Introduction
The African continent is home to the world’s second largest area of tropical forests which are, together with earth’s subtropical forests, by far the most floral-rich biomes on earth (Kier et al. Reference Kier, Mutke, Dinerstein, Ricketts, Küper, Kreft and Barthlott2005, Malhi et al. Reference Malhi, Adu-Bredu, Asare, Lewis and Mayaux2013). The African tropical forests appear to be less biodiverse than their Asian and American counterparts (Kier et al. Reference Kier, Mutke, Dinerstein, Ricketts, Küper, Kreft and Barthlott2005, Parmentier et al. Reference Parmentier, Malhi, Senterre, Whittaker, Alonso, Balinga, Bakayoko, Bongers, Chatelain, Comiskey, Cortay, Kandem, Doucet, Gautier, Hawthorne, Issembe, Kouamé, Kouka, Leal, Lejoly, Lewis, Nusbaumer, Parren, Peh, Phillips, Sheil, Sonké, Sosef, Sunderland, Stropp, Ter Steege, Swaine, Tchouto, Van Gemerden, Van Valkenburg and Wöll2007, Slik et al. Reference Slik, Arroyo-Rodríguez, Aiba, Alvarez-Loayza, Alves, Ashton, Balvanera, Bastian, Bellingham, Van Den Berg, Bernacci, Bispo, Blanc, Böhning-Gaese, Boeckx, Bongers, Boyle, Bradford, Brearley, Hockemba, Bunyavejchewin, Matos, Castillo-Santiago, Catharino, Chai, Chen, Colwell, Chazdon, Clark, Clark, Clark, Culmsee, Damas, Dattaraja, Dauby, Davidar, DeWalt, Doucet, Duque, Durigan, Eichhorn, Eisenlohr, Eler, Ewango, Farwig, Feeley, Ferreira, Field, Filho, Fletcher, Forshed, Franco, Fredriksson, Gillespie, Gillet, Amarnath, Griffith, Grogan, Gunatilleke, Harris, Harrison, Hector, Homeier, Imai, Itoh, Jansen, Joly, de Jong, Kartawinata, Kearsley, Kelly, Kenfack, Kessler, Kitayama, Kooyman, Larney, Laumonier, Laurance, Laurance, Lawes, do Amaral, Letcher, Lindesll, Lu, Mansor, Marjokorpi, Martin, Meilby, Melo, Metcalfe, Medjibe, Metzger, Millet, Mohandass, Montero, Valeriano, Mugerwa, Nagamasu, Nilus, Ochoa-Gaona, Onrizal, Page, Parolin, Parren, Parthasaranthy, Paudel, Permana, Piedade, Pitman, Poorter, Poulsen, Poulsen, Powers, Prasad, Puyravaud, Razafimahaimodison, Reitsma, dos Santos, Spironello, Romero-Saltos, Rovero, Rozak, Ruokolainen, Rutishauser, Saiter, Saner, Santos, Santos, Sarker, Satdichanh, Schmitt, Schöngart, Schulze, Suganuma, Sheil, Pinheiro, Sist, Stevart, Sukumar, Sun, Sunderland, Suresh, Suzuki, Tabarelli, Tang, Targhetta, Theilade, Thomas, Tchouto, Hurtado, Valencia, van Valkenburg, van Do, Vasquez, Verbeeck, Adekunle, Vieira, Webb, Whitfeld, Wich, Williams, Wittmann, Wöll, Yang, Yao, Yap, Yoneda, Zahawi, Zakaria, Zang, de Assis, Luize and Venticinque2015). Nevertheless, Sosef et al. (Reference Sosef, Dauby, Blach-Overgaard, van der Burgt, Catarino, Damen, Deblauwe, Dessein, Dransfield, Droissart, Duarte, Engledow, Fadeur, Figueira, Gereau, Hardy, Harris, de Heij, Janssens, Klomberg, Ley, Mackinder, Meerts, van de Poel, Sonké, Stévart, Stoffelen, Svenning, Sepulchre, Zaiss, Wieringa and Couvreur2017) recorded 22,577 plant species in tropical African forests, of which 30% are endemic, and concluded that much botanical exploration still remains to be done. The forests of the Congo Basin, which account for 89% of the African tropical forests, are also extremely important as carbon sinks, removing up to six times more carbon than the forests in the Amazon region (Harris et al. Reference Harris, Gibbs, Baccini, Birdsey, de Bruin, Farina, Fatoyinbo, Hansen, Herold, Houghton, Potapov, Suarez, Roman-Cuesta, Saatchi, Slay, Turubanova and Tyukavina2021, Malhi et al. Reference Malhi, Adu-Bredu, Asare, Lewis and Mayaux2013). Furthermore, the African carbon sink has been stable for the past three decades until 2015, while the Amazonian carbon sink has been declining since the 1990s (Hubau et al. Reference Hubau, Lewis, Phillips, Affum-Baffoe, Beeckman, Cuní-Sanchez, Daniels, Ewango, Fauset, Mukinzi, Sheil, Sonké, Sullivan, Sunderland, Taedoumg, Thomas, White, Abernethy, Adu-Bredu, Amani, Baker, Banin, Baya, Begne, Bennett, Benedet, Bitariho, Bocko, Boeckx, Boundja, Brienen, Brncic, Chezeaux, Chuyong, Clark, Collins, Comiskey, Coomes, Dargie, de Haulleville, Kamdem, Doucet, Esquivel-Muelbert, Feldpausch, Fofanah, Foli, Gilpin, Gloor, Gonmadje, Gourlet-Fleury, Hall, Hamilton, Harris, Hart, Hockemba, Hladik, Ifo, Jeffery, Jucker, Yakusu, Kearsley, Kenfack, Koch, Leal, Levesley, Lindsell, Lisingo, Lopez-Gonzalez, Lovett, Makana, Malhi, Marshall, Martin, Martin, Mbayu, Medjibe, Mihindou, Mitchard, Moore, Munishi, Bengone, Ojo, Ondo, Peh, Pickavance, Poulsen, Poulsen, Qie, Reitsma, Rovero, Swaine, Talbot, Taplin, Taylor, Thomas, Toirambe, Mukendi, Tuagben, Umunay, van der Heijden, Verbeeck, Vleminckx, Willcock, Wöll, Woods and Zemagho2020).
