Northwest Arkansas has built one of the most talked-about trail networks in the US, anchored by the 40-mile Razorback Greenway and dozens of paved multi-use trails. But counting who rides them is harder than building them. Fourteen permanent counters cover two counties, and adding more is expensive. That leaves planners with limited evidence on where ridership is highest, where to prioritize maintenance, and where to expand.
Researchers at the Center for Advanced Spatial Technologies (CAST) at the University of Arkansas set out to close that gap. Their study combines Strava Metro data from 2019 to 2023 with ground-truth counts from 12 permanent EcoCounter sites. They added weather, demographic, topographic and infrastructure data, then trained and compared three modeling approaches. To test how well each one would work where no counter exists, they held out each counter site in turn and asked the model to predict it blind.
The best performer was a Random Forest model, with a median error of 18.5% across the 12 sites. Eight of the 12 were predicted within 25% of their true ridership. Strava Metro activity was the single strongest predictor, accounting for more than half of the model's explanatory power. A simpler ratio-based model offered an easy-to-explain cross-check, with a mean error of about 40%. Training on several years of data clearly beat calibrating on a single year.
The result is an estimate of average annual daily bicycle counts for all 13,163 segments in the network, published through ArcGIS Online. The highest volumes cluster along the Razorback Greenway and its feeder trails connecting Bentonville, Rogers and Fayetteville. About half of all segments see fewer than 30 riders a day, while the busiest fifth carry most of the activity. That helps planners focus maintenance on the corridors that matter most and supports grant applications that need ridership documentation.
The study is clear about what the estimates are for. They are planning-grade, suited to ranking corridors, forecasting demand and before-and-after comparisons, not a substitute for site-specific counts where precise volumes are needed. The four sites the model predicted least well share a common profile, with lower Strava participation relative to total trail use. The authors recommend keeping all existing counters, adding two to three new ones at high-Strava, uncovered locations, and trying rotating short-term counters. Quarterly deployments of two to three weeks matched the performance of a continuous four-month window with about 30% fewer counter-weeks.
Accuracy also improved each year, with median error falling from 37% in 2019 to 20% in 2023.
The team plans to refresh the model annually as new counter data and Strava Metro releases arrive.
Read the full study HERE (Brian Culpepper, Katie Wyatt and Jack Cothren - 2026)
Center for Advanced Spatial Technologies (CAST), University of Arkansas
