Sage Journals: Discover world-class research

Abstract

Cyclist noise exposure has implications for health, comfort, and safety. Methods used for in situ measurement of on-road noise levels for cyclists vary, and the effects of key study design factors have not been investigated. To enable reliable research into cyclist noise exposure, this study aims to determine the accuracy of smartphone noise measurements in comparison with a sound level meter (SLM) reference instrument, and how noise levels are affected by travel speed, air speed, sensor placement, and use of a windscreen. Field data were collected with paired instruments in a typical urban cycling scenario, and comparisons made varying one design factor at a time (smartphone versus SLM, with versus without windscreen, handlebar versus shoulder placement, etc.). Results show that smartphones can generate reliable measurements (compared with SLM) of high-resolution (1-s) cyclist exposure for C-weighted noise, but not A-weighted noise. Sensor placement and windscreen have small effects on noise readings, but air speed and travel speed greatly affect measured noise levels. Future studies measuring on-road noise must consider the effects of wind- and bicycle-generated noise to ensure internal validity. Studies should also consider both study objectives and instrumentation when selecting a noise measure (frequency weighting). Research is needed into bicycle noise generation and perception of traffic noise by cyclists to enhance the reliability of future studies.

Keywords

pedestrians bicycles human factors planning and analysis highway noise pedestrian and bicycle data

Promoting safe and comfortable cycling has become a goal of governments aiming to reduce auto dependence and combat climate change, congestion, obesity, and other challenges. To support this goal, a body of research has developed over several decades examining cyclist exposure to motor vehicle traffic and hazardous environmental conditions, the effects of those exposures on key outcomes (comfort, satisfaction, cycling activity, health, etc.), and strategies to mitigate exposures (protected facilities, networks, etc.). This research uses a variety of methods to quantify exposure, usually either estimation from spatial exposure data or in situ measurement with instrumented cyclists.

In the context of this research, on-road noise (sound level) measurement for cyclists can serve several types of investigation. Noise exposure itself is an environmental hazard commonly targeted for mitigation in highway design ( 1 – 4 ), although adverse health effects from noise exposure during cycling specifically have not been established. Traffic noise is also a source of annoyance and discomfort, and noise exposure has been investigated with an interest in the effects on travel behavior ( 5 – 7 ). Additionally, because motor vehicle traffic is a main source of on-road noise, noise measurements have been used as a proxy for exposure to other effects of motor vehicle traffic, particularly crash and air pollution risks ( 8 – 10 ).

An important advance in the tools of cycling research is the availability of increasingly powerful, sensor-laden, nearly ubiquitous smartphones. Smartphones provide transportation researchers with an accessible means to couple detailed location information with data from the phones’ internal sensors (accelerometer, barometer, microphone, camera, etc.), paired instruments with additional sensors, and traveler-reported input ( 11 – 13 ). Particularly for active travel research, smartphones enable researchers to increase data resolution and accuracy while decreasing participant burden and potentially observation bias.

The objective of this study is to facilitate robust consideration of noise exposure in cycling research. Although several studies have reported smartphone-based cyclist noise exposure, the method has not yet been validated. We seek to evaluate the accuracy of on-bicycle smartphone measurements of noise in comparison with reference instruments, and also to investigate other instrumentation issues for collecting reliable cyclist noise exposure data in situ: the impacts of air speed, travel speed, sensor placement, windscreens, and sound level weighting methods on noise measurements.

Literature Review

Cyclist Noise Exposure

Variations in noise exposure occur over both time and frequency spectra, and there are several methods of aggregating raw data on sound pressure levels to noise readings. Temporal aggregation is most often time-averaged, but alternative measures include minimum, maximum, and peak values within a time interval. Other temporal adjustments translate noise samples into long-term environmental exposures. The sound pressure level across the frequency spectrum at one moment in time is typically averaged using Z-weighting (equal across all frequencies) or A- or C-weighting (weighting curves designed to represent human hearing at low and high sound levels, respectively). Different measures provide relevant information for different types of application ( 4 , 8 , 14 , 15 ).

A recent systematic review identifies 24 published studies reporting cyclist noise exposure levels ( 10 ). As a few examples, a study in Montreal, Canada measured 1-min mean cyclist noise exposure of 71 dB-A (A-weighted noise level in decibels), ranging from 55 to 88 dB-A, using an SLM (Brüel & Kjaer Type 4448) ( 16 ). Subsequent studies using the same SLM in Ho Chi Minh City, Vietnam ( 17 ), Paris, France ( 18 ), and Copenhagen, Denmark ( 19 ) reported 1-min mean cyclist noise exposures of 79, 72, and 68 dB-A, respectively. The study in Paris ( 18 ) explicitly reported using a windscreen and placing the device at the cyclists’ shoulders, whereas other studies do not give full instrumentation details. A study by a different group of researchers in Montreal, Canada used a Brüel & Kjaer Type 4448 SLM to measure 1-s noise levels at a cyclist’s ear level, reporting mean exposure of 73 dB-A ( 20 ). These noise levels are comparable to a study in Chengdu, China that recorded 1-s dB-A using a NoisePro DLX Dosimeter SLM, reporting mean cyclist noise exposure of 74 dB-A ( 21 ). Several of these studies modeled relationships between cyclist noise exposure and environmental factors, reporting significant effects for facility type, land use, building density, parks, and ambient wind speed among other factors ( 18 , 20 ). A wind tunnel study of only wind noise reported noise levels of around 85 to 89 dB at cycling-relevant wind speeds of 16 to 24 km/h, increasing proportionally with wind speed at rate of around 0.4 dB per km/h ( 22 ).

Relationships Among Exposures and Outcomes

Cyclist noise exposure is most often assessed as a component of multi-pollutant studies ( 10 ). Various studies have investigated the relationship between noise and air pollutant concentrations during cycling, partly to determine whether noise measurements can be used as a proxy for other exposures ( 23 , 24 ). Noise as a proxy for air pollution is an attractive option because the instrumentation required for on-bicycle noise measurement is generally easier to use and lower-cost than pollutant-specific research-grade air quality monitors. Understanding cyclist air-pollution risks requires dynamic on-road measurements because on-road exposure is poorly represented by background urban air-quality monitoring data ( 10 , 25 ). In addition, cyclist intake and uptake of air pollutants is highly dependent on interaction between dynamic breathing (ventilation) and exposure levels over the course of a trip, and so time-resolved data are critical for accurate quantification ( 25 – 27 ).

