Trends and knowledge gaps in field research investigating effects of anthropogenic noise

Anthropogenic noise is a globally widespread sensory pollutant, recognized as having potentially adverse effects on function, demography, and physiology in wild animals. Human population growth and associated changes in urbanization, transportation, and resource extraction all contribute to anthropogenic noise and are predicted to increase in the coming decades. Wildlife exposure to anthropogenic noise is expected to rise correspondingly. Data collected through field research are uniquely important in advancing understanding of the real‐world repercussions of human activity on wildlife. We, therefore, performed a systematic review of literature published from 2008 to 2018 that reported on field investigations of anthropogenic noise impacts. We evaluated publication metrics (e.g., publication rates and journal type), geographical distribution of studies, study subject, and methods used. Research activity increased markedly over the assessment period. However, there was a pronounced geographical bias in research, with most being conducted in North America or Europe, and a notable focus on terrestrial environments. Fewer than one‐fifth of terrestrial studies were located in rural areas likely to experience urbanization by 2030, meaning data on ecosystems most likely to be affected by future changes are not being gathered. There was also bias in the taxonomic groups investigated. Most research was conducted on birds and aquatic mammals, whereas terrestrial mammals, reptiles, amphibians, fish, and invertebrates received limited attention. Almost all terrestrial studies examined diurnal species, despite evidence that nocturnality is the prevailing animal activity pattern. Nearly half the studies investigated effects of road or urban noise; the bulk of research was restricted to functional, rather than physiological or demographic consequences. Few experimental studies addressed repercussions of long‐term exposure to anthropogenic noise or long‐term postexposure effects, and multiple noise types or levels were rarely compared. Tackling these knowledge gaps will be vital for successful management of the effects of increasing wildlife exposure to anthropogenic noise.


Introduction
Anthropogenic noise is a globally widespread sensory pollutant recognized as having potentially adverse effects on multiple aspects of function, demography, and physiology of wild animals (Slabbekoorn & Ripmeester 2008;Barber et al. 2010;Blickley & Patricelli 2010;Kight & Swaddle 2011;Morley et al. 2013;Halfwerk & Slabbekoorn 2014;Williams et al. 2015;Gomez et al. 2016;Kunc et al. 2016;Shannon et al. 2016;Rosa & Koper 2018) Although there are numerous sources, anthropogenic noise is primarily associated with 4 main interrelated factors-human population growth, urbanization, transportation, and resource extraction-all of which are predicted to increase markedly in coming decades. The global human population is expected to rise by 1 billion by 2030 (UN Department of Economic and Social Affairs 2017). To accommodate population growth, urban land cover is forecast to double by 2040 in countries with emerging or developing economies and to expand by more than half in nations with advanced economies (Angel et al. 2011). On land, the number of cars is set to double by 2050, again with the largest gains (2.8-3.7 times) in the developing world (World Energy Council 2011). Air passenger numbers and container or bulk carrier shipping fleets are expected to increase by comparable margins over even shorter periods (Kaplan & Solomon 2016;International Air Transport Association 2017). Similarly, global extraction of biomass, fossil fuels, metals, and nonmetallic minerals is projected to grow from 79 to 167 GT by 2060 (Organisation for Economic Co-operation and Development 2019). Wildlife exposure to anthropogenic noise is expected to increase correspondingly. Data collected through field research are uniquely important in advancing understanding of the repercussions of human activity on wildlife. Although laboratory studies contribute valuable information (Blickley & Patricelli 2010), captivity can induce chronic stress in wildcaught and captive-bred animals (Bolasina 2011;Terio et al. 2013), which may interact with or mask effects of noise. Moreover, selective pressures in captivity differ from those in the wild (Frankham 2008) and can potentially lead to phenotypic and genetic differences between captive and wild populations (Johnsson et al. 2011;Fraser et al. 2019). Thus, field data are vital to determining real-world responses of wildlife to environmental change.
Given this situation, quantifying the characteristics of the field research would be of substantial value in identifying trends and knowledge gaps. We, therefore, performed a systematic review of literature published from 2008 to 2018 that reported results of field investigations into the impacts of anthropogenic noise. We evaluated article publication metrics, geographical distribution, study subject, and methods to provide an exhaustive overview of recent research efforts and to draw conclusions on approaches most likely to yield valuable advances.

