Predictive mapping of tree species assemblages in an African montane rainforest

doi:10.21203/rs.3.rs-2444058/v1

Conservation planning of montane areas requires knowledge of vegetation composition and distribution patterns that is often lacking in rugged terrain. We mapped the “Afromontane” vegetation of Bwindi Impenetrable Forest, Uganda, represented as tree species assemblages. Given the wide variation in elevation, topography and human impacts across Bwindi, we hypothesized that these key factors, either acting in isolation or in combination, would influence the tree species distribution patterns. To test this, detailed tree surveys were carried out using stratified random sampling. These data were evaluated against bioclimatic variables and elevation as potential predictive attributes using multivariate statistical techniques. Cluster analysis identified six tree species assemblages that were confirmed by MRPP, Mantel’s test, and Indicator Species Analysis to possess distinct compositions. Ordination analysis showed that elevation was the main factor explaining variation in tree species between forest sites but other factors including longitude, human disturbance, ridge-top, temperature, and precipitation were also significant. Random Forests modeling predicted and spatially mapped the tree species assemblages with elevation showing the highest predictive power. These results confirm our hypothesis that the combined effects of elevation, topography, climate, and human disturbance showed consistent and non-random relationships to tree species composition and distribution but the individual contributions could not be analytically separated. Nevertheless, we show that it is possible to predict the spatial pattern of species-rich but disturbed tree communities based on a limited sample across a complex landscape. Such methods and results offer accurate mapping of composition that will be of great value to managers and conservationists.

Bwindi Impenetrable Forest

elevation

human disturbance

Random Forests

Vegetation mapping

Montane forest is the rarest vegetation type on the African continent (White 1983). However, this biologically rich and unique “Afromontane” ecosystem is increasingly threatened by deforestation, fragmentation, and degradation (Balmford et al. 2001; Brooks et al. 2002; Mittermeier et al. 2004; Burgess et al. 2007; Cordeiro et al. 2007; Plumptre et al. 2007), while climate change has added new threats (Parmesan 2006; Sala et al. 2000; Carr et al. 2013; Ayebare et al. 2018; Wright et al. 2022). Many species confined to montane forest are highly susceptible to the impacts of climate change and human influences (Carr et al. 2013; Ayebare et al. 2018; Wright et al. 2022). This calls for prioritization of conservation actions to protect exceptional montane forest species and habitats that are prone to extinction driven by anthropogenic disturbance and climate change. Mapping of forest vegetation is one approach that can help to assess, plan, and guide conservation management in addressing these challenges (Zimmermann and Kienast 1999; Dias et al. 2004; Fjeldsa 2007; Brinkmann et al. 2011).

Predictive vegetation mapping is defined as predicting the vegetation composition across a landscape from mapped environmental variables (Franklin 1995). It is normally done using a combination of field data with digital maps of topography, as well as climatic limiting factors, facilitated by flexible, algorithmic modeling approaches and is driven by the need to map vegetation over large areas for conservation planning (Pfeffer et al. 2003; Evans and Cushman 2009). Using non-parametric machine learning data-mining techniques can readily uncover complex patterns in species data and relationships between species and physical features of the environment (Brieman 2001; Iverson et al. 2004; Prasad et al. 2006; Cutler et al. 2007; Evans and Cushman 2009).

The aim of the study was to predict and map the distribution of tree communities across Bwindi Impenetrable National Park, (Bwindi), an Afromontane forest in southwest Uganda. Because Bwindi is primarily forest (Howard 1991), we focus on trees to determine differences in the vegetation. There is a dearth in recent information on the distribution of tree species and communities across the Park. Accurate maps of vegetation can aid management and contribute to various research as well as identifying restricted and vulnerable communities and habitats to support priority conservation planning (McDonald and Boucher 1999; Pfeffer et al. 2003; Brinkmann et al. 2011). For example, the mountain gorilla (Gorilla beringei beringei) is a highly endangered flagship subspecies, and the population in Bwindi is one of just two in Africa (McNeilage et al. 1997). Despite their conservation (Butynski 1985) and economic (Macfie and Williamson 2010) importance, the present distribution patterns of gorillas related to forest composition are poorly understood because there is no up-to-date map of the forest vegetation (Guschanski et al. 2008).

Mountains create complex environmental gradients often covering broad range in temperatures, terrains, moisture, soils, illumination, and human impacts. Many environmental factors vary with elevation, such that the conditions required for species to persist are restricted to specific elevations leading to some characteristic zonation along elevation gradients (Hamilton 1975; Richards 1996; Lovett et al. 2001; Eilu et al. 2004; Ghazoul and Sheil 2010; Schmitt et al. 2010). Plant communities are further structured at a local scale by topography, reflecting differences in soil depth, structure, moisture, or nutrients (Leggat and Osmaston 1961; Hamilton 1969; Ghazoul and Sheil 2010; Eilu et al. 2004). Tropical species are known to be highly sensitive to disturbance and shifts in environmental conditions (Sheil 2016; Ssali et al. 2017a). For example, human disturbance has been shown to relax the constraints imposed by competition and extend effective elevation ranges of some species, particularly those in secondary forest, to warmer and cooler climates (Muñoz Mazón et al. 2019). Thus, montane vegetation spatial patterns are a result of complex interactions between topography, past and recent natural and human disturbance factors as well as interactions among species such as competition (Hoersch et al. 2002). This makes the understanding of relationships between physical features of the environment and vegetation a great challenge. Given the wide variation in elevation, topography and past human disturbance across Bwindi Forest and evidence from aforementioned studies, we hypothesized that elevation, topography, and human disturbance, either acting in isolation or in combination, greatly influence the composition and distribution of tree species and communities.

