Forest Ecology and Management 337 (2015) 61–66

Contents lists available at ScienceDirect

Forest Ecology and Management

journal homepage: www.elsevier .com/locate / foreco

Forest structure and composition of previously selectively loggedand non-logged montane forests at Mt. Kilimanjaro

http://dx.doi.org/10.1016/j.foreco.2014.10.0360378-1127/� 2014 Published by Elsevier B.V.

⇑ Corresponding author. Tel.: +41 31 631 49 33.E-mail address: ggpmrutten@gmail.com (G. Rutten).

Gemma Rutten a,⇑, Andreas Ensslin a, Andreas Hemp b, Markus Fischer a,c

a Institute of Plant Sciences, University of Bern, Altenbergrain 21, 3013 Bern, Switzerlandb Department of Plant Systematics, University of Bayreuth, Universitätsstrasse 30, 95440 Bayreuth, Germanyc Senckenberg Gesellschaft für Naturforschung, Biodiversity and Climate Research Centre, Frankfurt, Germany

a r t i c l e i n f o

Article history:Received 11 July 2014Received in revised form 26 October 2014Accepted 30 October 2014

Keywords:Afro-montane rainforestEast AfricaTree inventoryOcotea usambarensisHistorical loggingForest regeneration

a b s t r a c t

The montane forests of Mount Kilimanjaro in Tanzania have been subjected to a long history of selectivelogging. However, since 1984 logging of indigenous trees is prohibited. Today, these forests allow us toevaluate the long-term effects of selective logging. We mapped the height and diameter at breast height(DBH) of all trees >10 cm DBH on 10 sites of 0.25 ha. Five sites represent non-logged forests, another fiveselectively logged forests. We tested whether forests were still visibly affected 30–40 years after selectivelogging in terms of their forest structure and tree diversity. Additionally we compared tree densities ofdifferent species guilds, including disturbance-indicator species, late-successional species and main tim-ber species. Furthermore, we specifically compared the community size distributions of selectivelylogged and non-logged forests, first across all species and then for the most important timber species,Ocotea usambarensis, alone. 30–40 years after selective logging forests still showed a higher overall stemdensity, mainly due to higher relative abundances of small trees (<50 cm DBH) in general, and higherdensities of small size class stems of late-successional species specifically. For O. usambarensis, the selec-tively logged sites harboured higher relative abundances of small trees and lower relative abundances ofharvestable trees. The higher relative abundance of small O. usambarensis-stems in selectively logged for-ests appears promising for future forest recovery. Thus, outside protected areas, selective logging may bea sustainable management option if logging cycles are considerably longer than 40 years, enough largesource trees remain, and the recruiting O. usambarensis individuals find open space for theirestablishment.

� 2014 Published by Elsevier B.V.

1. Introduction

Undisturbed tropical forests have become extremely rare(Gardner et al., 2009). Thus, forest management should focus onmaximizing the conservation values of human-modified forest(Gardner et al., 2009). In this context, selective logging was pro-posed as a management option to maintain conservation values,in terms of carbon stocks and biodiversity, as well as the economicvalue of a once-logged forest (Putz et al., 2012).

Nowadays, a typical selective-logging cycle occurs at 30–40 years intervals even though selectively logged forest might stillbe degraded after 100 years (Bonnell et al., 2011). However, theeffects of selective logging on the structure, dynamics and recoveryof forests remain uncertain (Bonnell et al., 2011). This uncertaintycalls for a thorough understanding of the effects of selective

logging on forest dynamics and regeneration, both in terms ofstructure and composition. A degraded forest may be recognisedby changed recruitment resulting in smaller tree size and a higherstem density than in an undisturbed forest. Additionally, degradedforests are likely to have less abundant late-successional and moreabundant disturbance-indicator species.

