Amphibian Diversity in the Matrix of a Fragmented Landscape Around Ranomafana in Madagascar Depends on Matrix Quality

Study System

Field work was conducted in the Ranomafana National Park (RNP, 21°02′−21°24′S, 47°20′−47°35′E), East Madagascar, in forest fragments located east of RNP, and in different matrix sites located between these forest fragments (Figure 1). More detailed information on the location of the sites in RNP and the fragments is provided in Riemann et al. (2015). Most of the remaining rainforest habitat in the Ranomafana region is located within RNP, a 43,500 ha mid-altitude montane rainforest (500–1,300 m a.s.l.) with an annual precipitation between 1,700 and 4,300 mm (Wright & Andriamihaja, 2003). As a result of traditional Malagasy slash and burn agriculture (“tavy”), the remaining forest fragments around RNP are embedded in a matrix of cultivated land (e.g., banana plantations and rice fields) and secondary vegetation (i.e., grasslands with bush and shrub vegetation). OPEN IN VIEWER

We categorized the matrix habitat as (a) “matrix stream,” that is, streams crossing nonforested areas with adjacent riparian vegetation consisting of trees and bushes (between 1 and 5 m on each stream bank), (b) “banana plantation,” that is, plantations of bananas with some understory plants and nonpermanent rills or irrigation ditches, relatively structurally rich, (c) “secondary vegetation,” that is, abandoned fields or paddy fields dominated by grasses, shrubs, and bushes (mainly guavas), at least 50 m distant to the next stream, and (d) “rice fields” characterized by stagnant water and irrigation ditches or small streams but little or no higher vegetation strata. We chose sites within RNP and nine different forest fragments that range in size between 2 and 16.5 ha (for details see Riemann et al., 2015) as forest control sites. Five of these forest fragments comprise streams, including one fragment with two streams. Aerial photographs show that all but two fragments were separated from continuous forest already in 1950, and interviews with local people revealed that all studied fragments were at least 50 years separated from RNP. The Ranomafana region is globally exceptional by its amphibian diversity with almost 120 taxa known (Glaw & Vences, 2007; Vieites et al., 2009; own unpublished data).

Sampling Design

We determined species richness, relative abundance, and species composition on transects distributed along all four matrix categories, and compared these with data collected along streams and in terrestrial habitats inside RNP and forest fragments using the identical methods (22 transects inside RNP or continuous forest and 16 transects in 9 different forest fragments; from Riemann et al., 2015). We established a total of 18 independent line transects (50 × 2 m) (Marsh & Haywood, 2010): five transects were matrix streams, four were banana plantations, four were secondary vegetation, and five were in rice fields. In RNP and fragmented forests transects, we differentiated between stream and terrestrial habitats to equally account for stream depending species (either semiaquatic or stream breeding) and for species that reproduce independent from streams (i.e., phytotelmata, pond or terrestrial breeders; Riemann et al., 2015). Accordingly, in this study, we followed the sampling scheme of terrestrial transects (searching a band of 2 m width) and included in matrix stream transects 1 m riparian vegetation on each stream bank in addition to the water body.

Each transect was sampled nine times using standardized visual and acoustic transect sampling methods (Rödel & Ernst, 2004). A combination of acoustic and visual methods is recommended to detect all major ecological guilds of frogs (Rödel & Ernst, 2004; Vences et al., 2008). Visual sampling was most suitable for terrestrial leaf litter species, and acoustic sampling was particularly useful to detect arboreal species and species calling from hidden places, such as tree holes, phytotelmata or dense leaf litter. Transects were sampled during day and night as in the study area nocturnal as well as diurnal species occur (Glaw & Vences, 2007). Each transect was surveyed twice (once by day, once by night) between May and June 2010, and seven times (three times by day, four times by night) between January (beginning of the heavy rainy season) and June (drier season) 2011.

