Ecological impacts of water-based recreational activities on freshwater ecosystems: a global meta-analysis

Abstract

Human presence at water bodies can have a range of ecological impacts, creating trade-offs between recreation as an ecosystem service and conservation. Conservation policies could be improved by relying on robust knowledge about the relative ecological impacts of water-based recreation. We present the first global synthesis on recreation ecology in aquatic ecosystems, differentiating the ecological impacts of shore use, (shoreline) angling, swimming and boating. Impacts were assessed at three levels of biological organization (individuals, populations and communities) for several taxa. We screened over 13 000 articles and identified 94 suitable studies that met the inclusion criteria, providing 701 effect sizes. Impacts of boating and shore use resulted in consistently negative, significant ecological impacts across all levels of biological organization. The results were less consistent for angling and swimming. The strongest negative effects were observed in invertebrates and plants. Recreational impacts on birds were most pronounced at the individual level, but not significant at the community level. Due to publication bias and knowledge gaps, generalizations of the ecological impacts of aquatic recreation are challenging. Impacts depend less on the form of recreation. Thus, selectively constraining specific types of recreation may have little conservation value, as long as other forms of water-based recreation continue.

1. Introduction

Freshwater ecosystems provide many services to humans [1] and generally have a high recreation value [2,3], which is reflected in higher property values for sites with freshwater access or views, for example [4]. Aquatic recreation, such as boating, swimming or fishing, satisfies many psychosocial benefits and generates relevant economic outcomes [2,5]. Recreation also provides important health benefits [6]. Therefore, limiting access to water bodies for conservation reasons can negatively affect human welfare [7,8].

However, recreational activities such as fishing, swimming or boating may negatively impact biodiversity or affect ecosystem functioning [2,9,10]. For example, recreational angling facilitates the spread of invasive species [11]. Disturbances by recreationists may also negatively affect wildlife or habitat quality [12,13], for example by causing flight reactions in birds [14]. In fact, escape behaviour could have strong fitness implications and affect population size [14]. Relatedly, shoreline access may cause trampling effects that reduce vegetation cover and compact soil [15], boating can reduce vegetation cover [16]. Human presence at the waterside particularly affects disturbance-sensitive taxa [17] and may also affect ecosystem functioning [18] and ecosystem state, such as water quality [19]. Despite a long history of recreation ecology [20], a systematic synthesis of its effect sizes is currently lacking for aquatic ecosystems. This gap of synthetic knowledge complicates the design of conservation policies for aquatic recreation and may fuel stakeholder conflicts.

The impacts of recreation can be measured by magnitude, duration and frequency, all of which can have a series of ecological effects. We define ecological impacts as any visitor-related, biophysical changes to individuals, populations, communities or habitats [21]. Recreational impacts can scale across levels of biological organization (figure 1) [22]. The lowest level where consequences of recreation can emerge is the level of an individual organism, which includes behavioural, physiological, energetic and genetic impacts, such as increased vigilance [23], flight responses [24] and injuries [25]. Collective impacts resulting from elevated mortality or impaired recruitment can have population-level impacts, such as a reduced number of breeding pairs [26], reduced breeding success [27] or reduced abundance of macrophytes [28]. Community-level impacts are indicated by changing relative abundances of certain taxa, such as a reduced species richness [29], a change in the species composition [30] or the introduction of invasive species [11,31]. To provide a comprehensive picture, we analysed the recreational impacts of water-based recreation across all three levels of biological organization in a transparent and structured synthesis of ecological impacts of water-based recreation using comparable effect size metrics. In addition, we also examined impacts on environmental quality, such as water turbidity [19], littering and pollution of sediments [30], soil compaction [15] and reduced vegetation height or cover [16]. These aspects can indirectly affect species and are aesthetically relevant for humans and thus policy-making. Therefore, our work extends previous systematic reviews [20] that did not analyse the effect sizes of published works. Our synthesis aims to inform policy-making regarding the governance and management of water-based recreation.

Figure 1.

Conceptual framework: natural or anthropogenic impacts can cause stimuli for a reaction of individuals. Impacts can have direct consequences or reactions can cause indirect consequences, which can lead to effects on different levels of biological organization. Within the normative framework effects can be evaluated as negligible or relevant. If effects are evaluated as relevant they can be considered ‘disturbances' and protection measures are needed to reduce impacts. Adapted after Stock et al. [22]. (Online version in colour.)

Conflicts among conservation interests and outdoor recreational activities are well documented [32,33]. Disturbances to certain protected habitats or threatened species induced by recreation may justify conservation interventions that constrain, redirect or even ban certain or all forms of recreation [34]. In particular, the effects of aquatic recreation on disturbance-sensitive waterfowl species generate continuous conservation conflicts and have motivated far-reaching conservation actions, including access bans in many areas of the world [7,35]. In terrestrial ecosystems, a recent review of tourism-based effects on wildlife reported surprisingly few documented effects beyond individual reactions; hence, impacts associated with terrestrial recreation may be ‘over-reported’ [36]. Indeed, there could be knowledge gaps of human-induced impacts, particularly at higher levels of biological organization, such as populations or communities, because these are much more difficult to study than, for example, individual flight reactions [37]. However, these higher levels of biological organizations often determine whether conservation action to constrain recreation is warranted, which is both a legal and a normative question (figure 1).

Our objectives were to compare the relative magnitude of effects caused by different water-based recreational activities and examine for variation across different taxa. To that end, we conducted a meta-analysis, which is a structured and replicable method to compare the magnitude of effects derived from scientific literature using effect size metrics [38]. By synthesizing global knowledge in highly disparate literature, our meta-analysis provides the first comprehensive scientific synthesis of the ecological impacts of freshwater-based recreation. We examined the following hypotheses:

• (1) Ecological impacts of water-based recreation vary by type of recreation, because activities vary in their degree of disturbance (e.g. boating versus shore-based activities [39]).
• (2) Ecological impacts increase from shore to open water because open water activities involve intensified interaction with the whole system [21].
• (3) Ecological impacts decay towards higher levels of biological organization (from individual towards population and community levels), because density dependence and other compensation processes buffer anthropogenic disturbances at the population or community levels [22,40].

