Study region
Military training operations at NASWI originate from two primary airfields on Whidbey Island, Washington State, USA (Fig. 1). Ault Field is located approximately 5 km from the city of Oak Harbor, the largest community in Island County, while Outlying Landing Field (OLF) Coupeville is located 4 km from the town of Coupeville. Aircraft operations conducted at NASWI range from sessions of repeated closed-pattern routines (including “touch-and-go” field carrier landing practice, FCLP), to interfacility transfers and arrivals from and departures to off-station areas, including the Olympic Military Operations Area (MOA) on the Olympic Peninsula (the primary location of electronic warfare and air-to-air combat training). The flight paths for these operations extend across northwestern Washington, from the Pacific coast to the Cascade Mountains, encompassing the counties of Clallam, Jefferson, Island, San Juan, Skagit, and Snohomish, in Washington State, USA.
Analysis workflow
Acoustic metrics characterizing individual aircraft noise events and cumulative exposure levels were derived from acoustic data recorded at monitoring locations and used to validate a model simulating noise exposure across the entire study region. Modeled spatial predictions, expressed as noise contours, were overlaid with a dasymetric population density map to estimate population noise exposures at a fine spatial scale. Established thresholds and exposure-response functions were used to estimate the effect of the noise regime on multiple population health outcomes. This analysis workflow is detailed in Fig. 2.
Acoustic monitoring data and metrics
Acoustic monitoring data consisted of sound pressure level (SPL) measurements collected during previous investigations into military aircraft noise, primarily from locations near the Ault and Coupeville airfields or their associated flight paths, both on Whidbey Island and throughout the surrounding region. Congress passed unique legislation in 2019 requiring the Navy to conduct acoustic monitoring around NASWI during four discrete weeks in 2020–2021, and at one location within the Olympic MOA for 365 days [33]. These data were obtained from the Naval Facilities Engineering Systems Command [34]. Additional monitoring data from 2015 to 2019 were obtained from JGL Acoustics, Inc. and the National Park Service Night Skies and Natural Sounds Division [21, 35, 36]. In total, 20 unique locations were examined (Supplementary Table S1). SPL measurements were conducted with a class 1 sound level meter at a 1 Hz sampling rate and included A frequency-weighted equivalent continuous SPL LAeq and, where available, A-weighted fast time-weighted maximum SPL LAFmax and peak C-weighted LCpeak. Frequency spectrum measurements consisted of Z-weighted LZeq in one-third octave bands and were only available for a subset of locations. Further details regarding data collection can be found in the relevant references [21, 35,36,37].
We calculated a suite of acoustic metrics to characterize noise from single overflight events and cumulative noise levels associated with aircraft operations. Metrics were selected for their ubiquity in domestic and international standards and policy for land use compatibility, and because they provide the basis for exposure-response relationships concerning human health impacts [7, 9, 11, 12, 38,39,40]. All metrics throughout this study use A frequency weighting unless otherwise specified.
Single event metrics included the sound exposure level LE (also referred to as SEL), the 1-second average event maximum Lmax, the fast time-weighted maximum LFmax, and (when available) the instantaneous C-weighted peak sound pressure level LCpeak. All metrics were calculated in accordance with standards established by the International Organization for Standardization (ISO) and the Navy [11, 37]. The spectral content of noise events was measured in one-third octave frequency bands for a subset of monitoring locations near Coupeville airfield (locations 6-10) having a high prevalence of FCLP aircraft events. Spectrums were energy-averaged for individual events, then energy-averaged within sites to yield a representative FCLP for each location.
Overflight events were detected from continuous SPL time-series data according to guidelines established in ISO 20906 and the SAE Aerospace Recommended Practice [38, 41], and following the approach used by the Navy for noise monitoring [37]. A 10 second moving average was applied to each SPL time-series, smoothing the signal and reducing small variations that might otherwise be incorrectly labeled as events. An individual event was detected when this level exceeded a threshold varying with ambient conditions; ISO procedures recommend estimating background sound by the 95% exceedance level of total sound L95, and aircraft maxima should measure at least 15 dB above residual sound [38]. We note that some time-series data were collected only during periods of active aircraft operations (Supplementary Table S1) and lacked a representative reference background. The threshold for event detection for these time-series was the maximum value between the L95 of the hour (+/−30 min) and a baseline 35 dB + 15 = 50 dB ambient value for each second. An event was determined to terminate when the level fell and remained below the threshold for 5 s. Detected events containing multiple peaks above a local exceedance threshold (e.g., due to rapid flybys or multiple aircraft operating simultaneously) were subdivided into individual events corresponding with each peak. Detected acoustic events at locations 1–12 were cross-referenced against reported events from the Navy [34] and verified as military aircraft events accordingly. Detected events at locations 13–20 were manually verified by a trained observer [21, 35, 36].
