Logging impacts on avian species richness and composition differ across latitudes and foraging and breeding habitat preferences

(1) Literature search and data collection

We systematically searched for studies published within the past 60 years (1955–2015) examining the response of bird communities to logging. We looked for suitable peer-reviewed papers using the terms ‘birds’ and ‘logging’ on the ISI Web of Science online bibliographic database (accessed September 7, 2014), and also searched references within these papers. Limiting our search to published material potentially introduced a bias (Møller & Jennions, 2001) but also provided a level of quality control (Gray et al., 2007). We included experiments that compared avian species composition before and after logging as well as observational studies that compared recently logged forests to nearby undisturbed controls. Studies were included only if logging was the sole difference between treatment and control sites and forests were allowed to regrow following logging (i.e. we did not examine burning or agricultural effects). Many different forms of logging exist (e.g. clearcut, shelter-wood, selective, etc.), and these practices differ primarily in the amount of basal area removed/retained on the landscape. Thus, measurements of proportion of timber basal area retained on the landscape following logging (hereafter basal area retention) were used to control for variable logging intensities across studies. Regrowth since logging occurred is also likely to influence diversity responses to logging. Thus, the number of years of successional regrowth since logging occurred (hereafter years since logging) was used to control for variable ages of logging treatments. Because these two factors (basal area retention and years since logging occurred) are expected drastically to influence responses of diversity to logging, we limited our meta-analysis to studies that provided these measurements. We further limited our search to studies that surveyed bird communities within core logged areas to avoid the confounding influence of edge effects. We also had to limit our meta-analysis to studies that provided either a measure of species turnover (β-diversity) or complete species lists from unlogged and logged treatments from which β-diversity could be calculated (see below).

We defined regional species richness (γ-diversity) as all bird species detected in unlogged and logged forests. Local site richness (α-diversity) was defined as the identities of bird species in either unlogged or logged forest, respectively. We then calculated the per cent change in species richness following logging as (S_logged − S_unlogged)/S_unlogged, where S_logged is the number of species in the logged community and S_unlogged is the number of species in the unlogged community. Thus, positive values indicate a net gain of species following logging and negative values indicate a net loss of species.

We measured the change in species composition in response to logging with two β-diversity metrics: (i) the Jaccard dissimilarity index (β_J), and (ii) the Bray–Curtis dissimilarity index (β_BC). β_J measures the loss and gain of species because it only incorporates information about species presence and absence in either habitat. β_BC measures changes in the identity and relative abundance of species in multivariate space. These two metrics are recommended for examining differences in species composition among two or more sites (Anderson et al., 2011; Kraft et al., 2011; Myers et al., 2013), and both were calculated using the ‘vegan’ package in R (Oksanen et al., 2015). Both metrics range from 0 to 1, with higher values of either β-diversity metric representing greater change in species composition between logged and unlogged communities. Moreover, comparing these two measures provides additional information about logging effects on avian communities (Anderson et al., 2011). For example, logging in a given region may result in a high value (near 1) for β_BC but a relatively low value (near zero) for β_J. This pattern indicates that species are not lost or gained after logging, but indicates that logging strongly alters their relative abundances. However, because β_BC requires abundance data and β_J does not, we could only calculate β_BC for a subset (66%) of studies.

For those studies that provided abundance data, we also used null-models to calculate the change in species composition between logged and unlogged communities not explained by differences in regional species pools among latitudes (Kraft et al., 2011; Myers et al., 2013). All individuals encountered in logged and unlogged forest were randomly distributed among logging treatments (i.e. logged or unlogged), while preserving the total number of individuals in each treatment (Kraft et al., 2011; Myers et al., 2013). Thus, these null assemblages were only the product of stochastic assembly from the species pool, and all other mechanisms that might cause additional clumping (e.g. habitat specialization, local interactions among species, dispersal limitation) were removed. The pairwise dissimilarities of these simulated communities (β_SIM) were then compared to the observed dissimilarities (β_BC) relative to the standard deviation of β_SIM (σ_SIM) after 1000 iterations, and a standardized effect size of the difference was calculated as: β_SES = (β_BC − β_SIM)/σ_SIM. Therefore, β_SES represents the β-diversity that remains unexplained by stochastic assembly from the regional species pool (Kraft et al., 2011).

