Ground vibrations recorded by fiber-optic cables reveal traffic response to COVID-19 lockdown measures in Pasadena, California

Transportation activity reflects societal behavior across multiple temporal and spatial scales, making traffic monitoring a valuable proxy for assessing societal dynamics. The COVID-19 pandemic and associated public health measures, including social distancing and statewide lockdowns, drastically altered daily life, work, and mobility. Traditional traffic monitoring approaches include stationary sensors (radar, counters, embedded roadway sensors, cameras) that provide high-resolution data but are spatially sparse due to cost, and onboard GPS or mobile phone location data that offer broader coverage but suffer from low sampling rates, potential biases, and privacy concerns. Distributed acoustic sensing (DAS) can transform standard single-mode optical fibers into dense arrays that capture ground vibrations from vehicles with meter-scale spatial and sub-second temporal resolution over tens of kilometers. Leveraging two strands of unused 37 km telecommunication fiber in Pasadena, CA as a city-wide DAS array of ~5,000 channels, the study aims to quantify changes in traffic volume and mean speed from December 2019 through August 2020 and assess how traffic patterns responded to COVID-19 lockdown measures across multiple scales.

Prior work on traffic monitoring spans stationary sensors and mobile device-based systems, each with limitations related to spatial coverage, sampling rates, costs, privacy, and demographic biases. Telematics-based mobility reports (e.g., Apple and Google) showed large declines in movement during early lockdowns, but lack direct vehicle counts and fine spatiotemporal resolution. Independent seismological studies documented a global reduction (~50%) in high-frequency anthropogenic seismic noise during lockdowns, though such measurements depend strongly on station siting and local geology. DAS has emerged as a tool to capture traffic-related signals, including quasi-static deformation from vehicle loading and higher-frequency surface waves, and has been used to estimate vehicle counts and speeds. A related city-scale dark-fiber DAS study in Palo Alto during early lockdowns demonstrated feasibility but was limited to a few locations and provided only volume estimates using template matching. The present study extends this literature by estimating both traffic volume and mean speed city-wide, enabling congestion analysis and multi-timescale assessment over a longer period.

Data were acquired using the Pasadena DAS array: two strands of ~37 km unused telecommunication fiber instrumented as a DAS array with ~8–10 m channel spacing, totaling ~5,000 channels, continuously recording from December 2019 to August 2020 (>50 TB). For analysis, the continuous time series were divided into 10-minute segments with 25% overlap. Standard preprocessing removed linear trends, mean, and common-mode noise. A 0.25–2.0 Hz bandpass was applied, as spectrograms showed clear traffic energy in this band. The array was partitioned into overlapping subarrays of 21 channels sliding every 5 channels. Quality control excluded channels whose cross-correlation with adjacent channels was <0.5. Vehicle speeds were estimated via 4th-root slant stacking over 15–80 mph (24–128 km/h). Rather than tracking individual vehicles, the approach emphasizes statistical estimation per 10-minute window: local maxima in beamformed outputs above average amplitude were used to estimate vehicle counts and mean speed. To mitigate undercounting during closely spaced vehicles (e.g., rush hour), the duration of signal above a threshold (median value of minimal identified peaks) was converted to traffic volume using an empirically derived average single-vehicle duration. Traffic parameters were computed across the city by sliding subarrays along the fiber. Validation used Department of Transportation (Pasadena) Traffic Count Database System data at sparse locations and dates (2010–2020). Pre-lockdown DAS data (Dec 1, 2019–Mar 19, 2020) were aggregated into mean working-day diurnal profiles at nearby DAS channels for comparison at 10 representative DOT locations. Agreement was generally good, with discrepancies attributed to differing count dates, long-term trends, short-term variability (e.g., weather), and dark-fiber installation constraints (coupling, cable position, multi-lane roads versus a single roadside fiber).