The montane forests of East Africa have been subjected to a longhistory of forest degradation, mostly due to selective logging(Ndangalasi et al., 2007; Bonnell et al., 2011; Hemp, 2006b;Bussmann, 1996; Kleinschroth et al., 2013; Persha and Blomley,2009). The main target timber species for selective logging wasthe East-African camphor tree, Ocotea usambarensis (Bussmann,1996; Kleinschroth et al., 2013), because it yields excellent timberwhich is resistant to fungal damage (Schulman et al., 1998). Har-vesting stems between 50 and 90 cm DBH provided the most effi-cient wood extraction from forests as smaller trees yield littletimber and larger trees require too much effort (Persha andBlomley, 2009). Selective logging was thus driven by the need for

62 G. Rutten et al. / Forest Ecology and Management 337 (2015) 61–66

timber even before selective logging became a widely recom-mended management alternative to clear-cutting. The long historyof selective logging in East African forests renders them ideal studysystems when evaluating the long-term effects of selective loggingand to inform future management decisions.

Kilimanjaro’s montane forests have been subject to selectivelogging, especially before 1984, when severe forest destructionled to the banning of logging (Agrawala et al., 2003). To studythe effect of past selective logging on these forests, we measuredvariables of vegetation structure, population structure and spatialarrangement of individual stems for non-logged forest sites andsites which were selectively logged 30–40 years ago. We expectedformerly selectively logged forests to still have lower mean treesize, basal area, density of large trees and a higher overall stemdensity, as well as reduced tree species richness, Shannon diversityand pairwise Sørensen diversities compared with non-logged for-ests. Furthermore, we expected selective logging to result in higherdensities of disturbance-indicator species but lower densities oflate-successional species and main timber species. As a certain sizeclass is subjected to selective logging we expected degraded treesize distributions in selectively logged forests compared withnon-logged forests.

2. Methods

2.1. Study area and sites

Mt. Kilimanjaro, Tanzania, is located 300 km south of the equa-tor and stretches from the savannas at 800 m above sea level (asl)to the snow-covered summit at 5895 m asl. Due to a wide precip-itation gradient the Southern and Eastern slopes of Mt. Kilimanjarohave a wide spectrum of habitats in distinct vegetation zones.These include a distinct forest belt which ranges from about1800 m to 3200 m asl (Hemp, 2006a). The forests in the middlemontane zone (2100–2800 m asl) are dominated by O. usambaren-sis (Lauraceae; Hemp, 2006a), which due to its high commercialvalue was the main target for selective logging. Selective loggingin our study area, the southern slope, was done from several saw-mills located inside the forest belt. The number of stems extractedwas usually quite low and very unlikely to have exceeded a dozentrees per hectare. However, on the drier eastern slope overexploi-tation has resulted in forests free of mature Ocotea with the samestructure and otherwise the same species composition. Thesepotential montane Ocotea-forests cover an area of about 110 km2.This means that one third of the actual montane Ocotea-forest isalready depleted of Ocotea (Hemp, 2006b). In 1984 logging wasbanned and in 2005 these forests were protected by inclusion inthe National Park. Although there is still illegal selective loggingin many areas of the mountain (Lambrechts et al., 2002), mostselective logging along our study transects took place before1984 (Agrawala et al., 2003). Our study sites were located in themontane zone at a mean annual temperature of 15–18 �C andmean annual precipitation of 2700 mm/year at 2200 m asl(Hemp, 2006a).

Between September 2011 and March 2012 we established0.25 ha sites at five non-logged sites and at five sites which hadbeen selectively logged 30–40 years ago. We chose the selectivelylogged sites for our study based on long-term expertise and localcontacts in this specific area. While we are certain that these siteswere selectively logged more than 30 years ago, the absence ofyoung stumps indicated that there was no further logging sincethen. The sites were distributed over a 31 km east–west stretchalong the southern, south-eastern slope and at elevations between2120 m and 2750 m asl. The average distance between sites was14 km with a minimum distance of 300 m.

