USU researchers tap machine learning, game cameras to inform recreation area management

LOGAN — Your bag is packed, car is loaded and you're finally on your way to one of Utah's many recreation areas.

Anticipation builds throughout the drive. You likely have a certain climb, river, lake, trail or general area in mind, and you don't want to settle for anything else. When you arrive, the excitement falls off a cliff upon the realization that there's nowhere to park.

The lot is filled, the pull-outs on the side of the road are filled, and you can either compromise or head home. Unfortunately, it's a reality everyone who recreates throughout Utah and the West understands all too well.

It's also a problem that has long plagued land managers faced with the difficult decisions on how to best manage popular recreation areas for the near-term while also being mindful of future growth.

But to help get to the root of this conundrum, researchers at Utah State University's Institute of Outdoor Recreation and Tourism Management combined run-of-the-mill game cameras and machine learning tools to answer some of the more pressing questions related to managing recreation sites.

"None of those (land management) agencies really have a good understanding of what's actually happening at the parking location. That's really what we're trying to solve here," said Jordan Smith, the director of the institute. "This is the critical pitch point for so many of our outdoor recreation experiences: getting people to the places so they can go launch into the experiences that they want to have."

Smith and Chase Lamoborn, the assistant director of the institute, partnered with the U.S. Forest Service to test their approach along the heavily used Payette River recreation corridor in Idaho.

To do so, they used high-resolution aerial imagery from the National Agricultural Inventory Program to map where parking was available across the corridor.

But measuring capacity is only part of the equation, Smith said.

Getting an accurate idea of how many people were using the recreation sites and how close parking locations were actually getting to capacity required installing game cameras at recreation sites along the river to photograph activity over time (without collecting personal or identifying information, Smith pointed out).

Computer vision tools — algorithms that identify and count objects in images and video — were trained and processed nearly a half-million images to identify the presence of vehicles in parking areas and watercraft on the river.

This information, Smith said, gives land managers the ability to "triage" their actions with the knowledge of the current supply and demand for a particular recreation site, as well as what's possible at a site, considering the potential supply.

Essentially, it could tell managers where people are concentrated, when visitation is at its peak, and where bottlenecks happen — whether they be on the road, trail or river.

"Both of the processes, both the supply and the demand, are driven by these machine learning algorithms that, when we put them together, are able to solve a very practical problem for land managers," Smith said.

Utilizing machine learning for the monitoring system is also much more efficient, practical and inexpensive than it would be through traditional methods, Lamborn said.

Think a graduate student collecting counts of people on a trail or vehicles in a parking lot. It could be helpful, but it likely only captures that moment in time without the monitoring aspect.

"What this is allowing us to do is to collect large amounts of data, and then being able to process it. We have had projects in the past where we'd used game cameras, but a lot of it was having graduate students click through photos and then entering into a spreadsheet what those pictures were showing us," Lamborn said. "Now, our ability to process massive amounts of information, very inexpensively, is what really is the innovative side of this."

Smith put it plainly, saying what used to take a graduate student an entire master's thesis to complete can now be done in a few hours.

Still, they were adamant that the new approach improves the product without replacing people.

"We hire the same number of staff, often students, to provide the same hands-on training and professional experience within roughly the same budget. The difference is that we can now collect and process far more information than was previously possible. It gives our funding partners better information, while still providing valuable work experience," Lamborn said.

Moving the research from concept to practice, Lamborn said the institute is launching a project with Cache County to help their "very limited" staff better understand how recreation occurs throughout the county.

"These types of projects have pretty much been out of reach to them because of the cost. But now, you know, we can hire some undergrad interns, use these methods and provide a pretty good picture of what is occurring," Lamborn said. "At a fundamental level, we're really trying to provide a cost-effective service to any outdoor recreation manager or provider."

Additionally, the team launched a Recreation Capacity Platform to share the approach more broadly. The platform combines aerial maps, camera counts, and environmental conditions into intuitive digital dashboards.

"As outdoor recreation participation continues to grow, the public needs smarter systems to know when and where to have the kind of recreation experiences that they want," Smith said.

Related topics