2. Methods

We conducted a systematic literature search from November 2018 until February 2019, following the guidelines of Siddaway et al. [41]. Both peer-reviewed and grey literature were researched in different languages to compile a comprehensive evidence base using a transparent and replicable literature screening approach. We spent a considerable time developing and testing the search term, which included synonyms for recreation, aquatic, impacts and environment (see electronic supplementary material, S1). We used seven different literature databases and yielded 13 115 articles: Web of Science Core Collection n = 6937, Scopus n = 4206, BioOne n = 1056, Conservation Evidence n = 39, BASE n = 596 and ProQuest n = 45. The German literature database ‘Natursport’, which specifically compiles the ecological impacts of recreational activities, was used to identify grey literature. Here, we obtained 236 articles by selecting keywords with aquatic references, such as aquatic species, aquatic habitats and aquatic recreational activities (see electronic supplementary material, S1). We used Google and Google Scholar as additional sources with reduced search terms and obtained nine further articles in Google and 181 articles in Google Scholar. After removing duplicates, 11 919 articles were left for in-depth screening. An additional 1884 references were obtained from the reference lists of acquired full-text articles (electronic supplementary material, figure S1, S1).

(a) . Screening and exclusion criteria

By screening titles, abstracts and full texts, we excluded articles that did not study (i) the impacts of recreational activities, (ii) the impacts on aquatic ecosystems or (iii) the ecological impacts (e.g. human health issues from Escherichia coli). In addition, we excluded articles of languages other than English, German, French or Spanish, as these were not accessible to the authors. For consumptive activities (angling and hunting) we excluded impacts on target species (e.g. the impacts on fish as target species for angling) to ensure comparability of effect sizes.

Screening and exclusion were performed by three trained raters. To assure reproducibility, we calculated Fleiss's κ [42] to assess interrater reliability with a subsample of 88 articles. κ = 1 indicates complete, and κ ≤ 0 indicates no agreement among raters. We obtained free-marginal κ = 0.48 (95% CI [0.35, 0.62]) at 74% overall agreement, indicating moderate agreement. The first author tended to retain more articles that the other two excluded. Further inspection of these excluded articles proved the more exclusive approach to be more time-effective, and ultimately all raters reached a consensus on which articles to retain.

In total, we excluded 6283 and 3758 articles during the title and abstract screening, respectively. Because of concern that there might be a deficit of studies focusing on freshwater ecosystems, we did not exclude marine studies during the title and abstract screening, but we intended to focus on freshwater ecosystems in the first place. During full-text screening, however, we decided to focus solely on freshwater systems, as the sample size of suitable articles deemed sufficient. As result, we excluded 3526 marine articles. This decision is warranted because the species and environmental conditions in marine and freshwater ecosystems differ, hence, meta-analysis results from marine systems may not hold for freshwater systems and vice versa.

Studies were retained for in-depth analysis and effect size extraction that compared: (i) organism/ecosystem responses to the presence of recreational activities (impact) versus lack of recreational activity (control) using various designs, (ii) organism/ecosystem responses to different intensities of recreational activity (typically observational studies) and (iii) organism/ecosystem responses to a specific compound/substance/aspect associated with a specific recreational activity compared to a control within an experiment in situ or in mesocosms. After in-depth text retrieval, further studies were excluded, providing only aggregated information for multiple recreational activities, which did not allow breaking effect sizes down to specific activities. Another 141 articles had to be excluded because effect size estimation was not possible due to lacking control or comparator information or missing data that could not be retrieved, even after contacting the authors. The final meta-analysis was completed on 94 articles that met the inclusion criteria, from which effect sizes were derived.

(b) . Data extraction

Data from the 94 articles were extracted by three trained researchers (M.S., B.W., N.M.) from tables, results and figures (using software webplot digitizer: https://automeris.io/WebPlotDigitizer/) to calculate effect sizes [38] (sample size, mean, standard deviation, standard error, test statistics or raw data). If important metrics were missing, the corresponding author was contacted to obtain the missing data. Every article was reassessed by the first author to ensure a uniform coding procedure and comparability between the effect sizes of extracted values and coded categories.

During coding, we differentiated four categories of water-based recreational activities characterized by increasing interaction strength with water from shore to open water to test hypotheses 1 and 2: (i) shore activities with (almost) no direct interaction with water (walking, dog walking, biking, horse riding, picnicking, camping, hunting, wildlife viewing), (ii) shoreline recreational angling with direct interaction with water, but the activity itself being situated at the shoreline, (iii) swimming, which has direct interaction with water but predominantly occurs in the near shore zone, often combined with shoreline stays (including swimming, snorkeling and diving) and (iv) boating as an (almost) exclusively open water activity (including motor boating, jet skiing, water skiing, sailing, rowing, paddling, kite surfing, wind surfing, stand up paddling and boat angling).

To examine which taxa were most affected by recreation and at which level of biological organization (hypotheses 1–3), we extracted information on taxonomic group (invertebrates, fish, amphibians, reptiles, birds, mammals, plants, phytoplankton and zooplankton) and species, type of recreational activity, the specific recreational impact studied (human presence, noise, trampling, pollution, toxicity, eutrophication, damage, injury, invasive species release) and the response variable measured. Response metrics included avoidance (e.g. flight initiation distances (FIDs), swimming speed), time budgets (e.g. time spent vigilant, time spent feeding), physiological responses (e.g. weight, heart rate, injuries), abundance (e.g. number of individuals, density or biomass), reproduction (e.g. number of nests, number of clutch losses, sum of offspring), biodiversity (e.g. species richness, α-diversity indices), community structure (changes in species composition, β-diversity indices), habitat structure (number of strata), water quality (e.g. pH, nutrient concentration, turbidity), pollution (number or weight of litter items) and soil compaction (bulk density, barren area). These response variables were assigned to levels of biological organization (individual, population or community) or to environmental quality. Impacts of recreation at the individual level included behavioural and physiological responses. Impacts were categorized at the population level whenever measures of abundance or total reproductive success were reported. Note that animal abundance as a response measure can be categorized as the impact on the individual level if it indicates avoidance behaviour of individuals, as in a study by Knight et al. [43], where the number of individual birds was counted when an angler was present or absent within the same season. By contrast, if the abundance of animals or plants was compared across systems (e.g. different lakes) relative to controls (control versus impact) or before and after a ban on recreational use (before versus after design), impacts were categorized at the population level. Impacts that altered the diversity of species or functional groups and species composition were categorized at the community level. Impacts on habitat structure (plant height or vegetation cover) or general environmental components, such as water, sediments or soil, were classified as impacts on the environment of animals, plants and humans. We further identified the location of the activity and the affected habitat type to examine whether certain habitat types were more affected than others.

