FSC-certified forest management benefits large mammals compared to non-FSC

Data for: 
Zwerts, J.A., Sterck, E.H.M., Verweij, P.A. et al. FSC-certified forest management benefits large mammals compared to non-FSC. Nature 628, 563–568 (2024). https://doi.org/10.1038/s41586-024-07257-8

Abstract
Over a quarter of the world's tropical forests are exploited for timber. This has extensive impacts on biodiversity and ecosystems, primarily through the creation of roads which facilitates hunting for wildlife over extensive areas. Forest management certification schemes such as the Forest Stewardship Council (FSC) are expected to mitigate impacts on biodiversity but to date very little is known about the effectiveness of FSC certification due to research design challenges, predominantly due to limited sample sizes. Here we provide this evidence by using 1.3 million camera trap photos of 55 mammal species in 14 logging concessions in Western Equatorial Africa. We observed higher mammal encounter rates in FSC-certified than in non-FSC logging concessions. The effect was most pronounced for species weighing over 10 kg, and for species of high conservation priority such as the Critically Endangered forest elephant and western lowland gorilla. Across the whole mammal community, non-FSC concessions contained proportionally more rodents and other small species than FSC-certified concessions. Our findings provide convincing data that FSC-certified forest management is less damaging to the mammal community than non-FSC forest management. This study provides strong evidence that FSC-certified forest management, or equivalently stringent requirements and controlling mechanisms, should become the norm for timber extraction to avoid half-empty forests dominated by rodents and other small species.          

Data collection

We set up arrays of camera traps from 2018 to 2021 in 14 logging concessions owned by 11 different companies (5 FSC and 6 non-FSC) in Gabon and the Republic of Congo. Seven FSC-certified concessions were each paired to the closest non-FSC concession that was similar in terms of terrain and forest type1. All concessions are situated in a matrix of connected forests. Within each pair of concessions, camera traps (Bushnell Trophy Cam HD for pairs 1 - 6 and Browning 2018 Spec Ops Advantage for pair 7) were deployed simultaneously to account for seasonal differences, for two to three months. There was one exception where Covid restrictions obliged the cameras to remain in place for longer. Camera trap grid locations within each pair of concessions were chosen based on similarity between potential drivers of mammal abundance, including distance to settlements, roads, rivers, protected areas, elevation and time since logging (2-10 years before our study), although some camera grids overlapped older logging blocks. Camera traps were set out in systematic, one-kilometre spaced grids with a random start point. Upon reaching the predetermined GPS locations, the first potential installation location was used where cameras had at least 4 metres of visibility. This ensured that each grid was representative of environmental heterogeneity: i.e. not specifically targeting nor ignoring trails or other landscape elements that could influence detection2. The one-kilometre inter-camera distance exceeds most species' home range sizes to avoid spatial autocorrelation. Neither were species expected to migrate within the sampling duration of the study. Between 28 to 36 cameras were deployed in each concession, totalling up to 474 camera traps, distributed over 474 km2. Cameras were installed at a height of 30 cm to enable observations of mammals of all sizes while ensuring that each camera had at least four metres visibility in front of it. Cameras were programmed to take bursts of three photos to maximize the chance of detection and to take a photo every 12 hours for correct calculation of active days in the event of a defect before the end of the deployment period. For each camera, we recorded whether there was an elephant path skidder trail, small wildlife trail, or none of the above within each camera's field of view. We also visually estimated forest visibility (0-10m / 11-20m / >20m), slope (0-5° / 5-20° / >20°), presence of fruiting trees within 30 m and presence of small water courses within 50 m. When approaching each predefined camera point, we counted cartridges, snares and hunting camps from 500 m before the camera up to its location. Various field teams were employed in different sites and hence there may be some influence interobserver bias of hunting observations between sites.

Photo processing and data analysis

Camera trap efforts yielded 1,278,853 photos, including 645,165 photos with animals. All photos were annotated in the program Wild.ID, version 1.0.1. We identified animals up to the species level if photo quality permitted and otherwise designated the species as 'indet'3. As reliable species identification of small mammals is difficult, they were grouped into squirrels, rats and mice and shrews. Rare observations of humans, birds, bats, reptiles and domestic dogs were excluded from the analyses.

Observations of the same species that were at least 10 minutes apart were considered as separate detections. We assessed the influence of this threshold by calculating the number of detections for intervals of 10, 30, 60 and 1440 minutes, which all yielded similar numbers of observations. When multiple animals were observed, the number of individuals was determined by taking the highest number of individuals in a photo within the 10-minute threshold. Sampling effort was defined as the number of camera days minus down time due to malfunctioning cameras or obstruction of vision by vegetation. 

Mammal behaviour may be different in hunted concessions, as mammals may be shyer for non-natural objects like camera traps, which would in turn negatively affect their probability of detection. If this dynamic indeed existed, this shyness was assumed to fade over time with habituation to the materials, resulting in an increase of observations over time. We tested for an interaction between certification status and the number of observations over time using a linear model with a log transformed number of observations for the first 65 days of all deployments, as that was the shortest concession deployment period, ensuring that all concessions were equally represented, but did not find that FSC certification was related to a trend in observations over time. We recognize however that other factors may exist that may have influenced detection probability, such as movement rates, which may be affected by hunting.

For each species for each concession, we calculated encounter rate, weighted by group size, as the number of observations divided by the sampling effort and we reported all findings using the metric "Observations / day". Encounter rate was calculated for all species combined, per weight class, per taxonomic group, per IUCN Red List category and within taxonomic groups for large versus small forest antelopes, carnivores, primates and pangolins. Body weight of each species was determined by taking the mean across sexes3. Taxonomic groups Hyracoidea and Tubulidentata were excluded from the taxonomic analysis due to low sample sizes. Shrews were included as rodents in the taxonomy analysis even though they are formally not, because they are difficult to distinguish from mice. We consider this acceptable given that shrews are functionally very similar to rodents in the light of this study. Lastly, to study the impact of certification on total estimated faunal biomass, the encounter rate of each species was multiplied by its average weight divided by the sampling effort.

To assess whether encounter rates varied between FSC and non-FSC concessions, we quantified the means of the paired concessions using linear mixed-effects models with concession pairs, concessions and cameras as random effects, whereby cameras were nested within concessions within concession pairs, in a multi-level random effect structure. We allowed the means of concession pairs to vary between weight class, taxonomic group and IUCN Red List category if supported by model selection. We tested whether potential drivers of mammal abundance were important using a model-selection approach based on minimization of Bayesian Information Criterion (BIC) values. We found that the inclusion of geographic covariates did not substantially improve the model for weight classes, taxonomic groups and IUCN categories. Only for all mammals pooled together , the inclusion of elevation and distance to rivers resulted in slightly improved models, but differences were negligible and did not support strong evidence for a significant influence of these covariates4.  Quadratic geographic covariate terms and camera trap site covariates did not result in better models. Post hoc comparisons were multivariate t adjusted. We used two-sided Wilcoxon signed-rank tests for all other analyses. Statistical analyses were performed in R version 4.2.2.

References

1. Grantham, H. S. et al. Spatial priorities for conserving the most intact biodiverse forests within Central Africa. Environmental Research Letters 15, 0940b5 (2020).
2. Zwerts, J. A. et al. Methods for wildlife monitoring in tropical forests: Comparing human observations, camera traps, and passive acoustic sensors. Conserv Sci Pract 3, e568 (2021).
3. Kingdon, J. The Kingdon field guide to African mammals. (Bloomsbury Publishing, 2015).
4. Anderson, D. R. Model based inference in the life sciences: a primer on evidence. vol. 31 (Springer, 2008).