2.2. Tree inventory

In each site we mapped all trees larger than >10 cm diameter atbreast height (DBH) by measuring the distance and direction toknown coordinates with an ultrasonic range finder (Haglöf VertexIV Hypsometer, Langsele, Sweden) and a compass (Suunto KB-14precision compass). For each tree we measured DBH, height andcrown extension. We measured DBH with a tape measure and,for large trees with buttresses only, we used a laser dendrometer(Criterion RD 1000; TruPulse 200/200, Centennial, USA) to measurethe diameter directly above the buttresses. To estimate treeheights and crown extensions in all four cardinal directions (N, E,S, W) we used ultrasonic rangefinders. Finally, we identified eachtree to the species level, and assigned the species to disturbance-indicator species and late-successional species (Table S2) basedon species occurrence in a large survey of several hundred vegeta-tion records across the mountain (A. Hemp, unpublished data).

2.3. Analysis

We assessed the impact of past selective logging on plot-basedmeasures of vegetation structure and diversity (assessed as meanDBH, mean tree height, stem density per ha, basal area per ha,crown area per square meter, tree species richness and tree speciesShannon diversity and the species guilds) with one-way analysis ofvariance (ANOVA). We used Tukey’s Honest Significant Differencepost hoc tests within species guilds to test for differences betweentree sizes.

To test the impact of logging on species composition, we calcu-lated Shannon’s diversity index and the pair-wise Sørensen’s indexusing the diversity-function and the vegdist- and betadisper-func-tions in the ‘Vegan 2.0-8’ package (Oksanen et al., 2013), respec-tively. Then, we used an ordination analysis (with the metaMDS-function in the ‘Vegan 2.0-8’ package) to test whether loggingaffected community dissimilarity as assessed by the pair-wiseSørensen’s index. In this analysis we used elevation, precipitation,UTM X- and Y-coordinates, slope, disturbance and a landscapeindex as predictor variables (Table S2). For each plot, slope wasdefined as the average inclination from the centre of the sites toall trees at distances between 5 and 15 m. The landscape index rep-resents the proportion of man-made habitats (tree plantations,cropland, grasslands and traditional home gardens) within a1.5 km diameter around the target site as assessed from satelliteimages (Nauss et al., 2014).

To assess differences in population structure, for all species andfor O. usambarensis only, we classified the trees into small (<50 cmDBH), harvestable (50–89.9 cm DBH) and large (>90 cm DBH) trees(Persha and Blomley, 2009). We used a Chi-squared test (as imple-mented as chisq.test-function in R; R Development Core Team,2014) to test for differences between logged and non-logged for-ests in numbers of stems per size class. Then we performed 2-sam-ple tests for equality of proportions to test differences in numbersof stems between logged and non-logged forests separately foreach size class (using the prop.test-function in R; R DevelopmentCore Team, 2014).

Additionally, we assessed the direct neighbourhood of largetrees in order to reveal patterns at smaller spatial scales. We calcu-lated the percentage of small trees, disturbance-indicator speciesand late-successional species (see above and Table S1) in theneighbourhood of each large tree, for which we could considerthe full 5 m radius without interference by the site borders. The5 m radius had been proposed earlier as the minimum distance rel-evant for neighbour effects (Stoll and Newbery, 2005). For eachsite, we grouped the points in each neighbourhood into twogroups per pattern (small/large size, disturbance indicator/non-disturbance indicator and late-successional/non-late-successional)

G. Rutten et al. / Forest Ecology and Management 337 (2015) 61–66 63

using the ‘spatstat’-package (Baddeley and Turner, 2005). Then weused ANOVAs to assess the impact of logging on the percentages ofthese variables. All data were analysed with R, Version 3.0.3(R Development Core Team, 2014).

3. Results

3.1. Forest structure

The selectively logged sites had 1.6 times higher stem densitiesthan non-logged sites (Fig. 1). The other forest structure variables(DBH, basal area and crown area) did not significantly differbetween selectively logged and non-logged sites (Table 1).

3.2. Tree diversity

The study sites harboured a total of 34 tree species (Table S1).The selectively logged and non-logged sites did not significantlydiffer in their mean tree species richness, Shannon index or Søren-sen index (Tables 1 and S1). Species similarity between plots didnot depend on logging activities (multiple regression of loggingon the first two axes of a NMDS ordination, R2 = 0.18, P = 0.507;Table S2), but was affected by elevation (R2 = 0.75, P = 0.013;Table S2) and the landscape index (R2 = 0.63, P = 0.032; Table S2).