We recorded latitude and longitude of each study, and classified studies as tropical if they were above the tropic of Capricorn and below the tropic of Cancer. All other studies were classified as north or south temperate (including logging studies from boreal forests). This distinction was used to test for non-linear differences among tropical and temperate communities (e.g. habitat specialization) not linearly associated with latitude. Sample size (number of point counts, replicate plots, etc.) and spatial extent of each replicate sample (ha) were also recorded for each study. Because some studies used point counts (larger sample sizes, but relatively small spatial extent per sample) and others used territory-mapping techniques (smaller sample sizes, but relatively large spatial extent per sample), sample size and spatial extent per sample needed to be multiplied to calculate the total spatial extent of each study. We considered this measure of total spatial extent as the most appropriate way to compare studies that differ drastically in the spatial extent of each replicate sample. We also recorded the study method used for surveying birds (point-count, transect, spot-mapping, and mist-netting).

Bird species were classified into guilds based on the vegetation layer (or forest strata) in which they prefer to nest (cavity, canopy, understorey/shrub, or ground) and forage (aerial, canopy, understorey/shrub, or ground), based on their dietary preferences (frugivore, granivore, insectivore, nectarivore, or omnivore), and based on their body size using the Handbook of the Birds of the World (Del Hoyo et al., 2014) and Cornell's Birds of North America (Poole, 2005). Accounts of all species from our selected studies were found in one of these two resources. We classified species into one of four body-mass guilds based on the median and first and third quartiles of body mass among species: (i) species in the lower 25th percentile of body mass (<11.3 g); (ii) species in the 25–50th percentile of body mass (11.3–22.7 g); (iii) species in the 50–75th percentile of body mass (22.7–57.9 g); and (iv) species with body masses greater than the 75th percentile (>57.9 g).

(2) Statistical analyses

We used linear models to examine responses of avian species richness and composition to logging. If a given study examined multiple unique logging treatments (e.g. clearcut, 25, and 50% retention), then each treatment was included as a sample in our meta-analysis provided that each treatment was compared to control (undisturbed) forest. Instead of including a random effect of study (i.e. study site) to group logging treatments in our analyses, we included the relevant continuous variables of interest that differed among study sites: latitude and longitude. Specifically, we were interested in different responses of diversity to logging across latitudes. Because 75% of studies presented results from only one or two logging treatments, a random effect of study site is nearly perfectly correlated with latitude and therefore inappropriate to include in a model testing for a latitude effect (Clark & Linzer, 2015). Including latitude in all of our models appropriately groups studies by the relevant variable of theoretical interest, and no additional random effect of study site was necessary. However, logging treatments from the same study site may be spatially auto-correlated. Thus, we tested for any spatial auto-correlation in our data set using Mantel's test (‘mantel.rtest’ function from R package ‘ade4;’ (Dray & Dufour, 2007).

We began by verifying our measure of γ-diversity by testing for the familiar pattern of decreasing diversity with the absolute value of latitude (hereafter absolute latitude). This measure of γ-diversity may be confounded with sample effort, but total spatial extent only explained <10% of variation in γ-diversity across studies (r² = 0.099, P = 0.001). We then examined factors influencing the change in local species richness or composition (i.e. β_J, β_BC, and β_SES) following logging with a linear model that included a categorical variable indicating tropical or temperate forest (hereafter called latitudinal zone; tropical forests were defined as being at or below the Tropics of Capricorn or Cancer), absolute latitude (the absolute value of latitude), basal area retention, years since logging, γ-diversity, and interactions between latitudinal zone and absolute latitude, basal area retention, and years since logging. Clearcut harvests removed >90% of tree basal area, but generally removed all standing trees. Thus, this logging technique may have drastically different effects on vegetation and animal communities compared to selective harvest (removal of <90% basal area). Moreover, clearcut harvests were only prevalent in the temperate zone. Thus, we split temperate logging into clearcut and selective harvest but kept basal area retention in the model to account for substantial variation in retention among selective harvests. Finally, before implementing each model for species richness or composition, we tested for simple correlations between total spatial extent and the change in species richness or composition (Hill & Hamer, 2004). If a relationship was detected, we incorporated total spatial extent directly into the model. If a relationship was absent, we weighted each study by their total spatial extent in order to favour those studies with greater sampling effort and precision. Variables were log-transformed if their distributions were heavily right-skewed. We then sequentially removed insignificant (P > 0.05) interactions from this full model to arrive at a more parsimonious model. We also performed funnel plots, which test for a significant correlation between model residuals and sampling effort, to assess publication bias (Egger et al., 1997). If bias is present, model residuals will be significantly correlated with sampling effort. If bias is absent, model residuals should converge to zero for those studies with the greatest sampling effort, forming a funnel shape. We also used a likelihood-ratio test to assess the effect of avian study method (i.e. point-count, transect, mist-netting, or spot-mapping), but study method did not influence our models for species richness (χ² = 3.6, P = 0.31), β_J (χ² = 1.8, P = 0.62), or β_BC (χ² = 4.0, P = 0.26).