Literature Search
To characterize the current research landscape in field studies investigating effects of anthropogenic noise on wildlife, we created a database through a systematic review of relevant literature published from 2008 to 2018. We used a Web of Science "all database" search (Clarivate Analytics 2019) to generate our list of candidate studies because of its broad interdisciplinary coverage and ease with which citing and cited articles can be determined. We performed a search with the following terms: ("anthropogenic noise * " OR "noise * pollution" OR "sound * pollution") AND (wild animal * OR "free-living" animal * OR wild mammal * OR "freeliving" mammal * OR wild bird * OR "free-living" bird * OR wild reptile * OR "free-living" reptile * OR wild amphibian * OR "free-living" amphibian * OR wild fish * OR "free-living" fish * OR wild invertebrate * OR "freeliving" invertebrate * ), where an asterisk indicates truncation wildcards and quotation marks encapsulate search phrases. The terms wild and free-living were used to filter laboratory and theoretical studies from the search results. Preliminary investigations indicated that without this filter, it would have been necessary to screen an unfeasibly large number of articles (around 18,000), most of which would have been outside the scope of our review because they were not field studies. After the initial search, we identified papers cited by, or citing, articles listed in the search results and added these to the database (citing papers were identified using the Web of Science "times cited" function between 5 and 6 November 2018).

Study Screening
Inclusion criteria were first applied to titles and abstracts. Full texts were obtained for articles with titles and abstracts that appeared to meet the inclusion criteria or lacked the information required to make a judgment. The inclusion criteria were then reapplied to the fulltexts to confirm eligibility.
We used a PICO framework (Frampton et al. 2017) to define our inclusion criteria. Articles within the database were included in our analyses if they were judged to report primary research addressing the question, What are the effects of anthropogenic noise on wild animals.? We specified the PICO components for this question as follows: population, wild animals; intervention, anthropogenic noise; comparators, absence or differing types or levels of anthropogenic noise; and outcomes, functional, demographic, or physiological effects of anthropogenic noise. Hence, articles were considered eligible for inclusion only if the reported study met all of the following criteria: investigated populations of free-living animals in natural or urban environments; examined effects of anthropogenic noise; compared anthropogenic noise with no-noise controls or different types or levels of anthropogenic noise; and assessed functional, demographic, or physiological outcomes of anthropogenic noise exposure (see Data Extraction for definitions).

Data Extraction
Data relating to literature characteristics, geography, subject, and methodology were extracted from each included study.
To assess changes in the level of interest in this research area over time, we calculated annual publication rates. Also, for each included paper, we noted the journal title and publication-year impact factor (Clarivate Analytics 2019) as simple proxies for the type of interest in this research area (i.e., applied vs. fundamental), and the relative importance of the subject matter within the academic community (notwithstanding the debate about the relevance of impact factors as an indicator of individual study value [Seglen 1997]). Contributing journals were categorized as dealing primarily with applied research (as opposed to fundamental) if the words conservation, applied, management, planning, monitoring or their synonyms appeared in the journal title. Studies from journals not listed in Incites Journal Citation Reports were excluded from impact-factor analyses (n = 9 out of 267). Where journal impact factor was not available for the year of publication but was available for other years (n = 44 out of 258 included full-texts with impact factor available), the impact factor for the nearest year was used (mean [SD] difference between year of publication and impact factor year = 1.41 [0.81]).
The geographic distribution of included studies was evaluated by documenting the continent and country of the study. Additionally, country economic class (advanced or emerging or developing as defined in World Economic Outlook Database [International Monetary Fund 2018]) and 2 aspects of the studied habitat type were recorded. We defined habitat for all studies as either terrestrial, marine, or freshwater. For simplicity, marine studies were related to the nearest country without formal reference to territorial water ownership. For terrestrial studies, habitat was further categorized as urban, rural, or both urban and rural. Individual study sites used in terrestrial studies were also compared with the probability of their location becoming urbanized by 2030 based on global 5-km resolution projections developed by Seto et al. (2012). Sites were classified as urban or rural, with rural sites being assigned an urbanization probability of 0.0-1.0, depending on the value of the 5-km grid square in which they were situated. Only sites described in sufficient detail to be located with ≤5-km accuracy were included in this analysis (n = 127). Multiple sites used in the same study that were <5 km apart were treated as a single site located at the central point between them. Study subject was characterized by species taxonomic group (mammal, bird, reptile, amphibian, fish, or invertebrate) and diel activity pattern (diurnal, nocturnal, crepuscular, or cathemeral). Data on diel activity pattern for each subject species were compiled from peer-reviewed primary and secondary research papers, books, and online databases. Where data for a particular species were not available, activity pattern was assumed from the timing of data collection reported in the study's methods and labeled as such (n = 30). Activity pattern was defined only for terrestrial species because insufficient data were available from the literature on aquatic species. Taxon-specific diel activity pattern distributions were compared between included studies and those reported in the literature for mammals (Bennie et al. 2014), birds, and amphibians (Anderson & Wiens 2017). Such comparisons were not possible for reptiles or invertebrates, owing to inadequate numbers of included studies and a lack of published activity-pattern distribution data, respectively.
The methods for all included studies were categorized as either observational or experimental. Also, the type of noise to which subjects were exposed and the effect of noise exposure assessed were recorded. Effects were classified according to the categories in Table 1. For experimental studies, we also noted noise exposure duration (short, seconds, minutes, or hours; medium, days or weeks; long, months or years); response assessment timing (during, after, or during and after); postexposure response assessment duration (short, medium, or long, as defined above); and number of experimental noise types and levels of noise to which subjects were exposed. A single noise exposure level represents one level of playback amplification, rather than actual noise amplitude, which may vary continuously with many of the experimental noise types used (e.g., road noise).
Where a study fulfilled more than one level of a category (e.g., it was conducted across multiple countries), each level of that category was counted as an independent data point in analyses. Therefore, sample size is reported separately for each analysis and is in some cases larger than the number of included studies. Where data appeared to be shared between studies (e.g., breeding-success metrics for the same species from the same year at the same location), one publication was selected at random from the group, and the others were excluded.