3.3. Species guilds and O. usambarensis

In contrast to our expectations, the selectively logged sites har-boured a higher density of, albeit smaller, trees of late-successionalspecies (123.6; Fig. 2) than the non-logged sites (74.4). However,tree density of disturbance-indicator species was similar betweenlogged and non-logged sites (Table 1). Also, the densities of O.usambarensis did not significantly differ between logged and non-logged sites (Table 1), possibly because two selectively logged siteshad been completely depleted of Ocotea-trees, which led to largervariation in Ocotea stem densities among the selectively loggedsites than among the non-logged ones (F1,9 = 9.855, P = 0.012).

3.4. Size class distribution

The percentage of small trees <50 cm DBH in the selectivelylogged sites was higher, and the percentage of trees between 50and 89.9 cm DBH was lower than in the non-logged sites (Fig. 3a).

Fig. 1. Stem density per ha (mean ± SE) of all species in non-logged (light bars) andselectively logged montane forest sites (dark bars) at Mt. Kilimanjaro. Asterisksindicate levels of statistical significance based on ANOVA (⁄p < .05, ⁄⁄p < .01).

The percentage of small trees of O. usambarensis in the selec-tively logged sites was higher, and the percentage of trees between50 and 89.9 cm DBH lower (Fig. 3b) than in the non-logged sites.

3.5. Neighbourhood analysis

The percentage of disturbance indicator tree species and thepercentage of small stems in the neighbourhood of large trees werelarger in selectively logged sites than in non-logged sites (Table 1,Fig. 4). The percentage of late-successional species in the neigh-bourhood of the large individuals did not differ significantlybetween selectively logged and non-logged sites (Table 1).

4. Discussion

4.1. Effects of selective logging on forest structure and composition

Even 30–40 years after selective logging, we found a degradedforest structure in the selectively logged sites compared with thenon-logged sites. Our findings confirm the concerns raised in otherstudies that potential harvest cycles of 30 years for selective log-ging are too short to allow for a complete recovery of forest struc-ture (Hawthorne et al., 2012; Bonnell et al., 2011; Chapman andChapman, 2004). Selectively logged forests had a higher stem den-sity than non-logged forests, which is in line with the higher stemdensities found in disturbed montane forests in the West Usambar-a Mountains of Tanzania (Persha and Blomley, 2009). Their stemdensities (of 406 and 587 stems per ha) in natural and disturbedforests correspond well with our estimates (of 377 and 620 stemsper ha). In lowland forest in the Central African Republic, higherstem densities were also found for selectively logged forests thanfor non-logged forests (Hall et al., 2003). Whereas stem densitieswere higher in selectively logged sites, the other forest structureparameters (DBH, basal area and crown area) did not differbetween selectively logged and non-logged sites. Other studiesshow that this is often the case: In the West Usambara Mountains,basal area was also similar for selectively logged and non-loggedmontane forests and estimates of basal area were similar to ourmeasures (Persha and Blomley, 2009). Basal area was also foundto be similar in selectively logged and non-logged lowland forestsin the Central African Republic (Hall et al., 2003). While differencesin DBH, basal area and stem density between forest sites maystrongly depend on logging intensity (Bonnell et al., 2011 for mon-tane forests in Uganda and Kenya) it appears that stem densityshows the most consistent response to selective logging.

Past selective logging did not significantly affect tree speciescomposition. Likewise, tree diversity and species composition weresimilar in non-logged and selectively logged lowland forests inCentral Africa (Hall et al., 2003). Nevertheless, elevation and thepercentage of human-used habitats in a 1.5 km landscape aroundeach of our study plots (landscape index) explained significant var-iation in the community composition of our plots (R2 = 0.63;P = 0.038), suggesting that elevation and human influence mattereven though selective logging itself had no impact. Possibly theuse of non-timber forest products (NTFP), including pole cuttings,fire wood and medicinal plant collection, which occurs both inunprotected and in protected forest areas across East-Africa(Ndangalasi et al., 2007), have affected the structure and diversityof the forests. While the exact intensity of NTFP harvesting in Kili-manjaro National Park, and thus also in our sites, is unknown, it islikely to be related to the distance to the nearest settlements(Ndangalasi et al., 2007; Toivonen et al., 2011). We suggest takingNTFP into account in future studies of logging impacts.