Despite their key role in biodiversity conservation and carbon sequestration, forests in tropical Africa are increasingly threatened by human activities (Tyukavina et al. Reference Tyukavina, Hansen, Potapov, Parker, Okpa, Stehman, Kommareddy and Turubanova2018). Between 1990 and 2000, the annual net loss of forests in Africa corresponded to 3.3 million ha yr−1, increasing to 3.4 million ha yr−1 between 2000 and 2010, and further to 3.9 million ha yr−1 between 2010 and 2020 (FAO & UNEP 2020). Africa is currently the only region where annual forest change rates are still increasing (FAO & UNEP 2020). The deforestation on the African continent is mainly caused by small-scale subsistence agriculture (Curtis et al. Reference Curtis, Slay, Harris, Tyukavina and Hansen2018, Jayathilake et al. Reference Jayathilake, Prescott, Carrasco, Rao and Symes2021, Tegegne et al. Reference Tegegne, Lindner, Fobissie and Kanninen2016, Tyukavina et al. Reference Tyukavina, Hansen, Potapov, Parker, Okpa, Stehman, Kommareddy and Turubanova2018), driven by very low agricultural yields (Carrasco et al. Reference Carrasco, Webb, Symes, Koh and Sodhi2017, West et al. Reference West, Gibbs, Monfreda, Wagner, Barford, Carpenter and Foley2010), and a rapidly growing human population (Gerland et al. Reference Gerland, Raftery, Ševčíková, Li, Gu, Spoorenberg, Alkema, Fosdick, Chunn, Lalic, Bay, Buettner, Heilig and Wilmoth2014).
Although deforestation is a very conspicuous and severe threat to biodiversity, degradation of tropical forests may even be more detrimental, based on the large spatial scales at which it occurs (Barlow et al. Reference Barlow, Lennox, Ferreira, Berenguer, Lees, Nally, Thomson, Ferraz, Louzada, Oliveira, Parry, Ribeiro de Castro Solar, Vieira, Aragão, Begotti, Braga, Cardoso, Cosme de Oliveira Junior, Souza, Moura, Nunes, Siqueira, Pardini, Silveira, Vaz-de-Mello, Veiga, Venturieri and Gardner2016). Forest degradation refers to within-forest disturbance and mainly includes selective logging, removal of understorey vegetation, and overexploitation of specific species (Sasaki & Putz Reference Sasaki and Putz2009, Chaturvedi et al. Reference Chaturvedi, Raghubanshi and Singh2012, Tyukavina et al. Reference Tyukavina, Hansen, Potapov, Parker, Okpa, Stehman, Kommareddy and Turubanova2018). In tropical Africa, forest degradation is primarily driven by selective logging, mainly for the small-scale production of charcoal which is still the most commonly used energy source in many Central African countries (Hosonuma et al. Reference Hosonuma, Herold, De Sy, De Fries, Brockhaus, Verchot, Angelsen and Romijn2012, Jayathilake et al. Reference Jayathilake, Prescott, Carrasco, Rao and Symes2021, Tyukavina et al. Reference Tyukavina, Hansen, Potapov, Parker, Okpa, Stehman, Kommareddy and Turubanova2018). The degradation rate of forests in the Congo Basin has been estimated to be 317,000 ha yr−1 between 2000 and 2005 (Ernst et al. Reference Ernst, Philippe, Astrid, Catherine, Musampa and Pierre2013). Lambin et al. (Reference Lambin, Geist and Lepers2003) estimated that the annual rate of human-mediated forest degradation in Africa corresponds to approximately 50% of its annual rate of deforestation. This would imply that 1.95 million ha of African forest has been degraded between 2010 and 2020. However, Shapiro et al. (Reference Shapiro, Grantham, Aguilar-Amuchastegui, Murray, Gond, Bonfils and Rickenbach2021) recently reported that this assumption may drastically underestimate the actual rate. Using a forest condition index, they estimated that between 2000 and 2016, over 23 million ha of forest has been degraded in the Congo Basin alone.
Although remote-sensing techniques are constantly improving, it remains challenging to monitor the extent and impact of forest degradation due to the interference with the often untouched canopy in degraded forests (Matricardi et al. Reference Matricardi, Skole, Costa, Pedlowski, Samek and Miguel2020, Peres et al. Reference Peres, Barlow and Laurance2006). Evaluation of the extent and impact of forest degradation must therefore largely rely on vegetation surveys on the ground (Heym et al. Reference Heym, Uhl, Moshammer, Dieler, Stimm and Pretzsch2021). Impacts of forest degradation on biodiversity have been frequently studied in the Neotropics (e.g. Carreño-Rocabado et al. Reference Carreño-Rocabado, Peña-Claros, Bongers, Alarcón, Licona and Poorter2012, Mirabel et al. Reference Mirabel, Hérault and Marcon2020), Asia (e.g. Baboo et al. Reference Baboo, Sagar, Bargali and Verma2017, Millet et al. Reference Millet, Pascal and Kiet2010), and other parts of Africa (e.g. Bobo et al. Reference Bobo, Waltert, Sainge, Njokagbor, Fermon and Mühlenberg2006, Oyugi et al. Reference Oyugi, Brown and Whelan2008), but similar studies are rare in the Congo Basin (but see Hall et al. Reference Hall, Harris, Medjibe and Ashton2003, Ifo et al. Reference Ifo, Moutsambote, Koubouana, Yoka, Ndzai, Bouetou-Kadilamio, Mampouya, Jourdain, Bocko, Mantota, Mbemba, Mouanga-Sokath, Odende, Mondzali, Wenina, Ouissika and Joel2016), and almost lacking in the Democratic Republic of Congo (DR Congo) (but see Makana & Thomas Reference Makana and Thomas2006). Yet, as DR Congo harbours almost 60% of the African tropical forests (Malhi et al. Reference Malhi, Adu-Bredu, Asare, Lewis and Mayaux2013), it is essential to comprehend how plant community composition is affected by forest disturbance.