A recent review summarized the evidence for noise–air pollution relationships as: “the majority of studies report low correlations (<0.25)…and in more rare cases, average correlations (<0.6)” ( 10 ). An early study on relationships among exposures on predefined cycling routes in the Netherlands measured A-weighted noise using a Rion NA-26 sound level meter (SLM) with the microphone placed behind the cyclist at ear level ( 24 ). They report moderate correlation between 1-min averaged noise and particle number concentration (PNC), but low correlation for 1-s averaged noise and PNC, and no correlation between noise and fine particle concentration (PM_2.5) at either aggregation.

Dekoninck et al. ( 23 ) investigated the relationship between noise level and black carbon (BC) measurements with a Svantek 959 SLM mounted on the bicycle handlebars. Rather than the common A-weighting, they investigated alternative noise measures (low frequency, high frequency, and deviation) as indicators of engine noise, tire noise, and passing vehicles to use as a proxy for on-road BC exposure. A subsequent study ( 8 ) found noise measurements to be more temporally stable than BC measurements, recommending the use of a low-frequency noise measure (requiring 4 to 10 repeated measurements with 2 dB of accuracy) to map long-term (annual average) BC exposure levels.

Outside of air pollution relationships, a European study compared A-weighted average noise for cyclists to physiological stress indicators (skin conductivity) ( 28 ). They found a 2% increase in the probability of measuring a “stressful event” for every 1 dB-A increase in noise. However, a related study in São Carlos, Brazil found only a small, non-significant effect of loud noises (over 75 dB) on stress responses ( 29 ). A recent study in the United States compared A-weighted average noise (as a proxy for traffic exposure) to cyclist safety but found no significant relationship ( 9 ).

Smartphone Noise Measurement

Although most cycling studies employ professional SLM instruments, some have used smartphones for noise monitoring, with either the internal microphone or an attached external microphone ( 9 , 30 , 31 ). A recent U.S. study on noise-safety relationships ( 9 ) used smartphones for noise data collection, and comparison with a professional SLM for a small validation data set (just 11 observations) showed that smartphone readings were significantly higher (by 7 dB-A), but well correlated (Kendall rank correlation of 0.88). The authors noted that the difference may be partially attributed to the foam windscreen and recommended further validation of smartphones for cyclist noise exposure measurement.

Other cycling studies have not evaluated smartphone noise reading accuracy. A study on environmental monitoring with smartphones did include a subset of noise data collected while cycling ( 32 ). A small (single-trip) data subset was used to compare smartphones with and without an external mic (with a windscreen) to an SLM, generating correlations with the SLM readings of 0.79 for the smartphone with the external microphone and windscreen, and 0.68 for the smartphone with only the internal microphone.

Several studies from outside of the transportation research field have investigated the accuracy of smartphone-based sound level measurements. Regulatory and testing guidance generally indicates a required precision of around 1 to 2 dB for SLM, depending on the frequency, application, and other factors ( 15 , 33 ). In 2013, Kardous and Shaw ( 34 ) tested 14 smartphone applications on nine devices in a laboratory setting and found mean differences from reference SLM ranging from −12.2 to 6.8 dB (unweighted) and from −13.2 to 3.6 dB (A-weighted) across devices and applications. Performance varied substantially across 168 testing conditions (around ±8 dB). A follow-up study tested the four best-performing applications in iPhone devices (with mean error under 2 dB) with calibrated external microphones, again in a sound laboratory setting ( 33 ). They found substantially improved performance, with mean errors under 1 dB, and errors within ±2 dB across testing conditions.

A subsequent study tested a custom, calibrated smartphone application with an external microphone and showed further improved performance and compliance with IEC and ANSI testing standards ( 15 ). The application was tested against an SLM in five community settings (a coffee shop, office, restaurant, commuter train, and spin class) using iPhones without calibration or external microphones, and the results showed low bias (<2 dB-A mean difference) in the three loud (>75 dB-A) locations, but smartphone over-estimation of noise in the coffee shop and office, and individual sample (∼45 min each) differences that ranged from −22 and 10 dB-A ( 35 ). These studies showed that smartphone noise measurement can be accurate, but smartphones have yet to be validated for ambulatory (on-road) use, or in varying air flow (wind) conditions.

Summary and Research Questions

In summary, cyclist noise exposure is an active area of research, with implications for health, comfort, and safety. But the methods used for in situ measurement of on-road noise levels for cyclists vary, and the effects of different instrumentation and study design decisions (such as the frequency weighting method or sensor placement) have not been determined. In addition, the reliability of using smartphones for cyclist noise exposure measurement is unknown. Laboratory studies have shown that smartphones can provide accurate sound level readings in that environment, but limited evidence suggests they may not in other settings.

A few key sources of potential systematic measurement error for cyclist noise exposure have received little attention in the literature. First, the bicycle itself generates noise during travel, which may vary with travel speed and roadway surface. In addition, air flow (the combined effects of local ambient wind, the cyclist’s speed, and traffic-induced air flow/turbulence) generates noise, which may or may not be included as part of the intended study framework. Wind noise is predominantly low frequency, and so the effects of wind/air flow on noise readings likely depend on the frequency weighting method ( 36 ). Little is known about bicycle-generated noise, but it is likely unevenly distributed across the frequency spectrum, and so the impact would also depend on noise weighting.

Motivated by these substantial gaps in existing knowledge that impede robust research on cyclist noise exposure, this study aims to address the following questions: (1) how accurately can smartphones measure cyclist noise exposure (in comparison with SLM); (2) how does accuracy vary by frequency weighting method and temporal aggregation; and (3) how is noise accuracy affected by travel speed, air speed, sensor placement, and use of a windscreen? The answers to these questions will help inform future studies on cyclist noise exposure, in addition to further research on improving noise instrumentation methods.