As the forest structure, richness and diversity and environmen-tal parameters varied considerably between our sites, suggesting

Table 1Forest structure, tree diversity, species guilds, population structure and neighbourhood statistics for non-logged and selectively logged montane forests at Mt. Kilimanjaro. Mean(+sd) along with F-values and P-values from ANOVA-analysis. Lines with significant P-values are highlighted in bold.

Forest structure Non-logged (sd) Logged (sd) F1,8 P-value

Stem density (stems/ha) 377.60 (92.42) 620.80 (173.10) 7.68 0.024DBH (cm) 31.91 (2.60) 29.46 (3.14) 1.82 0.214Basal area (m2/ha) 48.76 (13.59) 62.86 (6.15) 4.47 0.068Crown area (m2/m2) 1.71 (0.46) 2.00 (0.90) 0.41 0.541Height (m) 14.55 (1.33) 16.39 (1.83) 3.33 0.106

Tree diversityRichness (number of species) 8.60 (2.07) 11.40 (2.70) 3.38 0.103Shannon index 1.45 (0.15) 1.67 (0.36) 1.56 0.247Beta diversity (pairwise Sørensen) – – – – 0.49 0.504

Species guilds (stems/0.25 ha)Ocotea usambarensis 36.40 (15.37) 47.80 (48.26) 0.25 0.628Late-succession speciesa 74.40 (13.58) 123.60 (20.55) 18.4 <0.001

Disturbance-indicator speciesa 24.00 (17.20) 43.40 (44.89) 0.81 0.393Population structure (stems/0.25 ha) v2 P-valueSmall (<50 cm DBH) 75.60 (20.02) 138.20 (49.46) 18.5 <0.001Harvestable(50–89.9 cm DBH) 16.00 (4.18) 14.40 (10.21) 15.4 <0.001Large (>90 cm DBH) 4.67 (1.53) 2.60 (1.34) 1.7 0.187

Percentage in neighborhood (5 m radius around large trees in plot center)Small trees 69.38 (11.52) 90.29 (7.23) 11.8 0.009Disturbance-indicator speciesa 8.76 (6.14) 31.34 (19.12) 6.32 0.036Late-succession speciesa 6.84 (6.97) 20.12 (25.42) 1.27 0.293

a Late-successional species and disturbance indicator species classified by expert knowledge (Table S1; pers. comm. Hemp).

Fig. 2. Stem density per ha (mean ± SE) of late-successional species for large andsmall stems, in non-logged (light bars) and selectively logged montane forest sites(dark bars) at Mt. Kilimanjaro. Asterisks indicate levels of statistical significancebased on ANOVA (⁄p < .05, ⁄⁄p < .01), the upper asterisks denote the difference inoverall stem densities between non-logged and selectively logged sites and lowerasterisks denote this difference separately for small and large stems.

64 G. Rutten et al. / Forest Ecology and Management 337 (2015) 61–66

that the detected differences between selectively logged and non-logged sites can be considered as a very robust result of selectivelogging across realistic environments.

4.2. Effects of selective logging on species guilds and size classdistributions

To further explore which species guilds add to the higher stemdensity in selectively logged forests, we compared densities of dif-ferent species guilds between selectively logged and non-loggedforests. We included disturbance-indicator species, late-successional species and the main timber species O. usambarensis.Contrary to our initial expectation, we found higher densities oflate-successional species in selectively logged forests than innon-logged forests. Possibly, post-logged forests, where few late-

successional trees remain, show increased regeneration as theseremaining trees serve as seed sources (Duah-Gyamfi et al., 2014;Ndangalasi et al., 2007; Magnusson et al., 1999; Chapman andChapman, 1997). Therefore, we assessed regeneration and popula-tion structures of selectively logged and non-logged forests by test-ing differences in their size class distributions. Higher relativefrequencies of small trees with stems between 10 and 50 cmDBH can be considered as advanced regeneration, a phase treesreach after initial seedling regeneration. Such advanced regenera-tion, specifically for late-successional species, was higher in ourselectively logged sites than in the non-logged sites, indicating thatforest recovery is on the way but may not be realized for severalmore decades.