To determine study validity, we systematically extracted information on study design, the number of water bodies examined as controls and treatments separately, whether a comparison was based on spatial or temporal separation of control and treatment and whether randomization was applied. We examined whether the study was controlled for possible confounding factors, how the control or comparator was specified and on which scale the study was conducted (comparison of water bodies or comparison of zones within a water body). These variables were used to assess the validity of each study and as weighting factors to control for study quality (see the section on sensitivity analysis and table 1). Additionally, we also noted author names, date of publication, dates of start and end of data collection, longitude and latitude of the study site or laboratory, language, name of journal and type of publication (scientific journal, book, conference proceedings, unpublished). We classified studies as peer-reviewed or not and as academic, governmental, non-profit or other to test with moderator analysis whether these factors influenced the results.

Table 1.

Study quality measures and according weights and categories. A main criterion to assess study validity was study design. ‘Replication of sampling units' as whole water bodies and ‘control of confounding factors’ could lead to higher-quality weight (maximum 1). Weight was halved if low and high intensities were compared instead of a real control and also if effects within a water body were studied instead of a comparison between water bodies. A criterion for randomization was only included to identify high quality studies as a qualitative measure. Weights suggested by Norris et al. [44] and Christie et al. [45].

study design type	Norris et al. [46]	Christie et al. [47]	quality weights	Study quality
A—after impact only	1	0.0904	0.0904	low
CI—control versus impact	2	0.206	0.206	moderate
BA—before versus after	2	0.226	0.226	moderate
G—gradient-response	3		0.3	moderate
BACI—before–after–control–impact	4	$1+{\mathrm{e}}^{\genfrac{}{}{0ex}{}{1/-(0.661+0.0153\cdot \mathrm{In}(n_{\mathrm{impact}\hspace{0.17em}\mathrm{sites}})+0.0647\cdot \mathrm{In}(n_{\mathrm{control}\hspace{0.17em}\mathrm{sites}})}{+0.111\cdot \mathrm{In}(n_{\mathrm{control}\hspace{0.17em}\mathrm{sites}})\cdot \mathrm{In}(n_{\mathrm{impact}\hspace{0.17em}\mathrm{sites}}))}}$	0.4	high
number of reference/control sampling units
0	0		+0	low
1	2		+0.2	moderate
>1	3		+0.3	high
number of impact/treatment sampling units
0	0		+0	low
1			+0.1	moderate
>1			+0.2	high
2	2
>2	3
replication of gradient–response models
<4	0			low
4	2		+0.2	moderate
5	4		+0.4	moderate
>5	6		+0.6	high
controlled for confounding factors
1—no			+0	low
2—yes			+0.1	high
randomized
1—no				moderate
2—yes				high
real control
1—yes			1/1	high
2—low versus high			1/2	low
scale
1—water body			1/1	high
2—zone			1/2	low

(c) . Statistical analysis

To quantify the ecological response to the impacts of water-based recreational activities we used Hedges's g to measure standardized mean differences [38]. Hedges's g is a correction from Cohen's d for small or unequal sample sizes. Both estimates compare the means of the treatment group (e.g. recreation site) and the comparator group (e.g. control site), standardized by dividing the pooled standard deviation of the response variable:

$$d=\frac{{\overline{x}}_2-{\overline{x}}_1}{S_{\mathrm{pooled}}},$$

where ${\overline{x}}_1$ is the mean of the control/comparator group and ${\overline{x}}_2$ is the mean of the treatment group and S_pooled is the pooled standard deviation defined as S²:

$$S_{\mathrm{pooled}}=\sqrt{\frac{(n_2-1)S_2^2+(n_1-1)S_1^2}{n_1+n_2-2}},$$

where n₁ and n₂ are the sample sizes and $S_1^2$ and $S_2^2$ are the standard deviations of the comparator and treatment, respectively.

To convert d to g, a correction factor J was used to avoid small sample size bias:

$$J=1-\frac{3}{4(n_1+n_2-2)-1}$$

$$g=d\hspace{0.17em}\times \hspace{0.17em}J.$$

Zero effect size indicates no effect of recreation on the response variable. The more distant the effect size is from zero, the larger the effect. Effect sizes were coded so that negative values indicate negative ecological impacts and positive values show positive impacts of aquatic recreation. Whereas the direction of the effect size for mortality and survival might be quite intuitive, the interpretation of measures of FIDs is less clear. Larger impacts of anthropogenic and natural sources are associated with higher FIDs [36]. However, habituation effects in areas with high recreational activity can systematically lower FIDs [36]. In particular, study designs with spatial separation would, therefore, lead to positive effect sizes comparing treatment and controls when we assume lower FIDs to be a positive effect of recreation. Because comparing different measures and recreational activities would not be reasonable with this approach, we coded the habituation effects as an indicator for the degree of impact, even though habituation itself can be positively interpreted as it supports the coexistence of humans and wildlife [36].

