Niche dynamics of Memecylon in Sri Lanka: Distribution patterns, climate change effects, and conservation priorities

Abstract Recent climate projections have shown that the distribution of organisms in island biotas is highly affected by climate change. Here, we present the result of the analysis of niche dynamics of a plant group, Memecylon, in Sri Lanka, an island, using species occurrences and climate data. We aim to determine which climate variables explain current distribution, model how climate change impacts the availability of suitable habitat for Memecylon, and determine conservation priority areas for Sri Lankan Memecylon. We used georeferenced occurrence data of Sri Lankan Memecylon to develop ecological niche models and assess both current and future potential distributions under six climate change scenarios in 2041–2060 and 2061–2080. We also overlaid land cover and protected area maps and performed a gap analysis to understand the impacts of land‐cover changes on Memecylon distributions and propose new areas for conservation. Differences among suitable habitats of Memecylon were found to be related to patterns of endemism. Under varying future climate scenarios, endemic groups were predicted to experience habitat shifts, gains, or losses. The narrow endemic Memecylon restricted to the montane zone were predicted to be the most impacted by climate change. Projections also indicated that changes in species’ habitats can be expected as early as 2041–2060. Gap analysis showed that while narrow endemic categories are considerably protected as demonstrated by their overlap with protected areas, more conservation efforts in Sri Lankan forests containing wide endemic and nonendemic Memecylon are needed. This research helped clarify general patterns of responses of Sri Lankan Memecylon to global climate change. Data from this study are useful for designing measures aimed at filling the gaps in forest conservation on this island.


| INTRODUC TI ON
Island ecosystems have received much attention in climate change research because they are considered to be among the most vulnerable (Harter et al., 2015;Leclerc et al., 2020;Taylor & Kumar, 2016).
Such island vulnerability can be results of extreme weather events that lead to the displacement of suitable habitats of species and the island conditions (e.g., size and restriction from the surrounding ocean) that limit organisms' response to climate change (Harter et al., 2015;Veron et al., 2019). Therefore, understanding how climate change and human-induced habitat loss will influence the risk of species extinction is critical for informing environmental policies (Cooper et al., 2011;Sinervo et al., 2010) on these islands.
Occurrence data on species distribution and high-resolution spatial data on climate can be integrated to predict climatic dimensions of a species niche, which is known as Ecological Niche Modeling (ENM) (Hijmans et al., 2005;Peterson, 2011). These models use a growing number of quantitative approaches to approximate fundamental niches of species in relation to temperature, precipitation, and other associated climate and topographic variables (Elith & Leathwick, 2009;Randin et al., 2009), and these ENMs can be used to develop conservation strategies (Andrade-Díaz et al., 2019;Carroll, 2010). Using these tools, we examined ENMs of a woody plant group, Memecylon L. in Melastomataceae (Figure 1), on a continental island, Sri Lanka, and predicted its response to future climate scenarios in order to inform conservation priorities.

| Climate, vegetation, and land use of Sri Lanka
The climate of Sri Lanka is dominated by two monsoon seasons known as southwest and northeast monsoons (Bonnefille et al., 1999). Based on the distribution of mean annual rainfall, three major precipitation zones have been recognized (  (Ashton et al., 1997). Additionally, two small areas at the extreme northwest and southeast of the island have arid climates. Sri Lanka's near-equator position provides a tropical climate with a mean annual temperature ranging from 15°C in high to 28°C in low elevations (Silva & Sonnadara, 2016).
The varied climates in Sri Lanka have resulted in different ecosystems that harbor a wide range of floristic diversity (Ashton et al., 1997). For example, the wet zone contains tropical rainforests, with broad-leaved trees and lianas; the intermediate zone contains broadleaf forest with an undergrowth of shrubs; dry zone forests consist of shrubs with deciduous leaves; and the arid zone contains open scrubland with scattered trees (Erdelen, 1988). A significant feature of plant diversity in Sri Lanka is the remarkably endemic angiosperm flora with approximately 863 endemic species out of circa 3087 (National Red List, 2020). Although the close biotic affinities between Sri Lanka and India have been identified (Pethiyagoda & Sudasinghe, 2017), there is also evidence for a distinct biotic component uniquely assembled on the island (Bossuyt, 2004).
Similar to other tropical forests around the world, those in Sri Lanka are threatened by the impacts of land-use changes (Dissanayake, 2020;Dissanayake et al., 2019). Yet, the ecological consequences of these anthropogenic changes are understudied relative to other forests of the New World and Old World Tropics (Gopal, 2013). Approximately, one-third of the total land area in Sri Lanka is used for agriculture, and locations that are in close proximity to district capitals are impacted due to increasing population pressure and urbanization. The reduction of forest cover has accelerated due to infrastructure enhancement, economic reforms, and population redistribution (Rathnayake et al., 2020). However, limited data on vegetation cover to compare with recent land-use information impede effective environmental management and planning on this island.

