Every Tree Is Precious: Biodiversity Suffers From Even Minimal Logging

From the set of 34 candidate models proposed to explain the difference in species richness between a logged forest and an equivalent primary forest site, the most parsimonious one was model M13 (richness ∼taxon ^∗ logging intensity). This model had the lowest AICc (Akaike Information Criterion corrected for a small sample size) value and explained 30% of deviance in the data relative to the null model, M0 (Table 1) . Model M13 indicates that logging intensity, within the range considered in this study (0.7 m³ ha⁻¹ to 200 m³ ha⁻¹), is a good predictor of the impact of selective logging on species richness. Additionally, the nature of this relationship differs among the four taxonomic groups we considered (mammals, amphibians, invertebrates, and birds).

Table 1Candidate Models Explaining the Relative Species Richness in Logged Tropical Forests, Ranked According to Increasing Value of ΔAICc

Model	Equation	log(L)a	Kb	ΔAICcc	wid	DE (%)e
13	richness ∼taxon ∗ intensity	−405.75	9	0	0.52	30
29	richness ∼taxon ∗ continent ∗ intensity	−388.46	21	0.28	0.46	52
11	richness ∼taxon ∗ continent	−406.45	11	6.5	0.02	29
15	richness ∼taxon ∗ distance to unlogged	−411.98	9	12.46	0	20
3	richness ∼logged area	−420.33	3	15.28	0	5
0	richness ∼1	−422.49	2	17.44	0	0
33	richness ∼taxon ∗ continent ∗ intensity + taxon ∗ time + altitude + distance to unlogged + logged area + forest area + stems	−391.79	24	17.57	0	48
30	richness ∼taxon ∗ continent ∗ time	−397.51	21	18.37	0	42
5	richness ∼intensity	−421.97	3	18.54	0	1
2	richness ∼continent	−420.94	4	18.67	0	3
8	richness ∼distance to unlogged	−422.14	3	18.89	0	1
7	richness ∼altitude	−422.18	3	18.97	0	1
9	richness ∼proportion logged	−422.2	3	19.02	0	1
4	richness ∼time	−422.27	3	19.15	0	0
6	richness ∼intensity	−422.31	3	19.23	0	0
20	richness ∼intensity + time	−421.8	4	20.39	0	1
22	richness ∼time + proportion logged	−421.82	4	20.42	0	1
25	richness ∼intensity + stems	−421.82	4	20.44	0	1
23	richness ∼intensity + proportion logged	−421.9	4	20.6	0	1
21	richness ∼time + distance to unlogged	−421.96	4	20.71	0	1
1	richness ∼taxon	−421.54	5	22.12	0	2
27	richness ∼intensity + time + stems	−421.67	5	22.36	0	2
12	richness ∼taxon + intensity	−420.88	6	23.08	0	3
10	richness ∼taxon + continent	−419.76	7	23.18	0	6
14	richness ∼taxon + distance to unlogged	−421.15	6	23.62	0	3
24	richness ∼taxon ∗ altitude	−417.72	9	23.94	0	10
18	richness ∼taxon + proportion logged	−421.32	6	23.95	0	2
16	richness ∼taxon + time	−421.35	6	24.02	0	2
26	richness ∼taxon + proportion logged + continent	−419.07	8	24.2	0	7
28	richness ∼taxon + intensity + stems	−420.83	7	25.31	0	4
31	richness ∼taxon + time + stems	−421.73	7	25.81	0	4
17	richness ∼taxon ∗ time	−419.17	9	26.85	0	7
19	richness ∼taxon ∗ proportion logged	−420.21	9	28.92	0	5
32	richness ∼taxon + continent + time + intensity + altitude + distance to unlogged + logged area + forest area + stems	−416.44	14	34.61	0	12

Response variable is the relative total species richness for all taxa. Variable codes re described in the Experimental Procedures. See also Table S1.

aMaximum log likelihood.

bNumber of parameters.

cΔAICc_i = AICc_i – AICc_min.

dAkaike weights.

eDeviance explained by the model .

The second most parsimonious model was model M29 (richness ∼taxon ^∗ logging intensity ^∗ continent), which also had a relatively low AICc value (ΔAICc = 0.28) and explained 52% of deviance in the data (Table 1). This more complex model further implies that the relationship between species richness, logging intensity, and taxonomic group also varies according to the continental location of the study site (Afrotropical, Indomalayan, or Neotropical).