(d) . Subgroup analysis

To compare the magnitude of ecological effects reported for different forms of the aquatic recreation and across different taxa and levels of biological organization and to test hypotheses 1–3, we calculated summary effect sizes for subgroups of the dataset (e.g. for boaters and anglers or by taxon) and analysed the effect sizes using multilevel modelling [48] in R (R v. 4.0.5, R Core Team, 2017). Most studies yielded multiple effect sizes because responses were recorded separately for multiple recreational impact intensities, sites, species or at multiple time points. We considered sample dependencies among effect sizes through the multilevel models by nesting study, species and response measure (i.e. random = ∼1|Study_ID/taxa/Response.measured) to calculate mean effect sizes over multiple studies using the function ma.rmv (metafor package [48]).

(e) . Weighting of effect sizes

To calculate summary effect sizes in the meta-analysis, each effect size from the original studies is usually weighted by its inverse variance or by a measure of study quality [38]. We weighted the effect sizes by a quantitative measure of study validity, because inverse variance was not appropriate as the sample size was often obscured by pseudoreplication in original studies. Following suggestions by Norris et al. [44] and Christie et al. [45], we assessed study validity by eight criteria to obtain weights for each single effect size. These criteria addressed the study design and number of sampling units (as whole water bodies), providing higher weights to a more robust study design (e.g. before–after–control–impact is considered of higher quality than before–after) or to studies that involve the randomization of study sites. Further details of the weighting procedure are provided in table 1.

To assess the robustness of our weighting approach, we compared the results with analyses in which weights were calculated by the inverse variance of effect size. In addition, we tested a qualitative validity measure to examine differences among low-, medium- and high-validity studies (table 1). A study was given a low overall validity if it scored ‘low’ on one or more criteria provided in table 1. If a study scored ‘high’ in each of the criteria, it was given high validity. All other studies were given medium validity.

(f) . Moderator analysis

Finally, we analysed the possible influences of so-called moderators as confounding factors, such as publication type, peer review, year of publication, study design, habitat type, taxa and type of water body (lentic or lotic), with multilevel metaregression models with analysis of variance (Q-test), which is an approach to explain heterogeneity in effect sizes [38]. For categorical variables, we used only those as moderators if sufficient datasets (i.e. greater than 2 effect sizes from greater than or equal to 2 independent studies) for each moderator category were present.

(g) . Publication bias

Studies with significant results and therefore large effects are more likely to be published [49], and this publication bias can influence the summary effect size of meta-analyses [49]. To test whether publication bias might have affected the results and as a further robustness check of our conclusions, we performed Egger's regression test [50] to assess whether a relationship between effect size and the variance of effect size was present in the four subgroups of recreational activities. In addition, we used fail-safe n (fsn function in metafor package) as a second measure to detect publication bias. This indicates the number of studies/effect sizes needed to reduce the significance level of the summary effect size. The fail-safe number (FSN) is considered robust when it is above 5k + 10, (with k = number of effect sizes) [51].

3. Results

We identified 178 studies (electronic supplementary material, table S2) investigating the ecological impacts of specific aquatic recreational activities (shore use 53, angling 52, swimming 11, boating 76). Effect sizes could be derived from 94 studies, with 31 (33%), 23 (24%), 8 (8%) and 36 (38%) emphasizing the effects of shore activities, (shoreline) angling, swimming and boating, respectively. These studies generated 701 effect sizes. Studies from Western Europe and North America dominated (with 31 (33%), 21 (22%), 12 (13%) and 8 (9%) studies conducted in the USA, Germany, United Kingdom and Canada, respectively; electronic supplementary material, S2).

Ecological impacts were reported for a range of taxa (invertebrates, fish, amphibians, reptiles, birds, mammals and plants) and different ecosystem compartments (soil, sediments and water). Of the 94 studies (electronic supplementary material, table S2) included in the meta-analysis, those emphasizing the impacts of recreational activities on individuals were most prevalent, especially in the early literature (figure 2b). During the last two decades, the number of studies targeting impacts on populations and communities has increased, evening out the distribution of studies across the three levels of biological organization that interested us. Forty-eight out of 94 studies analysed focused on recreational impacts on birds (figure 2c). The quality of studies in our meta-analysis was mainly rated low (figure 2d). However, the proportion of studies with medium study quality increased over time. No studies of high quality could be identified using a fully controlled experimental approach in the wild.

Figure 2.

Number of studies over decades per (a) recreational activity, (b) level of biological organization, (c) taxon, (d) study quality and (e) study design (CI = control versus impact, BA = before versus after impact, G = gradient (correlation)) and BACI = before–after–control–impact design). (Online version in colour.)

(a) . Ecological impacts of freshwater recreation overall and by recreational type

The overall summary effect size of all 94 studies was −0.62 [−0.83;−0.41] (mean, 95% confidence interval) for ecological impacts of water-based recreation. That means the ecological impacts of freshwater-based recreation were significantly negative pooled across all taxa, levels of biological organization and types of recreational activities. Most variability of the pooled effect sizes originated from random effects where the response variable was nested in taxa and study (σ² = 0.26), followed by study (σ² = 0.11) and taxa nested in the study (σ² = 0.02).

Across taxa and levels of biological organization, the four subgroups of recreational activities studied did not differ in their pooled effect size, as indicated by overlapping confidence intervals (figure 3). The pooled effect sizes were significantly negative for three of the four types of recreational activities, indicating rather negative ecological impacts of aquatic recreation. In line with expectations, boating showed the strongest significantly negative pooled effect size, followed by shore use and recreational angling (figure 3). Although swimming also showed on average a negative summary effect size, confidence intervals overlapped zero (figure 3), indicating non-significance. However, the sample size was particularly small for swimming.

Figure 3.

Forest plot with mean effect sizes of recreational activities. Shown are mean summary effect sizes (mean), 95% confidence intervals (95% CI), number of studies (N) and number of effect sizes (k). Effects are significant if 95% CI (horizontal lines) do not overlap the vertical zero line. (Online version in colour.)