4.3. Effects of selective logging on O. usambarensis

For the most important timber species, O. usambarensis, we didnot detect differences in overall stem density between selectivelylogged and non-logged sites. However, logged sites had higher rel-ative abundances of small trees and lower relative abundances ofharvestable O. usambarensis trees. This suggests that selective log-ging cycles of 30–40 years deplete forests of harvestable stems.However, the high relative abundances of small trees suggest thatconsiderably longer cycles might provide sustainable timberharvests.

Independent of logging, the high average density of Ocotea indi-viduals indicates that Ocotea regeneration is not as limited at thesouthern slopes of Mt. Kilimanjaro as at Mt. Kenya (Kleinschrothet al., 2013) or in the Eastern Arc mountains (Persha andBlomley, 2009). These inter-mountain differences may beexplained by many mountain-specific characteristics. Mt. Kili-manjaro’s montane forests have an average O. usambarensis densityof 168.4 trees per ha, almost 5 times higher than the stem densityfound at Mt. Kenya (Kleinschroth et al., 2013), and even 12 timeshigher than in the Eastern Arc mountains (Persha and Blomley,2009). As there is no reason to believe that montane forests offermore open gaps for regeneration at Mt. Kilimanjaro than at theseother mountains, this suggests that the high density of source treesat Mt. Kilimanjaro could be essential for regeneration via seeds or

Fig. 3. Stem size class distributions of all species (a) and of O. usambarensis (b).Mean relative frequency (±SE) of small trees (<50 cm DBH), harvestable trees (50–89.9 cm DBH) and large trees (>90 cm DBH) in non-logged sites (light bars) andselectively logged sites (dark bars) at Mt. Kilimanjaro. Asterisks indicate levels ofsignificance based on individual 2-sample tests for equality of proportions for eachof the classes (⁄⁄⁄p < .001).

Fig. 4. Percentage of small stems (a) and of stems of disturbance-indicators species(b) in the neighbourhood (5 m radius) of (remaining) large trees in non-logged(light bars) and selectively logged (dark bars) sites. Asterisks indicate levels ofsignificance based on ANOVA (⁄p < .05, ⁄⁄p < .01).

G. Rutten et al. / Forest Ecology and Management 337 (2015) 61–66 65

vegetative root sprouts, whereas the much lower O. usambarensisdensities on other East-African mountains (Kleinschroth et al.,2013; Persha and Blomley, 2009) result in an extremely limitedregeneration. Moreover, elephants generally decrease forest regen-eration (Lawes & Chapman, 2006; Struhsaker et al., 1996; Hemp,2006b), and the regeneration of O. usambarensis in particular(Kleinschroth et al., 2013). As elephants are not found onKilimanjaro’s southern slopes but they occur on the slopes ofMt. Kenya, this may further add to the different densities of sourcetrees at these mountains and thus to differences in regeneration. Inaddition, rainfall inside the southern forest belt of Kilimanjaromarkedly exceeds precipitation on other East African highmountains (Hemp, 2006a), possibly enhancing the regenerationof O. usambarensis. Together many of these factors could beresponsible for the high O. usambarensis density found on MountKilimanjaro, and finding out which are most important could assist

in forest regeneration and recovery measures of all degraded EastAfrican montane forests.