While conservation of existing tropical forests should still be the priority, the restoration of deforested land and degraded forests can aid in the recovery of tropical biodiversity and the recovery of ecosystem services (Brudvig Reference Brudvig2011, FAO & UNEP 2020). Between 2000 and 2019, 0.7 million ha of forest has been actively or passively restored in Sub-Saharan Africa (FAO & UNEP 2020). By 2030, the African Forest Landscape Restoration Initiative (AFR100) aims to increase the land surface under restoration to 100 million ha. DR Congo has committed to restore 8 million ha (AFR100 2021). Because of the high variability in restoration outcomes, more research is crucial to clear the path for restoration (Chazdon Reference Chazdon2008). Again, studies on the biodiversity consequences of tropical forest recovery are more commonly found in the Neotropics (e.g. Letcher & Chazdon Reference Letcher and Chazdon2009, Oberleitner et al. Reference Oberleitner, Egger, Oberdorfer, Dullinger, Wanek and Hietz2021) and tropical Asia (e.g. Hayward et al. Reference Hayward, Banin, Burslem, Chapman, Philipson, Cutler, Reynolds, Nilus and Dent2021, Mahayani et al. Reference Mahayani, Slik, Savini, Webb and Gale2020), but are rare in the Congo Basin (but see Bauters et al. Reference Bauters, Vercleyen, Vanlauwe, Six, Bonyoma, Badjoko, Hubau, Hoyt, Boudin, Verbeeck and Boeckx2019, Makelele et al. Reference Makelele, Verheyen, Boeckx, Cizungu Ntaboba, Mujinya Bazirake, Ewango and Bauters2021).
In this study, we established a set of forest inventory plots in the Yangambi region in DR Congo and aimed to find associations between both anthropogenic disturbance through selective logging and forest recovery on the one side, and a range of forest and tree community characteristics on the other side, as compared to reference undisturbed old-growth forest. Based on 125 sampling quadrats across 25 forest inventory plots, we specifically quantified differences among (i) disturbed old-growth forest, (ii) regrowth forest, and (iii) undisturbed old-growth forest in terms of tree species diversity, community traits, community composition, forest structure, and canopy cover. We hypothesise that (i) selective logging results in the decreasing abundance of some specific tree species and in shifts in community composition, without having a profound effect on the overall tree species diversity, (ii) that selective logging results in a lower community-weighted mean wood density and in lower crown closure, and (iii) that forest recovery following agricultural abandonment will be fast for secondary tree species, with low wood density, but slower for primary forest tree species, with higher wood density, implying that the overall tree species diversity will have recovered, but that the community composition will be different from undisturbed old-growth forest.
Methods
Study site and set up
Vegetation sampling was conducted in the Yangambi region, in the Tshopo province in North-Eastern DR Congo, approximately 100 km west of Kisangani between November 2020 and January 2021. In Yangambi, the climate is characterised by two drier seasons (December–March & June–July) and two rainy seasons (April–May & August–November) (van Vliet et al. Reference van Vliet, Muhindo, Kambale Nyumu, Mushagalusa and Nasi2018). The general landscape type in this area is typical for the Congo Basin and is characterised by a mosaic of land tenures: the Yangambi Man and Biosphere Reserve created in 1979; the Ngazi Forest Reserve; a logging concession; and customary land. To date, there is no official management plan (van Vliet et al. Reference van Vliet, Muhindo, Kambale Nyumu, Mushagalusa and Nasi2018). Prior to the Congolese independence in 1960, Yangambi housed a research station of INEAC (Institut National pour l’Etude Agronomique du Congo Belge), with multiple pastures for research and breeding activities. However, in 1962 the INEAC changed into INERA (Institut National des Etudes et Recherches Agronomiques) in 1962. Since then, the research activities at Yangambi started to diminish (Vanden Abeele et al. Reference Vanden Abeele, Janssens, Asimonyio Anio, Bawin, Depecker, Kambale, Mwanga Mwanga, Ebele, Ntore, Stoffelen and Vandelook2021), and as a result, many agricultural fields were abandoned, allowing forest to recover on these lands.
We established 25 forest inventory plots of 125 m × 125 m (1.56 ha), covering an area of ca. 50 by 20 km, just North of the Congo River (Figure 1). Plots were assigned to three different forest categories by georeferencing (WGS84) them onto historical land-use maps, which were issued by Evrard (Reference Evrard1954). Plots located on historical agricultural land from INERA were classified as regrowth forest (R). Although we could not exactly date the start of the forest recovery, our best estimate based on local sources is that the agricultural land was abandoned somewhere between 40 and 60 years ago. Plots located in historically old-growth forest areas but where we observed indications of considerable small-scale selective logging through the presence of tree stumps and other remnants of logging such as parts of trees or active logging were considered as disturbed old-growth forest (DO). These areas were typically located near regrowth forest and close to places with relatively higher population density. The remaining plots that were located in areas with historical old-growth forests, and where no signs of selective logging were present, were categorised as undisturbed old-growth forest (UO). The current UO and DO forests have previously been classified by Evrard (Reference Evrard1954) as heterogenous forests with primitive character. These old vegetation maps have been shown to be very reliable (Vancutsem et al. Reference Vancutsem, Pekel, Evrard, Malaisse and Defourny2009), and we may therefore assume there were no pre-existing differences between both forest categories.