The final, best model, obtained from the model averaging of the two most parsimonious models, can be expressed as

(Equation 1)

where , , and and where the relative species richness, S, of taxonomic group, i, on continent, j, depends on logging intensity, L, taxon- and continent-specific intercept, α, and slope, β (Table S3).Our final model suggests that logging intensity, taxonomic group, and continental location are more important for predicting logging impacts than are the other explanatory variables on which we have sufficient information, such as area of forest fragment, isolation from primary forests, time since disturbance, or proportion of the forest that has been logged [16, 17, 18]. This is also shown by the high cumulative Akaike weight of the two best models (0.98).

Our results from model averaging show that there are more species of mammals, amphibians, and invertebrates in lightly logged forests than in primary forests (Figures 2 and S2 and Table S3) . However, as logging intensity increases, the relative species richness of these three taxonomic groups decreases linearly, dropping below their primary forest richness at logging thresholds of 10 m³ ha⁻¹ for mammals (95% confidence interval [CI]: 4–18 m³ ha⁻¹), 23 m³ ha⁻¹ for amphibians (95% CI: 0–49 m³ ha⁻¹), and 41 m³ ha⁻¹ for invertebrates (95% CI: 0–78 m³ ha⁻¹). Crucially, mammals and amphibians would drop below 50% of their species richness in primary forests at the logging intensities of 38 m³ ha⁻¹ and 63 m³ ha⁻¹, respectively, if the linear trend of decreasing richness continues (95% prediction interval: 18–140 m³ ha⁻¹ for mammals and 3–200 m³ ha⁻¹ for amphibians).

We found that mammals and amphibians are the most sensitive groups to selective logging in terms of the rate of decline in species richness (Figure 2 and Table S3). Every increase of 20 m³ ha⁻¹ in logging intensity results in an approximately 35% decrease in mammal species richness. In contrast, recent reviews conclude that mammals are the least sensitive taxonomic group to logging [2, 12]. Whereas most previous studies focused on analyzing the average response of species to selective logging as a uniform land use (mean effect size), we explicitly quantified the responses of species along a gradient of logging intensity.

The high sensitivity of mammals to selective logging may be due to the increased hunting and poaching pressure, often associated with the increased incursion into forests through logging roads [19]. Although we excluded studies that confirmed hunting in either the logged or control study sites (see the Experimental Procedures), it is nevertheless possible that hunting might be underreported in some logged forests [20, 21].

Primates are not included in our data set, as the vast majority of studies on the impact of logging on primates measures primate abundance, rather than richness, which is typically very low [22].

The steep decline of amphibian species richness in logged areas may be due to their high sensitivity to hotter and drier microclimates, which are created by selective logging [23, 24]. This effect of climate might be compounded by the lower vagility of amphibians [25]. Additionally, direct mortality of amphibians during a logging operation might be higher than that for other, more vagile, taxa. In broader and global studies of forest conversion, including secondary forests and agroforestry, amphibian richness has been found to decrease with decreasing canopy cover and leaf litter depth [21, 26].

Invertebrates showed a relatively gradual decrease in species richness with logging intensity. The invertebrate data set included mainly studies on butterflies, dung beetles, ants, and a few other terrestrial taxa. Butterfly richness in disturbed and secondary forests has been previously found to depend on the spatial scale of the study [27], and dung beetle richness in secondary forests depends in part on the forest fragment size [28]. However, focusing exclusively on selectively logged forests, we found that logging intensity is the most important determinant of invertebrate species richness. In temperate forests, a similar relationship of species richness declining with decreasing forest retention levels has been shown for certain invertebrate groups [29, 30].

Birds exhibit a contrasting trend, whereby their species richness gradually increases with logging intensity (Figure 2 and Table S3). An equivalent pattern of increasing bird species richness with increasing logging intensity has been found in Ghana, where, however, only low logging intensities were considered [14]. Similarly, in temperate forests, increasing intensity of forest thinning has also been found to be correlated with higher bird species richness [31].

Our analysis of the forest specialist bird species shows that when total species richness of birds increases above the control site richness, the richness of forest specialist birds, S_FS, decreases (Figure 3) . The relative richness of habitat generalists, S_FG, increases linearly with the total relative species richness of all birds (Figure 3). This means that whenever logging leads to higher total bird species richness, the number of forest specialist species markedly decreases, which is outweighed by the strong increase in the richness of habitat generalists.

This suggests that the seemingly low susceptibility of birds to selective logging, when all species are considered, is due to an influx of habitat generalists (Figure 2 and 3). A similar pattern has been suggested by numerous studies in both temperate and tropical forests under various forms of degradation [15, 21, 32] but had previously not been quantified in logged tropical forests.