Methodology

The study objectives are addressed using in situ paired-instrument contrasts for a typical urban cycling scenario. On-road noise level readings are compared for two instruments in the same environment, varying one design factor at a time (smartphone versus SLM, with versus without windscreen, handlebar versus shoulder placement, etc.). We include A and C sound level weightings because both are readily available from SLM and smartphone applications, they are expected to be affected differently by wind-, bicycle-, and traffic-generated noise, and cycling studies are interested not only in sound levels pertaining to human hearing (represented by A-weighting) but also sound levels as a proxy for motor vehicle traffic (generating low-frequency noise better represented by C-weighting). Note that the study objective is not to study cyclist noise exposure per se, or its relationship with environmental factors, but rather to study the impacts of instrumentation decisions. This will generate the knowledge needed to inform study methods for larger (multi-season, multi-rider, multi-area) data collection campaigns.

Field Data Collection

Noise data were collected simultaneously using both smartphones and professional sound level meters (SLM) mounted on the handlebars of a bicycle (Gitane Verso Sport) along with an anemometer (Kestrel 5500 Weather Meter). The data collection set-up is shown in Figure 1. The two smartphones (Apple iPhone 11 and XR, respectively) were used with the “Gauges” version 4.2.2. and “Decibel Meter Sound Detector” version 2.6 applications to record 1-s location and speed (from GPS data) and A- and C-weighted noise levels (from the internal MEMS microphones). The two SLM (TSI Quest SoundPro DL) were calibrated before each data collection following the manufacturer’s instructions.

Figure 1.

Instrument set-up for field data collection.

To test for windscreen effects, one SLM had a windscreen while the other did not. The two smartphones were used to test average A-weighted and C-weighted noise levels, respectively. To test for the effect of sensor placement on noise readings, a second instrument set-up was used in which the microphone from one of the two SLMs was located at the cyclist’s shoulder (see Figure 1) while the other remained at the handlebars (neither with a windscreen). A 24-km route was ridden by a single researcher in metropolitan Vancouver, Canada on three fair-weather days June 2021. The route included a typical combination of urban cycling facilities (on-street bicycle lanes and off-street/multi-use paths) and was mostly flat to allow for consistent pedaling as the experimenter rode at their preferred travel speed.

Data Analysis

Data log files from each instrument were merged by timestamp. Because anemometer data could only be logged at 2-s intervals, 1-s air speed was interpolated from the adjacent observations. Observations without complete information from all four sound level instruments were removed.

Mean percent difference (MPD) between paired instruments is reported to represent the measurement bias between two instrumentation methods, while mean absolute percent difference (MAPD) represents the systematic distance between the methods. In both cases, percent difference is calculated using a mid-point reference (i.e., $\frac{A - B}{(A + B) / 2} \times 100 %$ ). Linear correlation coefficients are reported to represent the extent to which the readings from two instrumentation methods co-vary.

Relationships between noise measures and travel and air speed are investigated using univariate regression analysis. To account for the serial correlation in the time-series data, first-order autoregressive error is included in a generalized least squares model specification. All statistical tests use a p < 0.05 significance threshold. Data and regression analysis was undertaken in the R software environment, using the “nlme” package ( 37 , 38 ).

Results

We first present in Table 1 a descriptive summary of the on-road noise and speed data collected, followed by the results of instrumentation comparisons in the following sub-sections. After data cleaning, 18,311 complete 1-s observations remained, representing 305 min of data. The noise level readings ranged up to 120 dB, with central values around 75 dB. The temperature during data collection averaged 25°C (range 18–34), with mean barometric pressure of 1,014 mbar (range 1,010–1,018) and mean relative humidity of 47% (range 30%–83%), according to Kestrel instrument measurements. The observed travel speeds encompass an appropriately wide range of dynamic speeds, although the mean travel speed of 10 km/h is at the low end of typical trip speed distributions, which average around 15 to 18 km/h with a standard deviation of 5 km/h ( 11 , 39 ).

Table 1.

Summary Statistics

Measurement	1st quartile	Median	Mean	3rd quartile	Maximum
Travel speed (km/h)	0.0	10.7	9.7	15.7	37.8
Air speed (km/h)	3.3	9.9	9.3	14.0	31.8
Smartphone (dB-A, average)	63.7	71.6	69.7	77.2	96.1
Smartphone (dB-C, average)	71.9	76.6	74.4	78.9	87.5
SLM, unscreened (dB-A, average)	61.9	66.2	65.9	70.5	106.9
SLM, unscreened (dB-A, peak)	77.3	82.5	82.5	87.4	120.4
SLM, unscreened (dB-C, average)	73.2	77.9	77.2	82.5	106.4
SLM, unscreened (dB-C, peak)	85.8	90.9	90.3	95.6	119.8
SLM, screened (dB-A, average)	61.8	65.6	65.6	69.9	102.5
SLM, screened (dB-A, peak)	76.7	81.7	82.2	87.2	117.8
SLM, screened (dB-C, average)	71.7	75.1	74.4	77.9	102.9
SLM, screened (dB-C, peak)	83.9	87.7	87.6	91.3	119.3

Note: SLM = sound level meter.

Impacts of Windscreen and Sensor Placement

Table 2 gives the results of comparing synchronous noise readings between the paired SLM with and without a windscreen in the first four rows. Neglecting a windscreen inflates C-weighted noise by 3 dB (4%) but has a smaller effect (<1%) on A-weighted noise. Noise readings with and without a windscreen still highly correlate, and the MAPD is ≤6% for all four measures, although the effect on variability is greater for peak than average noise measures. Figure 2 shows the distribution of the percent difference between the SLM without versus with a windscreen, revealing the asymmetry of higher unscreened noise readings, particularly for dB-C.

Table 2.