Figure 1. Location of the 25 forest inventory plots across the three forest categories in the Yangambi region (DR Congo).
Vegetation survey and plant trait quantification
Five quadrats (25 m × 25 m) were randomly established in each forest inventory plot, resulting in a total of 125 sampling quadrats. In each quadrat, the abundance and diameter at breast height (DBH) of all woody species with a DBH ≥ 5 cm were measured. The quadrats were geolocated (Garmin GPSMAP 64 series), and crown closure (%) was quantified from four readings in the cardinal directions with a model A convex spherical densiometer.
The specific leaf area (SLA) of all taxa identified at species level was assessed based on dry herbarium specimens from the Meise Botanic Garden herbarium following Pérez-Harguindeguy et al. (Reference Pérez-Harguindeguy, Díaz, Garnier, Lavorel, Poorter, Jaureguiberry, Bret-Harte, Cornwell, Craine, Gurvich, Urcelay, Veneklaas, Reich, Poorter, Wright, Ray, Enrico, Pausas, de Vos, Buchmann, Funes, Quétier, Hodgson, Thompson, Morgan, ter Steege, van der Heijden, Sack, Blonder, Poschlod, Vaieretti, Conti, Staver, Aquino and Cornelissen2013). Perez et al. (Reference Perez, Rodriguez and Heberling2020) provided evidence that the SLA can reliably be estimated from herbarium-based measurements. Leaf area was assessed using leaf disks (4.9 cm2), to minimise damage to herbarium specimens, or otherwise by a whole leaf scan using ImageJ (Rueden et al. Reference Rueden, Schindelin, Hiner, Dezonia, Walter, Arena and Eliceiri2017). The obtained disks and whole leaves were weighted afterwards. Five replicates were used per species. In addition to the SLA, the wood density (WD) of 141 species was compiled from the literature (Table S1) (RMCA 2021, Carsan et al. Reference Carsan, Orwa, Harwood, Kindt, Stroebel, Neufeldt and Jamnadass2012, Harja et al. Reference Harja, Rahayu and Pambudi2021).
Data analyses
Statistical analyses were performed using R software (R Core Team 2020). The data of the five quadrats were summed into one value per plot. For each plot, the number of trees, species richness (Hill’s N0), species diversity (Hill’s N1 or the exponential of the Shannon entropy), and evenness (Hill’s N1/log(Hill’s N0)) were then calculated. These numbers were then standardised for sample completeness, as sample sizes differed between plots. Sample completeness represents the proportion of the total number of individuals in a unit that belongs to the species in the sample, contrary to sample size standardisation which allows less biased comparisons (Chao & Jost Reference Chao and Jost2012). Extrapolation of the estimates was done to twice the reference sample size per plot, following Chao & Jost (Reference Chao and Jost2012). Calculations were based on the collected species abundance data with 9999 bootstraps using the iNEXT function of the iNEXT package (Hsieh et al. Reference Hsieh, Ma and Chao2016). Subsequently, the different variables and the crown closure were tested for differences between the three forest categories using Kruskal-Wallis rank sum tests and pairwise Wilcoxon rank sum tests with correction for multiple testing, as the assumptions for parametric testing were not met.
Dissimilarity of the tree community composition across the forest categories was visualised using a non-metric multidimensional scaling (NMDS), based on the Bray-Curtis distance matrix, the most suitable distance for abundance data (Ricotta et al. Reference Ricotta, Podani and Uv2017). A two-axis solution provided a stress level (0.13) well below the 0.20 threshold, indicating a good representation in reduced dimensions. The ordination was performed with the metaMDS function in the vegan package (Oksanen et al. Reference Oksanen, Blanchet, Friendly, Kindt, Legendre, Mcglinn, Minchin, Hara, Simpson, Solymos, Stevens, Szoecs and Wagner2020). Subsequently, a Mantel test with 9999 permutations was performed to test the effect of geographic distance on the dissimilarity of the tree community composition using the mantel function in the Vegan package (Oksanen et al. Reference Oksanen, Blanchet, Friendly, Kindt, Legendre, Mcglinn, Minchin, Hara, Simpson, Solymos, Stevens, Szoecs and Wagner2020). In addition to the NMDS, a permutational multivariate analysis of variance with 9999 permutations was performed to test for differences in community composition between the three different forest categories using the adonis function in the Vegan package (Oksanen et al. Reference Oksanen, Blanchet, Friendly, Kindt, Legendre, Mcglinn, Minchin, Hara, Simpson, Solymos, Stevens, Szoecs and Wagner2020). Pairwise comparisons of the different forest groups with 9999 permutations and Bonferroni correction were performed with the pairwise.adonis.function in the pairwiseAdonis package (Arbizu Reference Arbizu2017). Tree species indicative for each of the three forest categories were identified based on their indicator value (Dufrene & Legendre Reference Dufrene and Legendre1997). This value was calculated using the multipatt function in the indicspecies package (De Cáceres & Legendre Reference De Cáceres and Legendre2009), with 9999 permutations.
Community-weighted means (CWM) for each plot were calculated for the SLA and the WD, using the number of individuals as weight. The community-weighted means are known to be very responsive to environmental change and highly sensitive to disturbance (Carreño-Rocabado et al. Reference Carreño-Rocabado, Peña-Claros, Bongers, Alarcón, Licona and Poorter2012). Both variables met the assumptions for parametric testing, and to evaluate differences in community-weighted mean SLA and WD among forest categories, a one-way analysis of variance (ANOVA) was conducted followed by pairwise Tukey post hoc tests.
