Large‐scale impacts of selective logging on canopy tree beta‐diversity in the Brazilian Amazon

Tropical forests support extensive biodiversity (Pimm & Raven, 2000) and maintain globally significant carbon stocks (Pan et al., 2011). Yet these are subject to widespread selective logging (Blaser 2011), which is the primary driver of tropical forest degradation (Hosonuma 2012; Pearson 2017). Selective involves harvesting commercially valuable species a minimum cutting size, leaving behind non-commercial species, smaller individuals, degraded structure with fragmented canopy patches intense sunlight (Bousfield 2020). Despite this disturbance, selectively logged often retain most their ecosystem services relative (Edwards, Tobias, 2014; Putz 2012) much more biodiverse than competing land-uses, especially agriculture (Gibson Harvest intensity plays an important role in determining severity impacts on biodiversity, higher intensities resulting greater losses tree richness (Martin 2015). Large, trees targeted by vulnerable local population collapses (Richardson Peres, 2016), yet play irreplaceable ecological roles (Lindenmayer Pinho They provide food habitat for multitude organisms, giving that provides humid, dark, cool microclimates, account large proportion total biomass (Nascimento Laurance, 2002). thus also drives favourable microclimates short term (Mollinari 2019) above-ground over longer timescales (Sist 2014). Understanding diversity has tended focus alpha (i.e. richness) associated changes community composition measured at small spatial scales (Cazzolla Gatti 2015; Gourlet-Fleury 2013; Martin However, focusing across areas within few relatively plots (1 ha or smaller) risks underestimating larger scales. Logging activity not evenly distributed forest, creating heterogenous mosaic undergo different levels disturbance (Cannon 1994). Such heterogeneity conditions could lead shifts landscape-level detected local-scale assessments (Berry 2008), observed responses being heavily impacted scale sampling (Dumbrell 2008; Hill Hamer, 2004). Furthermore, grain size units) can influence variation (Barton 2013). A key question, therefore, how varying grains. Beta diversity—the component regional (gamma) accumulates due variations assemblages (Socolar 2016)—could be through two processes. Firstly, heterogenization, whereby communities become increasingly from each other, as same set high-value shared consistently entire landscape. Alternatively, homogenization, similar left behind. beta-diversity partitioned into separate components, turnover nestedness, both selected (Baselga, 2010). Turnover occurs when present one site, but lost another where they replaced species. nestedness site contains only subset site. Our understanding potential limited. In Borneo, had increased compared unlogged forests, suggesting logging-induced heterogenization 2008). Amazon, patterns were ~10 years after logging, recovered towards pre-logging state 25 (Gaui 2019). studies relied number 1 (30 12, respectively) dispersed landscape, fails capture substantial hyperdiverse disturbances affect landscapes, particularly considered. contiguous high importance policy conservation efforts aiming 2016). study, we tackle question We do so leveraging detailed distribution maps containing information 155,000 (DBH >35 cm) 3100 continuous Amazonian concession. The Amazon (Ter Steege 2013), including some highly timber (e.g. big-leaf mahogany- Swietenia, ipê- Tabebuia, jatobá- Hymenaea; Schulze largest unexploited (Merry 2009). Within Brazilian Forestry Service (Serviço Florestal Brasileiro; SFB) been undertaking ambitious programme since 2006 create network long-term legal concessions National State Forests (FLONAs FLOTAs). combination other designated sustainable use, up 35–50 million hectares potentially available harvest coming decades (de Marques 2016; Sist 2021). Ensuring does erode issue. generate simulations gradient ask, wider classes: (1) whether affected mechanism heterogenisation homogenization); (2) primarily responsible explaining change; (3) effects observed. study area consists wet lowland northern Basin Guianan Shields, Vale Jari region, Pará, Brazil (1°13′12″S 52°33′36″W). was granted Grupo Orsa forestry company ‘reduced-impact logging’ (RIL) concession management Institute Environment Renewable Natural Resources (IBAMA). Between 2001 2003, complete census undertaken, spatially mapping 5-m positional accuracy all stems ≥35 cm diameter breast height (DBH), identified level team experienced parabotanists. This five people led Mr. Domingos Sanches, expert parabotanist who working mid-1980s completely familiar flora. species-level field identifications enhanced cross-referencing either fertile infertile vouchers collected situ those deposited herbaria Embrapa Amazônia Oriental Museu Paraense Emílio Goeldi, later Herbarium Felisberto Camargo Universidade Federal Rural da Amazônia. credibility further strengthened collections wood samples cross-referenced Xylotheque, houses 4672 397 (Ferreira, Spatial conducted manually, operating 12.5 m width bands, covering 5083 undisturbed excluded riparian areas. Fieldwork permission required study. inventory data based carried out Annual Production Units (UPAs) 2, spanned 1635 3448 ha, respectively, included 291,027 mapped individual DBH representing 377 196 genera, 56 families. Species nomenclature standardised adhere Plant List database (TPL, Before harvest, mean density, basal area, standing volume 49.7 ± 9.2 ha−1, 10.46 2.35 m3 81.2 27.2 49.2 16.2 maximum 254 cm. To plots, overlaid 3-ha grid 5100-ha then manually shifting and/or rotating cells order fit irregular shaped boundaries resulted 1036, (total sample = 3108 ha) density 151 (range 48–244), 156,601 (Figure 1). tested intensity. first, lower-intensity scenario, occurred average 20 represents typical Amazon. second higher-intensity 40 exceeding limit 30 ha−1 scenario high-intensity beyond limits. landscape (at low intensity) unlogged, randomly allocated third 1036 three treatments any impact random allocation found treatment type, repeated 999 iterations new suite class. simulate cycle used financial Bousfield al. (2021) estimate value (accounting estimated tree, species-specific processing yields selling prices). Assuming loggers would preferentially log ranked (≥50 following limits) value. assigned until cumulative met quota (either forest). destruction caused creation road facilitate extraction adopted methods (2022), provided (AMATA Rondônia) extent networks concessions. combined deck construction 1.82% line estimates (Carvalho 2017) global median 1.7% (Kleinschroth Healey, damage cause community, plot deemed inside killed during process. skid trails trail (AMATA) harvest. Using maps, cover 1.6% 12 (again close previous estimates; Carvalho Readjusting simulated here, 2.66% 5.32% land (20 ha−1) (40 intensity, respectively. damage, trails, assumed (35 ≤ < 50 construction. As per standard practice concessionaries, > located simply avoided costs labour requirement involved removing them, did assign them precaution, reanalysis destroyed, minimal results S7). After simulating extraction, residual depicted surviving harvests occurred. analyse gradients employed generalised dissimilarity modelling (Ferrier 2007) R version 4.1.0 using GDM package (Fitzpatrick method extension matrix regression, designed specifically accommodate common types non-linearity larger-scaled sets: curvilinear relationship between increasing distance, compositional sites; rate positions along environmental 2007). approach fits non-linear functions predictor variable flexible i-splines, ‘ecological distance’ pair cells. i-spline indicates amount while holding variables constant. allowed us quantify driving patterns, (.g. geographic separation) constant, assess (from 0 ha−1). pairwise Bray–Curtis pairs our measure (>500,000 comparisons), distance predictors. fitted basis quantified randomising variable's position turn permutations noting loss explanatory power. dissimilarities Jaccard index, square root transformed, presence-absence transformed area. raw abundance-based trees) measures outcomes (see Figures S3–S6 comparison). ascertain vary classes, grouped medium <50 cm), (50 <110 very ≥110 overall 339, 336 59, above analysis separately attributable type (unlogged, low-intensity logged), trees, class, bray.part function betapart (Baselga Here, entirety concession, such there repeats original underwent treatment. Pairwise comparisons made, dissimilarity, its constituent parts (nestedness turnover) calculated. test observed, 5 scales: 1, 3, 5, 10 ha. imposing allocating grids bounded entirely varied, assessed remained felling 4.1 0.2 13.1 0.3 under m2 harvests, 2.3 0.02 4.9 0.04 low- respectively 22% 47% before logging). These accurately represent real-world occur S1). Residual 4.4 11.4 scenarios, 8.9 0.4 22.9 0.5% community. Across composition, accounting 22.4 1.3% explained 77.6 separation. considering dominant 89.2 2.2% 10.8 2). Similarly, accounted 72.4 3.4% 27.6 distance. For medium-sized effect beta-diversity, 7.5 0.8% variation. classes (mean harmonic p-value 0.001; Wilson, 2019, see Table S1 model summaries). logging-intensity gradient. whole no <20 ha−1), >30 ha−1; Figure emergent strongly altered even (<20 shift ~15% larger. little (>20 started stronger Removing slightly reduced S2). When instead abundance, low, change S3). Different demonstrated Bray-Curtis presented here (Figures S4–S6). turnover, plots. pattern lower- 3a). There increase highest intensities, slight 3 S8). although it communities. (both extremely 3d). Total communities, main types, 3b,c). all, separation constant dropping 4a). By contrast, became 4c). 69.4 3.1% 1-ha 61.4 9.2% 25-ha whereas 17.7 5.7% 41.3 11.3% Geographic grains 0.001, S2 4b), well S9), stable sizes, weakened 30.6 38.6 82.3 58.7 4d). 0.001). degradation, occurring vast areas, Focusing ~3100 show range principal forest. and, lesser extent, 110 cm). Given critical ecosystems, improved policies prevent large-scale suggests current (typically (≥35 DBH) term. conclusions supported longer-term, small-scale experimental demonstrate recovery Borneo have ~150 (Fisher 2011) far disturbed Indeed, RIL regimes (as here), Bornean closely resembled conventionally (Imai 2012). point Partitioning result High diverse protecting beneficial targeting species-rich sites similarly suggest land-sharing style Gilroy, 2014) less detrimental maintaining sparing, supporting theoretical (Ramage While profitable 2021), damaging biological groups birds dung-beetles França 2017), highlighting trade-off management. Although limited lower increases limits concern given prevalence illegal (Brancalion 2018; Finer sizeable huge footprint (Matricardi 2020), if simulated. alters butterflies (Montejo-Kovacevich 2018), bats (Peters 2006), dung beetles (França Neotropical fauna apparently sensitive analogues Afrotropical Indomalayan (Burivalova Logging-induced complex wide taxa must considered assessing contextually relevant 2015), world's store as-of-yet 2009), context tropics low. Commercial availability law restrict South-East Asia regularly exceeded 100 differ significantly Africa leading marked reduction vertical persist regions generally (Putz future research. targets large, strong signal subtractive Along gradient, likely prized case medium- large-sized making susceptible valued removed