team is specifically pairing camera-based observations with AI to overcome the many challenges of direct observation and individual device measurement. You can find out more information here, and check back for updates on our progress as we determine the accuracy of this method for measuring park and playground-based use and physical activity.

Jordan Carlson has also worked in this space. You can read about this work here and here.

Thank you for a great presentation! Observed race and ethnicity data have long been recognized as problematic. For example, someone who appears ‘brown’ may not self‑identify as Hispanic/Latino, and someone perceived as ‘white’ may not identify as being of European descent. This has been a persistent challenge in tools like SOPARC, which rely on observer-coded categories. Now, with the growing use of AI—and the well‑documented risks of algorithmic bias—how can we more responsibly address these forms of misclassification and the broader biases they introduce into observational research? (Also, now I am craving BBQ).

Great and tough question. We decided to not present the results of the race and ethnicity data that we collected using SOPARC for the reasons you stated. I hope we also made it clear that Placer.ai does not measure or provide race and ethnicity data of the mobile device user. It is unclear if these data are collected by the data providers of the cell phone data.

But where we agree this would be a concern is in the use of large language models (LLMs) for determining the race or ethnicity of park or playground users from video camera data. We have yet to label, train, and test an LLM for classifying the race or ethnicity of park users from camera data ourselves to know the potential misclassification. We believe this is a similar concern as with SOPARC observational data. And similar to our decision with SOPARC data, we plan to stay clear of using AI data for measuring race and ethnicity given the current inaccuracy of existing LLMs. Instead, we will continue to utilize census estimates of neighborhood data or seek to measure participant-reported data. These have limitations of course, but align with data equity principles.