Forest structure was evaluated using the measured DBHs. Trees were categorised in ten size classes according to their DBH. An interval of 10 cm was used in the first nine classes, starting from 5 cm DBH. The tenth class had a wider interval, aggregating individuals with a DBH over 95 cm. Subsequently, the diameter frequency distribution was compared among the three forest categories with Pearson’s chi-squared tests with simulated p-values based on 9999 replicates using the chisq.test function in the stats package (R Core Team 2020). The chi-square test was preferred over the frequently used Kolmogorov-Smirnov test, because the former is more flexible than the latter (Mitchell Reference Mitchell1971). The diameter frequency distribution comparison was complemented by the calculation of the structural diversity, which is a single measure summarising the diameter distribution (McElhinny et al. Reference McElhinny, Gibbons, Brack and Bauhus2005). Structural diversity was quantified by the first order Hill number which was standardised for sample completeness following the same methodology as outlined above. Calculation and extrapolation of structural diversity was done with the iNEXT function with 9999 bootstraps in the iNEXT package, as above. All figures were made using the ggplot2 package (Wickham Reference Wickham2016).
Results
Species richness and diversity
A total of 7375 woody plant individuals (DBH ≥ 5 cm) from 211 taxa were recorded across the 25 forest inventory plots. Among the 211 species, 79.15% were identified at species level, 14.22% at genus level, and 6.63% were unidentified species. Crown closure (χ 2 = 2.21, P > 0.05) and hill’s N0 (χ 2 = 1.98, P > 0.05) were not significantly different among the three forest categories, whereas the tree abundance (χ 2 = 10.90, P < 0.01), Hill’s N1 (χ 2 = 9.78, P < 0.01) and evenness (χ 2 = 10.06, P < 0.01) did differ between the forest categories (Figure 2). Tree abundance was 17.2% lower (P < 0.05) in DO forests and 24.3% (P < 0.05) lower in R forests, when compared to UO forests. Species diversity and evenness were 33.6% (P < 0.001) and 31.9% (P < 0.001) higher, respectively, in DO forests as compared to UO forests (Table S2).
Figure 2. Comparisons between tree abundance and tree species diversity metrics across old-growth undisturbed forest (OU), old-growth disturbed forest (OD), and regrowth forest (R) in the Yangambi region (DR Congo). Hinges represent the 25th, 50th, and 75th percentiles, respectively. Whiskers extend to maximum 1.5 times the interquartile range. Letters code for significant differences between groups. (a) Tree abundance, (b) species richness, (c) species diversity, (d) evenness.
Community composition
The NMDS of the 25 forest inventory plots showed a spatial separation of the three forest categories (Figure 3). This was supported by the permutational multivariate analysis of variance which indicated differences between categories (R 2 = 0.253; P < 0.001). Specifically, community composition significantly differed between DO forests and UO forests (R 2 = 0.207; P < 0.001), and between R forests and UO forests (R 2 = 0.267; P < 0.001). The Mantel test between the geographic distance matrix and the Bray-Curtis distance matrix showed a significant but weak correlation (r = 0.200; P < 0.05). Across the 25 forest inventory plots, 27 out of the 167 at species level identified trees were indicative based on their indicator value (both P < 0.05 and indicator value > 0.20). Seventeen indicator species were indicative for UO forest (e.g. Cola griseiflora), four for DO forest (e.g. Leplaea cedrata), and six for R forest (e.g. Heisteria parvifolia) (Table S3).
Figure 3. Non-metric multidimensional scaling (NMDS) ordination of 25 forest inventory plots in the Yangambi region (DR Congo) across the three forest categories: undisturbed old-growth forest (UO), disturbed old-growth forest (DO), and regrowth forest (R). Ordination is based on the Bray-Curtis distance matrix calculated from the tree species abundances and represents the woody species (DBH ≥ 5 cm) community composition.
Specific leaf area and wood density
The community-weighted means for SLA did not differ between forest categories (F = 1.38, P > 0.05); however, the community-weighted mean WD exhibited a significant difference (F = 9.03, P < 0.01) (Figure 4). In DO forests, community-weighted mean WD was 5.0 % lower (P < 0.05), whereas in R forests it was 6.2 % lower (P < 0.01) as compared to UO forests.
Figure 4. Comparisons between the community-weighted means of specific leaf area (SLA) and wood density (WD) across the three forest categories. Hinges represent the 25th, 50th, and 75th percentiles, respectively. Whiskers extend to maximum 1.5 times the interquartile range. Letters code for significant differences between groups.
Forest structure
The diameter distribution showed that there were more small-sized trees in UO forests than in DO forests (χ 2 = 34.51; P < 0.001) and in R forests (χ 2 = 61.54; P < 0.001) (Figure 5). These differences between categories were also present for the structural diversity, represented by Hill’s N1, (χ 2 = 13.76; P < 0.01), where UO forests differed from DO forests (P < 0.01), and from R forests (P < 0.01).
Figure 5. Observed frequencies among the ten DBH size categories compared among the three forest categories. Size classes one to nine have a 10 cm interval, while the tenth has a wider interval, grouping individuals with a DBH of over 95 cm.
Discussion
Understanding how anthropogenic disturbance affects forest structure, tree species richness, and community composition is of major importance for biodiversity conservation in globally underexplored biodiversity hotspots such as the Congo Basin. The goal of our study was to find associations between both anthropogenic disturbance through selective logging and forest recovery on the one side, and a range of forest community and tree community trait characteristics on the other side, as compared to reference undisturbed old-growth forest. Although our three delineated forest categories certainly enveloped some within-variation in terms of time since abandonment (for the regrowth forest plots) and in terms of degree of disturbance (for the disturbed old-growth forest plots), we were able to identify differences in community composition, forest structure, and community traits among forest categories.