Windscreen and Placement Effects on Sound Level Meter (SLM) Noise Readings

Comparison	Sound level measure	Mean difference (dB)	Mean percent difference (%)	Mean absolute difference (%)	Correlation
Windscreen effect¹	A-weighted, average	0.3	0.5	3.5	0.88
	A-weighted, peak	0.3	0.4	4.5	0.74
	C-weighted, average	2.8	3.9	5.4	0.82
	C-weighted, peak	2.7	3.6	6.0	0.70
Placement effect²	A-weighted, average	−1.3	−2.0	8.8	0.55
	A-weighted, peak	0.6	0.6	9.3	0.44
	C-weighted, average	−1.7	−2.4	7.1	0.80
	C-weighted, peak	−1.6	−2.0	8.2	0.66

Unscreened versus screened, so a higher unscreened reading would be >0.

Shoulder versus handlebar, so a higher shoulder reading would be >0.

Figure 2.

Distribution of sound level meter (SLM) noise readings without versus with a windscreen.

The last four rows of Table 2 give the results of comparing synchronous noise readings between the paired SLM located at the cyclist’s shoulder versus at the handlebars (without windscreens). Sensor placement has a small bias effect of reducing measured noise at the shoulder by about 2% (except for A-weighted peak readings, which are slightly higher by <1%). Sensor placement has a bigger effect on variability, with MAPD of 7% to 9% between placement locations, and a correlation between the locations as low as 0.44. The placement effect on variability is greater for peak than average noise measures, and for A-weighted than C-weighted noise.

Validation of Smartphone Noise Readings

The validation focuses on time-average noise readings, consistent with most cycling studies, using the SLM (at the handlebars, with and without a windscreen) as the reference instrument. Table 3 gives the results of comparing synchronous noise readings between the smartphone and SLM instruments. Smartphone accuracy is better for C-weighted than A-weighted noise. Compared with the unscreened SLM, the smartphone readings are biased high for A-weighted noise and low for C-weighted noise. Figure 3 shows the distribution of the percentage difference between the smartphone and the SLM without a windscreen, illustrating substantial differences between A-weighted and C-weighted noise results.

Table 3.

Comparison of Smartphone Versus Sound Level Meter (SLM) Noise Readings

Sound level measure	Comparison SLM	Mean difference (dB)¹	Mean difference (%)	Mean absolute difference (%)	Correlation
A-weighted, average	Unscreened	4.2	5.7	11.8	0.53
A-weighted, average	Screened	4.5	6.1	12.4	0.48
C-weighted, average	Unscreened	−2.8	−3.6	5.4	0.81
C-weighted, average	Screened	<0.1	<0.1	4.3	0.79

Smartphone versus SLM, so a higher smartphone reading would be >0

Figure 3.

Distribution of smartphone versus unscreened sound level meter (SLM) noise readings.

Considering the screened SLM as the most accurate reference readings, the C-weighted smartphone noise readings can be considered accurate with a mean difference of <0.1 dB (<0.1% MPD, 4% MAPD, 0.8 correlation), but A-weighted noise readings are not with a mean difference of 4.5 dB (6% MPD, 12% MAPD, 0.5 correlation). Smartphone noise readings are higher when compared with the screened than the unscreened SLM, which increases accuracy for C-weighted noise, but decreases accuracy for A-weighted noise (which is already high-biased). The result differences using screened versus unscreened SLM are greater for C-weighted noise because the windscreen more greatly affects C-weighted noise readings (see Table 2).

Impacts of Travel and Air Speed

Dividing the data between stationary (N = 4,623) and moving (N = 10,794) subsets (based on travel speed >0) reveals substantial differences in smartphone accuracy for A-weighted noise. Smartphone accuracy (compared with SLM) for A-weighted noise is substantially lower for moving than stationary observations, with around twice the mean difference from SLM, and around a 50% increase in the MAPD. For C-weighted noise, the difference between moving and stationary smartphone accuracy is much smaller, suggesting that A-weighted smartphone readings are substantially more affected by bicycle- and wind-generated noise. Figure 4 shows a scatterplot of smartphone dB-A readings versus travel speed for the moving observations, suggesting a clear relationship with bicycle travel speed.

Figure 4.

Travel speed impact on smartphone noise readings.

Table 4 gives the results of the univariate regression analyses (using only moving observations), for each combination of independent variable (travel or air speed) and noise measure (six instrument-weighting combinations). The estimated phi values (autocorrelation parameters) are all high, indicating the importance of accounting for serial correlation in the regression analysis (and other analyses with high-resolution on-road noise data). The speed coefficients in Table 4 indicate the strength of linear relationships between measured noise level and air or travel speed; higher magnitudes for these parameters indicate that the measured noise increases more with speed. The speed coefficients are all statistically significant (at p < 0.05), indicating that noise readings increase by 0.1 to 0.3 dB per km/h of travel speed and by 0.3 to 1.1 dB per km/h of air speed. The strength of relationships between speed and noise (as indicated by the magnitudes of speed coefficients in Table 4) are greater for air than travel speed, for unscreened than screened SLM, and for smartphone than SLM instruments. The differences in speed effects by noise weighting method are smaller and less consistent, with generally greater relationships with speed (i.e., higher speed coefficients) for A-weighted smartphone noise readings but C-weighted SLM noise readings.

Table 4.

Regression Results for Travel and Air Speed Effects on Noise Readings

Dependent variable (dB)	Noise weighting	Speed variable (km/h)	Intercept	Speed coefficient	Phi¹
Smartphone	A	Travel	70.54*	0.304*	0.905
SLM unscreened	A	Travel	64.74*	0.162*	0.860
SLM screened	A	Travel	65.21*	0.088*	0.871
Smartphone	C	Travel	73.80*	0.217*	0.869
SLM unscreened	C	Travel	77.17*	0.175*	0.889
SLM screened	C	Travel	74.61*	0.081*	0.880
Smartphone	A	Air	62.66*	1.060*	0.797
SLM unscreened	A	Air	62.94*	0.329*	0.855
SLM screened	A	Air	62.88*	0.289*	0.866
Smartphone	C	Air	71.86*	0.402*	0.843
SLM unscreened	C	Air	73.91*	0.461*	0.868
SLM screened	C	Air	71.74*	0.325*	0.867

Note: SLM = sound level meter.

statistically significant at p<0.05.