Statistical Analyses
We used R version 3.4.1 (R Core Team 2017) to conduct all our analyses. Relationships between the number of included papers published annually (response variable) and the year of publication were analyzed using Conservation Biology Volume 35, No. 1, 2021 generalized linear models (glm command, R Base package), with a Poisson error distribution and log-link function. The relationship between journal impact factor (response variable) and year of publication was analyzed using a general linear model (lm command, R Base package). Model assumptions (linearity, normal distribution of residuals, homoscedasticity, and leverage) were checked visually and found to have been met. Effect size calculations followed Nakagawa and Cuthill (2007). Model predictions were plotted with visreg (visreg version 2.5-0 [Breheny & Burchett 2016]), and all other visualizations were generated with ggplot (ggplot2 version 2.2.1 [Wickham 2009]). Comparisons of activity-pattern distributions between studies and taxonomic groups and comparisons of overall methodological approach distributions with distributions within subsets of the data (e.g., urban, or bird studies) were made with multinomial goodness-of-fit tests. In these tests, we used chisquare as the test statistic to calculate p values (xmulti, XNomial version 1.0.4 [Engels 2015]). Effect sizes for the goodness-of-fit analyses were calculated as Craven's V with cramerVFit (rcompanion version 2.3.7 [Mangiafico 2019]). Maps were created in QGIS version 3.4.9 (QGIS Development Team 2018).

Systematic Review
We identified 1721 unique articles, of which 276 papers met the inclusion criteria (Supporting Information). Nine full-texts publications were removed from the included group because they shared data with other included papers, leaving 267 papers for inclusion in the analyses.

Literature
The number of included papers published annually increased between 2008 and 2018 (Z 1,9 = 6.17, p < 0.0001, effect size r = 0.89 [95% CI 0.62, 2.22], Fig. 1a). In 2008, 7 papers meeting the inclusion criteria were published, but in 2017 this increased more than 6-fold to 43. Data presented for 2018 are incomplete because our database search was carried out in October of that year.
Of 88 journals with papers included in this study, 12 titles published >5 papers each (Fig. 1c), together providing 43.8% (117/267) of the total. The remainder were split across 76 titles. The 6 titles providing the  of papers published in these journals did not change over time (Z 1,7 = 1.29, p < 0.19) (Fig. 2a).