| Memecylon in Sri Lanka
There are 32 Memecylon species in Sri Lanka, and of them, 25 are reported as endemic (Bremer, 1988). Sri Lankan Memecylon is dis- Due to its diversity across the island and high regional endemism, this plant group represents an ideal model to investigate niche differentiation and the possible impact of climate change on vegetation in different climate zones. Currently, Sri Lanka has set aside a considerable proportion of land for conservation (UNEP-WCMC & IUCN, 2020). However, it is important to understand if these conserved areas capture diverse habitats of rare and endemic species.
Memecylon, which has a relatively large number of species characterized by high endemism, is suitable as a model system to investigate whether the current conservation measures adequately capture the diversity of areas and estimate the potential loss of species' suitable areas under varying climate change scenarios.
To date, research on Memecylon has mainly focused on its taxonomy, ethnobotany, and evolution (e.g., Amarasinghe, Joshi, et al., 2021;Bremer, 1988;Sivu et al., 2013) and studies to examine how climate change potentially impacts its distribution are completely lacking. In this study, we compiled occurrences of Memecylon from diverse sources and bioclimatic data associated with various climate scenarios to address three key questions: (1) What are the patterns of distribution of Sri Lankan Memecylon under current climate conditions? (2) How will climate change impact the availability of suitable habitats of Memecylon, and what will be the main driving factors of these changes? And (3) Which areas should be targeted for conservation as indicated by Memecylon distribution in the face of climate change in Sri Lanka? Endemic species are reported as highly vulnerable to climate change (Loarie et al., 2008;Manes et al., 2021;Thuiller et al., 2006) due to occupying specialized niches, limited dispersal capabilities, and reduced adaptive capacities when compared to nonendemic species (Chichorro et al., 2019;Staude et al., 2020). Further, organisms that have more restricted geographical ranges are at greater risk (Elsen & Tingley, 2015;Stubbs et al., 2018) compared to species with large geographical ranges, which may find refugia in parts of their range (Lucas et al., 2019). Therefore, we hypothesized that endemic Memecylon with restricted distribution ranges in Sri Lanka may be at more risk with climate change compared to nonendemics and species with large geographic ranges. Here, we tested the null hypothesis that endemic and nonendemic Memecylon in Sri Lanka with all types of geographical ranges are equally vulnerable.