Visual sampling covered all visible individuals on the ground and in arboreal structures and in addition the aquatic habitat on stream transects. Logs and rocks were not displaced, neither was vegetation pulled apart (e.g., Pandanus leafs) to keep disturbances of the study system as low as possible and to ensure equal sampling effort in all sites. All visually detected individuals were captured, measured, sexed, marked via toe clipping (no functionally important toes, following the recommendations by Grafe, Stewart, Lampert, & Rödel, 2011), and identified by morphology in the field (according to Glaw & Vences, 2007). After data sampling, all individuals were immediately released at point of capture, except some vouchers that were collected to ascertain species affiliation (see Appendix 1). Visual transect sampling was performed at constant speed (approx. 2.5 m/min), but interrupted during handling time and for acoustic sampling at four fixed points (start, 12.5 m, 25 m, 37.5 m). At each point, we registered all calling individuals in a 12.5 m distance (Rödel & Ernst, 2004) straight, left, and right for 5 min and identified them to species level. Two species (one found in secondary vegetation: Plethodontohyla notosticta; one from continuous forest: Gephyromantis depressiceps) were singletons for the complete data set and were therefore excluded from all analyses.

Voucher specimens were euthanized, preserved in 75% ethanol, and deposited at the Mention Zoologie et Biodiversité Animale, Faculté des Sciences, University of Antananarivo, Madagascar (UADBA) and the Museum für Naturkunde Berlin, Germany (ZMB). All toe clips were collected as tissue samples and stored in pure ethanol for DNA barcoding (or dried and stored for further analyses). We used barcoding to confirm identification to species level based on a fragment of the mitochondrial 16S rRNA gene (Vences, Thomas, van der Meijden, Chiari, & Vieites, 2005; Vences et al., 2008), following the molecular protocol described in Ndriantsoa et al. (2013). We determined sequences at least once for species that where assumed to be identifiable by a distinct morphology. Barcoding was repeated for each transect sampling for highly cryptic species (e.g., taxa of the subgenera Chonomantis and Ochthomantis) and supposed undescribed taxa. Obtained sequences were compared with published data on GenBank or own reference sequences (Appendix 1).

Data Analyses

We compared local species richness (absolute number of species detected) and relative abundance (number of frog individuals of all species) per transect between the four matrix categories and to the continuous forest and forest fragments for all habitats using Kruskal–Wallis ANOVA tests. p values of post hoc pairwise comparisons using Wilcoxon rank sum test were adjusted according to Benjamini and Hochberg (1995). For illustrating total species richness in relation to sampling effort (i.e., number of transects), sample-based rarefaction curves were calculated. We further compared patterns of species composition between matrix, continuous forest, and forest fragment sites. We used nonmetric multidimensional scaling (NMDS) to visualize and evaluate patterns of dissimilarity among transects based on their species composition. The ordination was constructed from a Bray-Curtis dissimilarity matrix using species presence or absence data. We then performed permutational multivariate analyses of variance (perMANOVA; Anderson, 2001; McArdle & Anderson, 2001) to test the hypothesis of differences in species composition between the four matrix categories, and continuous forest and forest fragments for terrestrial and stream habitats. This nonparametric permutation-based variant of MANOVA partitions sums of squares of multivariate data equivalent to univariate ANOVA and the pseudo F statistic can be calculated directly from any distance measure (Anderson, 2001). We performed perMANOVA based on Bray-Curtis dissimilarities using species presence or absence data. p values were obtained from 9,999 permutations. p values for post hoc pairwise comparisons were Bonferroni corrected, that is, multiplied by the number of pairwise tests. All statistical analyses, except calculation of species accumulation curves (PAST 3.13 statistics; Hammer, Harper, & Ryan, 2001), were performed in R version 3.2.2 (R Core Team, 2015).

General Results

With a combination of visual and acoustic encounter sampling (and DNA barcoding for species identification), we detected a total of 43 species over all matrix habitats (on 18 transects; Appendix 1). This is comparable to total species richness detected in continuous forest (42 species, on 22 transects, terrestrial and stream) and forest fragments (40 species, on 16 transects, terrestrial and stream) by Riemann et al. (2015). Combined, we found 62 frog species on all 56 transects in RNP and the adjacent forest fragments and matrix sites. Total species richness within the matrix was highest along streams (35), followed by banana plantations (18), and lowest in secondary vegetation (7) and rice fields (8). Between stream habitats, when visually comparing estimated species numbers in relation to standardized sampling effort (Figure 2), there was no obvious difference in total species richness (SR; i.e., matrix vs. forest fragment streams vs. continuous forest streams). However, in nonstream habitats, the total SR was reduced in highly disturbed matrix habitats (secondary vegetation, rice fields) compared with less-disturbed (continuous forest, forest fragments) habitats or disturbed but structurally rich (banana plantations) matrix habitats (Figure 2). OPEN IN VIEWER