(b) . Ecological impacts of freshwater-based recreation across levels of biological organization and on the environment

At all three levels of biological organization (individuals, populations and communities) the summary effect size of freshwater-based recreation was negative for all four recreational activities, specifically for boating and shore use, but effects were not always significant for recreational angling and swimming (figure 4). Specifically, the negative ecological effects of angling activities were significant at the population level but not at the individual and community level. The summary effect sizes of swimming were significantly negative at the population and community level, but the latter effect size originated from a single study and thus remains uncertain. Swimming impacts at the individual level were close to zero and not significant. Moderator analysis showed that for all recreational activities, the level of biological organization were not a significant moderator, either as a continuous (levels 1–3, with 1 = individual level, 2 = population level and 3 = community level) or categorical variable (Q-test, p > 0.1). Ecosystem and habitat effects were significant only for shore use, while effects were not significantly different from zero for angling, boating and swimming.

Figure 4.

Forest plot with mean effect sizes per level of biological organization (individual, population, community and ecosystem) of recreational activities. Shown are the mean summary effect sizes (mean), 95% confidence intervals (95% CI), number of studies (N) and number of effect sizes (k). Effects are significant if 95% CI (horizontal lines) do not overlap the vertical zero line. (Online version in colour.)

(c) . Ecological impacts of freshwater-based recreation across taxonomic groups

The summary effect size of recreational impacts on each taxonomic group was, on average, negative but not always significant (figure 5). The most negative effect sizes were observed for recreational impacts on invertebrates and plants, particularly for shore use and boating. Regarding shore activities, significantly negative ecological impacts were documented for invertebrates, birds and plants. The most significant negative summary effect size of recreational angling was observed in amphibians. Other significant impacts of angling were found for birds and reptiles (only one study), and less pronounced and sometimes non-significant for other taxa (e.g. invertebrates and plants). Significantly negative impacts of swimming were reported only for invertebrates and fish, with summary effect sizes for invertebrates being close to zero and for fish close to significance. Negative ecological impacts of boating activities were revealed for invertebrates, fish, reptiles, birds and plants. All summary effect sizes of boating impacts were significant except for plants. Despite some significant differences in effect sizes between taxonomic groups for shore use and boating, there were only moderate differences in ecological impacts of different recreational activities within taxonomic groups, as indicated by largely overlapping confidence intervals and mostly negative effect sizes.

Figure 5.

Forest plot with mean effect sizes per taxon of recreational activities. Shown are the mean summary effect sizes (mean), 95% confidence intervals (95% CI), number of studies (N) and number of effect sizes (k). Effects are significant if 95% CI (horizontal lines) do not overlap the vertical zero line. Values based on low sample size (N < 2 or k < 3) are shown in grey. If k = 2 the two effect sizes are given instead of 95% CI. If k = 1 the one effect size is given instead of mean summary effect size. (Online version in colour.)

(d) . Ecological impacts of freshwater-associated recreation on birds

The impacts of shore use on individual birds (e.g. flight reactions) and bird populations (only one study) were significantly negative (figure 6). Given the lack of suitable studies, no summary effect size could be calculated for the impacts of shore use on bird communities (k = 1). The summary effect size of boating on birds was negative and significant at the individuals level. Two single effect sizes derived from studies of boating impacts on bird populations were both slightly negative, but at the community level, the summary effect size was non-significant (only one study). The impacts of recreational angling on individual birds were, on average, more negative than the population effects, but not significant at both levels and less pronounced at the bird community level. No studies of swimming impacts on birds were identified in our meta-analysis.

Figure 6.

Forest plot with mean effect sizes per level of biological of birds (individual birds, bird populations and bird communities) of the four recreational activities. Shown are the mean summary effect sizes (mean), 95% confidence intervals (95% CI), number of studies (N) and number of effect sizes (k). Effects are significant if 95% CI (horizontal lines) do not overlap the vertical zero line. Values based on low sample size (N < 2 or k < 3) are shown in grey. If k = 2 the two effect sizes are given instead of 95% CI. If k = 1 the one effect size is given instead of mean summary effect size. (Online version in colour.)

(e) . Moderator analysis

The analysis of moderators to explain heterogeneity in effect sizes revealed no significant influences of water type (lentic and lotic) in studies on the impacts of shore use, angling and boating. Swimming in lentic waters tended to have more negative effect sizes than in lotic waters (Q-test, N = 6, k = 66, QM = 3.55, p = 0.06), but the sample size was low (three effect sizes out of two independent studies for lentic and 63 effect sizes from four studies for lotic waters). Habitat type as a moderator tended to influence impacts of shore use (Q-test, N = 20, k = 131, QM = 10.38, p = 0.07) and boating (Q-test, N = 32, k = 185, QM = 6.64, p = 0.08) with beach and benthic regions being the most affected habitat types, respectively. By contrast, habitat type was not a significant moderator of angling. The sample size for habitat impacts from swimming was too low to test habitat type as a moderator.

The scale of the study design (comparison of water bodies versus comparison of different zones within a water body) was not a significant moderator for the ecological impacts estimated for shore use, angling and swimming (p > 0.1). However, there was a trend for boating impacts being less negative in studies of different waters compared to studies of different zones within one water (N = 34, k = 198, QM = 3.66, p = 0.06). Only 22 studies examined more than one water body as control and more than one water body as treatment of recreational use. Forty-one studies simply compared different zones within the same water body. There was no influence of the study scale on the impacts of shore use and angling (p > 0.1).

Peer review was a significant moderator for the studies on impacts of angling (N = 20, k = 135, QE = 4526 (p = 0.00), QM = 6.69 (p = 0.01)), with peer-reviewed articles showing less negative effect sizes than non-peer-reviewed studies. By contrast, year of publication and publication type were not significant moderators for any of the four recreational activity categories (p > 0.1).

We compared the effect sizes of studying the impact of the presence of a recreational activity compared to sites without recreation (control versus impact) and the impact of an increase in the intensity of recreational activities (low versus high impact). The aggregated effect sizes of these two different approaches were very similar (see electronic supplementary material, figure in S1). The ecological impacts of increased intensity of recreational activity (low versus high impact) appeared slightly stronger than effects of those calculated from studies comparing the presence versus absence of water-based recreation (except for swimming, where the summary effect of low versus high impact relied on only one study).