Cumulative metrics quantify noise exposure over periods of time and form the basis of most community or public health impact assessments. Calculated cumulative metrics included: Ldn, the day-night average sound level (also referred to as DNL), with a +10 dB penalty applied to nighttime periods (22:00-07:00); Lden, the day-evening-night average sound level, with a penalty of +5 and +10 dB applied to evening (19:00-22:00) and nighttime (22:00-07:00) periods, respectively; Lnight, the equivalent continuous sound pressure level during nighttime hours; and LeqH, the equivalent continuous sound pressure level over a specified time period H, such as 24 h. Cumulative noise exposure within the Olympic MOA was quantified only with Ldnmr, the onset-rate adjusted monthly day-night average sound level, as it is conventionally used to account for the sporadic nature and potentially high onset rates of noise within special-use airspace [37].
Cumulative acoustic metrics were calculated for every monitoring location and date, including Ldn, Lden, Lnight, Leq24h, and hourly Leq. These metrics were computed directly from continuous time-series measurements Leq,1s, rather than an aggregation of individual noise events LE, in accordance with ISO standards [11] and to enable direct comparisons of ambient noise levels on days with and without flight operations.
Aircraft operations data and simulation models
Detailed flight operations records were obtained from the Naval Facilities Engineering Systems Command for the four weeklong monitoring periods in 2020 and 2021, which were designed to capture “a range of flight operations across a range of seasonal weather conditions… during periods of high, medium, and low flight activity” [34, 37]. These records documented flight profile and track activity from Ault Field and OLF Coupeville, as well as maintenance and engine run-up operations. Records for training routes and airspace profiles within the Olympic MOA were also obtained for a 365-day period within 2020 and 2021. These data were originally collected for the Navy Real-Time Aircraft Sound Monitoring Study [33] and presented a unique opportunity to investigate direct links between military aircraft operations and the noise regime.
We used the Noisemap software suite to simulate and spatially map noise exposure across the study region [42]. Noisemap is a noise modeling tool approved by the United States Department of Defense and used by the Navy to predict noise from flight operations. It integrates airfield operational data, flight profile specifications (including track, altitude, and speed), and a library of reference noise measurements with environmental terrain data to simulate the acoustic propagation of generated noise and resulting exposure at a grid of points on the ground level. The number of operations used by Noisemap is based on the average annual day, or the average number of airfield operations that would occur during a single day assuming 365 days of flying per year [37]. The average number of total operations during the four discrete monitoring periods was approximately 83% of the projected total operations for an “average year” at NASWI for 2021 [30], thus underestimating true flight activity at the annual scale.
Operations data were summarized as the total number of operations per flight profile for each period, and the mean number of operations per flight profile was calculated across all monitoring periods. This yielded a final model representing average flight activity across all periods throughout the year. Noisemap then simulated this activity, including additional noise due to maintenance and preflight ground run-up operations, such that the total predicted aircraft noise exposure was the accumulated noise exposure generated from all active operations of aircraft on all flight profiles [42].
The Noisemap model produced noise exposure contours in 1 dB increments for Ldn, Lnight, and Leq24h from a grid of points spaced evenly at a standard distance of 914 m, or 3000 ft. The model also calculated noise exposure at specific locations corresponding to monitoring locations 1–11 to enable comparison of simulated noise metrics with those empirically measured by acoustic monitoring in the field. A second simulation was created to estimate noise exposure within the Olympic MOA using operations data averaged across the year.
Lastly, we applied the models to simulate the health impacts of alternative noise regimes by scaling the relative quantity of total flight operations across the range of 50–150%, ultimately projecting the response of population health outcomes to decreases or increases in aircraft activity. While this included estimates for the total number of operations projected for 2021 from the Navy environmental impact statement, it should be noted that this simple scaling of operations quantities from the four discrete monitoring periods does not accurately reflect the true operations and fleet composition active throughout 2021, and the projected population impact estimates are not representative, but rather demonstrative.