Estimated autocorrelation parameter.

Table 5 gives the results of the univariate regression analyses (using only moving observations), for each combination of independent variable (travel or air speed) and noise comparison measure (smartphone accuracy or windscreen effect). The speed coefficients in Table 5 indicate the strength of linear relationships between air or travel speed and three different instrumentation comparisons (i.e., percent differences in measured noise levels); higher magnitudes for these parameters indicate larger differences between noise readings with increasing speed. The speed coefficients for all but one of the comparison measures are statistically significant (at p < 0.05), and indicate that the windscreen effect (i.e., difference between unscreened and screened noise readings) increases by 0.1% to 0.3% per km/h of travel or air speed. The strength of relationships between speed and windscreen effects (as indicated by the magnitudes of speed coefficients) are greater for air than travel speed, and for C-weighted than A-weighted noise. The speed coefficients in Table 5 also indicate that the differentials between smartphone and SLM noise readings increase by up to 1.0% per km/h of travel or air speed. The relationships between speed and smartphone noise reading accuracy compared with SLM are greater (i.e., higher speed coefficients) for A-weighted than C-weighted noise, for air than travel speed (mostly), and for screened than unscreened SLM comparisons.

Table 5.

Regression Results for Travel and Air Speed Effects on Noise Reading Differentials

Dependent variable (% difference)	Noise weighting	Speed variable (km/h)	Intercept	Speed coefficient	Phi¹
Smartphone versus unscreened SLM	A	Travel	7.24*	0.312*	0.840
Smartphone versus screened SLM	A	Travel	6.95*	0.392*	0.845
Unscreened versus screened SLM	A	Travel	−0.80*	0.117*	0.618
Smartphone versus unscreened SLM	C	Travel	−2.93*	−0.036	0.791
Smartphone versus screened SLM	C	Travel	−0.82*	0.162*	0.745
Unscreened versus screened SLM	C	Travel	2.30*	0.221*	0.788
Smartphone versus unscreened SLM	A	Air	0.34	0.978*	0.801
Smartphone versus screened SLM	A	Air	0.97	0.983*	0.810
Unscreened versus screened SLM	A	Air	−0.76*	0.128*	0.618
Smartphone versus unscreened SLM	C	Air	−1.95*	−0.120*	0.789
Smartphone versus screened SLM	C	Air	−0.33	0.142*	0.743
Unscreened versus screened SLM	C	Air	2.04*	0.270*	0.788

Note: SLM = sound level meter.

statistically significant at p<0.05.

Estimated autocorrelation parameter.

Figure 5 illustrates the effect of air speed on smartphone dB-A noise reading accuracy as the percent difference from unscreened SLM (moving observations only). Because smartphone noise readings are more sensitive than SLM to air speed, even compared with the unscreened SLM, the noise reading differential increases clearly with measured air speed. At low air speeds below 5 km/h, the smartphone readings are slightly low-biased. But the difference increases consistently with speed, and for air speeds of 10 km/h and above the smartphone readings have a consistent and increasingly strong positive bias. A similar image emerges compared with screened SLM readings.

Figure 5.

Air speed impact on smartphone versus unscreened sound level meter (SLM) dB-A noise readings.

Impact of Data Aggregation

Depending on the application, cycling exposure studies may be interested in more temporally aggregate noise measures. Generally, data aggregation reduces variable difference between measurements (MAPD), but not bias (MPD). Figure 6 shows smartphone A- and C-weighted noise accuracy (versus screened SLM) as MPD and MAPD at aggregations of up to 5 min. The 1-s bars are equivalent to the results reported in Table 3. As expected, MPD is almost unchanged by aggregation while MAPD decreases substantially. At 5-min aggregation, the C-weighted smartphone readings have an MPD of −0.3% (mean difference of −0.1 dB-C) and a MAPD of just 2.3%.

Figure 6.

Effect of temporal aggregation on smartphone versus sound level meter (SLM) noise readings.

Discussion

Findings

The results above indicate good smartphone performance in measuring high-resolution (1-s) C-weighted noise for an on-road cyclist, with a mean difference from the screened SLM readings of <0.1% (<0.1 dB-C) and a mean absolute difference of 4.3%. Smartphone performance for measuring A-weighted noise was poorer, with a mean difference of 6.1% (4.5 dB-A) and a mean absolute difference of 12.4%. Temporal aggregation up to 5-min intervals decreases MAPD to 2.3% and 8.0% for C-weighted and A-weighted noise, respectively, and so segment-level analysis using smartphone-generated data will be more precise than point- or event-level analysis (e.g., studying on the effect of passing cars).

On-bicycle smartphone noise readings increase substantially with air and travel speed, by up to 1.1 dB per km/h air speed (for smartphone-measured A-weighted noise). This effect was also observed, but to a lesser extent, for the SLM A- and C-weighted noise readings both with and without a windscreen, but the windscreen muted the speed effect by 10% to 50%. Because smartphone noise readings are two to three times more affected by travel/air speed than SLM readings, smartphone accuracy (in comparison with the SLM reference instrument) degrades with travel/air speed. This is especially true for A-weighted noise, for which smartphone readings systematically diverge from SLM readings by up to 1% per km/h.

The travel/air speed effects on measured noise are likely a combination of bicycle- and wind-generated noise effects, which could not be separated in this study. The stronger relationship with air speed (Table 4 and Table 5) suggests the primary effect is from wind, and that the observed effects are unlikely because of a relationship between travel speed and environmental noise. Note that the measured air speed effects are with respect to the cyclist and microphone, and do not represent the effects of ambient wind speed on noise exposure; past research found no significant effect of ambient wind on noise exposure ( 16 , 18 ), likely because ambient wind has only a minor effect on air speed for a traveling cyclist ( 40 ).

With regard to the study instrumentation design questions, we find that a SLM windscreen has a small effect on noise readings (up to about 5%), which increases with travel/air speed (increasing by about 2% per 10 km/h). Sensor placement has a smaller effect on noise readings (up to about 2%), but introduces greater variability, possibly given differential influences of bicycle- and wind-generated noise at the cyclist’s shoulder versus handlebar.