(f) . Considerations of study quality

We conducted rigorous tests on whether the weighting procedure affected the study findings. For all four types of recreational activities, our quantitative measure of study quality (table 1) did not explain variation in effect sizes (Q-test; shore use: QM = 0.71, p = 0.40; angling: QM = 2.85, p = 0.09; swimming: QM = 0.34, p = 0.56; boating: QM = 0.76, p = 0.38). Weighting according to study quality yielded slightly less negative effect sizes for impacts of shore use, angling and swimming (without changing the level of significance) and created slightly more negative effect sizes for impacts of boating compared to default weighting by inverse variance (electronic supplementary material, figure in S2). Low and medium study quality also did not explain the variation of effect sizes (p > 0.1). However, descriptive visualization of low quality and medium quality as subgroups revealed that summary effect sizes of medium-quality studies showed less negative mean effect sizes than summary effects of low-quality studies in all four recreational activities (electronic supplementary material, figure S2). In addition, summary effect sizes for the impacts of shore use, angling and boating were only significant in low-quality studies but not in medium-quality studies, indicating that low-quality studies overrepresent negative impacts of recreation.

(g) . Examining publication bias

Egger's test revealed a highly significant publication bias in each of the four recreation categories (shore use: QM = 17.11, p < 0.01; angling: QM = 6.52, p = 0.01; swimming: QM = 19.30, p < 0.01; boating: QM = 7.79, p = 0.01). The FSN indicated that even if a publication bias affected the meta-analysis, the true effect sizes of shore use, angling and boating would be significant because the FSN was greater than 5k + 10 (shore use: 49 081 (greater than 1045); angling: 42 167 (greater than 870); boating 70 511 (greater than 1055)). The results can, therefore, be considered quite robust against publication bias, except the summary effect size of swimming (FSN: 89 (less than 585)).

4. Discussion

We found mixed support for the first study hypothesis that the ecological impacts of freshwater-related reactions vary with the type of activity. When aggregated across taxa and levels of biological organization, all four examined freshwater-related recreational activities were found to have significant negative ecological impacts. By contrast, impacts on the environment were only significant for shore use. Examining recreational impacts in more detail for specific taxa and levels of the biological organization revealed differences in effect sizes and levels of significance among the four recreational types. Although the impacts of boating were consistently negative, we rejected the second hypothesis as we could not reveal a consistent signal that increasing interaction strength with the aquatic ecosystem from shore use to boating also increased the ecological impact. In fact, impacts varied with the taxon and the level of biological organization, suggesting the potential for strong contextual effects. It is also possible that our categorization into four distinct recreational types failed to properly capture the specific disturbance potential of a given recreational activity (e.g. degree of water body affected, sound levels, number of people involved). We were unable to derive quantitative information on user densities and other exact metrics of disturbance, which is why we opted for a broad classification of recreational types. Finally, we did not find support for the third hypothesis, that the ecological impacts of aquatic recreation should be strongest at the individual level and decay towards the population and community levels. In general, and despite the evidence for publication bias and low study quality, we conclude that aquatic recreation can be expected to negatively affect ecological traits, but the effects depend upon context, activity, taxa and most likely use intensity and thus should not be generalized.

(a) . Ecological impacts of specific recreational activities

Aggregated across all response variables and taxa, all four recreational types exerted similarly negative ecological impacts. However, when examining specific taxa and ecological responses, individual activities varied in their ecological outcomes. In some cases, study numbers were not sufficient to estimate effect sizes or confidence intervals overlapped zero, indicating that the current body of literature would not support general conclusions. Other findings were unequivocal. For example, shore use consistently caused negative trampling effects on riparian soil and vegetation, which can lead to decreased plant height and cover and a shift in species composition towards tolerant plant species and thus affected the environment as a whole [15]. There was also strong evidence that shore use disturbed mobile taxa such as birds. This finding can be explained by riparian and near shore areas being preferred breeding, foraging and migratory stopover habitats of many birds [52] while at the same time being preferred recreation sites. Even though intensities of shore use are expected to increase on warm and sunny days [2], some of the shore uses, such as dog walking, are less seasonal. Shore activities might, therefore, be a prevalent activity along freshwaters and during sensitive periods, such as territory establishment and spring breeding season [53]. We suggest that the best outcomes for conservation may be achieved by temporal zoning. By contrast, selectively constraining or banning single activities while allowing others will unlikely generate substantial conservation gains but fuel stakeholder conflicts instead.

Although (shoreline) angling caused overall significant ecological impacts across all taxa when effect sizes were pooled, these impacts were not consistent across the three biological levels and across taxa and were not significant for environmental response variables. While anglers have been proposed to be a particularly salient source of ecological disturbance, especially for sensitive bird species, because of their long stays in sensitive riparian habitats and their presence overnight [29], we found that the impacts of angling consistently less negative and generally milder than those of (non-angling) shore uses. One explanation might be that most anglers practise fishing in a rather solitary and relatively quiet fashion, thereby minimizing disturbances [54]. Angling restrictions to reduce wildlife disturbance might, therefore, not be as effective as commonly believed. The greatest impacts of angling were observed for amphibians, but these impacts can probably be explained as indirect effect from predation by fish [55]. Angler-managed lakes tend to host a more diverse fish community, including many larger-bodied individuals that predate on tadpoles [55], than unmanaged lakes [46]. Impacts related to recreational angling that were not observed for other recreational activities were mortality and injuries in wildlife related to angling gear [47,56]. Although littering is of the general concern in recreation [57], angling gear exposes particular threat to wildlife because of the use of hooks, non-degradable fishing lines and toxic material such as lead [47,56]. Less environmentally harmful angling gear alternatives (e.g. degradable fishing lines and lead-free weights) can be recommended from a conservation perspective.

