Methodology
How we transform raw GPS observations into infrastructure-grade traffic intelligence.
Processing Pipeline
Six stages from raw signal to delivered estimate.
Data Collection
Anonymized GPS signals from consumer devices generate billions of location observations daily across CONUS. No hardware deployment required.
Normalization
Bias correction accounts for device penetration rate variation across geographies, demographics, and time periods. Raw counts are scaled to represent total traffic.
Aggregation
Observations are aggregated spatially to the road-segment level and temporally to hourly and daily intervals. All outputs represent groups, never individuals.
Confidence Scoring
Each estimate carries a confidence interval derived from sample size, local penetration rate, and temporal consistency. Downstream users can filter by confidence threshold.
Latency
Approximately 4 days from observation to delivery. Processing, quality checks, and normalization run continuously, not in annual batch cycles.
Vehicle Classification
Passenger vehicle and Class-8 truck volumes are separated using movement pattern analysis: speed profiles, stop patterns, route selection, and dwell behavior.
Confidence Scoring
Every Cohort Atlas traffic estimate ships with a confidence interval, so you always know the reliability of the data you're using. Higher sample density and temporal coverage produce tighter bounds — giving planners and engineers the certainty they need for funding decisions and safety analyses.
Example: AADT Estimate — Route US-50, Segment 4021
Per-Segment
Confidence intervals for every road segment, not just statewide aggregates.
Sample-Aware
Bounds tighten automatically as GPS sample density increases.
Decision-Ready
Know when data quality meets your project's statistical threshold.
Why Legacy HPMS Relies on Models
The traditional Highway Performance Monitoring System was designed around physical infrastructure constraints that no longer represent the state of the art.
Sparse primary sensors
Fixed-location counters cover a fraction of the network.
Periodic manual/automatic counts
Short-duration counts extrapolated to represent annual conditions.
Statistical expansion and smoothing
Models fill gaps between observed and unobserved segments.
Release cycles measured in years
Data reflects conditions 12-18 months prior to publication.