Comparison to Literature

The on-road cyclist noise exposure levels reported here (central values around 75 dB) are consistent with past studies from cities around the world ( 16 , 18 , 19 , 21 , 28 ). Past research comparing smartphone to SLM noise readings in laboratory and (indoor) community settings reported higher A-weighted accuracy (<2 dB-A mean difference) ( 15 , 33 , 35 ); those studies did not report C-weighted noise comparisons. Lower accuracy in an on-road setting is consistent with our finding that smartphone accuracy degrades with travel/air speed. Past on-road comparisons between smartphone and SLM noise readings reported 7 dB-A difference and correlations of 0.7 to 0.9 ( 9 , 32 ); our results show a smaller mean difference but lower correlation for dB-A noise. Results of past wind tunnel testing showed increases in wind noise for cyclists of around 0.4 dB per km/h air speed ( 22 ), which is similar to the air speed coefficient for C-weighted noise readings in Table 4.

Limitations

This study examined specific issues of interest for collecting cyclist noise exposure data with off-the-shelf smartphones, but could not exhaustively test all instrumentation factors. Only iPhones were used, which performed well in past research ( 34 ), but we cannot comment on the accuracy of other smartphone makes. We also did not test external microphones, which (unsurprisingly) improve smartphone noise measurement accuracy over internal microphones ( 33 ). The arrangement of instruments on the handlebars and shoulder may have affected the noise level readings because of wind flow around other instruments or the head (which would likely increase with speed). Other possibly varying data collection decisions we did not examine include: type of windscreen, smartphone application, handlebar mounting method, and orientation of smartphone and bicycle. Our study measured performance in typical cycling conditions in Vancouver, Canada, but smartphone accuracy may vary in other cities given different ambient noise, motor vehicle fleet, ambient wind, and so forth.

Conclusion

The findings above indicate that smartphones can generate reliable measurements of high-resolution cyclist exposure for C-weighted noise, but not A-weighted noise. C-weighted smartphone noise measures are also more robust to the effects of travel and air speed on noise readings. In designing future studies of on-road noise, researchers should consider the noise weighting most relevant for their study objectives as well as the accuracy and reliability of the instruments available.

Most research into cyclist noise exposure has focused on A-weighted noise, because it represents human sound perception and is the basis for most regulation and evidence on negative health impacts ( 41 ). Some research using on-road noise as a proxy measure for motor vehicle traffic has focused on other frequency measures, and smartphones may be more useful for this type of research. The most appropriate frequency weighting for comfort and stress research is an open question requiring investigation. If stress responses are the consequence of exposure to road traffic stimuli, then the most appropriate frequency weighting may be a combination of the human auditory response and traffic-specific noise generation (perhaps requiring a new frequency weighting for perceived traffic noise).

Whether using smartphones or professional SLMs, studies employing in situ on-bicycle noise measurement must consider the effects of wind- and bicycle-generated noise to ensure internal validity. The study framework should define the noise sources of interest. If the total sound exposure level for a cyclist is the quantity of interest, then wind- and bicycle-generated noise should be included. If motor vehicle traffic-generated noise is the more precise quantity of interest, then wind- and bicycle-generated noise should be excluded. Similarly, if bicycles are being used as probes for mobile noise measurement (to create a noise map, for example), then again wind- and bicycle-generated noise should be excluded. Even if wind noise is within the study scope, wind effects on measurements from a mounted microphone are likely different from wind noise in a cyclist’s ear, because wind noise is induced by turbulence ( 42 ).

Because the wind- and bicycle-generated noise effects increase with speed, failing to account for them can substantially bias study results, especially for dB-A. At a typical travel speed of 20 km/h, smartphone and SLM noise readings could be biased high by 6.0 and 3.5 dB, respectively, and smartphone inaccuracy increased by 8%. A substantial risk for confounding arises because bicycle speeds tend to vary systematically on a network: at intersections, hilly terrain, crowded facilities, horizontal curves, and so forth. Inflated noise measures where cyclists tend to travel faster will bias related coefficients estimated in a regression analysis. Noise levels would also be biased high where there is systematically more bicycle-generated noise for other reasons, such as on rougher pavement, or more local wind, such as in a street canyon.

To inform more robust analysis of cyclist noise exposure, future research should investigate possible strategies to account for wind- and bicycle-generated noise in on-road studies, such as frequency filters, correction factors, or instrumentation improvements (better windscreens, mounting apparatus, in-ear measurement, etc.). Given the importance of air speed for measured noise levels, future research should validate instrument arrangements in a laboratory with controlled air flow before field data collection. Controlled field studies could then measure bicycle-generated noise as a function of speed for different bicycles, pavements, and environmental conditions. Systematic noise measurements in different traffic environments would allow decomposition of traffic sources of cyclist noise exposure. Another area needing further research is the perception of noise while cycling, and how it may vary for different source types: wind, motor vehicles, ambient, and so forth. We hope the findings in this study are used to inform future study designs investigating both on-road cyclist noise exposure and instrumentation for ambulatory in situ noise measurement.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Alexander Bigazzi, Maria Albitar; data collection: Maria Albitar; analysis and interpretation of results: Maria Albitar, Alexander Bigazzi; draft manuscript preparation: Maria Albitar, Alexander Bigazzi. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was enabled by support from the Canadian Foundation for Innovation John R. Evans Leaders Fund (#37145) and the B.C. Knowledge Development Fund.