Selective logging
As hypothesised, our survey showed that there were clear differences in community composition and forest structure between plots in undisturbed old-growth forest and plots in disturbed old-growth forest. There were no differences in the number of woody species between any forest category; yet contrary to our expectation, species diversity was higher in plots in disturbed old-growth forest. This higher species diversity likely indicates that the species abundance distribution is less skewed in disturbed forests. Indeed, the evenness index showed that the species abundance is more evenly distributed in plots in disturbed forest than in plots in undisturbed old-growth forest. Combined with the observation that plots in disturbed old-growth forests contained significantly less tree individuals, these findings are in line with our hypothesis and might suggest that selective logging, in which exploitable species are selectively removed, is closely associated with these patterns. In the Yangambi area, timber is removed by local villagers using, for example, machetes, mainly for construction material and cooking purposes (Depecker pers. obs.). The stems that are removed are usually from species from which the wood is exploitable. We expect that selective logging at first only results in a shift in densities, but that it will ultimately result in the loss of certain species as selective loggers will specifically target species with stems that are exploitable. For example, Prioria oxyphylla and Scorodophloeus zenkeri were much less abundant in disturbed forests and are used as construction material. Moreover, in DR Congo the latter species is also exploited for its medicinal properties and is used to make charcoal (PROTA 2021). Other species that were highly reduced in numbers were Dialium pachyphyllum, Greenwayodendron suaveolens, and Strombosia pustulata, which all bear seeds that are dispersed by animals (PROTA 2021, Meunier et al. Reference Meunier, Doucet and Moumbogou2015). Animal-dispersed seeded tree species tend to have higher wood densities than abiotic-dispersed seeded tree species (Osuri et al. Reference Osuri, Ratnam, Varnam, Alvarez-Loayza, Astaiza, Bradford, Fletcher, Ndoundou-Hockemba, Jansen, Kenfack, Marshall, Ramesh, Rovero and Sankaran2016), possibly indicating that loggers specifically target species with high WD, suggesting that species with animal-dispersed seeds are more vulnerable to selective logging. Careful monitoring of tree species is needed to unravel the specific effects of selective logging at the species level.
Plots in disturbed old-growth forest on the other hand still contained many shade-tolerant tree species of which the wood is non-exploitable, such as species of the genus Cola which have a low DBH and highly irregular trunks. Cola acuminata also produces highly valued nuts with stimulating and medicinal properties (Meunier et al. Reference Meunier, Doucet and Moumbogou2015). Likewise, the abundance of most Diospyros spp. is indicative for old-growth forests, of which only Diospyros crassiflora is exploited for its wood (Meunier et al. Reference Meunier, Doucet and Moumbogou2015).
Although a clear difference was found in community composition between forest categories in our study, the response in terms of species diversity and species richness was more variable. An earlier study on selective logging in the Congo Basin compared disturbed and undisturbed forests in terms of species diversity in two distinct locations (Makana & Thomas Reference Makana and Thomas2006). Although a significant influence of disturbance was detected in the first location, similar values for species richness and species diversity were detected in the second location. Furthermore, a clear effect of disturbance, primarily selective logging, on species diversity was detected in tropical forests in the Republic of the Congo (Ifo et al. Reference Ifo, Moutsambote, Koubouana, Yoka, Ndzai, Bouetou-Kadilamio, Mampouya, Jourdain, Bocko, Mantota, Mbemba, Mouanga-Sokath, Odende, Mondzali, Wenina, Ouissika and Joel2016) and India (Baboo et al. Reference Baboo, Sagar, Bargali and Verma2017), but not in a tropical forest in Bolivia (Carreño-Rocabado et al. Reference Carreño-Rocabado, Peña-Claros, Bongers, Alarcón, Licona and Poorter2012). It is clear that solely assessing the possible impact of anthropogenic disturbance through species diversity can mask its impact and that community composition is a better indicator of the possible impacts of forest degradation (Edwards et al. Reference Edwards, Tobias, Sheil, Meijaard and Laurance2014, Imai et al. Reference Imai, Tanaka, Samejima, Sugau, Pereira, Titin, Kurniawan and Kitayama2014). Indeed, although several studies across different tropical regions have reported varying responses of species diversity to forest disturbance, they all reported significant shifts in the community composition following anthropogenic disturbance (Makana & Thomas Reference Makana and Thomas2006, Baboo et al. Reference Baboo, Sagar, Bargali and Verma2017, Carreño-Rocabado et al. Reference Carreño-Rocabado, Peña-Claros, Bongers, Alarcón, Licona and Poorter2012, Ifo et al. Reference Ifo, Moutsambote, Koubouana, Yoka, Ndzai, Bouetou-Kadilamio, Mampouya, Jourdain, Bocko, Mantota, Mbemba, Mouanga-Sokath, Odende, Mondzali, Wenina, Ouissika and Joel2016)
Because ecosystem functioning is mainly mediated by species’ functional traits (Chaturvedi et al. Reference Chaturvedi, Tripathi, Raghubanshi and Singh2021, de Bello et al. Reference de Bello, Lavorel, Hallett, Valencia, Garnier, Roscher, Conti, Galland, Goberna, Májeková, Montesinos-Navarro, Pausas, Verdú, E-Vojtkó, Götzenberger and Lepš2021), community-weighted mean functional traits can provide better ecologically information than only using taxonomic data (Katovai et al. Reference Katovai, Burley and Mayfield2012). We hypothesised that we would find lower community-weighted mean wood density and a lower canopy closure in plots in disturbed old-growth forest, as compared to plots in undisturbed old-growth forest. In terms of canopy closure, we did not find differences between any forest category. Consequently, the light availability in the plots remained unchanged, which presumably clarifies why we observed no significant differences in community-weighted mean specific leaf area between forest categories. It has indeed been established by Carreño-Rocabado et al. (Reference Carreño-Rocabado, Peña-Claros, Bongers, Alarcón, Licona and Poorter2012) that increased light availability can induce changes in community-weighted mean functional traits in tropical forests. Furthermore, although disturbed and undisturbed old-growth forests were characterised by different indicator species assemblages, the majority can still be classified as shade-tolerant, which is commonly associated with low SLA (Poorter Reference Poorter2009). Nonetheless, the absence of differences between forest categories should be interpreted cautiously as SLA is highly responsive to environmental variation, such as sun exposure. In contrast to community-weighted mean SLA, community-weighted mean WD did significantly differ between plots in disturbed and plots in undisturbed old-growth forest. As anticipated, a decrease in community-weighted mean WD was observed in plots in disturbed old-growth forest, which indicates that trees with high WD and of high qualities as construction material have been selectively removed (Saranpää Reference Saranpää, Barnett and Jeronimidis2003).