Study area

Bwindi Impenetrable National Park (henceforth “Bwindi”) is located in SW Uganda (Fig. 1), in the Kigezi Highlands at the eastern edge of the Albertine Rift (latitude 0º53´- 0º8´S and longitude 29º35´- 29º50´E). Covering an area of 331 km², Bwindi lies at the north west end of the Kigezi Highlands which are associated with the up-warping and faulting during the formation of the Albertine Rift and are underlain by Precambrian shale, phyllite, quartz, quartzite, schist and granite of the Karagwe-Ankolean System (Leggat and Osmaston 1961). The Atlas of Uganda (1967) classified the soils of Bwindi as belonging to the “non-differentiated humic ferrallitic” types and having been derived from the foregoing geological formation, and they comprise of mainly tropical red earths with an overlying layer of brown to black spongy humus. The erosive action of the numerous rivers in their youth stage within the Highlands has caused the topography of the park to be extremely rugged with narrow, steep-sided valleys and deep gullies that run in various directions, bound by emergent hill crests lying between 1,190 m in the northwest and 2,607 m a.s.l. in the southeast (Howard 1991). The forest has been classified as a ‘moist lower montane forest’ (Hamilton 1982), ranging from near the upper boundary of lowland forest to montane forest (Hamilton 1975). It is one of the few forests in all of Africa where lowland and montane forests are in a continuum (Hamilton 1989). This exceptional feature is very important for studies of elevational variations in rainforest ecosystems in tropical Africa. Hamilton (1982, 1989) postulated that the Bwindi was richer in tree species than the other forests of western Uganda because of proximity to the Pleistocene refuge. However, recent studies attribute the rich tree diversity of Bwindi to high rainfall and soil characteristics (Eilu et al. 2004), highlighting the importance of studying shifts in floristic diversity, composition, and distribution along environmental gradients. The park lies in a region with the highest rural human population density in Africa (Codeiro et al. 2007). Because of this, little natural forest persists outside the boundaries of the park (Twongyirwe et al. 2011). Before it became a park, the forest was subjected to various forms of human disturbance with pit-sawing and poaching occurring throughout the park (Albrechet 1976; Harcourt 1981; Butynski 1984; Howard 1991; McNeilage et al. 1998), while mining (gold and tungsten) was concentrated in specific locations (Harcourt 1981; Butynski 1984). Human-induced wildfires impacted the forest periphery (Leggat and Osmaston 1961; Butynski 1984), and numerous human trails and two public roads traverse the forest (Harcourt 1981; Butynski 1984). Ongoing disturbance processes include localized landslides, elephants (Loxodonta africana), occasional wildfires following long periods of drought, harvesting of poles and wire snaring (Babaasa 2000; Babaasa et al. 2004; McNeilage et al. 2001; McNeilage et al. 2006; Olupot et al. 2009; Ssali et al. 2012; Hickey et al. 2019), and a public road cutting through the narrow forest corridor and high elevation zone (Barr et al. 2015). This complexity of factors contributes to making Bwindi a challenging site for investigations of plant communities and consequently attempts to map Bwindi vegetation have been limited.

Information on Bwindi’s vegetation cover is scarce and limited. A forest type map prepared by Cahusac (1961) for managing timber harvesting operations is now outdated given the management history of Bwindi that included over 40 years of intense and forest-wide timber harvesting (Butynski 1984; Howard 1991), whose long-term impact was the creation of a diverse patch mosaic of vegetation types differing in age, hence succession, and broken canopy cover (Babaasa et al. 2004; Sheil 2012). Incessant, heavy human and natural disturbances have prevented reclosure of the forest canopy (Ssali et al. 2017a, b). Previous descriptions of the tree flora of Bwindi were limited to a small area of high elevation in the south-east (Leggat and Osmaston 1961; Hamilton 1969, 1974; Lind and Morrison 1974; Howard 1991) or low elevation in the north (Eilu et al. 2004). Later work on trees (Davenport et al. 1996), though covering more representative areas of the forest, recorded only the species encountered but did not geo-reference the sample sites, preventing spatial analyses and extrapolation. Until now, it had remained unclear how forest-wide ﬂoristic patterns were spatially structured in relation to the environment. The results of this study were urgently needed by park management for conservation planning of Bwindi.

Study design

To account for the different environmental conditions, we employed a stratified random sampling approach to guide sampling of tree species. The park was divided into five strata based on geological formations visible on the Digital Elevation Model (DEM) of Bwindi and the starting points on the boundary of the DEM in each stratum selected randomly with the random point function within ArcGIS (version 10.2; ESRI, Redlands, CA, USA). Line transects were purposely drawn on the DEM in each stratum from the random starting points to traverse the topographic positions of the ridges (Fig. 1; Table 1). The number and length of the transects selected varied with area, accessibility and shape of the strata. We then superimposed the transect drawings on high-resolution (0.5m) true color, digital aerial photographs of Bwindi. The aerial photos were visually interpreted along the transects by drawing polygons around areas perceived to be of uniform tree community structure based on differences in tone and texture. This allowed the sample plots to be placed in what we perceived to be different tree communities.

Tree species sampling

We carried a printed copy of the digitized polygons, overlaid with a coordinate grid, and used a hand-held Global Positioning System (GPS) device to locate the digitized polygons in the field. A single random point within each digitized polygon was selected for tree species sampling. At each random sample point, we used the point-to-tree distance technique or plotless sampling method to sample trees (Hall 1991; Sheil et al. 2003; Klein and Vilcko 2006). This technique involved selecting the nearest 15 trees (≥20 cm dbh) around the random center-point. The selected trees were identified to species level and we measured the diameter-at-breast-height (dbh) of each individual. We named the tree species following nomenclature used in Kalema and Hamilton (2020). The distance from the center-point to the 15^th farthest tree was measured and regarded as the radius of circular plot. This procedure is especially suitable for rapid and robust assessments of tropical rain forest composition compared to fixed area plot methods (Sheil et al. 2003; Klein and Vilcko 2006). At each center-point, eight environmental attributes were recorded: aspect – as the compass direction facing down slope; and steepness of the slope using a clinometer. Untransformed aspect and slope are poor for quantitative analysis, so slope was transformed to a more suitable index by taking the sine of the slope in degrees; aspect was also transformed into a suitable index by taking the negative cosine of the angle in degrees minus 35 (McCune and Grace 2002; Cushman and Wallin 2002). Four physiographic positions of valley, hillside, ridge tops and gully were simply recorded as “1” if the plot was in that physiographical class and “0” otherwise. The final recorded site characteristics consisted of spatial variables – the Universal Transverse Mercator (UTM) coordinates of easting and northing (datum WGS 84) using a hand-held GPS unit, standardized to zero mean and unit variance.

Biophysical predictor variables

We acquired twenty-one digitally mapped biophysical variables (Table 2). These were: the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM) 30×30 m Version 2 (http://gdem.ersdac.jspacesystems.or.jp/), and nineteen 1-km-scale (30 sec) bioclimatic variables, that include 11 temperature and 8 precipitation variables from WorldClim version 2 (http://worldclim.org/version2; Fick and Hijmans 2017). We clipped the ASTER GDEM and the 19 bioclimatic rasters to a window covering Bwindi Forest only. Lastly, the human disturbance across the park was based on the combined relative encounter of human activity signs that were likely to have an impact on vegetation – wood cutting, bee hives, old pitsaw sites, disused mineral extraction pits, snares, and fireplaces (McNeilage et al. 1998). We used the geo-references of all the 289 plots to extract site values from the predictor rasters. We constructed an environmental matrix of extracted variable values together with site measurements (aspect, slope, topographic position and x,y point coordinates) and tested for pairwise collinearity and one of any pair of highly correlated variables (Pearson r ≥0.75) discarded (Table 2). We ended with an environmental matrix of 289 plots by 12 least correlated environmental variables.