ORCID iD

Alexander Bigazzi

References

Singh

Kumari

Sharma

A Review of Adverse Effects of Road Traffic Noise on Human Health. Fluctuation and Noise Letters, Vol. 17, No. 01, 2018, p. 1830001. https://doi.org/10.1142/S021947751830001X

Babisch

Updated Exposure-Response Relationship between Road Traffic Noise and Coronary Heart Diseases: A Meta-Analysis. Noise and Health, Vol. 16, No. 68, 2014, p. 1. https://doi.org/10.4103/1463-1741.127847

Ibili

Adanu

E. K.

Adams

C. A.

Andam-Akorful

S. A.

Turay

S. S.

Ajayi

S. A.

Traffic Noise Models and Noise Guidelines: A Review. Noise & Vibration Worldwide, Vol. 53, No. 1–2, 2022, pp. 65–79. https://doi.org/10.1177/09574565211052693

Brown

A. L.

Van Kamp

WHO Environmental Noise Guidelines for the European Region: A Systematic Review of Transport Noise Interventions and Their Impacts on Health. International Journal of Environmental Research and Public Health, Vol. 14, No. 8, 2017, p. 873. https://doi.org/10.3390/ijerph14080873

Stansfeld

Clark

Smuk

Gallacher

Babisch

Road Traffic Noise, Noise Sensitivity, Noise Annoyance, Psychological and Physical Health and Mortality. Environmental Health, Vol. 20, No. 1, 2021, p. 32. https://doi.org/10.1186/s12940-021-00720-3

Winters

Davidson

Kao

Teschke

Motivators and Deterrents of Bicycling: Comparing Influences on Decisions to Ride. Transportation, Vol. 38, No. 1, 2011, pp. 153–168. https://doi.org/10.1007/s11116-010-9284-y

Gössling

Humpe

Litman

Metzler

Effects of Perceived Traffic Risks, Noise, and Exhaust Smells on Bicyclist Behaviour: An Economic Evaluation. Sustainability, Vol. 11, No. 2, 2019, p. 408. https://doi.org/10.3390/su11020408

Dekoninck

Botteldooren

Int Panis

Using City-Wide Mobile Noise Assessments to Estimate Bicycle Trip Annual Exposure to Black Carbon. Environment International, Vol. 83, 2015, pp. 192–201. https://doi.org/10.1016/j.envint.2015.07.001

Griffin

G. P.

Hankey

S. C.

Simek

C. L.

Buehler

Street Noise Relationship to Bicycling Road User Safety. Report No. Report 02-27. Safe-D: Safety Through Disruption, Wise, Virginia, 2018.

10.

Gelb

Apparicio

Cyclists’ Exposure to Atmospheric and Noise Pollution: A Systematic Literature Review. Transport Reviews, Vol. 41, No. 6, 2021, pp. 742–765. https://doi.org/10.1080/01441647.2021.1895361

11.

Mohamed

Bigazzi

Speed and Road Grade Dynamics of Urban Trips on Electric and Conventional Bicycles. Transportmetrica B: Transport Dynamics, Vol. 7, No. 1, 2019, pp. 1467–1480. https://doi.org/10.1080/21680566.2019.1630691

12.

Haddad

de Nazelle

The Role of Personal Air Pollution Sensors and Smartphone Technology in Changing Travel Behaviour. Journal of Transport & Health, Vol. 11, 2018, pp. 230–243. https://doi.org/10.1016/j.jth.2018.08.001

13.

Nuñez

J. Y. M.

Bisconsini

D. R.

Rodrigues da Silva

A. N.

Combining Environmental Quality Assessment of Bicycle Infrastructures with Vertical Acceleration Measurements. Transportation Research Part A: Policy and Practice, Vol. 137, 2020, pp. 447–458. https://doi.org/10.1016/j.tra.2018.10.032

14.

Ramos

T. da C.

da Silva

A. N. R.

de Souza

L. C. L.

Dekoninck

Botteldooren

Teixeira

I. P.

Cyclists’ Noise Exposure in a Brazilian Medium-Sized City. Ciencia & Saude Coletiva, Vol. 25, No. 7, 2020, pp. 2891–2902. https://doi.org/10.1590/1413-81232020257.157522018

15.

Celestina

Hrovat

Kardous

C. A.

Smartphone-Based Sound Level Measurement Apps: Evaluation of Compliance with International Sound Level Meter Standards. Applied Acoustics, Vol. 139, 2018, pp. 119–128. https://doi.org/10.1016/j.apacoust.2018.04.011

16.

Apparicio

Carrier

Gelb

Séguin

A. -M.

Kingham

Cyclists’ Exposure to Air Pollution and Road Traffic Noise in Central City Neighbourhoods of Montreal. Journal of Transport Geography, Vol. 57, 2016, pp. 63–69. https://doi.org/10.1016/j.jtrangeo.2016.09.014

17.

Gelb

Apparicio

Noise Exposure of Cyclists in Ho Chi Minh City: A Spatio-Temporal Analysis Using Non-Linear Models. Applied Acoustics, Vol. 148, 2019, pp. 332–343. https://doi.org/10.1016/j.apacoust.2018.12.031

18.

Gelb

Apparicio

Modelling Cyclists’ Multi-Exposure to Air and Noise Pollution with Low-Cost Sensors—The Case of Paris. Atmosphere, Vol. 11, No. 4, 2020, p. 422. https://doi.org/10.3390/atmos11040422

19.

Apparicio

Gelb

Cyclists′ Exposure to Road Traffic Noise: A Comparison of Three North American and European Cities. Acoustics, Vol. 2, No. 1, 2020, pp. 73–86. https://doi.org/10.3390/acoustics2010006

20.

Minet

Stokes

Scott

Weichenthal

Hatzopoulou

Should Traffic-Related Air Pollution and Noise Be Considered When Designing Urban Bicycle Networks?

Transportation Research Part D: Transport and Environment, Vol. 65, 2018, pp. 736–749. https://doi.org/10.1016/j.trd.2018.10.012

21.

Liu

Lan

Shirai

Austin

Yang

Seto

Exposures to Air Pollution and Noise from Multi-Modal Commuting in a Chinese City. International Journal of Environmental Research and Public Health, Vol. 16, No. 14, 2019, p. 2539. https://doi.org/10.3390/ijerph16142539

22.

Seidman

M. D.

Wertz

A. G.

Smith

M. M.

Jacob

Ahsan

S. F.

Evaluation of Noise Exposure Secondary to Wind Noise in Cyclists. Otolaryngology–Head and Neck Surgery, Vol. 157, No. 5, 2017, pp. 848–852. https://doi.org/10.1177/0194599817715250

23.