Local Species Richness

Local species richness, that is, the number of species encountered on a transect during the 2-year study period, differed significantly between matrix sites, continuous forest, and forest fragments (Kruskal-Wallis, χ²= 45.1, DF = 7, N = 56, p < 0.001; Figure 3). Species richness was significantly higher in stream habitats compared with all other habitats; however, species richness did not differ between streams in the matrix, in forest fragments, or continuous forest (see Appendix 2 for post hoc comparisons). In nonstream habitats, local SR was significantly different between the habitat types, being lower in highly disturbed matrix habitats (i.e., secondary vegetation, rice fields) than in forest habitats (i.e., continuous forest, forest fragments) and banana plantations (Figure 3). OPEN IN VIEWER

Local Relative Abundance

Due to possible double counts in acoustic records, we analyzed the number of frog individuals per transect (i.e., local relative abundance) separately for individuals detected by visual encounter surveys (VES) and acoustic encounter surveys (AES). Local relative abundance, that is, the number of frog individuals on a transect during the 2-year study period divided by the number of surveys per transect (i.e., nine surveys), differed significantly between matrix sites, continuous forest, and forest fragments when including only VES data (Kruskal-Wallis, χ²= 44.1, DF = 7, N = 56, p < 0.001; Figure 4) and also AES data (Kruskal-Wallis, χ²= 20.7, DF = 7, N = 56, p = 0.004), although post hoc comparisons did not reveal any differences for AES data between habitats (Appendix 3). In stream habitats, local relative abundance did not differ between streams in the matrix, in forest fragments, or continuous forest, neither when including only visually detected frogs (i.e., applying VES; Figure 4; see Appendix 4 for post hoc comparisons) nor when including only acoustically detected frogs (i.e., applying AES; see Appendix 3 for post hoc comparisons). In nonstream habitats, there was a significant difference between the habitat types when including visually detected frogs (i.e., VES), with relative local abundance being lower in highly disturbed matrix habitats (i.e., secondary vegetation) than in forest habitats (i.e., continuous forest, forest fragments), however, being not lower in banana plantations and rice fields (Figure 4; see Appendix 4 for post hoc comparisons). We detected the same pattern but no statistical significance when including only acoustically detected frogs (Appendix 3). OPEN IN VIEWER

Species Composition

The species composition differed generally between the habitat types (perMANOVA, Bray-Curtis dissimilarity using presence or absence data, 9999 permutations, F_7,48 = 9.2, p < 0.0001). Matrix streams, banana plantations, and rice fields showed clear clusters in the NMDS that were not overlapping (Figure 5). Variation in species composition between transects in secondary vegetation habitat was highest, and some of these transects overlapped with banana plantation transects. Multiple post hoc comparisons between all matrix and forest types showed significant differences in assemblage composition between most matrix types and of matrix types to most forest and fragment habitat types (Appendix 5). OPEN IN VIEWER

The Innate Value of Matrix Habitats for Amphibian Diversity

Matrix habitats vary greatly in terms of their capacity to support forest species (Lawton, Naeem, Thompson, Hector, & Crawley, 1998), with sites with higher disturbance having usually lower species richness (Gascon et al., 1999; Tocher, Gascon, & Zimmerman, 1997; Vasconcelos, Sorenson, Watanabe, Foguel, & Palacios, 2015). In our study, this was only true for highly disturbed habitats with highly degraded vertical structuring, namely in secondary vegetation (i.e., dominated by bushes and small guava trees) and rice fields. However, in stream habitats, we surprisingly detected that both species richness and relative abundance of frogs were independent of where the streams were located (in continuous forest, forest fragments or matrix), that is, in a small area around streams with only a few meters of gallery forest at each side embedded in a matrix of rice fields and highly disturbed secondary vegetation, occurred as many frog species as around streams surrounded by either small and partly disturbed (fragments) or large and undisturbed (RNP) forest patches.