Geography
Studies were carried out on all continents except Antarctica. Most were based in North America and Europe (47.6% [128 of 269] and 32.0% [86 of 269], respectively), whereas Africa and Asia accounted for 3.3% (9 of 269) of the total. Overall, research was conducted in 42 countries (or their adjacent waters) (Figs. 2a & 2b & Supporting Information). The United States was the most actively researched country (29.7%; 81 of 273), with more than twice the number of studies as Canada, the next most actively researched country. Fewer than one-quarter of the studies were conducted in countries with emerging or developing economies.
More studies took place in terrestrial habitats (70%; 187 of 267) than in aquatic environments (30%; 80 of 267), with all but one aquatic study being conducted in the marine environment. Although most terrestrial studies were carried out in North America (31.6%; 59 of 187) ( Fig. 2a & Supporting Information), most marine studies (43.9%; 35 of 80) were conducted in European waters ( Fig. 2b & Supporting Information). None of the aquatic studies took place in or off Asia.
Terrestrial studies were predominantly conducted in entirely rural environments (47.0%; 88 of 187). Of the remainder, there were similar numbers of urban-rural habitat comparisons (27.8%; 52 of 187) and exclusively urban studies (24.0%; 45 of 187). Within the terrestrial studies, 246 sites (from 127 studies) were described in sufficient detail to assess their probability of urbanization by 2030 (covered in Methods). Only 19.5% (48 of 246) of these sites were in rural areas with some probability of urbanization (<50% probability = 7.3%; 18 of 247, >50% probability = 12.1%; 30 of 247) (Fig. 2c). Approximately half the sites (48.4%; 119 of 246) were in rural areas with 0% likelihood of becoming urbanized during this period. The remaining one-third (32.1%; 79 of 246) were situated on land that was already urban.

Subject
Birds and mammals were the 2 most intensively studied taxonomic groups, together being the subject of almost 90% of all included studies (birds = 57.2%; 155 of 271, mammals = 31.4%; 85 of 271) (Fig. 3a). Within the mammal studies, 82.3% (70 of 85) investigated marine species. Fish and invertebrates were the subject of 7 studies each, and just 1 study considered a reptile species.
Road and urban noise were the most frequently examined types of anthropogenic noise, together accounting for 48.1% (140 of 291) of the total (Fig. 4). Studies examining effects of urban, oil or gas extraction, and wind turbine noise were mainly observational. None of the limited number of investigations evaluating effects of aircraft (n = 6) or rail noise (n = 1) employed an experimental approach.

. (a) Duration of noise exposure (short-seconds, minutes, or hours; medium-days or weeks; long-months or years), (b) timing of response assessment, (c) assessment duration, (d) number of noise types, and (e) noise exposure levels assessed relative to numbers of included experimental studies investigating effects of anthropogenic noise on wild animals.
Of the 132 experimental studies, approximately twothirds examined consequences of short-term (≤24 h) noise exposure, whereas long-term (≥1 month) effects were rarely investigated (Fig. 5a). Just over half the studies assessed responses both during and after exposure; 43.2% (57 of 132) assessed responses only during expo-sure (Fig. 5b). Of the 75 studies assessing responses after exposure, this was usually for short periods (≤24 h); fewer than 10% examined responses more than a month after exposure (Fig. 5c). Almost two-thirds of the experimental studies considered a single noise type, and only 6 studies compared more than 4 types (Fig. 5d) most experimental studies (84.4%) used a single noise level (Fig. 5e).

Discussion
Our systematic review identified clear patterns in the publication metrics, geographical distribution, subject, and methodological characteristics of field research investigating the effects of anthropogenic noise. The number of included studies published annually increased markedly over the assessment period, whereas the impact factor of publishing journals and the proportion of studies published in specialist applied journals did not change. Most studies were carried out in North America or Europe and were conducted in terrestrial environments. Fewer than one-fifth of terrestrial studies were located in rural areas likely to experience urbanization by 2030, meaning data on ecosystems most likely to be affected by future changes are not being gathered. Birds and aquatic mammals were the most investigated taxonomic groups, with terrestrial mammals, reptiles, amphibians, fish, and invertebrates receiving only a relatively small fraction of total research attention. Almost all terrestrial studies examined diurnal species. This was not solely a function of the large number of bird studies, because terrestrial mammals studied were also largely diurnal, despite the group being predominantly nocturnal or crepuscular. Nearly half the studies investigated effects of road or urban noise, with the bulk of research effort across all studies being directed toward assessment of functional (rather than physiological or demographic) consequences. Among experimental studies, few addressed repercussions of longer term exposure to anthropogenic noise or longer term postexposure effects, and multiple noise types or levels were rarely compared. Consequently, considerable knowledge gaps remain that will need to be tackled to ensure successful management of the effects of increasing wildlife exposure to anthropogenic noise.