Dekoninck

Botteldooren

Int Panis

An Instantaneous Spatiotemporal Model to Predict a Bicyclist’s Black Carbon Exposure Based on Mobile Noise Measurements. Atmospheric Environment, Vol. 79, 2013, pp. 623–631. https://doi.org/10.1016/j.atmosenv.2013.06.054

24.

Boogaard

Borgman

Kamminga

Hoek

Exposure to Ultrafine and Fine Particles and Noise during Cycling and Driving in 11 Dutch Cities. Atmospheric Environment, Vol. 43, No. 27, 2009, pp. 4234–4242. https://doi.org/10.1016/j.atmosenv.2009.05.035

25.

Bigazzi

A. Y.

Figliozzi

M. A.

Review of Urban Bicyclists’ Intake and Uptake of Traffic-Related Air Pollution. Transport Reviews, Vol. 34, No. 2, 2014, pp. 221–245. https://doi.org/10.1080/01441647.2014.897772

26.

Nyhan

McNabola

Misstear

Evaluating Artificial Neural Networks for Predicting Minute Ventilation and Lung Deposited Dose in Commuting Cyclists. Journal of Transport & Health, Vol. 1, No. 4, 2014, pp. 305–315. https://doi.org/10.1016/j.jth.2014.09.014

27.

Bigazzi

A. Y.

Figliozzi

M. A.

Dynamic Ventilation and Power Output of Urban Bicyclists. Transportation Research Record: Journal of the Transportation Research Board, 2015. 2520: 52–60. https://doi.org/10.3141/2520-07

28.

Teixeira

I. P.

Rodrigues da Silva

A. N.

Schwanen

Manzato

G. G.

Dörrzapf

Zeile

Dekoninck

Botteldooren

Does Cycling Infrastructure Reduce Stress Biomarkers in Commuting Cyclists? A Comparison of Five European Cities. Journal of Transport Geography, Vol. 88, 2020, p. 102830. https://doi.org/10.1016/j.jtrangeo.2020.102830

29.

Nuñez

J. Y. M.

Teixeira

I. P.

da Silva

A. N. R.

Zeile

Dekoninck

Botteldooren

The Influence of Noise, Vibration, Cycle Paths, and Period of Day on Stress Experienced by Cyclists. Sustainability, Vol. 10, No. 7, 2018, p. 2379. https://doi.org/10.3390/su10072379

30.

Ueberham

Schlink

Dijst

Weiland

Cyclists’ Multiple Environmental Urban Exposures—Comparing Subjective and Objective Measurements. Sustainability, Vol. 11, No. 5, 2019, p. 1412. https://doi.org/10.3390/su11051412

31.

Ueberham

Schmidt

Schlink

Advanced Smartphone-Based Sensing with Open-Source Task Automation. Sensors, Vol. 18, No. 8, 2018, p. 2456. https://doi.org/10.3390/s18082456

32.

Zuo

Xia

Liu

Qiao

Mapping Urban Environmental Noise Using Smartphones. Sensors, Vol. 16, No. 10, 2016, p. 1692. https://doi.org/10.3390/s16101692

33.

Kardous

C. A.

Shaw

P. B.

Evaluation of Smartphone Sound Measurement Applications (Apps) Using External Microphones—A Follow-up Study. The Journal of the Acoustical Society of America, Vol. 140, No. 4, 2016, pp. EL327–EL333. https://doi.org/10.1121/1.4964639

34.

Kardous

C. A.

Shaw

P. B.

Evaluation of Smartphone Sound Measurement Applications. The Journal of the Acoustical Society of America, Vol. 135, No. 4, 2014, pp. EL186–EL192. https://doi.org/10.1121/1.4865269

35.

Jacobs

Roberts

Reamer

Mathis

Gaffney

Neitzel

Noise Exposures in Different Community Settings Measured by Traditional Dosimeter and Smartphone App. Applied Acoustics, Vol. 167, 2020, p. 107408. https://doi.org/10.1016/j.apacoust.2020.107408

36.

RODE. Audio For Film 101: Ways to Reduce Wind Noise When Recording Outside. rode.com. https://rode.com/en/about/news-info/audio-for-film-101-ways-to-reduce-wind-noise-when-recording-outside. Accessed August 8, 2021.

37.

Pinheiro

Bates

DebRoy

Sarkar

, and R Core Team. Nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1-122, 2015. https://cran.r-project.org/web/packages/nlme/index.html

38.

R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2019.

39.

Flügel

Hulleberg

Fyhri

Weber

Ævarsson

Empirical Speed Models for Cycling in the Oslo Road Network. Transportation, Vol. 46, 2017, pp. 1395–1419. https://doi.org/10.1007/s11116-017-9841-8

40.

Tengattini

Bigazzi

A. Y.

Validation of an Outdoor Coast-Down Test to Measure Bicycle Resistance Parameters. Journal of Transportation Engineering, Part A: Systems, Vol. 144, No. 7, 2018, p. 04018031. https://doi.org/10.1061/JTEPBS.0000152

41.

Pulsar Instruments. Understanding A-C-Z Noise Frequency Weightings. Pulsar Instruments, Hunmanby, England, 2021.

42.

van den Berg

G. P.

Wind-Induced Noise in a Screened Microphone. The Journal of the Acoustical Society of America, Vol. 119, No. 2, 2006, pp. 824–833. https://doi.org/10.1121/1.2146085

Instrumentation to Measure On-Road Cyclist Noise Exposure: Considerations for Study Design with Smartphones and Sound Level Meters

Abstract

Keywords

Literature Review

Cyclist Noise Exposure

Relationships Among Exposures and Outcomes

Smartphone Noise Measurement

Summary and Research Questions

Methodology

Field Data Collection

Data Analysis

Results

Impacts of Windscreen and Sensor Placement

Validation of Smartphone Noise Readings

Impacts of Travel and Air Speed

Impact of Data Aggregation

Discussion

Findings

Comparison to Literature

Limitations

Conclusion

Footnotes

Author Contributions

Declaration of Conflicting Interests

Funding

ORCID iD

References