Data analyses

Based on field data, a plots-versus-species matrix (using basal area values of each tree species relative to the area of the plot [m²ha^-1]) was created. We only considered tree species occurring in more than four plots, resulting in a 289 plot by 89 tree species matrix. The data were natural log-transformed following the generalized procedure (McCune and Grace 2002) to minimize the influence of large trees. We subjected the data matrix to a varius multivariate techniques to identify associations and spatial patterns among the tree species using R software (version 4.2.1; R Core Team 2022). A polythetic, agglomerative hierarchical clustering (HC) using the flexible beta linkage method (β = -0.25; Lance and Williams 1966; Legendre and Legendre 1998) with the agnes() function and Bray-Curtis as a distance measure was used as a method for cluster formation. We applied the Indicator Species Analysis (ISA), described below, to determine the optimal level of pruning the HC dendrogram. Because we had numerous plots (n = 289) we simplified the presentation using composite plots from the original plots by computing the centroid of each of the cluster formed by HC analysis as the mean of the basal area/ha of each tree species per cluster.

We used Multi-Response Permutation Procedures (MRPP) and a Mantel’s test to test for differences in composition between the HC cluster groups. We also utilized ISA, a method that combines frequency and mean abundance tables, to identify the characteristic tree species for each cluster. Statistical significance of the indicator tree species was determined by random permutations for each species at p<0.05 significance level. Indicator values vary from 0 (no indication) to 1 (perfect indication).

A Kruskal’s non-metric multidimensional scaling (NMDS) based on Bray-Curtis coefficient was used to evaluate the ecological tendencies reflected in the HC dendrogram and the relationship between plots, clusters and environmental variables (Table 2). The plots were ordinated, then overlaid with the cluster centroids from HC analysis with surrounding confidence ellipses at ±2 SD from the mean (enclosing approximately 95% of plots within each cluster). Lastly, 12 least correlated site environmental variables (Table 2) were fitted onto the ordinations using 1,000 permutations.

We used Random Forests technique (Breiman 2001; Liaw and Wiener 2002) to predict and map the spatial distribution of the HC clusters. The Random Forests algorithm has been shown to produce very accurate predictions without overfitting models to the data (Breiman 2001). In Random Forests, bootstrap samples are drawn to construct multiple trees. Each tree is grown with a randomised subset of predictors and a large number (500-2,000) of trees are grown (hence 'forest' of trees). The trees are grown to maximum size without pruning and aggregation takes place by averaging the trees. It is not possible to interpret each of the trees in Random Forests, but the procedure does provide tables on relative importance among predictor variables. The proportion of the data not drawn (out-of-bag or OOB) in each bootstrap replicate is used to calculate an unbiased error rate and variable importance, eliminating the need for a test set or cross-validation. Since our response variable, the clusters, was a factor (categorical), we performed a classification procedure in the analysis. Six least correlated variables (Pearson’s r <0.75) that are digitally mapped for the whole park: Isothermality, temperature seasonality, Minimum temperature of the coldest month, Annual precipitation, Mean temperature of the coldest quarter and elevation were tested as predictors for cluster distribution.

Cluster analysis

The HC dendrogram (Fig. 2) grouped the plots into six clusters based on composition as it is the cluster-stage that yielded the lowest average summed p-values as well as highest number of significant indicator species based on ISA for 2 to 16 clusters (Fig. 3). The six groups were further grouped in three pairs, with clusters 1 and 5 being most similar in composition, while clusters 2 and 6 were more distinct.

Multi-Response Permutation Procedures and Mantel’s test

The Multi-Response Permutation Procedures (MRPP) test showed that the clusters were significantly different from random association (observed delta = 0.79, expected delta = 0.89, p<0.001). The Mantel test (r = 0.32 p<0.001) also indicated that the six clusters differed significantly in composition.

Indicator Species Analysis

In total, 63 indicator tree species proved significant for the six cluster groups (Table 3). All the species, with the exception of Gambeya albida in cluster 4, had indicator values of less than 50% implying that that they occurred also, at lower abundance and frequency, in other clusters. The clusters had numerous indicator species with clusters 2, 5 and 6 having the largest number – 14, 14 and 18, respectively, while clusters 1, 3 and 4 had the fewest – 6, 7 and 4, respectively. Twenty-four species were pioneers (early successional) and mid-successional species (non-pioneer light demanders, NPLDS), spread across the clusters. Five indicator species in clusters 1, 2 and 5 are classified as vulnerable (VU) in IUCN Red List Categories of threat.

Combined NMDS ordination and HC analyses

NMDS resulted in a 3-axis optimal solution and a good fit with a clear positive relationship between observed community dissimilarity and ordination distances. The NMDS ordination diagram produced a dense cluster of plot points and HC analysis confidence ellipses that greatly overlapped on all the three axes (Fig. 4a and b). The plot points that were positioned where the ellipses intersected did not belong exclusively to one cluster but had compositions indicative of two or more clusters. The overall patterns show a complex range of variables that show consistent and non-random relationships to species composition and distribution. The strongest correlations were observed for elevation and longitude, but ridge-top, precipitation, temperature and human disturbance factors also showed a non-random role (Fig. 5a and b). Since elevation and longitude arrows were close and pointed in the same direction, it means they are positively correlated implying that clusters 2, 3, 4 and 6 positioned along the direction of the arrows are depicted to be at high elevation to the east of the park.

Tree species assemblage spatial distribution pattern

The Random Forests predictive map for the six clusters (Fig. 6) and description of the clusters in Table 3, revealed distinct vegetation geographically arranged along the north-west and south-east latitudinal and longitudinal axes of Bwindi. Cluster 1 (white tone) and 5 (light green tone) were largely in areas north and west of Bwindi, while cluster group 2 (pink tone) and 6 (dark green tone) were primarily in the extreme southeasterly areas. Cluster groups 3 (yellow orange tone) and 4 (yellow green tone) were in the middle of the forest. However, in some areas, the clusters mosaicked, with patches of cluster group 6 within cluster groups 2 and 5, while those of cluster group 2 were within cluster group 5, and those of cluster group 1 within cluster groups 2 and 5. We found no evidence of any of our six clusters being more or less associated with forest edges.

Variable importance estimates (Fig. 7) revealed that elevation had the greatest estimated importance in terms of both accuracy and Gini score in predicting cluster composition and spatial distribution. Only two other variables had importance of nearly similar magnitude for predictive accuracy: Precipitation of Driest Month (prec_min) and Annual Precipitation (prec_ann). The OOB estimate of the error rate was 50.7% (Table 4).

Our findings demonstrate that the Random Forests model, supported by multivariate statistical techniques, was effective in delineating tree communities quantitatively and predictively mapping them in response to complex topographic gradients, compounded by vegetation with a heavy and long disturbance history. For the final result to be optimal, all our data were used in the model instead of forfeiting them for an independent validation. We chose to use the OOB estimate of the error rate provided in Random Forests. Our OOB error statistic was moderately high. Brinkmann et al. (2011) attributes inaccuracies of predicted maps to the predictor variables not being able to differentiate between certain vegetation units, poor resolution of, and errors in, available maps of predictor variables, among others. More important could be Volkov et al. (2003) observation that the distribution of plant abundances in natural communities is largely random. Pfeffer et al. (2003) suggest that several plant species are tolerant to a wide range of varying conditions, and therefore, they can never be a perfect species-environment correlation. Additionally, species data tend to be noisy (ter Braak 1995) because of the complex interactions between several underlying factors (Hoersch et al. 2002). Nevertheless, the most significant variables generated by Random Forests model reflect the primary biophysical factors that would be expected to have a large influence on composition and distribution of tree species in a montane forest (Hamilton 1975; Richards 1996; Lovett et al. 2001; Eilu et al. 2004; Ghazoul and Sheil 2010; Schmitt et al. 2010) and this made it feasible to predictively map vegetation at unsampled locations. This underscores the statistical power of Random Forests to find ecological patterns among species and their relationships with the physical environment across our study site characterized by a complexity of an extremely rugged landscape with high topo-diversity and a patchwork of microclimatic habitats as well as a continuum of species-rich but heavily disturbed vegetation of our study site.