The mean diameter distribution in the forests in the Yangambi region has previously been described by the reverse J-shaped model (Fayolle et al. Reference Fayolle, Picard, Doucet, Swaine, Bayol, Bénédet and Gourlet-Fleury2014). In our study, the reverse J-shape was more pronounced in plots in undisturbed old-growth forests than in plots in disturbed old-growth forests which contained significantly less trees in the first size class, suggesting less regeneration (Gebeyehu et al. Reference Gebeyehu, Soromessa, Bekele and Teketay2019). Similar results of reduced regeneration in forests were obtained in other tropical forests and were assigned to disturbance (Makana & Thomas Reference Makana and Thomas2006, Baboo et al. Reference Baboo, Sagar, Bargali and Verma2017, Hall et al. Reference Hall, Harris, Medjibe and Ashton2003).
Forest recovery following agricultural abandonment
Based on the historical land-use maps, our results indicated that after 40 to 60 years there were no significant differences between plots in regrowth forests and plots in undisturbed old-growth forest in terms of species diversity, which was hypothesised. They only differed in terms of the number of tree individuals and community-weighted mean WD. These findings are in line with other studies in the Congo Basin (Bauters et al. Reference Bauters, Vercleyen, Vanlauwe, Six, Bonyoma, Badjoko, Hubau, Hoyt, Boudin, Verbeeck and Boeckx2019, Makelele et al. Reference Makelele, Verheyen, Boeckx, Cizungu Ntaboba, Mujinya Bazirake, Ewango and Bauters2021) and in other tropical regions (Letcher & Chazdon Reference Letcher and Chazdon2009, Mahayani et al. Reference Mahayani, Slik, Savini, Webb and Gale2020). In contrast to species diversity, we found that the community composition in plots in regrowth forest was significantly different from those in undisturbed old-growth forest. These findings are in line with what was expected and might suggest that community composition needs more time to converge in similarity to plots in undisturbed old-growth forest. Similar to our results, the community composition of young regrowth forest (25–30 years) differed from old-growth forest in the Maringa-Lopori-Wamba forest landscape in DR Congo, whereas old regrowth forests (150–300 years) were very close to old-growth forests (Bauters et al. Reference Bauters, Vercleyen, Vanlauwe, Six, Bonyoma, Badjoko, Hubau, Hoyt, Boudin, Verbeeck and Boeckx2019). Likewise, a recent study in the Yoko forest reserve, at approximately 30 km Southeast of Kisangani, showed that the community composition of 60-year-old regrowth forests was still significantly different from local old-growth forests (Makelele et al. Reference Makelele, Verheyen, Boeckx, Cizungu Ntaboba, Mujinya Bazirake, Ewango and Bauters2021). Our results are furthermore in line with studies in the Neotropics (Oberleitner et al. Reference Oberleitner, Egger, Oberdorfer, Dullinger, Wanek and Hietz2021, Poorter et al. Reference Poorter, Craven, Jakovac, van der Sande, Amissah, Bongers, Chazdon, Farrior, Kambach, Meave, Muñoz, Norden, Rüger, van Breugel, Zambrano, Amani, Andrade, Brancalion, Broadbent, de Foresta, Dent, Derroire, DeWalt, Dupuy, Durán, Fantini, Finegan, Hernández-Jaramillo, Hernández-Stefanoni, Hietz, Junqueira, N'dja, Letcher, Lohbeck, López-Camacho, Martínez-Ramos, Melo, Mora, Müller, N'Guessan, Oberleitner, Ortiz-Malavassi, Pérez-García, Pinho, Piotto, Powers, Rodríguez-Buriticá, Rozendaal, Ruíz, Tabarelli, Teixeira, Sampaio, van der Wal, Villa, Fernandes, Santos, Aguilar-Cano, de Almeido-Cortez, Alvarez-Davila, Arreola-Villa, Balvanera, Becknell, Cabral, Castellanos-Castro, de Jong, Nieto, Espírito-Santo, Fandino, García, García-Villalobos, Hall, Idárraga, Jiménez-Montoya, Kennard, Marín-Spiotta, Mesquita, Nunes, Ochoa-Gaona, Peña-Claros, Pérez-Cárdenas, Rodríguez-Velázquez, Villanueva, Schwartz, Steininger, Veloso, Vester, Vieira, Williamson, Zanini and Hérault2021) and tropical Asia (Hayward et al. Reference Hayward, Banin, Burslem, Chapman, Philipson, Cutler, Reynolds, Nilus and Dent2021).