To some extent, the positive response of total bird species richness to logging intensity could be also explained by an observational bias. Birds are more easily detected in disturbed habitats, including logged forests [33]. Additionally, some bird species, being more mobile, might be overall less susceptible to logging, as they might be using different patches of (logged) forests for different resource needs. Certain frugivorous and nectarivorous species may successfully forage in forests logged at intermediate intensities, where food is more abundant, while still nesting in primary forests [34].

On the other hand, the decreasing richness of forest specialists at higher logging intensities might be further exacerbated by unreported hunting of large bird species, such as hornbills, macaws, or pigeons [35, 36]. Further, it is possible that historical hunting pressure had removed most hunted birds in some control and logged forests, which would mask some effects of logging [35, 37, 38].

Neotropical fauna appears to be more sensitive to logging intensity than Afrotropical and Indomalayan fauna, apart from mammals, which decline even more steeply in richness in Africa than in the Neotropics (Figure S2 and Table S3). There is a lack of studies of species richness in selectively logged forests on mammals and amphibians from the Indomalayan region. The differential responses of fauna to logging on the three continents might be connected with different total, per hectare of biomass, as well as with different logging practices.

Our conclusions cannot be extrapolated beyond the logging intensities considered in our analysis (approximately 0.2 to 20 trees ha⁻¹). Nonetheless, through our comprehensive review of logging studies, our analysis reflects a realistic range of logging intensity occurring in the tropics. We recognize that a weighted meta-analysis could have improved the confidence with which we can estimate our model parameters and logging intensity thresholds. However, inconsistent error reporting across individual logging studies precluded us from doing so. During the literature review, we faced a tradeoff between the number of data points that we include (over 200 studies could have potentially been included) and their quality. To maintain high standards of the meta-analysis, we have included only data that fulfilled our quality criteria, as described in the Experimental Procedures. We also recognize that some logging studies might suffer from pseudoreplication in their sampling design [7]. Nonetheless, this potential problem would not affect a meta-analysis of species richness, which we are primarily interested in (as opposed to similarity in species composition) [7]. While we realize that there are multiple metrics of biodiversity (e.g., beta diversity or functional diversity), we focused on local species richness (alpha diversity) because it is the most widely reported metric in the context of selective logging. Species richness has to be interpreted with caution, as shown also by our supplemental findings on forest specialist and generalist bird species.

The quest for finding meaningful thresholds of harvest intensity that might be crucial to biodiversity is not unique to tropical forest conservation. In temperate and boreal regions, retention forestry has become widely practiced over the last three decades. Whereas selective logging focuses mainly on how much timber is harvested, retention forestry prescribes how many trees remain after harvest without the intention of extracting them in the future. Despite different goals, parallels can be drawn. Several studies have for example examined how different levels (and spatial configuration) of retention logging affect biodiversity [29, 39, 40]. However, a consensus on optimal retention level thresholds for biodiversity conservation is yet to be found [30, 41].

Our results show that the volume of timber extracted is the most important predictor of species richness in logged forests. However, logging intensity should not be the sole basis of forest management strategies. First, a wealth of studies shows that reduced-impact logging is a desirable practice in tropical forestry for maintaining plant diversity, soil quality, and carbon accumulation [42, 43]. Second, there is mounting evidence that forests should be managed as complex adaptive systems [44]. Therefore, adaptation of logging intensities should be accompanied by, for example, the maintenance of ecosystem legacies, such as hollow trees that are not valuable to timber production but are crucial to biodiversity [21, 45]. Additionally, which individuals will be spared from harvest should not be dictated purely by the evolving marketability of tropical timber species.

Selectively logged forests are becoming an increasingly dominant component of many tropical landscapes [2]. And yet, the conservation value of selectively logged tropical forests is less understood, and most likely more overestimated, than those of more dramatic land cover changes, such as deforestation driven by agriculture or tree-plantation developments. Our analysis reveals that mammals and amphibians are particularly sensitive to logging intensity and would suffer halving of species richness at logging intensities of just 38 m³ ha⁻¹ and 63 m³ ha⁻¹, respectively. On the other hand, most taxonomic groups would be resilient to selective logging at intensities of less than 10 m³ ha⁻¹. Our study provides a quantitative analysis of the nuanced responses of species along a gradient of logging intensity that could help inform evidence-based sustainable logging practices from the perspective of biodiversity c