The most important factor for reproduction of amphibians in many systems is the availability of lentic or lotic habitats (Hillers, Veith, & Rödel, 2008; Hofer, Bersier, & Borcard, 2000; Pineda & Halffter, 2004). We could show that streams are important habitats to support high species richness not only in forest fragments and continuous forest (Riemann et al., 2015) but also in the matrix and further supported comparable relative abundances of frogs. The regional frog assemblage in eastern Madagascar includes a high proportion of species that depend on lotic systems for reproduction (i.e., many species of Mantellidae). Stream-independent breeding species included a wide variety of reproductive modes (e.g., phytotelmata breeder, terrestrial breeder, pond breeder) and were also found in stream habitats. However, corresponding to the availability of their respective breeding habitat, species living in open lentic systems (i.e., Heterixalus alboguttatus, Ptychadena mascareniensis) were predominantly found in rice fields (Figure 6). OPEN IN VIEWER

Disturbance had yet an effect on the species composition of frog assemblages, that is, a set of different species was found around streams in matrix, fragmented and continuous forest with stream assemblages of matrix and forest fragments being more similar, and assemblages in the continuous forest having the most distinct species composition. Accordingly, several species did not occur in highly disturbed matrix stream habitats, in particular, microhylid frogs usually breeding in phytotelmata (mainly tree holes, e.g., Platypelis grandis, Plethodontohyla inguinalis, Plethodontohyla brevipes), while there was a set of generalistic species that did occur in large abundances at all levels of disturbance along streams (e.g., Gephyromantis tschenki, Boophis madagascariensis).

In the matrix, in particular, streams with some surrounding riparian vegetation and banana plantations represent an important component of amphibian diversity in this fragmented landscape and might be themselves valuable habitats. Accordingly, these sites should be included into local conservation planning, in particular, in a landscape where not many alternatives of low-disturbed habitat are available (see also Nopper et al., in press, for an example of hedges as longitudinal structures that might be valuable as corridors for reptiles between dry forest habitats in Madagascar). However, it is yet unclear whether the amphibian populations in matrix habitat represent viable populations on a long term. There is no data available if the amphibian populations of the matrix streams are sustained only by recurring immigrations (i.e., source-sink dynamics) by either migrating frogs or tadpoles that are flushed downstream after heavy rains from forested parts on higher altitude.

Matrix as Corridor Between Natural Habitats

Matrix can act as a selective filter for migration through the landscape. It may facilitate animal movements among fragments of natural habitat by acting as a corridor of habitat that is at least suitable for travel (Antongiovanni & Metzger, 2005; Ricketts, 2001; Rosenberg, Noon, & Meslow, 1997; Selonen & Hanski, 2003), and might therefore connect two patches of natural habitat and subsequently decreasing presumably negative effects on small and isolated populations (e.g., decrease in genetic diversity, sensitivity to stochastic events). Alternatively, it may represent a true barrier for dispersal. The critical factor thereby is the suitability of the matrix for dispersal. Therefore, variation in matrix quality can affect matrix use by species living in isolated fragments (Norton et al., 2000; Sisk et al., 1997; Tubelis et al., 2004), which can ultimately affect assemblage composition (Richter-Boix et al., 2007) and population size and persistence in these fragments (Joly, Miaud, Lehmann, & Grolet, 2001; Vandermeer & Carvajal, 2001). In Brazil, Gascon et al. (1999) showed for various taxa (birds, frogs, mammals, ants) that a high proportion of nominally primary-forest species use matrix habitats for movements (and/or reproduction), as did Laurance (1991) for tropical rainforest mammals in Australia. In this study, the high frog diversity and abundance in and along matrix streams and adjacent narrow gallery forests indicate that this type of matrix habitat might be important as corridor between forest patches in the highly fragmented landscape of Madagascar. Migration in streams, or inadvertent flushing downstream after heavy rains, might happen in either the tadpole or the frog stage, and from fragment to fragment in or along these streams, or between the remaining rainforest (RNP) and adjacent forest fragments. In addition, terrestrial migrations might be facilitated along stream banks by relatively higher air humidity as compared with those matrix sites where open water is missing. Examples from western Africa show that typical rainforest frogs do indeed occur outside natural rainforest as long there is a considerable wide and dense riverine forest (Barej et al., 2015; Rödel, 2000).

Banana plantations in Madagascar are usually not homogeneous monocultural plantations but are a structurally rich habitat including a distinct understorey and scattered canopy trees. This habitat is used by a rather high diversity of frogs, indicating the general suitability as corridor for frog migration. However, these plantations are usually located only in the perimeter of forest fragments and are not connecting forest patches and are therefore not serving as corridors at present. Other types of matrix habitat are obviously not suitable corridors due to their low quality for most amphibian species (i.e., rice fields, secondary vegetation) and thus are likely to be barriers for migrating individuals of forest specialists.