| Occurrence data collection
The study area, Sri Lanka, is located between 5°55′-9°51′N latitude and 79°52′-81°51′E longitude (WGS 84-UTM Zone 44N) in the Indian Ocean and has a land extent of 65,525 km 2 (Rathnayake et al., 2020). It contains three elevation zones ( Figure 2) known as (1) lowland: up to 300 m above sea level; (2) upland: 300-1000 m; and (3) highland: >1000 m (Katupotha, 2013) where forests within them are broadly categorized as lowland and montane forests (Werner & Balasubramanium, 1992). Occurrence records of Memecylon from Sri Lanka were collected from the following herbaria: B, BM, BR, FLAS, K, L, M, MO, NY, PDA, SING, and US (acronyms: Thiers, 2020), GBIF, and published literature (Ekanayake et al., 2014;Gunathilaka, 2019;Madurapperuma et al., 2014;Medawatte et al., 2011). Upon review, F I G U R E 2 (a) Climate zones (b) elevation zones. Purple dots show Memecylon occurrences. Maroon circles show the field collection sites identifications of some herbarium specimens were found to be erroneous: ~15% of specimens in GBIF, ~13% deposited at the PDA, ~2% at the SING, ~5% at the US Herbaria. The likely reason for the misidentification of Memecylon specimens is due to many specimens being sterile because of seasonal and/or rarity of flowering events (Amarasinghe, Joshi, et al., 2021). Therefore, we corrected all misidentified specimens stored in the PDA, SING, and US herbaria by carefully studying the type specimens and taxonomic descriptions during the visits to these herbaria. Other herbaria stored mostly duplicates of specimens deposited at PDA; however, when nonduplicate specimens were found from the online databases of other herbaria, occurrence data were used only from correctly identified specimens. We also selected GBIF data points that were correctly identified, based on digitized specimens. Twenty-five specimens from fieldwork were deposited at PDA.
In total, 903 digitized herbarium specimens were georeferenced using GeoLocate, and locations were verified by cross-checking with Acme Mapper v2.2. (1991). All identical points (duplicate specimens which could not be detected manually), points without environmental data, and proximate data points (points that fall in the same raster cell, ~1 km 2 ) were removed using R packages spocc, scrubr, and spatstat (Baddeley et al., 2015;Chamberlain, 2020;Chamberlain et al., 2021) on R v3.3.1 (R Core Team, 2019). Finally, based on our knowledge of Memecylon, all occurrence data were visually examined in QGIS v3.3.3k to look for potential errors. Samples collected from sites that are likely to be visited more frequently (i.e., near roads, urban areas, and botanic gardens) may introduce bias because those occurrence points may not adequately capture the range of environmental conditions in which a species might occur (Rocchini & Garzon-Lopez, 2017).
We believe this bias is minimal in this study because specimens and field-collected data were primarily from locations within old-growth or secondary forests rather than readily accessible areas. Memecylon angustifolium Wight, Memecylon ellipticum Thwaites, Memecylon giganteum Alston, Memecylon leucanthemum Thwaites, Memecylon macrophyllum Thwaites, M. revolutum, and Memecylon wightii Thwaites, were removed due to insufficient sampling for model generation (fewer than 10 occurrences or prone to overfitting). We also removed two species (Memecylon gracilimum Alston and Memecylon macrocarpum Thwaites) in which occurrence data were found only from a single location. Using these filtering criteria, 21 Memecylon taxa were used for analyses (  (Fick & Hijmans, 2017). We established a buffer zone of 100 km around the occurrence data of each species separately using QGIS to generate calibration areas for the models (Barve et al., 2011). We then developed spatial analyses for each species using extents that included occurrence distributions and buffer area, as well as extents based on the entire island; these analyses were done with R packages maptools and mapproj (Bivand & Lewin-Koh, 2020;Brownrigg & Minka, 2020). Pairwise Pearson's correlation coefficients (r) for all bioclimatic variables within the species-specific buffer zones were estimated to avoid collinearity between them using R packages raster and rgdal (Hijmans, Etten, et al., 2021;Keitt et al., 2010); species-specific predictors were retained based on a threshold of |r| ≤ 0.65.
To model future climate scenarios, we used bioclimatic variables

| Statistical analysis
Ecological niche models were used to calculate the niche breadth (Connor et al., 2018). The measure of niche breadth derived by Levin's index shows the breadth of suitable climatic factors for a species at a 0-1 scale (Feinsinger et al., 1981). Here, values closer to 1 reflect generalist species with wide climatic tolerance, while values closer to 0 represent more specialized species (Feinsinger et al., 1981).
The models were converted into binary presence-absence maps with three threshold approaches: the minimal training presence threshold, the threshold that equalizes sensitivity and specificity, and the threshold that maximizes the sum of sensitivity and specificity of the binary maps using R packages scales (Wickham & Seidel, 2020