Population noise exposure
US population distributions are often derived from census units, which vary in geographic size based on population density. Units in urban areas are typically small with evenly distributed populations, while units in rural areas are larger with irregularly distributed populations. Using census units as a basis for population assessment can substantially limit the resolution of any spatial analysis of rural communities, and can reduce the accuracy of estimated impacts from socio-environmental problems [43, 44].
To overcome this limitation, we implemented a workflow established by Swanwick et al. to create a 30-m resolution population density estimate for the study area [45]. This approach dasymetrically distributed block-level population estimates across all non-transportation impervious surfaces for each census block in the study area. We used the same approach to estimate population density for federally- and state-recognized tribal reservations and tribal-designated statistical areas (TDSA). Population data were obtained from the US Census Bureau’s 2021 American Community Survey, and impervious surface area data from the most recently available 2019 National Land Cover Database [46]. Simulated noise contours produced from Noisemap were rasterized to the same 30-m resolution as the population density map and intersected to yield an estimate of the number of people exposed to noise levels at or above thresholds established by domestic policy and international guidelines and associated with a substantial risk of impact on human health.
The World Health Organization (WHO) strongly recommends reducing aircraft noise levels below 45 dB Lden, as aircraft noise above this level is associated with adverse health effects [9]. As such, we considered the 45 dB Ldn contour to represent the spatial extent of adverse cumulative noise exposure, and the population residing within this area was therefore exposed to quantities of noise known to be harmful to human health. Additional thresholds used to estimate the at-risk population included aircraft noise levels associated with annoyance (45 dB Lden) [9], adverse effects on sleep (40 dB Lnight) [9], a risk of hearing impairment over time (70 dBA Leq24) [3, 39], and land use incompatibility according to regulations set by the Navy, Federal Aviation Administration (FAA), and US Department of Housing and Urban Development (65 dB Ldn) [27, 47, 48]. The number of individuals predicted to be impacted by these health risks vary according to the relationships described in the following section.
Population health impacts
Population health impacts, evaluated according to the number of individuals estimated to experience an adverse health outcome due to noise exposure, were calculated using established exposure-response relationships for annoyance, sleep disturbance, and compromised childhood learning (Fig. 3). These health outcomes were selected because they serve as critical indicators of community health [2,3,4], they are ubiquitous in noise law (e.g. environmental assessment [30], land-use [27, 47, 48]), and they have published exposure-response relationships that are commonly implemented in domestic and international policy and standards to assess health outcomes from noise [9, 11, 16, 49]. In particular, WHO guidelines identify these outcomes as having sufficiently robust exposure-response relationships to support quantitative health assessment [9]. These outcomes are also the first responses in a stress-mediated chain of physiological effects that can lead to more severe health consequences. Noise exerts effects either directly though objective sound exposure (hearing impairment or sleep disturbance) or indirectly through the subjective emotional and cognitive perception of sound (annoyance) [1, 4, 50]. Both of these pathways elicit neurobiological stress responses that in turn promote cardiovascular risk factors (blood pressure, glucose levels) that can manifest in disease (hypertension, ischemic heart disease) [1, 4, 50, 51] or induce psychological effects that jeopardize mental health (anxiety, depression) [4, 50, 52].
These downstream health outcomes, namely cardiometabolic and psychological effects, were excluded from consideration in the present study because they currently lack generalized exposure-response relationships for public health assessment and are not widely used in domestic and international noise policy and guidelines. While relationships have been quantified for cardiometabolic and psychological effects [52,53,54], inconclusive empirical support and methodological differences between studies has precluded the development of robust generalized exposure-response relationships [53] and led to the exclusion of these health outcome assessments from WHO guidelines [9, 55, 56]. The chosen outcomes of annoyance, sleep disturbance, and childhood learning serve as proven indicators of community health that can be used to inform policy and prioritize future primary assessments of additional health outcomes from members of the population directly.
While most international noise policies and guidelines rely on Lden as the primary cumulative noise metric [9, 11, 49], a majority of US states (including Washington) do not apply a penalty to the evening time period, and instead rely on Ldn. As such, operational flight profile data from the simulation models were only available in day-night periods, and the following health analyses use Ldn in lieu of Lden. This is expected to result in slightly more conservative estimates than would be expected if Lden were available, given that aircraft flight operations were not uncommon during evening hours.