Literature
Comparing the 6-fold increase in anthropogenic noise articles published from 2008 through 2017 with the doubling over the same period for the whole field of biological sciences (Kelly 2018) confirms growing research effort toward effects on wildlife. Nonetheless, relatively few articles (2.6%) appear in higher ranking (impact factor ≥6) journals, suggesting that editorial interest remains comparatively restricted and perhaps indicating a need to better communicate the broad potential threat posed. Consistently low proportions of included studies appearing in specialist applied journals throughout the assessment period are notable, given the growth in knowledge concerning anthropogenic noise impacts since the 1990s (Morley et al. 2013;Williams et al. 2015;Shannon et al. 2016). However, most new information collected on the impacts of noise is behavioral (as demonstrated by the prominence of Behavioral Ecology and Animal Behaviour among contributing journals) and tends to be published in behavior-rather than conservation-journals. Hence, a strong research focus on identifying mitigation strategies appears yet to develop, although useful attempts have been made for the marine environment (McGregor et al. 2013;Perrow 2019).

Geography
The geographical distribution of study locations was similar to that reported by Shannon et al. (2016), indicating minimal change since 2013 (the end of that study's assessment period). That most studies were conducted in North America and Europe (79.6%) highlights continued lack of research attention toward nations with emerging or developing economies. This situation is further emphasized by the lack of terrestrial studies in areas likely to undergo urbanization, which predominantly occur in such countries (Angel et al. 2011;Seto et al. 2012, Fig. 2c). These states (particularly those in the tropics) are home to the majority of global biodiversity (Myers et al. 2000), which having experienced less historical exposure to anthropogenic noise is likely to be less tolerant of this disturbance than biodiversity in nations with advanced economies (Blickley & Patricelli 2010;Shannon et al. 2016). Consequently, there is a need to boost research effort in these areas to levels proportionate with their greater relative risk.
Relative to marine habitats, freshwater habitats were underrepresented by an order of magnitude, when compared with their relative biodiversity. Freshwater species are exposed to an exceptional range of anthropogenic noise because aquatic (e.g., boating or shipping, construction, and energy generation) and terrestrial (e.g., road traffic) noise penetrate freshwater environments (Holt & Johnston 2015;Mickle & Higgs 2018). Therefore, this omission represents a substantial missed opportunity-despite covering only approximately 0.8% of Earth's surface, freshwater environments hold 9.5% of all recognized species (Balian et al. 2008) and one-third of all vertebrate species (Strayer & Dudgeon 2010).

Subject
Although our results are consistent with previous reports that birds and marine mammals are generally overrepresented as research subjects (Slabbekoorn et al. 2010;Morley et al. 2013;Hawkins et al. 2015;Williams et al. 2015;Shannon et al. 2016), the corresponding underrepresentation of terrestrial mammals in noise research has received little comment. Almost all (97.5%) mammal species live on land (Cole et al. 1994 included mammal studies examined terrestrial species. Because most terrestrial mammals are nocturnal (Bennie et al. 2014) and the majority tend to be comparatively elusive irrespective of diel activity pattern (Couzens et al. 2017), this disparity may be a consequence of the difficulties presented by their study. Indeed, the added impracticalities of conducting research at night (Gaston 2019) almost certainly contributed to the low proportion of included mammal studies investigating nocturnal subjects. Nevertheless, such difficulties need to be overcome if a comprehensive account of the impacts of anthropogenic noise on wildlife is to be achieved. Not least because nocturnality may be the prevailing animal diel activity pattern (30% of vertebrates and 60% of invertebrates [Hölker et al. 2010]) and nocturnal noise exposure will increase as humans shift toward a 24-h society (Rosekind 2005).
More broadly, if taxon-group biomass is a reasonable proxy for potential exposure levels, then research effort must be substantially redirected to more accurately mirror this. Estimates suggest that fish and invertebrates make up approximately 90% of the total biomass attributed to the taxonomic groups examined in this review ( Bar-On et al. 2018). Yet, birds and marine mammals were the subject of 83% of studies included here (Supporting Information). Of the less well-studied groups, reptiles were the least studied. This may result from the historic preoccupation of anthropogenic noise research with effects on vocal communication (Rabin & McCowan 2003;Patricelli & Blickley 2006;McGregor et al. 2013;Shannon et al. 2016), which is comparatively rare in reptiles (Young et al. 2013). Regardless, most reptiles are capable of hearing to some extent (Dooling et al. 2000) and so could potentially be affected by anthropogenic noise (Simmons & Narins 2018). Also, noise effects on prey are likely to affect this group because most reptiles are active predators (Hutchins 2003).