We also tested for differential impacts of logging on species richness or composition of guilds that nest or forage in different forest strata, have different dietary preferences, or have different body mass. We used identical approaches to those above with an added fixed factor of guild and an interaction between guild and latitudinal-zone/logging-type (i.e. tropical selective harvest, temperate selective harvest, temperate clearcut harvest). Again, a random effect of study site was unnecessary because logging treatments were already grouped by location (i.e. latitude and longitude).

We then tested if the impacts of logging on local avian species richness or composition differed among north-temperate forest types (i.e. pure deciduous or conifer forest versus mixed conifer–deciduous forest). Identical approaches to those described above were used to test for a difference in species richness and composition across pure and mixed forest types. We also tested for interactions of forest type with logging type (clearcut versus selective harvest), basal area retention and years since logging to evaluate if forest type influenced diversity responses to logging. All analyses were performed using program R (R Core Development Team, 2014).

(1) Data description

Sixty-two studies matched our search criteria (Fig. 1, see online Supporting Information, Table S1). Unfortunately, Africa and temperate Eurasia were under-represented in our analyses because most studies in these regions focused solely on richness and few published the necessary data to calculate changes in species composition. Thirty-three of the 62 studies examined multiple logging treatments, resulting in 119 control–treatment replicates for our meta-analysis. These studies spanned a wide range of logging intensities (0.0–93.0% basal area retention; mean ± S.D. = 33.0 ± 27.8%) and years since logging (1.0–80.0 years; mean ± S.D. = 11.9 ± 15.2 years). Thirty-three replicates were from clearcut forests (all in the temperate zone) and 86 were from selectively logged forests (38 from tropical and 48 from temperate latitudes). Most studies were published to report logging effects on species richness, and 60% of studies reported statistically insignificant effects of logging on richness. Thus, publishing bias should not be a major factor influencing the meta-analyses. Indeed, funnel plot regression tests indicated that publication bias did not influence our global models for species richness (t = 0.38, P = 0.71) nor either compositional measure (β_J: t = 1.12, P = 0.27; β_BC: t = −0.63, P = 0.54) (see online Supporting Information, Fig. S1). Spatial auto-correlation was not present in datasets for changes in species richness (r = −0.064, P = 0.96) or species composition (β_J: r = 0.028, P = 0.21; β_BC: r = 0.013, P = 0.37). As expected, γ-diversity (r² = 0.460) decreased with increasing latitude (Fig. 2).

(2) Logging effects on species richness across latitudes

Effects of selective logging on bird species richness differed strongly among tropical and temperate forests and among harvest type in the temperate zone (Fig. 3). Logging reduced species richness across the tropics, with an average loss of 11% of species 6 years after a 50% basal-area retention harvest (Fig. 3A). Retaining less basal area increased these losses (Fig. 3B), with an average loss of 22% of species 1 year following low basal-area retention (∼5%) logging in the tropics. On average, all levels of basal-area retention in the tropics resulted in loss of species (Fig. 3B). More importantly, species richness never approached pre-logging levels in tropical forests over time, with an average loss of 10.3% of species 40 years after a 50% basal-area retention harvest (Table 1). In contrast to selective logging in the tropics, selective logging in temperate forests did not reduce bird species richness and corresponded with slightly increased richness if at least 60% of basal area was retained (Fig. 3B). Not surprisingly, temperate clearcut logging had strong impacts on bird diversity, with an average loss of 16% of species 6 years after harvest. These losses were even greater at higher latitudes, with an average loss of 27% of species 6 years following clearcuts at or above 50° latitude (Fig. 3A). Thus, logging generally reduced species richness in the tropics. Limited disturbance via logging yielded increases in species richness in temperate forests, but clearcut logging yielded reductions in richness.