To predict prevalence of high annoyance and high sleep disturbance throughout the population, associated exposure-response functions were used to obtain an estimated percentage of the population impacted from the noise exposure level at the 30 m2 spatial grain (raster). Levels exceeding the defined range of a function were capped at the maximum predicted response value, while levels below were assigned a value of zero. The estimated population of each raster was multiplied by this percentage and summed across all units within the study area to estimate the total population subject to each health outcome.
Annoyance
Exposure-response curves quantifying the relationship between aircraft noise exposure and human annoyance can differ dramatically by region, community, and type of aircraft and activity. Similarly, curves used in public health policy vary widely between nations. For example, the dose-response curve endorsed by the Federal Interagency Committee on Noise (FICON) [40] remains the current US standard for estimating community response to noise exposure, and is employed by the FAA and Navy. However, the recent comprehensive Neighborhood Environmental Survey (NES) conducted by the FAA found that this standard does not reflect the current US public perception of aviation noise and provided an updated and nationally representative exposure-response curve [12]. Exposure-response curves developed and recommended by the ISO and WHO represent intermediate responses for a given noise exposure level [9, 11].
Although these relationships are commonly applied in the implementation of health risk assessment and noise policy related to commercial and civil aircraft noise, there is evidence that they may underestimate impacts of noise from military aircraft due to the dramatic differences in the frequency and intensity of military aircraft events [12, 19, 20, 22, 57]. For these reasons, we include a unique exposure-response relationship developed by Yokoshima et al., based on a synthesis of individual studies on aircraft noise from US military and Japan Self-Defense Forces [19]. Collectively, these five exposure-response curves were used to assess the range of predicted impacts by relating aircraft noise Ldn to the probability of a population being highly annoyed (Fig. 3A).
Sleep disturbance
Substantial evidence supports the considerable and consistent effects of aircraft noise on sleep disturbance [9, 58]. These exposure-response relationships are based on survey and experimental assessments that identify aircraft noise as the cause of awakenings from sleep, the process of falling asleep, and/or sleep disturbance. Nighttime noise exposure near military airfields has been found to substantially reduce sleep quality [20, 59]. However, because these studies are highly limited in number, exposure-response curves relating sleep disturbance to military aircraft noise exposure are not available. As such, we employed two published exposure-response curves that relate nighttime aircraft noise Lnight to the probability of being highly sleep disturbed (Fig. 3B), namely, the guideline curve presented by the WHO and an updated version of this curve by Smith et al. that includes more recent survey data [9, 58]. As previously discussed, these curves are expected to result in conservative estimates of impacted populations.
Childhood learning
We investigated the noise exposure levels at geographic centers of public, private, and postsecondary schools within the study area, obtained from the National Center for Education Statistics [60]. Systematic reviews conducted by the WHO and National Academies of Sciences, Engineering, and Medicine found evidence for a negative effect of aircraft noise exposure on reading and oral comprehension, standardized assessment performance, and long-term and short-term memory in children at school [5, 61]. Specifically, WHO guidelines identify an increased risk of impaired reading and oral comprehension at 55 dB Lden, equating to a 1 month delay in reading age, and an additional 1–2 month delay for each 5 dB increase beyond 55 dB Lden [9]. As simulations produced estimates of Ldn for the average annual day, assuming 365 days of exposure, we derived this noise level specific risk for each school according to its level of equivalent continuous exposure over a school year duration of 180 days.
Hearing impairment
Environmental noise pollution associated with military airfields and military operating areas can occur at levels that can result in both short- and long-term hearing impairment [62,63,64]. An exposure-response curve directly relating cumulative noise exposure to hearing impairment has not been developed at the population scale. Instead, acute noise exposures that could impact hearing were calculated and compared against action levels for occupational noise according to protocols established by the Occupational Safety and Health Administration and the National Institute for Occupational Safety and Health [65, 66]. Because this analysis requires measurement of continuous sound levels over time as opposed to cumulative metrics, daily noise exposure doses using a 24-h reference duration (representing potential exposure experienced by residents) were calculated for monitoring locations only. Single-event noise levels were also compared against established thresholds for direct physiological impairment [24, 67].