A combined HC analysis and ISA identified six tree communities across Bwindi each with their own indicator species. The numerous indicator species for each tree community, without a single species being dominant, is indicative of a large proportion of the forest being secondary vegetation communities. In many Ugandan forests, single-species dominance is a characteristic of mature (climax) communities (Eggeling 1947). Furthermore, close to 40 percent of the indicator species, spread across the tree communities, were pioneers (early successional species that are shade intolerant and associated with disturbed areas such as Macaranga barteri at low elevation and M. capensis, Polyscias fulva and Neoboutonia macrocalyx at high elevation) and mid-successional or non-pioneer light demanders (NPLDs; shade tolerant as seedlings but requiring strong illumination for subsequent growth such as Gambeya albida, Myrianthus holstii and Newtonia buchananii). This is further evidence that considerable areas of the forest are in early succession stages. Human disturbances of the past, particularly pitsawing, were spread throughout the forest (Albrechet 1976; Butynski 1984; Howard 1991; McNeilage et al. 1998). However, the disturbances varied in intensity and were concentrated in different areas at different time periods (McNeilage et al. 2001). This led the forest vegetation to be structurally and compositionally complex, with a mosaic of patches of disturbed and regenerating forest, differing in size and succession stage, being superimposed on differences caused by topography and possibly soil characteristics. It could be a very long time before the forest regenerates to the primary (climax) stage.

Ordination analysis demonstrated that the tree communities were mainly arranged along two main gradients – elevation and space (longitude). Bwindi is elongated from north-west to south-east, exhibiting a gradual change in mean elevation in same direction. The elevation range of Bwindi is wide (slightly above 1,400 m), making it the strongest variable to influence shifts in tree species assemblage composition. Although elevation had the most important influence on tree composition and distribution, additional variability is explained by local topography (i.e., ridge-top), landscape climate factors (temperature and precipitation) and human disturbance. Local physiography affects vegetation composition as the soils of slopes and lower valleys are generally moister, better structured, and richer in plant nutrients than drought-prone ridge tops (Ghazoul and Sheil 2010). Our results show elevation and temperature variables were negatively correlated. Generally, temperature decreases in a linear fashion with elevation (Lieberman et al. 1996) with low-temperature tolerance probably playing a role in tree composition shifts along elevation gradients, with increasing cold tolerance offset by lower competitive ability under warmer conditions (Slik et al. 2009). The Random Forests model predicted that the amount of rainfall (Annual precipitation) and drought severity (Precipitation of Driest Month) contribute to shifts in composition. From this perspective, many variables influence tree community spatial patterns but in a complex manner in which individual contributions could not be analytically separated.

Whilst the influence of elevation on tree community composition and distribution was systematically identified and statistically demonstrated, the tree communities in Bwindi do not form distinct zones along the elevation gradient. There are multiple possible reasons for lack of elevation zoning. We highlight two: Bwindi is one of the few forests in East Africa where lowland and highland forests are in a continuum (Hamilton 1974, 1975). In a rather limited area, there is intermingling of lowland tree species that reach their upper boundary and highland tree species at their lower boundary. Our results showed an overlap of the cluster confidence ellipses on all the 3-dimensions of the ordination diagram and numerous plots and species positioned where the ellipses intersect indicating that they can reasonably belong to more than one community (Townsend and Walsh 2001; Kent 2012). Secondly, lack of perfect indicator species (having indicator value ≤0.5) implies that numerous tree species are spread widely but are most frequent and abundant within the tree community they indicate, presumably representing their ecological optimum (Mueller-Dombois and Ellenberg 2002; Austin 2013). The continuous nature of change of plant species in moist forest communities is related to a continuous change in environmental variables (i.e., precipitation and temperature) along the elevation gradient (Lovett and Lindberg 1993). Also, the high human disturbance of the past that involved mainly large-scale species-selective cutting and removal of large hardwoods from the forest could have relaxed the constraints imposed by competition, making it possible for some tree species to colonize the disturbed areas resulting in extension of their elevation range limits (Sheil 2016; Muñoz Mazon et al. 2019). Under varying intensities of heavy pit-sawing and over-exploitation, different successional cycles may have occurred in different parts of the forest (Chapman and Chapman 2004), masking the natural vegetation patterns that may have existed.

We show that montane vegetation is the result of a complex interaction between elevation, space, topography, climate as well as past and recent human and natural disturbance factors and it is very difficult to fully explain the response of the vegetation to a limited set of environmental characteristics. However, we establish that the complex montane forest vegetation in Bwindi can be represented as six characteristic tree communities that are in continuum. We found the spatial pattern of these communities across the landscape were largely predictable mainly from the interaction of elevation, temperature and precipitation variables, and this information was used to map montane vegetation. Our results demonstrate that the Random Forests modelling procedure is capable of mapping the tree species assemblages that accurately reflect the true nature of composition across complex topographic gradients with heavily disturbed vegetation. Such classiﬁed, community-level, patch-scale maps of vegetation have long been the foundation of natural resources management and the science of landscape ecology (Evans and Cushman 2009). The distinction among tree communities along an elevation gradient highlights the importance of Bwindi being protected along its entire span from lower elevation, sub-montane to montane forest habitats as it would conserve some species of conservation concern that may be restricted to narrow elevation ranges (Lovett et al. 2001; Hall et al. 2009). We found five indicator species that are classified as vulnerable in the IUCN Red List criteria of threat (Kalema and Hamilton 2020), to be most abundant and frequent in three tree communities. Also, it is highly likely that climate change is going to alter the habitat and composition of the vegetation and animal communities (Ayebare et al. 2018; Wright et al. 2022). Strong protection of contiguous vegetation communities along the entire elevation gradient in Bwindi will ensure that species can migrate to locations that are still suitable as the climate changes.

Acknowledgements

This study was supported by the Institute of Tropical Forest of Mbarara University of Science and Technology, Wildlife Conservation Society, the International Foundation for Science, the Mohammed bin Zayed Species Conservation Fund, the British Ecological Society, and the University of Massachusetts, Amherst.

Author contributions

DB designed the study and collected the data in the field; DB, JF, and CS collaborated on the data analyses and writing; TK and DS reviewed and edited the manuscript; all authors discussed and commented on the manuscript.