We were able to classify over 1450 bird species into guilds based on the vegetation layer (or forest strata) in which they nest or forage, dietary preferences, and body mass from 41 of the studies used above. This represented all species documented in these 41 studies. We were only able to classify species from these 41 studies because other studies failed to provide detailed species lists. Logging effects on species richness of these guilds differed across latitudes and harvest types (Fig. 4, Table 2). Logging in tropical forests generally decreased avian richness across all nesting, foraging, dietary, and body-mass guilds except for omnivores (Fig. 4). Ground-nesting, ground-foraging, and large-sized (>75th percentile body mass) guilds experienced the largest declines in species richness with tropical logging (Fig. 4). By contrast, nearly all guilds had increases in species richness following selective logging in temperate latitudes, especially ground- and aerial-foraging guilds, granivores, and birds between 22.7 and 57.9 g body mass (Fig. 4). Clearcut logging in temperate latitudes resulted in a net loss of species from nearly all guilds, especially frugivores, large-sized species, and guilds that nest or forage in the canopy (Fig. 4). Therefore, habitat and dietary preferences of species determined their responses to logging. Effects of logging on species richness of guilds occupying lower vegetation layers differed starkly among latitudes, with lower-strata tropical birds decreasing in richness and lower-strata temperate birds increasing in richness following logging (Fig. 4A, B).

(3) Logging effects on species composition across latitudes

The effect of logging on bird species composition also differed across latitudes and harvest types (Fig. 5, Table 3). β_J, which measures the loss and gain of species following logging, was nearly constant across tropical latitudes, dropped steeply at lower temperate latitudes, and then increased again at higher temperate latitudes (Fig. 5A). Thus, β_J was roughly constant following selective harvests in tropical and temperate latitudes except between 30° and 40° latitude. This pattern indicates that relatively fewer species were lost or gained following logging in lower-temperate latitudes compared with either tropical or higher-latitude temperate logging. Temperate clearcut harvests showed steeper increases in β_J with latitude than temperate selective harvests (Fig. 5A), corresponding with substantial loss of species following clearcut logging at higher latitudes (Fig. 3A).

By contrast, β_BC (which measures changes in species relative abundances) was approximately constant across latitudes following selective logging in tropical and temperate forests (Fig. 5B). This indicates that while few species were lost or gained following selective logging in lower-latitude temperate forests, the change in the relative abundances of these species was similar to that in tropical and higher-latitude regions. β_BC changed more dramatically following temperate clearcut harvests (Fig. 5B, Table 3). The change in bird species composition following temperate clearcut logging tended to converge with the change in bird species composition following temperate and tropical selective logging after roughly 30 years of forest regrowth (Fig. 6, Table 3). Nonetheless, species composition was still distinct from unlogged forests (i.e. β_BC > 0) even 80–90 years after logging (Fig. 6). Analyses of β_SES, which represents changes in species composition that remain unexplained by stochastic assembly from the regional species pool, revealed quantitatively similar patterns as analyses of β_BC (Table 3) and were positively correlated with β_BC (r = 0.60, P < 0.001). Thus, observed patterns in the change in species composition following logging across latitudes did not appear to be driven by differences in regional species pools across latitudes.

Effects of logging on species composition, especially β_BC, were strong but less variable among nest-site, forage-site, dietary, and body-mass guilds than effects of logging on species richness of these guilds (Fig. 7, Tables 4 and 5). Selective logging appeared disproportionately to alter the species composition of tropical canopy-nesting species, tropical ground-foraging species, tropical granivores, and tropical large-bodied species more than other guilds at any latitude (Fig. 7). Temperate clearcut harvesting had substantial impacts on avian species composition across guilds, but especially on cavity- and canopy-nesting species, canopy- and aerial-foraging species, frugivores, insectivores, and large-bodied species (Fig. 7). Thus, despite differential effects of logging on species richness across latitudes and guilds, the composition of bird species in all guilds changed substantially following logging at all latitudes.

(4) Logging effects differ among north-temperate forest types

Effects of logging on avian species richness and composition varied substantially among pure and mixed north-temperate forest types (Fig. 8). Logging in either pure conifer or deciduous forests types tended to reduce bird species richness on average, but logging in mixed conifer–deciduous forests resulted in no net change in species richness (Fig. 8A). Furthermore, forest type appeared to explain the decline in β_J at lower north-temperate latitudes. Logging caused more loss/gain of species (β_J) in higher-latitude north-temperate forests than in lower-latitude north-temperate forests (Fig. 5A). However, variation in β_J was better explained by forest type than by latitude in the north-temperate zone (Table 6, Figs 5 and 8B), suggesting that greater changes in species identities following logging at higher latitudes may be associated with the general transition from pure deciduous or conifer forest to mixed conifer–deciduous forest. However, the change in species relative abundances following logging (β_BC) did not differ among pure and mixed north-temperate forest types (Fig. 8C), indicating that pre- and post-logging communities in lower north-temperate latitudes had similar species but differed in the relative abundances of those species. Thus, the observed pattern of increasing β_J but constant β_BC across temperate latitudes may be caused by geographic shifts in the abundances of conifer and deciduous trees.