Defensible Probe-based traffic volumes, grounded in local realities: How to get started

You need traffic volumes on roads you don't count. Maybe it's a corridor study that covers 30 segments, but you only have counters on four. Maybe it's a safety analysis where half the network is missing AADT. Or a before-and-after evaluation where the "before" counts were never collected. Or you're facing federal reporting requirements like HPMS or MIRE and need defensible local AADT across more of your network than your count program currently covers.

Permanent stations and short-term deployments give you reliable numbers where they're installed, but they can't cover everything. Expanding a count program to fill those gaps means more equipment, more field crews, and more budget.

Probe data can help. It draws on anonymized connected-vehicle observations to estimate volumes across roads and time periods your counters don't reach. But a volume model is only as reliable as the data behind it. Generic models trained on national or statewide counts may work in aggregate, but they smooth over the differences that define your network: your seasonal patterns, commuter behaviour, road configurations. The only way to account for those conditions is to calibrate against ground-truth counts collected on your own roads.  

That's the problem local calibration solves, and it's what we do at SMATS. We build a model specifically for your jurisdiction, trained and validated on your permanent and temporary counts. It's a dedicated model that's recalibrated as new counts become available, so it improves as your count program grows and adapts as your network changes.

What local calibration changes for you

The core difference is defensibility. When volume estimates are trained on your own count data and validated against held-back observations from your network, you can point to the evidence behind them. That matters when you're submitting numbers for federal reporting, scoring a grant application, or justifying a project in a corridor study.

Estimates that hold up under scrutiny. Your road mix, seasonal patterns, freight corridors, and operating conditions shape the model directly. A locally calibrated model reflects what's happening in your area, across your road classes, urban and rural settings, and different times of day and year, rather than averaging over conditions from other regions.

A model that stays current. Traffic patterns shift as development, land use, and commuter behaviour change. The model is recalibrated as new counts come in, strengthening underrepresented areas and keeping estimates aligned with current conditions. It also shows you where additional field counts would improve confidence most.

Local calibration doesn't replace your count program. Physical counts remain the ground truth. They train, validate, and refine the model, and they provide direct confirmation where site-specific evidence is needed. What calibration does is extend the reach of those investments across roads and time periods your counters can't cover.

In practice: Maricopa Association of Governments

MAG needed volume estimates across a 10,000-square-mile planning area covering freeways, arterials, and intersections, far more than their existing count program could reach.  

Using locally calibrated models built from MAG's own ground-truth data, we generated bi-directional volume estimates at over 300 locations. The results: an overall R² of 0.98, average absolute error of 10.5%, and total estimated volumes within 1.5% of actual counts. Freeways hit R² ≈ 1.00 with just 4.3% error. Arterials, which naturally show more demand variability, still produced R² = 0.89. Time-of-day comparisons held within 1–4% across all periods.

MAG was pleased with the results, and the project has since grown to cover more local areas and road classes where count coverage was limited.

Read the full case study: [Click here]

How to get started  

The easiest first step is a conversation. Tell us what you need volumes for, share what count data you have, and we'll assess whether it can support a locally calibrated model. We'll identify coverage gaps, flag quality issues, and give you a clear picture of what's possible.  

From there, the process is straightforward. You share your existing count records, whether from permanent stations, automatic recorders, pneumatic tubes, radar, loops, video, or historical archives. We review them for completeness, quality, and location accuracy, and assess how well they represent your wider network. Then we build and validate a calibrated model using your data, with reporting that shows where the model performs well and where additional counts could improve confidence. Once complete, calibrated volumes are delivered through iNode, where you can access them by road segment, date range, time of day, and reporting interval.

You don't need modelling expertise or data-science resources on your end. We handle the calibration, validation, and delivery. As new count data becomes available or conditions change, we recalibrate.

If you want to prepare before reaching out, here's what helps:  

Review your counts. Look at how existing data is distributed across road classes, geographic areas, volume ranges, and seasons.  

Think about your use cases. Are you filling gaps for HPMS reporting? Supporting a corridor study? Building inputs for a travel demand model? The intended application shapes what outputs matter most.

‍Gather your records. Locations, dates, directions, intervals, and observed volumes. The more context, the faster we can assess coverage.  

Start with the data you already have. Contact us to review your traffic counts, identify coverage gaps, and map out next steps.

‍