Data availability

Data are available on request from Senior Author D. Babaasa.

Competing interests

The authors have no relevant financial or non-financial interests to disclose

Albrechet H (1976) Chimpanzees in Uganda. Oryx 13:357–361. https://doi.org/10.1017/S0030605300014083
Atlas of Uganda (1967) 2nd Edition. Lands and Survey Department, Kampala, Uganda
Austin MP (2013) Vegetation and environment: discontinuities and continuities. In: van der Maarel E, Franklin J (eds) Vegetation ecology, 2nd edn. Wiley-Blackwell, New York, pp 71–106
Ayebare S, Plumptre AJ, Kujirakwinja D, Segan D (2018) Conservation of the endemic species of the Albertine Rift under future climate change. Biol Conserv 220:67–75. https://doi.org/10.1016/j.biocon.2018.02.001
Babaasa D (2000) Habitat selection by elephants in Bwindi Impenetrable National Park, Uganda. Afr J Ecol 38:116–122. http://ir.must.ac.ug/xmlui/handle/123456789/1683
Babaasa D, Eilu G, Kasangaki A, Bitariho R, McNeilage A (2004) Gap characteristics and regeneration in Bwindi Impenetrable National Park, Uganda. Afr J Ecol 42:217–224. https://doi.org/10.1111/j.1365-2028.2004.00519.x
Balmford A, Moore JL, Brooks T, Burgess N, Hansen LA, Williams P, Rahbek C (2001) Conservation conflicts across Africa. Science 291:2616–2619. https://www.science.org/doi/ 10.1126/science.291.5513.2616
Barr R, Arrea IB, Asuma S, Masozera AB, Gray M (2015) Pave the Impenetrable? An economic analysis of potential Ikumba-Ruhija road alternatives in and around Uganda’s Bwindi Impenetrable National Park. Conservation Strategy Fund Technical Series No. 35. Sebastopol, CA
Breiman L (2001) Random forests. Mach Learn 45:5–32. https://doi.org/10.1023/A:1010933404324
Brinkmann K, Patzelt A, Schlecht E, Buerkert A (2011) Use of environmental predictors for vegetation mapping in semi-arid mountain rangelands and the determination of conservation hotspots. Appl Veg Sci 14:17–30. https://doi.org/10.1111/j.1654-109X.2010.01097.x
Brooks TM, Mittermeier RA, Mittermeier CG, da Fonseca GAB, Rylands AB, Konstant WR, Flick P, Pilgrim J, Oldfield A, Magin G, Hilton-Taylor C (2002) Habitat loss and extinction in the hotspots of biodiversity. Conserv Biol 16:909–923. https://doi.org/10.1046/j.1523-1739.2002.00530.x
Burgess ND, Balmford A, Cordeiro NJ, Fjeldsa J, Kuper W, Sanderson E, Scharlemann JPW, Sommer JH, Williams PH, Rahbek C (2007) Correlations between biodiversity value, human density and human infrastructure across the high biodiversity tropical mountains of Africa. Biol Conserv 134:164–177. https://doi.org/10.1016/j.biocon.2006.08.024
Butynski TM (1984) Ecological survey of the Impenetrable Forest (Bwindi) forest Uganda, and recommendations for its conservation and management. Unpublished report to the Uganda Government. New York Zoological Society, New York
Butynski TM (1985) Primates and their conservation in the Impenetrable (Bwindi) Forest, Uganda. Prim Conserv 6:68–72
Cahusac AB (1961) Impenetrable C.F.R. Forest Types (Interpretation Report 6). Reference Map No. KI/45 Appendix G. GJ Leggat GJ, Osmaston HA Working Plan for the Impenetrable CFR 1961–1971. Uganda Forest Department, Entebbe
Carr JA, Outhwaite WE, Goodman GL, Oldfield TEE, Foden WB (2013) Vital but vulnerable: Climate change vulnerability and human use of wildlife in Africa’s Albertine Rift. Occasional Paper of the IUCN Species Survival Commission No. 48. IUCN, Gland, Switzerland and Cambridge, UK
Chapman CA, Chapman LJ (2004) Unfavourable successional pathways and conservation value of logged tropical forest. Biodivers Conserv 13:2089–2105. https://doi.org/10.1023/B:BIOC.0000040002.54280.41
Cordeiro NJ, Burgess ND, Dovie DB, Kaplin BA, Plumptre AJ, Marrs R (2007) Conservation in areas of high population density in sub-Saharan Africa, Editorial. Biol Conserv 134:155–163. https://doi.org/10.1016/j.biocon.2006.08.023
Cushman SA, Wallin DO (2002) Separating the effects of environmental, spatial and disturbance factors on forest community structure in the Russian Far East. For Ecol Manag 168:201–215. https://doi.org/10.1016/S0378-1127(01)00744-7
Cutler DR, Edwards TC Jr, Beard KH, Cutler A, Hess KT, Gibson J, Lawler J (2007) Random forests for classification in ecology. Ecol 88:2783–2792. https://doi.org/10.1890/07-0539.1
Davenport T, Howard P, Matthews R (1996) Bwindi Impenetrable National Park Biodiversity Report. Forest Department, Kampala
Dias D, Elias RB, Nunes V (2004) Vegetation mapping and nature conservation: a case study in Terceira Island (Azores). Biodivers Conserv 13:1519–1539. https://doi.org/10.1023/B:BIOC.0000021326.50170.66
Eggeling WJ (1947) Observations on the ecology of the Budongo rain forest, Uganda. J Ecol 34:20–87. https://doi.org/10.2307/2256760
Eilu G, Hafashimana DL, Kasenene JM (2004) Density and species diversity of trees in four tropical forests of the Albertine rift, western Uganda. Divers Distrib 10:303–312. https://doi.org/10.1111/j.1366-9516.2004.00089.x
Evans JS, Cushman SA (2009) Gradient modeling of conifer species using random forests. Landsc Ecol 24:673–683. https://doi.org/10.1007/s10980-009-9341-0
Fjeldsa J (2007) How broad-scale studies of patterns and processes can serve to guide conservation planning in Africa. Conserv Biol 21:659–667. https://doi.org/10.1111/j.1523-1739.2007.00706.x