4.4. Neighbourhood patterns

While we did not detect differences between selectively loggedand non-logged forests in overall disturbance-indicator speciesdensity, there were more disturbance-indicator species in the close5 m-neighborhood of remaining large trees in selectively loggedforests. This suggests that disturbance-indicator species prefer spe-cific neighbourhoods. Late-successional tree species showed theopposite pattern of higher density in selectively logged than innon-logged sites, but of similar percentages of late-successionalstems in the neighbourhoods of large trees in selectively loggedand non-logged sites. Apparently the late-successional speciesprofited from increased regeneration after selective logging with-out specific local preference for neighbouring species. However,the disturbance-indicator species were not generally promoted,but regenerate more in specific local neighbourhoods, probablybecause the general level of disturbance after selective logging islower than the one after clear cutting or forest fire. To reveal more

66 G. Rutten et al. / Forest Ecology and Management 337 (2015) 61–66

specific mechanisms and species-level dynamics driving regenera-tion after disturbance in these montane forests, we recommendfocusing on the population dynamics and spatial patterns ofselected tree species and on specific guilds and functional traits,such as dispersal strategy. Such studies will require larger numbersof trees per species than what we found in our study of more gen-eral focus.

4.5. Management conclusions

In conclusion, the selectively logged montane forests at Mt. Kili-manjaro show degraded vegetation structures with higher stemdensities than in non-logged forests. This increased stem densityresulted from a higher density of small size class stems, and ahigher relative frequency of small trees around remaining, possibleseed source, trees. Interestingly, the selectively logged forests alsoharboured more stems, mainly of small size class, of late-succes-sional species. Managers considering selective logging shouldretain these possible seed sources by maintaining several largetrees per hectare.

For the most important timber species, O. usambarensis, we didnot detect differences in overall stem density between selectivelylogged and non-logged sites. However, logged sites had higher rel-ative abundances of small trees and lower relative abundances ofharvestable O. usambarensis trees. The relative high abundance ofsmall O. usambarensis-stems in selectively logged forests exempli-fies the generally enhanced regeneration in the logged forests ofMt. Kilimanjaro than in non-logged forests and this high abun-dance of small individuals appears promising for future forest con-ditions. Thus, outside protected areas, selective logging might be asustainable management option. However, this requires that log-ging cycles are longer than 40 years, that enough large treesremain, which could serve as sources of seeds or root suckers,and that the recruiting O. usambarensis individuals find open spacefor their establishment, if necessary assisted by removing trees ofearly successional species.

Acknowledgements

We thank U. Pommer, A. Mmari, J. Maruchu, A. Reinehr and M.Wachendorf for field assistance, C. Hemp for providing infrastruc-ture and P. Stoll, R. Delgado and two anonymous reviewers for use-ful comments on earlier versions of the manuscript. This study wassupported by the Swiss National Science Foundation (SNSF) in thecontext of Research-Unit 1246 of the German Research Foundation(DFG). The required research permit was granted by COSTECH(2010-340-ER-NA-96-44 and 2011-340-ER-96-44) and TANAPA(TNO/HQ/ C.10/13/VOL.III).

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.foreco.2014.10.036.

References

Agrawala, S., Moehner, A., Hemp, A., van Aalst, M., Hitz, S., Smith, J., et al., 2003.Development and Climate Change in Tanzania: Focus on Mount Kilimanjaro.Environment Directorate and Development Cooperation Directorate,Organisation for Economic Co-operation and Development, Paris.

Baddeley, A., Turner, R., 2005. Spatstat: an R package for analyzing spatial pointpatterns. J. Stat. Softw. 12, 1–42.

Bonnell, T.R., Reyna-Hurtado, R., Chapman, C.A., 2011. Post-logging recovery time islonger than expected in an East African tropical forest. For. Ecol. Manage. 261,855–864.

Bussmann, R.W., 1996. Destruction and management of Mount Kenya’s forests.Ambio 25, 314–317.

Chapman, C.A., Chapman, L.J., 1997. Forest regeneration in logged and unloggedforests of Kibale National Park, Uganda. Biotropica 29, 396–412.

Chapman, C.A., Chapman, L.J., 2004. Unfavorable successional pathways and theconservation value of logged tropical forest. Biodivers. Conserv. 13, 2089–2105.

Duah-Gyamfi, A., Swaine, E.K., Adam, K.A., Pinard, M.A., Swaine, M.D., 2014. Canharvesting for timber in tropical forest enhance timber tree regeneration? For.Ecol. Manage. 314, 26–37.