• City-wide, traffic volume declined and mean speed increased during the COVID-19 lockdown compared to pre-lockdown levels.
• On E. Colorado Blvd. (arterial street subarray), pre-lockdown average daily working-day traffic volume was ~24,500 vehicles. After rising COVID-19 cases and the March 19, 2020 statewide order, volume dropped abruptly by ~40% relative to pre-lockdown, with the decline starting several days before the order. Traffic remained low for ~3 weeks, then gradually increased from mid-April. By August 1, 2020, volume was ~75% of pre-lockdown.
• Weekly patterns shifted: pre-lockdown, volumes were stable Monday–Saturday with a Sunday drop; post-lockdown, peaks appeared Thursday–Saturday, especially pronounced in early April.
• Diurnal patterns: volume decreased across all hours post-lockdown, with the largest reductions (~65%) during morning and evening (commute periods); daytime drops averaged ~20%. Peak-hour estimates may be underestimated due to method saturation during heavy congestion.
• Spatial heterogeneity in volume changes: city-wide average daily traffic volume decreased by ~15% compared to pre-lockdown, but effects varied by road.
  • Near university (Caltech): average ~50% decrease, up to ~65% across the day.
  • Arterial streets: average ~30% decrease, with up to ~50% reductions in morning and night.
  • Busiest roads showed little decline, indicating continuity of essential activities.
  • Near the main hospital, traffic increased, particularly during daytime, consistent with increased demand.
• Mean speed changes: minimal on streets already near or slightly above speed limits; substantial increases where pre-lockdown congestion was common. Near Huntington Hospital, mean speed rose from below the speed limit to near the limit post-lockdown.
• Speed–volume relationships indicate reduced congestion city-wide during lockdown, with flows moving closer to uncongested, higher-speed conditions.
• DAS-based results are consistent with DOT measurements and mobility reports, while providing objective, high-resolution vehicle counts and speed estimates with extensive spatial coverage.

The study demonstrates that DAS can provide objective, high-resolution measurements of fundamental traffic parameters (volume and mean speed) across an urban area, enabling precise quantification of the multi-scale impacts of COVID-19 lockdown measures. The observed sharp drop in volume, altered weekly and diurnal patterns, and localized increases (e.g., near a hospital) validate that societal responses and essential activities are reflected in traffic behavior. Increased mean speeds in formerly congested areas confirm congestion relief during lockdown. Compared with telematics reports and high-frequency seismic noise proxies, the DAS approach offers real vehicle counts and fine spatiotemporal detail without privacy concerns, thereby overcoming biases and limitations of alternative datasets. Agreement with DOT counts and mobility trends supports reliability. Relative to earlier DAS studies limited to a few locations and volume-only estimates, this work delivers city-wide mapping of both volume and speed, enabling congestion analysis and assessment over extended periods, which is valuable for transportation planning, public health assessment, and economic recovery monitoring.

Using a 37 km city-wide DAS array on existing telecommunication fibers in Pasadena, the study quantified traffic volume and mean speed before and during COVID-19 lockdown, revealing a general decline in volume and an increase in speed, with strong spatial and temporal variability. The largest volume decreases occurred near university areas, modest decreases on arterials, minimal changes on some busy roads, and increases near the hospital; speed increases were concentrated in previously congested corridors. These results highlight DAS as a scalable, affordable, and continuous traffic monitoring solution with implications for public health, economic activity assessment, and transportation safety and efficiency. Future work could improve volume estimation during peak congestion by incorporating amplitude and higher-frequency signals, develop more sophisticated algorithms (e.g., machine learning) for vehicle detection and classification, and integrate DAS with conventional sensors for enhanced calibration and multimodal monitoring.

• Potential underestimation of traffic volume during peak congestion when vehicles are closely spaced; local maxima-based detection and duration-based corrections may saturate.
• Validation relied on non-concurrent DOT counts at sparse locations and varying dates, introducing differences due to long-term trends and short-term variability (e.g., weather, events).
• Dark fiber installation characteristics (coupling quality, cable position relative to lanes, roadside placement along multi-lane roads) can affect sensitivity and representativeness.
• The study focused on aggregate volume and mean speed rather than individual vehicle tracking; incorporating high-frequency signals or advanced methods may better capture amplitude information and improve accuracy.
• Lack of concurrently operated stationary sensors across the entire study period limits direct one-to-one validation.