Franklin J (1995) Predictive vegetation mapping: geographic modelling of biospatial patterns in relation to environmental gradients. Prog Phys Geogr 19:474–499. https://doi.org/10.1177/030913339501900403
Ghazoul J, Sheil D (2010) Tropical rain forest ecology, diversity and conservation. Oxford University Press, New York
Guschanski K, Caillaud D, Robbins MM, Vigilant L (2008) Females shape the genetic structure of a gorilla population. Curr Biol 18:1809–1814. https://doi.org/10.1016/j.cub.2008.10.031
Hall J (1991) Multiple-nearest-tree sampling in an ecological survey of Afromontane catchments forests. For Ecol Manag 42:245–266. https://doi.org/10.1016/0378-1127(91)90028-T
Hall J, Burgess ND, Lovett J, Mbilinyi B, Gereau RE (2009) Conservation implications of deforestation across an elevational gradient in the Eastern Arc Mountains, Tanzania. Biol Conserv 142:2510–2521. https://doi.org/10.1016/j.biocon.2009.05.028
Hamilton AC (1969) The vegetation of south-west Kigezi. Uganda J 33:175–199
Hamilton AC (1974) Distribution patterns of forest trees in Uganda and their historical significance. Vegetatio 29:29–32. https://www.jstor.org/stable/20036795
Hamilton AC (1975) A quantitative analysis of altitudinal zonation in Uganda forests. Vegetatio 30:99–106. https://doi.org/10.1007/BF02389611
Hamilton AC (1982) Environmental history of East Africa: A Study of the Quaternary. Academic Press, New York
Hamilton AC (1989) Tropical rain forest ecosystems. In: Leith H, Werger MJ (eds) Ecosystems of the world, vol 14B. Elsevier, Amsterdam, pp 155–182
Harcourt A (1981) Can Uganda's gorillas survive? A survey of the Bwindi Forest Reserve. Biol Conserv 19:269–282. https://doi.org/10.1016/0006-3207(81)90003-3
Hickey JR, Uzabaho E, Akantorana M, Arinaitwe J, Bakebwa I, Bitariho R, Eckardt W, Gilardi KV, Katutu J, Kayijamahe C, Kierepka EM, Mugabukomeye B, Musema A, Mutabaazi H, Robbins MM, Sacks BN, Zikusoka GK (2019) Bwindi Sarambwe 2018. Surveys: monitoring mountain gorillas, other select mammals, and human activities. GVTC, IGCP and partners, Kampala, Uganda
Howard PC (1991) Nature conservation in Uganda’s tropical forest reserves. IUCN, Gland, Switzerland
Hoersch B, Braun G, Schmidt U (2002) Relation between landform and vegetation in alpine regions of Wallis, Switzerland: A multiscale remote sensing and GIS approach. Comput Environ Urban Syst 26:113–139. https://doi.org/10.1016/S0198-9715(01)00039-4
Iverson LR, Prasad AM, Liaw A (2004) New machine learning tools for predictive vegetation mapping after climate change: Bagging and Random Forest perform better than Regression Tree Analysis. In: Smithers R (ed) Landscape ecology of trees and forests: Proceedings of the twelfth annual IALE (UK) conference; 2004 June 21–24; Cirencester, UK. International Association for Landscape Ecology, pp 317–320
Kalema J, Hamilton A (2020) Field guide to the forest trees of Uganda. CABI Oxfordshire
Kent M (2012) Vegetation description and data analysis: a practical approach, 2nd edn. Wiley-Blackwell, UK
Klein C, Vilcko F (2006) A new empirical approach for estimation in k-tree sampling. For Ecol Manag 237:522–533. https://doi.org/10.1016/j.foreco.2006.09.072
Lance GN, Williams WT (1966) A generalised sorting strategy for computer classification. Nature 211:218. https://doi.org/10.1038/212218a0
Legendre P, Legendre L (1998) Numerical Ecology, 2nd edn. Elsevier Sciences BV, Amsterdam, The Netherlands
Leggat GJ, Osmaston HA (1961) Working Plan for the Impenetrable Central Forest Reserve, Kigezi District, western Uganda, 1961–71. Uganda Forest Department, Entebbe
Liaw A, Wiener M (2002) Classification and regression by Random Forests. R News 2/3:18–22
Lieberman D, Lieberman M, Peralta R, Hartshorn GS (1996) Tropical forest structure and composition on a large-scale altitudinal gradient in Costa Rica. J Ecol 84:137–152. https://doi.org/10.2307/2261350
Lind EM, Morrison MES (1974) East African Vegetation. Longman, London
Lovett JC, Clarke GP, Moore R, Morrey GH (2001) Elevational distribution of restricted range forest tree taxa in eastern Tanzania. Biodivers Conserv 10:541–550. https://doi.org/10.1023/A:1016610526242
Lovett GM, Lindberg SE (1993) Atmospheric deposition and canopy interactions of nitrogen in forests. Can J For Res 23:1603–1616. https://doi.org/10.1139/x93-200
Macfie EJ, Williamson EA (2010) Best practice guidelines for Great Ape tourism. IUCN/SSC Primate Specialist Group (PSG), Gland, Switzerland
McCune B, Grace JB (2002) Analysis of ecological communities. MjM Software Design, Gleneden Beach, OR, US
McDonald DJ, Boucher C (1999) Towards mapping the fynbos for the revised vegetation map of South Africa. In: Timberlake J, Kativu S (eds) African Plants: Biodiversity, Taxonomy and Uses. Royal Botanic Gardens, Kew, pp 195–206
McNeilage A, Plumptre A, Brock-Doyle A, Vedder A (1998) Bwindi Impenetrable National Park, Uganda Gorilla and Large Mammal Census. WCS Working Paper No. 14, Wildlife Conservation Society, New York, US
McNeilage A, Plumptre AJ, Brock-Doyle A, Vedder A (2001) Bwindi Impenetrable National Park, Uganda: gorilla census, 1997. Oryx 35:39–47. https://doi:10.1046/j.1365-3008.2001.00154.x
McNeilage A, Robbins MM, Gray M, Olupot W, Babaasa D, Bitariho R, Kasangaki A, Rainer H, Asuma S, Mugiri G, Baker J (2006) Census of the mountain gorilla Gorilla beringei beringei population in Bwindi Impenetrable National Park, Uganda. Oryx 40:419–427. https://doi:10.1017/S0030605306001311