Gardner, T.A., Barlow, J., Chazdon, R., Ewers, R.M., Harvey, C.A., Peres, C.A., et al.,2009. Prospects for tropical forest biodiversity in a human-modified world. Ecol.Lett. 12, 561–582.

Hall, J.S., Harris, D.J., Medjibe, V., Ashton, P. Mark S., 2003. The effects of selectivelogging on forest structure and tree species composition in a Central Africanforest: implications for management of conservation areas. For. Ecol. Manage.183, 249–264.

Hawthorne, W.D., Sheil, D., Agyeman, V.K., Abu Juam, M., Marshall, C.A., 2012.Logging scars in Ghanaian high forest: towards improved models forsustainable production. For. Ecol. Manage. 271, 27–36.

Hemp, A., 2006a. Continuum or zonation? Altitudinal gradients in the forestvegetation of Mt. Kilimanjaro. Plant Ecol. 184, 27–42.

Hemp, A., 2006b. Vegetation of Mount Kilimanjaro: hidden endemics and missingbamboo. Afr. J. Ecol. 44, 305–328.

Kleinschroth, F., Schöning, C., Kung’u, J.B., Kowarik, I., Cierjacks, A., 2013.Regeneration of the East African timber tree Ocotea usambarensis in relationto historical logging. For. Ecol. Manage. 291, 396–403.

Lambrechts, C., Hemp, C., Nnyiti, P., Woodley, B., Hemp, A., 2002. Aerial survey ofthe threats to Mt. Kilimanjaro forests.

Lawes, M.J., Chapman, C.A., 2006. Does the herb Acanthus pubescens and/orelephants suppress tree regeneration in disturbed Afrotropical forest? For.Ecol. Manage. 221, 278–284.

Magnusson, W.E., de Lima, Odilon P., Quintiliano Reis, F., Higuchi, N., FerreiraRamos, J., 1999. Logging activity and tree regeneration in an Amazonian forest.For. Ecol. Manage. 113, 67–74.

Nauss, T., Detsch, F., Otte, I., Appelhans, T., 2014. Satellite-derived land- coverheterogeneity indices for various spatial buffers around 73 research plots on thesouthern slopes of Mt. Kilimanjaro. V 1.0. ZENODO. Available from: http://dx.doi.org/10.5281/zenodo.12540.

Ndangalasi, H.J., Bitariho, R., Dovie, D.B.K., 2007. Harvesting of non-timber forestproducts and implications for conservation in two montane forests of EastAfrica. Biol. Conserv. 134, 242–250.

Oksanen, J., Blanchet, F.G., Kindt, R., Oksanen, M.J., Suggests, M., 2013. Package‘vegan’. Commun. Ecol. Package Version 2.

Persha, L., Blomley, T., 2009. Management decentralization and montane forestconditions in Tanzania. Conserv. Biol. 23, 1485–1496.

Putz, F.E., Zuidema, P.A., Synnott, T., Peña-Claros, M., Pinard, M.A., Sheil, D., et al.,2012. Sustaining conservation values in selectively logged tropical forests: theattained and the attainable. Conserv. Lett. 5, 296–303.

R Developement Core Team, 2014. R: A language and environment for statisticalcomputing. ISBN 3-900051-07-0. R Foundation for Statistical Computing.Vienna, Austria, 2013. <http://www.R-project.org>.

Schulman, L., Junikka, L., Mndolwa, A., Rajabu, I., 1998. Trees of Amani NatureReserve, NE Tanzania. The Ministry of Natural Resources and Tourism, Tanzania,335pp.

Stoll, P., Newbery, D.M., 2005. Evidence of species-specific neighborhood effects inthe Dipterocarpaceae of a Bornean rain forest. Ecology 86, 3048–3062.

Struhsaker, T.T., Lwanga, J.S., Kasenene, J.M., 1996. Elephants, selective logging andforest regeneration in the Kibale Forest, Uganda. J. Trop. Ecol., 45–64.

Toivonen, J.M., Kessler, M., Ruokolainen, K., Hertel, D., 2011. Accessibility predictsstructural variation of Andean Polylepis forests. Biodivers. Conserv. 20, 1789–1802.