The establishment of corridors to increase the connectivity between habitat fragments is an often recommended idea in conservation management (Prevedello & Vieire, 2010; Rosenberg et al., 1997). In Madagascar, where the natural rainforest is highly fragmented and where realistic and pragmatic options for biodiversity conservation are urgently needed, the protection of streams including small stripes of gallery forest in matrix habitat, or replanting at streams where this riparian vegetation is missing, might be a suitable and achievable target for conservation outside officially protected areas.

Matrix Habitat as Buffer Zone

High-quality matrix reduces the isolation of patches by allowing species to expand their activities (e.g., feeding) outside of fragments, thus increasing effective fragment size (Tubelis et al., 2004). Beyond its direct effects, matrix structure can affect the degradation of fragments through its capacity to buffer fragments from deleterious edge effects (Gascon et al., 2000).

In Malagasy rainforest, banana plantations are often situated in the periphery of natural forest, mainly adjacent to forest fragments. They might be suitable buffer zones as they are structurally rich (i.e., comprising understorey vegetation, mid-storey banana plants and scattered canopy trees) and therefore buffer the microclimate (e.g., humidity, temperature) toward the center of the forest patch. Consequently, frog diversity is comparatively high, in particular, when the plantations are associated with lotic waters. If plantations are structurally rich (e.g., if they include shade trees) and if they contain suitable breeding habitats (e.g., streams for amphibians), they might be suitable habitat and buffer zone, for at least a set of the local assemblage. This was shown, for example, in amphibians in shaded coffee plantations in Mexico (Pineda & Halffter, 2004; Pineda, Moreno, Escobar, & Halffter, 2005) and for lizards and frogs in pine plantations in Uganda (Vonesh, 2001).

Implications for Conservation

Worldwide the size of the area of matrix habitat often exceeds the area of the original habitat or of fragments (Hannah et al., 2008). It is apparent that matrix quality can be an important driver of animal community structure in a fragmented landscape (Watling et al., 2011). Therefore, a better understanding of the matrix is needed by explicitly including environmental differences of the matrix, spatial context, and dispersal processes into studies on disturbed landscapes (Jules & Shahani, 2003; Leibold et al., 2004; Lindenmayer & Franklin, 2002; Watling et al., 2011). It is still ambiguous whether populations in matrix habitats, such as matrix streams and banana plantations in Madagascan rainforest, represent viable populations in the long term and whether matrix quality and structure allow for migration and genetic exchange between populations in different fragments or fragments and continuous forests. However, our study indicates that particular matrix types might serve indeed as suitable habitat (i.e., matrix streams, banana plantations), corridors between forest fragments (i.e., matrix streams), or buffer zones for natural forest patches (i.e., banana plantations), while other matrix types are not suitable habitat for most forest frog species or are dispersal barriers (i.e., secondary vegetation, rice fields; Figure 6). Since frogs have relatively short generation times and our study landscape was disturbed and fragmented at least 50 years ago (and probably much longer), it can be assumed that the observed patterns of frog diversity represent reliable patterns for fragmented rainforest landscapes in Madagascar.

Understanding the factors determining matrix quality for animals is a key step to guide conservation action in fragmented landscapes. As one important action, it was suggested to increase the structural similarity between matrix and natural habitat as a recommendation for management of the matrix in fragmented landscapes (Barnes, Emberson, Chapman, Krell, & Didham, 2014; Prevedello & Vieira, 2010). According to our data, we suggest to guide conservation actions in this system, in particular, to streams in the matrix and their adjacent gallery forests.

The remaining natural vegetation in and outside protected areas in Madagascar is decreasing and will be unable to preserve its biodiversity on a long term (Andreone et al., 2008). Therefore, we must understand responses of organisms to disturbance (Gardner et al., 2007) in order to create buffer zones and corridors combining both disturbed and natural habitats. In particular, more studies are needed that finely differentiate between different matrix types, and analyze which human-modified landscapes (e.g., disturbed primary habitat, secondary habitats, plantations, agricultural systems) have the most conservation potential, as well as studies identifying the proximate mechanisms of disturbance sensitivity.