Mittermeier RA, Robles Gil P, Hoffmann M, Pilgrim J, Brooks T, Mittermeier CG, Lamoreux J, Da Fonseca GA (2004) Hotspots revisited. CEMEX, Mexico City, MX
Mueller-Dombois D, Ellenberg H (2002) Aims and methods of vegetation ecology. The Blackburn Press, Caldwell, NJ, USA
Muñoz Mazón M, Klanderud K, Finegan B, Veintimilla D, Bermeo D, Murrieta E, Delgado D, Sheil D (2019) Disturbance and the elevation ranges of woody plant species in the mountains of Costa Rica. Ecol Evol 9:14330–14340. https://doi.org/10.1002/ece3.5870
Olupot W, Barigyira R, Chapman CA (2009) The status of anthropogenic threat at the people-park interface of Bwindi Impenetrable National Park, Uganda. Environ Conserv 36:41–50. https://doi:10.1017/S0376892909005347
Parmesan C (2006) Ecological and evolutionary responses to recent climate change. Annu Rev Ecol Evol 37:637–669. https://www.jstor.org/stable/30033846
Pfeffer K, Pebesma EJ, Burrough PA (2003) Mapping alpine vegetation observations and topographic attributes. Landsc Ecol 18:759–776. https://doi.org/10.1023/B:LAND.0000014471.78787.d0
Plumptre AJ, Davenport TRB, Behangana M, Kityo R, Eilu G, Ssegawa P, Ewango C, Meirte D, Kahindo C, Herremans M, Peterhans JK, Pilgrim JD, Wilson M, Languy M, Moyer D (2007) The biodiversity of the Albertine Rift. Biol Conserv 134:178–194. https://doi.org/10.1016/j.biocon.2006.08.021
Prasad AM, Iverson LR, Liaw A (2006) Newer classification and regression tree techniques: bagging and Random Forests for ecological prediction. Ecosystems 9:181–199. https://doi.org/10.1007/s10021-005-0054-1
R Core Team (2022) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/
Richards PW (1996) The tropical rain forest: An ecological study. Cambridge University Press, Cambridge, U.K
Sala OE, Stuart Chapin F III, Armesto JJ, Berlow E, Bloomfield J, Dirzo R, HuberSanwald E, Huenneke LF, Jackson RB, Kinzig A, Leemans R, Lodge DM, Mooney HA, Oesterheld M, Poff NL, Sykes MT, Walker BH, Walker M, Wall DH (2000) Global biodiversity scenarios for the year 2100. Science 287:1770–1774. https://doi.org/10.1126/science.287.5459.1770
Schmitt CB, Denich M, Demissew S, Friis I, Boechmer HJ (2010) Floristic diversity in fragmented Afromontane rainforests: altitudinal variation and conservation importance. Appl Veg Sci 13:291–304
https://doi.org/10.1111/j.1654-109X.2009.01067.x
Sheil D (2012) The impenetrable challenge of an overwhelming understorey. Bull Br Ecol Soc 43:45–47
Sheil D (2016) Disturbance and distributions: avoiding exclusion in a warming world. Ecol Soc 21:10. http://dx.doi.org/10.5751/ES-07920-210110
Sheil D, Ducey MJ, Sidiyasa K, Samsoedin I (2003) A new type of sample unit for the efficient assessment of diverse tree communities in complex forest landscapes. J Trop For Sci 15:117–135. https://www.jstor.org/stable/23616330
Slik JF, Raes N, Aiba SI, Brearley FQ, Cannon CH, Meijaard E, Nagamasu H, Nilus R, Paoli G, Poulsen AD, Sheil D (2009) Environmental correlates for tropical tree diversity and distribution patterns in Borneo. Divers Distrib 15:523–532. https://doi.org/10.1111/j.1472-4642.2009.00557.x
Ssali F, Moe SR, Sheil D (2017a) A first look at the impediments to forest recovery in bracken-dominated clearings in the African Highlands. For Ecol Manag 402:166–176. https://doi.org/10.1016/j.foreco.2017.07.050
Ssali F, Moe SR, Sheil D (2017b) Tree seed rain and seed removal, but not the seed bank, impede forest recovery in bracken (Pteridium aquilinum (L.) Kuhn)- dominated clearings in the African highlands. Ecol Evol 8:4224–4236. https://doi.org/10.1002/ece3.3944
Ssali F, Sheil D, Nkurunungi JB (2012) How selective are elephants as agents of forest tree damage in Bwindi Impenetrable National Park, Uganda? Afr J Ecol 51:55–65. https://doi.org/10.1111/aje.12006
ter Braak CJF (1995) Ordination. In: Jongman RHG, ter Braak CJF, van Tongeren OFR (eds) Data Analysis in Community and Landscape Ecology. Cambridge University Press, Cambridge, pp 91–173
Townsend P, Walsh S (2001) Remote sensing of forested wetlands: application of multitemporal and multispectral satellite imagery to determine plant community composition and structure in southeastern USA. Plant Ecol 157:129–149. https://doi.org/10.1023/A:1013999513172
Twongyirwe R, Majaliwa JGM, Ebanyat P, Tenywa MM, Sheil D, Heist MV, Oluka M, Kumar L (2011) Dynamics of forest cover conversion in and around Bwindi Impenetrable Forest, Southwestern Uganda. J Appl Sci Environ Manag 15:189–195. https://doi.org/10.4314/jasem.v15i1.68439
White F (1983) The Vegetation of Africa: a descriptive memoir to accompany the Unesco/AETFAT/UNSO vegetation map of Africa. Unesco, Paris
Wright E, Eckardt W, Refisch J, Bitariho R, Grueter CC, Ganas-Swaray J, Stoinski TS, Robbins MM (2022) Higher maximum temperature increases the frequency of water drinking in mountain gorillas (Gorilla beringei beringei). Front Conserv Sci 3:738820. https://doi.org/10.3389/fcosc2022.738820
Volkov I, Banavar JR, Hubbell SP, Maritan A (2003) Neutral theory and relative species abundance in ecology. Nature 424:1035–1037. https://doi.org/10.1038/nature01883
Zimmermann NE, Kienast F (1999) Predictive mapping of alpine grasslands in Switzerland: species versus community approach. J Veg Sci 10:469–482. https://doi.org/10.2307/3237182