The ecological impacts of swimming were generally not well documented, probably because of the difficulties in studying the impacts of swimming isolated from the multiple recreational activities happening jointly [58]. Studies of swimming as an isolated impact often concentrate on the toxicity of sunscreen compounds on biomass and mortality of e.g. molluscs [59]. Human discharge is an additional swimming-related impact source that can affect water quality by nutrient excretion or pathogens (not covered here). Surprisingly, we did not find literature on the disturbance effects of swimming on birds or mammals. When resting at banks or shore swimmers should have similar disturbing effects like other shore users. More studies on the impacts of swimming, snorkelling and diving are recommended to fill the knowledge gaps, especially because in temperate regions some lakes experience very high swimming uses during warm summers [2].

Boating showed the strongest and most consistent ecological impacts of the four activity types examined. We attribute our findings to the speed and noise levels, as well as wave actions and anchor impacts, associated with boating activities. Even though there are comparatively slow-moving and less noisy boating forms, such as paddling, many boating activities include methods of propulsion. Both speed and noise increase disturbance impacts [60] and explain the strong impacts of boating activities documented across all levels of biological organization. Specifically, boat propellers and anchoring can lead to damage to hard sediments and macrophytes, sediment resuspension thereby increasing turbidity and cause injuries in wildlife [9]. Additionally, wake wash, anchoring and mooring effects [30] and the introduction and translocation of invasive species by boats [31] are common ecological impacts associated with boating [9]. Reducing speed, noise and wake wash, as well as the use of zoning, might be effective in mitigating boating effects.

(b) . Impacts of recreational use of freshwaters across levels of biological organization

Early literature on the ecological impacts of recreational freshwater use focused on the individual level. In recent decades, ecological impacts at higher levels of biological organization (populations and communities) have also been increasingly studied. Methodological advances might have increased abilities to study the population or community effects of recreation. Importantly, ecological impacts at the population level might be more relevant for conservation than at the individual level [40], although highly threatened species provide exceptions for single individuals' protection. Therefore, from a normative perspective (figure 1) [22], recreation-induced impacts on the mortality or breeding success of animals may be considered more important than changes in behaviour, physiology or other individual-level traits. Therefore, we rejected hypothesis 3 that recreational impacts would dampen from individual to population and community level.

Our meta-analysis revealed no significant differences among the summary effects of water-based recreation impacts at different levels of biological organization (i.e. recreation had similar impacts at the individual, population and community level, but the impacts were not consistently negative and significant). Importantly, some activities had significant effect sizes at higher levels of the biological organization but not at lower levels. This finding was unexpected because, intuitively, ecological impacts should decrease in strength from the individual to community level (hypothesis 3). While angling showed mainly strong impacts on lower levels of biological organization, impacts on the community level were minor. For shore use and boating, impacts at the community level were similarly strong as those at other levels. These activities also showed very strong environmental impacts, such as increased turbidity by boat wakes, which in turn can affect individuals, populations and communities. In short, our review did not support that freshwater recreation mainly affects individual levels of biological organization, but impacts are context- and activity-dependent, and cannot be easily generalized.

(c) . Ecological impacts of freshwater recreation across taxonomic groups

Our taxonomic analyses revealed impacts of freshwater recreation varying by taxon and most pronounced for invertebrates and plants. Invertebrate communities serve as sensitive biomarkers for the ecological status of freshwaters [61] and can be influenced by changes to riparian vegetation and macrophyte cover as well as fish predation [62,63]. Reduced plant cover significantly impacts other species and ecosystem traits, such as increased turbidity and sediment resuspension [64], reduced fish reproduction [65], increased planktivory by fish [84] and loss of epiphyton and periphyton communities [66]. A sustained change in vegetation, as well as invertebrate abundance and species composition in response to recreation, can ultimately affect higher trophic levels, such as birds and fish as predators [61]. Importantly, not all recreational activities will have similar impacts. For example, our study failed to find significant effects of recreational angling on invertebrates and plants, while the impacts of boating and shore use were strong and consistently negative on these taxa. Again, a key message of our work is that although recreational impacts on specific taxa are conceivable, they should not be generalized across all forms of recreation.

In birds, we detected high variation in effect sizes when comparing different recreational activities. This finding might be due to differences in species’ susceptibility to human presence and local variation in habitat quality, cover and user behaviour [40]. Moreover, FIDs vary among species and individuals of the same species [67], and this popular metric to assess the individual-level flight reactions of birds is also sensitive to local habitat and habituation. Other reasons for inconsistent responses to individual recreational activities could involve large variation in life histories, diets, nesting preferences, etc. of bird communities in riparian and wetland areas. Despite their high variation in behavioural responses, birds are considered highly disturbance-sensitive taxa and of significant conservation concern, probably explaining why bird studies were initially dominant in the material that we reviewed. In addition, they are relatively easy to study, with large amounts of readily available data. However, experimental manipulations of birds in freshwater contexts are rare, and observational studies, especially those of low quality, were found to generate larger negative effect sizes than higher-quality studies. These patterns might explain why the impacts on birds are well documented and significant, but also quite variable and less strong relative to other taxa, especially when compared to impacts on plants and invertebrates.

(d) . Study limitations and future research needs

Our work constitutes the first systematic, comprehensive and transparent meta-analysis that summarized the current evidence about freshwater recreation impacts on species, populations and ecosystems. However, we identified a series of limitations that are worth noting. Most of the studies identified used study designs that we scored being of low validity. In particular, most studies were observational without controlling for confounding factors, which might be one explanation for high variances in effect sizes observed for some taxa, levels of biological organization or recreation types. Although we were able to derive more than 700 effect sizes, it was not possible to generate effect sizes for all combinations of taxon and recreational use. We also found studies of lower validity tended to report stronger negative effects than studies with medium validity, which is another indicator of publication bias contributing to overreporting of recreational impacts [36]. Although we have strong evidence for publication bias in our dataset, our robustness checks (e.g. FSN) indicated that impacts would still be significant for shore use, angling and boating. These results can, therefore, be considered robust despite the study limitations identified. We strongly recommend more experimental studies and better reporting of even basic data in future studies to allow the pool of effect sizes to be enlarged in the future.