Methodology
Although the overall proportions of experimental and observational research suggest an appropriately synthetic combination of methodological approaches (Tilman 1989), departures from this pattern within certain groups of included studies most likely relate to 2 factors. First, the preponderance of experimental mammal studies appears to be an artifact of bias toward marine mammals as subjects (82.4% of included mammal studies). The dominant interests of marine-mammal researchers were effects of seismic surveys, pile driving, sonar, and deterrent pingers (61% of included marine mammal studies). All these noise sources are intermittent and so facilitate experimental study, regardless of whether researchers have control over the noise source. Second, the high prevalence of observational studies among those carried out in urban environments and those with birds as subjects may all be linked to the difficulties of reproducing (and so manipulating) urban and traffic noise. Birds were the subject of most terrestrial studies, and most bird studies investigated urban or traffic noise. Hence, these groups effectively comprise the same studies. Neither birds nor terrestrial environments limit the capacity for experimental study. However, replicating the spectral composition and sound pressure-level attenuation of loud, low-frequency, anthropogenic noise via playback of recordings is fairly involved (Rosa et al. 2015) and especially challenging at large scales and over long periods (Blickley et al. 2012). Therefore, researchers may have used actual urban and traffic noise to avoid methodological complications and ensure stimulus validity, at the expense of being able to demonstrate causation. Studies investigating the effects of other, predominantly low-frequency noise types (e.g., aircraft, rail, oil and gas extraction, and wind turbine) were also mainly, or entirely, observational. In this sense, the development of user-friendly protocols for high-fidelity playback of such noise types would undoubtedly help facilitate experimental studies.
Noise types investigated in included studies largely matched expectations that the most extensive sources (i.e., those relating to transport networks [Barber et al. 2010;Blickley & Patricelli 2010]) should be the most widely examined. Even so, effects of aircraft (6 studies) and to a lesser extent, rail noise (one study) were both assessed surprisingly little. Aircraft noise is particularly omnipresent, being audible, for example, across virtually the entire continental United States, including protected natural areas (Barber et al. 2011) and underwater (Erbe et al. 2018). Also, rail noise is set to increase substantially across Europe, given European Commission commitments to triple the length of the existing high-speed rail network by 2030 (European Commission 2011).
One of the most conspicuous (and perhaps critical) knowledge gaps we identified is the lack of studies on physiological and demographic responses to anthropogenic noise. Effects of anthropogenic noise are likely to occur at all levels within biological systems, from cellular processes to population dynamics (Kight & Swaddle 2011). Accordingly, accurate predictions of how noise will affect wildlife (and therefore the prescription of effective management decisions) will depend on information from across this range. High costs frequently associated with mark-recapture schemes and large population studies (Skalski et al. 2005) probably limit demographic research on effects of anthropogenic noise, whereas the deficit of physiological research is more likely a consequence of practical difficulties inherent in acquiring such data in the field. Most commonly used methods require subjects to be trapped and handled, for blood or tissue sampling (Dantzer et al. 2014)  caught, making physiological field research impractical in many scenarios. But, while our results underline the need to overcome these problems, they should not be interpreted as demonstrating an overall lack of knowledge regarding physiological effects of anthropogenic noise. A sizeable body of data has been collected using captive animals (e.g., 38 of 49 physiological anthropogenic noise studies identified in Shannon et al. [2016] took place in the laboratory) and provides much useful information, despite limitations imposed by artificial environments.
A need to address biases in functional research effort is also evident from the high proportion of included papers investigating acoustic communication and movement behavior. Such focus is understandable, given the likelihood of interference when acoustic communication and anthropogenic noise share frequency ranges and the relative ease with which GPS tags allow tracking of marine mammal movements (81.5% [53 of 65] of movement behavior studies investigated marine mammals). Nonetheless, only by paying greater attention to the full range of potentially affected functions, such as reproduction and development, foraging and antipredatory behaviors, and general cognition (Kight & Swaddle 2011;Jafari et al. 2019), will it be possible to provide the comprehensive, integrated understanding necessary to generate effective mitigation strategies.
The shortage of long-term experimental studies, both in terms of exposure duration and post-exposure response assessment duration, is also striking and limiting, as is the lack of studies comparing noise types and levels. Long-term studies are often crucial in quantifying ecological impacts of environmental change (Lindenmayer et al. 2012), not least because shortterm responses may be misleading and can even suggest the exact opposite of eventual outcomes (Tilman 1989). Additionally, characterizing developmental effects of early-life anthropogenic noise exposure (Swaddle et al. 2015) will necessitate multiyear studies in many species. Securing long-term research funding is undoubtedly difficult, but addressing comparisons between effects of different noise types and levels of exposure is comparatively quick, straightforward, and immediately beneficial. Contrasting responses between different noise types establishes whether reactions are specific or relate only to general noise disturbance. And, comparing responses between multiple levels of exposure permits dose-response curves to be generated (Goudie & Jones 2004;Houser et al. 2013;Dunlop et al. 2018), allowing mitigation measures to be precisely optimized (Von Benda-Beckmann et al. 2014). In discussing noise types and levels, it is also worth noting that standards of reporting regarding noise playback are frequently insufficient to permit precise replication. Our original intention was to include basic descriptive data concerning experimental noise fre-quency range and peak amplitude in our analyses. Yet, of 132 included experimental studies, only 34 contained enough information to extract even these rudimentary details. More detailed reporting, including comparison of recording and playback frequency or power spectra, comprehensive playback-equipment specifications, and sharing of sound files, would be advantageous in this regard.