| Patterns in endemic categories
Sri Lankan Memecylon were classified into five categories based on endemism information (Bremer, 1988;Sivu et al., 2012;Subramanyam & Rao, 1949; The National Red List, 2020), area of suitable habitats, elevation, climate variables of the current distribution, and niche breadth; these categories were wide endemic, lowland narrow endemic, montane narrow endemic, dry zone nonendemic, and wet zone nonendemic. Endemic Memecylon that have >10,000 km 2 area of suitable habitats and >0.5 niche breadth were categorized as wide endemics. Memecylon were classified as being narrow endemics if their current geographic projection of ecological niche was <10,000 km 2 and niche breadth was <0.5.
This narrow endemic category was subclassified as montane narrow endemics (occurrence points are restricted to >300 m above sea level) and lowland narrow endemic (occurrence points are distributed in lowland, i.e., <300 m above sea level). Nonendemic Memecylon which contain suitable habitats within the wet zone of the island and showed a strong contribution of precipitationrelated bioclimatic variables were subclassified as wet zone nonendemics. Nonendemic Memecylon which contain suitable habitats within the dry zone and showed a strong contribution of temperature-related bioclimatic variables were subclassified as dry zone nonendemic. By the above-defined standards, we found four species to be wide endemics, eight as lowland narrow endemics, three as montane narrow endemics, and an additional three species each were categorized as wet-zone and dry-zone nonendemics, respectively.
In each endemic category, the threshold models of species were overlapped on each other and the overlapping area under the current condition was calculated using customized R scripts. These overlaps showed regions where niche conditions are suitable for two or more taxa. Threshold models of each taxon in 12 future models were also overlapped using the above procedure, and changes in the distribution of suitable habitat (km 2 ) for each species between current and future distributions were evaluated.

| Current suitable habitats
When niche space is projected into geographic space, Memecylon species tend to occupy distinct suitable habitats under current conditions. These ENMs, which approximate predicted fundamental niches of each species, showed different contributions of bioclimatic variables as indicated by area under the receiver operating characteristic curve (AUC: with and without the contributing variables) and permutation importance (Table 3). Mean temperature of the coldest quarter (bio11) is the variable, which contributed to models of most species (17 out of 21).
AUC scores of the resultant models are not reported in this study as they should be interpreted with caution because sampling bias can result in spatial clustering of points, which may affect model quality by inflating model accuracy (Veloz, 2009 since much of our analyses and comparisons were based on binary maps (i.e., threshold selected to convert mat to areas of suitable and not suitable habitat) and a threshold-dependent measure, like KAPPA, is more suitable for these maps. Of the three threshold approaches examined, we selected the threshold approach that equalized sensitivity and specificity because of higher overall model performance (see also Bean et al. (2012) and Shabani et al. (2018)). Performance scores of threshold ENMs (

| Future projections
Responses of Memecylon to different future climate scenarios are variable ( and losses) than BCC-CSM1-1 (Table 4). Interestingly, many areas currently unsuitable are predicted to become increasingly suitable for Memecylon, while some currently suitable regions will become unsuitable in the future (Figure 3 and Table 4). For instance, lowland narrow endemic, wet zone nonendemic, and wide endemic categories showed potential eastward habitat shifts where the species belonging to these categories are currently absent. The habitat changes explained above will occur as early as 2050. As expected, most species showed a greater percentage of habitat change in 2070 compared to 2050 (Table 4).

| Patterns in endemic categories
In all endemic categories, richness areas show habitats where all species overlap, based on the total number of species within a grid cell (e.g., richness area maps in Figure  Examining how climate change might impact richness maps of the various endemic categories, we found that for montane narrow endemics, there was a complete loss of suitable habitat even under the optimistic MIROC5 (RCP 2.6) model shown in Figure 3. For wide endemics, we found that there were reduced suitable habitats that captured areas of high species richness, especially in 2050 models.
For nonendemic categories, new areas that capture multiple species within those categories emerged, suggesting that environmental conditions improve for these species, assuming that they can disperse and track changes over time.