Table 1 Sampling design for trees (≥20cm dbh) for floristic gradient modeling in Bwindi Impenetrable National Park, Uganda.

Stratum	Transect elevation range (m a.s.l.)	No. of sample points
Stratum	Transect elevation range (m a.s.l.)	Valley	Hillside	Ridge	Gully	Total
Central	2,014 – 2,237	8	31	16	1	56
East	2,020 – 2,467	1	59	2	1	63
South	1,479 – 2,342	3	50	11	2	66
North	1,320 – 1,732	3	48	6	1	58
West	1,479 – 2,124	2	31	10	3	46
All	1,320 - 2,467	17	219	45	8	289

Table 2 Potential explanatory variable names and descriptions for floristic gradient modeling in Bwindi Impenetrable National Park, Uganda

Biophysical variable name	Code in the model	Retained after collinearity analysis?	Reference
BIO1 = Annual Mean Temperature	temp_mean	No	Fick and Hijmans 2017
BIO2 = Mean Diurnal Range (Mean of monthly (max temp-min temp))	temp_mdr	No	Fick and Hijmans 2017
BIO3 = Isothermality (bio02/bio07)(*100)	temp_iso	Yes	Fick and Hijmans 2017
BIO4 = Temperature Seasonality (standard deviation*100)	temp_seas	Yes	Fick and Hijmans 2017
BIO5 = Max Temperature of Warmest Month	temp_max	No	Fick and Hijmans 2017
BIO6 = Min Temperature of Coldest Month	temp_min	No	Fick and Hijmans 2017
BIO7 = Temperature Annual Range (bio05-bio06)	temp_rang	No	Fick and Hijmans 2017
BIO8 = Mean Temperature of Wettest Quarter	temp_wetq	No	Fick and Hijmans 2017
BIO9 = Mean Temperature of Driest Quarter	temp_drq	No	Fick and Hijmans 2017
BIO10 = Mean Temperature of Warmest Quarter	temp_waq	No	Fick and Hijmans 2017
BIO11 = Mean Temperature of Coldest Quarter	temp_coq	No	Fick and Hijmans 2017
BIO12 = Annual Precipitation	prec_ann	Yes	Fick and Hijmans 2017
BIO13 = Precipitation of Wettest Month	prec_max	No	Fick and Hijmans 2017
BIO14 = Precipitation of Driest Month	prec_min	No	Fick and Hijmans 2017
BIO15 = Precipitation Seasonality (Coefficient of Variation)	prec_seas	Yes	Fick and Hijmans 2017
BIO16 = Precipitation of Wettest Quarter	prec_wetq	No	Fick and Hijmans 2017
BIO17 = Precipitation of Driest Quarter	prec_drq	No	Fick and Hijmans 2017
BIO18 = Precipitation of Warmest Quarter	prec_waq	No	Fick and Hijmans 2017
BIO19 = Precipitation of Coldest Quarter	prec_coq	Yes	Fick and Hijmans 2017
30 m Digital Elevation Model	elevation	Yes	NASA and METI 2011
Human disturbance	disturbance	Yes	McNeilage et al. 1998
Steepness of the slope	slope	Yes	Field measurement
Aspect	aspect	Yes	Field measurement
Ridge tops	ridge	Yes	Field measurement
Valley	valley	No	Field measurement
Gully	gully	No	Field measurement
Hillside	hillside	No	Field measurement
Latitude	latitude	No	Field measurement
Longitude	longitude	Yes	Field measurement

Table 3 Indicator tree species for the cluster groups in Bwindi Impenetrable National Park, Uganda. The cluster number corresponds to those in Tables 3 and 4, Figures 2, 4a, 4b, 5a, 5b and 6. Early-successional (pioneer) species and mid-successional or non-pioneer light demanders (NPLDs) species follow Kalema and Hamilton (2020)

Cluster number	No. of sites	Elevation range; mean elevation (m a.s.l.); and location of cluster group	Indicator tree species	Successional guild	Indicator value	p-value
1	58	1,372 – 2,350; 1,956 Mid-elevation transitional to the west of the park	Strombosia scheffleri	NPLD	0.3108	0.001
			Cassipourea congoensis		0.1535	0.002
			Entandrophragma excelsum (VU)	NPLD	0.1135	0.003
			E. cylindricum (VU)	NPLD	0.0944	0.004
			Trilepisium madagascariense		0.0924	0.012
			Leplaea mayombensis (VU)		0.0623	0.043
2	35	1,778 – 2,407; 2,195 Submontane forest on ridge-tops in the centre-south of the park	Prunus africana (VU)		0.2174	0.001
			Alangium chinense Harms	Pioneer	0.4070	0.001
			Neoboutonia macrocalyx	Pioneer	0.2706	0.001
			Croton macrostachyus	NPLD	0.2669	0.001
			Maesa lanceolata	Pioneer	0.2004	0.001
			Dombeya torrida	Pioneer	0.1882	0.001
			Allophylus ferrugineus		0.1643	0.001
			Bersama abyssinica	NPLD	0.1358	0.001
			Tabernaemontana pachysiphon		0.1358	0.001
			Zanthoxylum rubescens		0.1329	0.001
			Myriathus holstii	NPLD	0.0895	0.025
			Croton megalocarpus		0.0749	0.019
			Zanthoxylum gilletti		0.0677	0.037
			Casearia battiscombei		0.0665	0.007
3	45	1,479 – 2,217; 2023 Mid-elevation forest in the centre-south of the park	Ficalhoa laurifolia	Pioneer	0.4209	0.001
			Syzygium guineense	Pioneer	0.3842	0.001
			Symphonia globulifera		0.1714	0.002
			Ilex mitis		0.1323	0.006
			Albizia gummifera		0.0661	0.050
			Synsepalum brevipes		0.0654	0.013
			Syzygium cordatum		0.0520	0.042
4	33	1,773 – 2,311; 2,188 Sub-montane forest in the centre-south of the park	Gambeya albida	NPLD	0.6552	0.001
			Cassipourea gummiflua		0.2547	0.001
			Drypetes gerrardii		0.2547	0.001
			Newtonia buchananii	NPLD	0.1280	0.005
5	75	1,320 – 2,131; 1,696 Low-elevation forest in the north, forest corridor and west of the park	Leptonychia mildbraedii		0.3913	0.001
			Shirakiopsis elliptica	Pioneer	0.3170	0.001
			Parinari excelsa		0.3152	0.001
			Carapa grandiflora	NPLD	0.1443	0.003
			Funtumia africana		0.1366	0.001
			Maesopsis eminii	Pioneer	0.1253	0.002
			Macaranga barteri	Pioneer	0.1233	0.001
			Ocotea kenyensis (VU)		0.0875	0.016
			Tetrorchidium didymostemon		0.0751	0.022
			Allanblackia kimbiliensis		0.0674	0.024
			Trichilia dregeana		0.0667	0.010
			Millettia dura	Pioneer	0.0586	0.047
			Strombosiopsis tetrandra		0.0569	0.044
			Celtis philippensis		0.0533	0.031
6	43	1,416 – 2,467; 2,203 High elevation montane forest to the southeast of the park	Olinia rochetiana		0.3882	0.001
			Macaranga capensis	Pioneer	0.2806	0.001
			Agarista salicifolia		0.2791	0.001
			Rapenea melanophloeos		0.2291	0.001
			Nuxia congesta		0.2208	0.001
			Pittosporum viridiflorum	Pioneer	0.2046	0.001
			Faurea wentzeliana		0.2032	0.001
			Olea capensis	NPLD	0.1992	0.001
			Xymalos monospora		0.1311	0.006
			Polyscias fulva	Pioneer	0.1259	0.005
			Podocarpus latifolius		0.1092	0.005
			Harungana madagascariensis	Pioneer	0.0854	0.024
			Ekebergia capensis		0.0806	0.015
			Vepris nobilis		0.0783	0.031
			Morella salicifolia		0.0639	0.032
			Mystroxylon aethiopicum		0.0604	0.036
			Anthocleista vogelii		0.0560	0.022
			Pauridiantha callicarpoides		0.0554	0.033

Table 4 Confusion matrix of OOB estimate of error rate for the predictive tree assemblage spatial distribution pattern map of Bwindi Impenetrable National Park, Uganda. The cluster number corresponds to those in Table 3, Figures 2, 4a, 4b, 5a, 5b and 6

Cluster group	1	2	3	4	5	6		Class error %
1	14	6	9	7	16	5		75.4
2	3	15	6	4	2	5		57.1
3	5	8	20	5	6	0		54.4
4	3	8	5	16	1	0		51.5
5	10	1	5	2	50	4		30.5
6	2	4	1	2	8	24		41.5
OOB estimate of error rate							50.7

No competing interests reported.

Predictive mapping of tree species assemblages in an African montane rainforest

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Results

Discussion

Conclusion

Declarations

References

Tables

Additional Declarations

Status:

Journal Publication

Version 1