Limitations
The validity of our conclusions depends on the reliability of our initial literature search. By employing a range of synonyms in our search terms, screening papers cited by, or citing, articles listed in the initial search results and using the 46-language default search, we aimed to provide the most exhaustive search possible. This strategy returned 1721 results, which were considered to constitute a representative proportion of the body of literature as a whole. Nevertheless, as with all systematic reviews, our choice of search terms means some relevant studies will not have been identified. In most cases, systematic bias is unlikely to result from the exclusion of such studies. However, we acknowledge the terms wild and free-living are less likely to be used in field studies published in applied journals when compared with fundamental journals. In addition, only 3 non-English language studies were included (2 Spanish and one Polish), representing just 1% of the total number. A recent survey of biodiversity conservation literature found over one-third of scientific documents were not published in English (Amano et al. 2016). Although the authors used Google Scholar, which includes nonpeer-reviewed gray literature and government and institutional reports not covered by Web of Science (Haddaway et al. 2015), these proportions suggest that our search may have failed to identify some relevant non-English papers. Such omissions would likely have the greatest influence on our geographical results, particularly in underestimating the number of studies conducted in countries with emerging or developing economies, where use of English may be restricted. But, even in the unlikely event that one-third of papers were omitted and all were from countries with emerging and developing economies, they would still constitute less than half the full total, so such nations would continue to be underrepresented when considering their higher probability of future exposure.

Recommendations
Given the vital roles of field research in understanding and mitigating effects of anthropogenic noise on wildlife, it is imperative that research efforts are commensurate to the potential threat and appropriately directed. In conducting this systematic review, we identified a range of key areas where increased attention would advance Conservation Biology Volume 35, No. 1, 2021 progress toward this goal. Based on these findings, our recommendations for future field research investigating effects of anthropogenic noise are as follows.
First, place increased emphasis on collecting data in locations likely to experience the greatest future exposure to anthropogenic noise (particularly areas predicted to become urbanized).
Second, pay greater regard to critically underrepresented habitats, taxonomic groups, and circadian habits-especially freshwater environments, invertebrates, fish, amphibians and reptiles, terrestrial mammals, and nocturnal species from all groups.
Third, make more effort to explore effects of important nonroad transport noise sources, such as aircraft and trains.
Fourth, increase vitally important focus on physiological and demographic responses, paying particular attention to overcoming practical limitations on physiological data collection in the natural environments.
Fifth, fully explore potential functional effects beyond acoustic communication and movement behavior.
Sixth, aim to gather evidence over longer time scales, assessing both long-term exposure and long-term consequences. And, seventh, make comparison of responses between multiple noise types and levels standard practice in experimental studies, alongside sufficiently detailed reporting of playback techniques to permit precise replication.