| Gap analysis
The overlay of the richness area map (the area that captures the maximum number of species within each category as shown in

| Response to climate change
The overall current distribution showed that about 60,016 km 2 of the total area in Sri Lanka is potentially suitable for Memecylon.
These suitable habitats cover all climate and elevation zones in Sri Lanka except the southeast dry zone (Figure 3,  Memecylon species occupying mountains will be especially affected by climate change. The unique climatic conditions in the mountains of Sri Lanka (Jayalal et al., 2017;Ruklani & Rubasinghe, 2021;Werner, 1995)   Under future climates, a significant reduction of richness areas within the current protected areas was observed for all endemic categories (Figures 4 and 5). Here, only the current suitability is F I G U R E 6 Areas recommended for conservation. (a) Conservation recommendation map: current richness areas (areas show habitats where the highest number of species overlap) of endemism categories, forest cover (source: Rathnayake et al., 2020), and protected areas maps (source: WDPA) are superimposed. The forests which require conservation are within the areas demarcated in magenta. These areas include both the purple areas within these demarcations represent high confidence richness areas (these are obtained from multiple model iterations and purple shows all model iterations for the highest number of species predict present) and richness area of dry-zone nonendemic Memecylon in the northern part of the island which is not within the high confidence richness areas. (b) Uncertainty map:

| Conservation prioritization
The uncertain richness areas (at least one model iteration for all species predicts presence) are shown in blue and the high confidence nonrichness areas (at least one species is predicted to be absent in all model iterations) are shown in white TA B L E 5 Gap analysis using protected area map, land-use map, and richness area (represent only those areas with a maximum number of overlapping species) maps Note: Richness areas = Richness area of the category. Extent within all protected = Extent of each category overlapped with all types of protected lands in the protected area map. Extent within conservation forest = Extent of each category overlapped with conservation forests in the protected area map. Extent within forests = Extent of each category within Forest in the land-cover map. N/A = not applicable due to total loss of overlapping areas.
Estimates are provided as area overlapping lands in each category and the extent of each category within protected lands and forests. Both protected area map and land-use map are used for current models, but only protected area map is used for future models.
considered for conservation recommendations because planning for future climate scenarios is problematic given uncertainty regarding models and policies that might mitigate (or not) climate change impacts. As the future predictions indicate habitat change of Memecylon in 2050, there is an urgent need to implement conservation management for vulnerable Memecylon categories. In particular, rare and endemic species of Sri Lankan Memecylon warrant conservation attention due to the predicted habitat loss inferred from this study. To address and mitigate these losses, various other conservation parameters, such as estimating the land cost, regional versus global conservation priorities, and conservation risks, should also be taken into account (Butt et al., 2020;Naidoo & Ricketts, 2006).
Moreover, as sample sizes were generally low for many species studied here, conservation planning and actions require further detailed spatial analyses to identify both problems and opportunities in a regional and local socio-ecological context.

| Future directions for studies of Memecylon ecological niches
We used only a subset of total Sri Lankan Memecylon as we eliminated species with few occurrence data and species found from a single national park. In addition, we identified a significant undiscovered diversity of Sri Lankan Memecylon during fieldwork. In future studies, this diversity should be taken into account. Also, low sample sizes and potential bias in sampling associated with inadequately capturing the environmental conditions in which the species occurs call for additional fieldwork and further spatial analyses.
Further, we used only 12 future climate models among all possible future scenarios (Hijmans et al., 2005). Moreover, examining these niche differences in the context of phylogenetic relationships may help us understand the factors that have led to the diversification of Memecylon within the island and to infer ancestral niches. However,

Sri Lankan Memecylon is not monophyletic; instead, it includes
Memecylon from India, Andaman, and the Seychelles (Amarasinghe, Joshi, et al., 2021). To understand the ancestral niches of Sri Lankan Memecylon, niche models should be generated from all these geographical regions in South Asia where Memecylon occurs and analyzed in a phylogenetic framework. However, incomplete information on occurrence data and identification errors of a large number of Memecylon specimens from the other South Asian regions impeded constructing niche models and understanding ancestral niches. Sri Lanka has a rich diversity of soils distributed across the island (Wimalasiri et al., 2020); however, information about the soil requirements of Memecylon is totally lacking. Therefore, future studies should also include soil data to understand the abiotic niches of

Memecylon.
Our findings will help clarify general patterns of woody plants occupying habitats in Sri Lanka and provide data to inform conservation strategies on this island. Given the expected significant changes in future suitable habitats of this plant group, the reduction of the area occupied by the species in the richness areas will be intensified unless species are able to adapt to the future climate change or conservation measures are implemented.

ACK N OWLED G M ENTS
We thank the following herbaria for assistance with accessing and de-