As another limitation most studies used a within-ecosystem design, comparing sites with and without recreation or along an activity gradient within in the same ecosystem. Such designs do not allow population- or community-level insights at the ecosystem scale. Even though the scale was not a significant moderator, there was a tendency that the ecological effects of recreation on the scale of whole water bodies were less pronounced than the effects on the scale of zones within the same water body. Thus, the interpretation of impacts at the scale of whole waters is limited. Experimental studies were generally very rare and very few used robust before–after–control–impact designs. This is a major shortcoming of the entire literature mainly relaying on observational data. Only experiments can identify true causalities and are thus strongly recommended for future studies.

Only 94 studies out of the 178 that were principally relevant could be used for this meta-analysis, because the rest did not allow the calculation of effect sizes as a result of missing data. We tried to be as inclusive as possible to maximize the sample size, by including studies with different scales and comparisons of different intensities, and used a series of robustness checks for our work. Importantly, we considered study quality using a weighting approach, which we found generated more conservative results and is thus considered superior to an unweighted approach. In this manner, we also addressed the problem of pseudoreplication in many studies by incorporating the number of studied water bodies in our validity measure. Nevertheless, the sample size for specific taxa and recreation type combinations was low and those average effect sizes should be treated with caution. In addition, the large number of bird studies compared to other taxa might have biased the overall summary effect size towards less negative. The detection of statistically significant but weak effects is typical of large sample sizes. Thus, many less prominent (as currently covered by literature) but nonetheless important ecological impacts might remain undetected because of low sample sizes. We strongly recommend performing more studies and engaging in more complete reporting to facilitate the extraction of data needed to calculate effect sizes in future research.

Even though this work can be considered a global meta-analysis because it incorporates studies from almost all continents (except Africa and Antarctica), the studies were conducted primarily in the Northern Hemisphere and mainly in only three countries. This might have introduced spatial and interpretative biases into the analyses that could not be avoided. More research on underrepresented continents is recommended. The bias towards wealthier countries should be addressed by supporting research in the global south. In addition, the large sample size regarding bird studies, compared to low sample sizes for other taxa might have biased the overall summary effect size to be less negative. The detection of statistically significant but not strong effects is typical of large sample sizes, and indicates that many less prominent but nonetheless very real impacts are also being experienced by other systems, but they are below our detection due to much smaller sample sizes. Further research is especially needed to study the impacts on reptiles, amphibians and in general population and community levels of all taxa. Emerging methods that might facilitate such research include drones, citizen science, remote cameras and sensor networks.

(e) . Conclusions and implications for conservation policy

Our synthesis revealed that recreation impacts on freshwater ecosystems are often negative, potentially widespread, but vary by recreational type, taxon and ecological response variable. Hence, ecological impacts of recreation can be expected, but they are context-specific and cannot be easily generalized. Therefore, policymakers and conservation managers need to engage in nuanced and contextualized evaluations about the impact of recreation rather than to bluntly assume that aquatic recreation necessarily constrains biodiversity or ecosystem status. Policymakers and managers are also recommended to carefully define conservation objectives, be explicit and transparent about the underlying normative decision framework, and engage proactively with stakeholders before implementing actions. Specifically, we recommend carefully defining which level of biological organization is considered ecologically or legally relevant when motivating conservation action, because our work has shown that outcomes of recreation are in some cases activity specific and vary from individuals to ecosystems. While individual-level conservation might be necessary for threatened species facing the risk of extinction, in most conservation contexts, conservation of population size, community integrity or habitat status is more relevant. Our work showed that ecological impacts of outdoor recreation may not be present in certain taxa and for certain levels of biological organization. In these situations, constraining recreation with the hope of achieving conservation gains may be an illusion that negatively affects human welfare without supporting conservation. Additionally, as the aggregated effect sizes of the various recreation types were often quite similar, we warn against conservation policies that are tailored to just one type of recreational use. Constraints on single-user groups may have no conservation benefit as long as other activities continue to use local ecosystems. In cases where decision-makers deem conservation action to be relevant, our work suggests that removing recreation and other human uses entirely from selected ecosystems may be most promising, followed by variants of spatio-temporal zoning of all uses. Clearly, removing access temporarily or permanently will strongly negatively affect human welfare, which has to be explicitly considered in the design and implementation of conservation policies to ideally achieve net gains for conservation and human welfare (e.g. by improving infrastructure in non-protected ecosystems as a compensatory measure). Any conservation measure should be carefully monitored, ideally using before–after–control–impact designs to generate the applied evidence base on whether managing recreation has the intended conservation benefits. In particular, monitoring the impacts of zoning and reduced access on the scale of whole waters is strongly recommended, as this practically relevant knowledge for conservation management is currently lacking.

Data accessibility

Data and R script are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.h18931zm3 [68]. Supplementary information are provided in the electronic supplementary material.

Authors' contributions

M.S., R.A. and C.W. conceptualized and designed the meta-analysis and the systematic review. Acquisition of data, literature screening and extraction of relevant information for effect size estimation was performed by M.S. with support from B.W. and N.M. M.S. and R.A. performed the analysis and drafted the manuscript. Interpretation of results, critical revision of the manuscript for important intellectual content and final approval was performed by all authors.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

The authors acknowledge funding by the Landesverband Sächsischer Angler e.V., the Landesfischereiverband Bayern e.V. and the Angler Association of Lower Saxony within the STÖRBAGGER-project (www.ifishman.de/en/projects/stoerbagger/). Additional funding came through the German Federal Ministry of Education and Research (BMBF) through the Aquatag project (grant no. 01LC1826E), BMBF together with the German Federal Agency for Nature Conservation (BfN) with funds granted by the German Federal Ministry for the Environment, Nature Conservation and Nuclear Safety (BMU) within the BAGGERSEE-Projekt (grant no. 01LC1320A; www.baggersee-forschung.de).