Despite having a similar community composition as plots in disturbed old-growth forests, plots in regrowth forests were characterised by a different indicator species assemblage, including Myrianthus arboreus, a late pioneer species (PROTA 2021). Additionally, Combretum lokele and Dacryodes edulis were also characteristic for regrowth forests. Both species are commonly found in villages and are indicative for the important anthropogenic influence in these forests (Meunier et al. Reference Meunier, Doucet and Moumbogou2015).
The forest structure in the plots in regrowth forests was comparable to plots in disturbed old-growth forest, but significantly different from plots in undisturbed old-growth forest. These differences were not caused by the lower abundance of large-sized trees in regrowth forests, as would be expected because these are slower to recover (Bauters et al. Reference Bauters, Vercleyen, Vanlauwe, Six, Bonyoma, Badjoko, Hubau, Hoyt, Boudin, Verbeeck and Boeckx2019, Makelele et al. Reference Makelele, Verheyen, Boeckx, Cizungu Ntaboba, Mujinya Bazirake, Ewango and Bauters2021). On the contrary, the observed differences can be explained by the higher frequency of small-sized trees in plots in undisturbed old-growth forests. This possibly implies that the tree regeneration could be hampered. Similar to other studies (Bauters et al. Reference Bauters, Vercleyen, Vanlauwe, Six, Bonyoma, Badjoko, Hubau, Hoyt, Boudin, Verbeeck and Boeckx2019, Makelele et al. Reference Makelele, Verheyen, Boeckx, Cizungu Ntaboba, Mujinya Bazirake, Ewango and Bauters2021), we suggest that forests in the Yangambi landscape initially recover rapidly to a close canopy forest with shade-tolerant understorey species following agricultural abandonment, but that the recruitment of primary forest species is slow, and that further forest succession may be hampered by ongoing anthropogenic disturbance in these regrowth forests.
Limitations
We acknowledge that our study is not flawless and that there are some limitations that have to be considered. Our plots are to a certain extent spatially clustered, which was not possible to avoid due to the location of the disturbance zone in the Yangambi area. Furthermore, due to reasons of feasibility, it was not possible to incorporate additional control plots to account for potential spatial variability (Davies & Gray Reference Davies and Gray2015). The Mantel test was used to test the effect of geographic distance on the dissimilarity of the tree community composition. It should be noted that the Mantel test showed a significant correlation between geographic distance and dissimilarity of community composition. This distance decay of similarity is commonly reported in tropical forests (Ganivet et al. Reference Ganivet, Unggang, Bodos, Demies, Ling, Sang and Bloomberg2020, Liebsch et al. Reference Liebsch, Marques and Goldenberg2008, Wittmann et al. Reference Wittmann, Schongart, Montero, Motzer, Junk, Piedade, Queiroz and Worbes2006). Pre-existing spatial variation seems to confound with the main treatment (logging), but the correlation is rather weak. Although we cannot conclude that the observed differences between sites are a direct effect of selective logging per se, we can interpret the results in light of our ecological understanding of the area and its activities (Oksanen Reference Oksanen2004, Davies & Gray Reference Davies and Gray2015).
Conclusion
We found that the tree community composition and forest structure were significantly different between forest categories and that the response of species diversity was more variable. Based on our knowledge of the activities in the Yangambi region, our results suggest that these differences in community composition found between UO forest and DO forest can be attributed to selective logging. Our data furthermore suggest that 40 to 60 years after agricultural abandonment, species diversity is able to recover, but community composition and forest structure are not. Given the limitations of our sampling design, we suggest vegetation surveys across larger geographic areas to deal with the spatial clustering of our survey plots within one disturbance category. Based on the known correlation between forest productivity and SLA, and between carbon storage and WD, our results also indicate that disturbed old-growth forest and regrowth forest are as productive as undisturbed old-growth forest but are not able to store as much carbon (Chave et al. Reference Chave, Coomes, Jansen, Lewis, Swenson and Zanne2009, Madani et al. Reference Madani, Kimball, Ballantyne, Affleck, van Bodegom, Reich, Kattge, Sala, Nazeri, Jones, Zhao and Running2018). Conservation of undisturbed old-growth forests is vital, although also regrowth of forests can play a crucial role in maintaining biodiversity and providing ecosystem services (Crouzeilles et al. Reference Crouzeilles, Curran, Ferreira, Lindenmayer, Grelle and Rey Benayas2016, Van de Perre et al. Reference Van de Perre, Willig, Presely, Andemwana, Beeckman, Boeckx, Cooleman, de Haan, De Kesel, Dessein, Grootaert, Huygens, Jansens, Kearsley, Kabeya, Leponce, Van den Broeck, Verbeeck, Würsten, Leirs and Verheyen2018).
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/S0266467422000347
Data Availability Statement
The data that support the findings of this study will be made openly available in Zenodo
Acknowledgments
We would like to thank the Ministère de L’Environnement et Développement Durable (MEDD) and Institut National pour l’Étude et la Recherche Agronomiques (INERA) and the FORETS project, which is financed by the European Union, for their help in the field. We would also like to thank both Hans Beeckman and Tom De Mil of the Royal Museum for Central Africa (RMCA) for their help in retrieving wood density data.
Author contributions
OH, FV, and JD designed this study. JD, JA, RM, YH, and JK participated in fieldwork. JD analysed the data. FV, OH, and JD wrote the manuscript. All authors contributed to finalising the manuscript.
Financial support
This study was funded by Research Foundation-Flanders, research mandate granted to JD (FWO; 1125221N) and research project granted to OH (FWO; G090719N), and the Foundation for the promotion of biodiversity research in Africa, granted to JD and YH (SBBOA, www.sbboa.be).
Conflict of interest
All authors confirm that there is no conflict of interest regarding the publication of this article.
Supporting Information
Additional supporting information can be found online in the Supporting Information section at the end of the article.
