A new species of the Boophis rappiodes group (Anura, Mantellidae) from the Sahamalaza Peninsula, northwest Madagascar, with acoustic monitoring of its nocturnal calling activity

Abstract A new species of treefrog of the Boophis rappiodes group (Anura, Mantellidae) is described from the Sahamalaza – Iles Radama National Park in northwest Madagascar. This new species is green in colour with bright red speckling across its head and dorsum; similar in morphology to other species of this group including: B. bottae, B. rappiodes, B. erythrodactylus and B. tasymena. The new species can be distinguished by its advertisement call and by a genetic divergence of more than 4.9% in the analysed mitochondrial 16S rRNA gene fragment. Its call consists of two note types: a trill and a click; although similar sounding to B. bottae, the trill note of the new species has a faster pulse rate while the click note is predominantly two-pulsed rather than three. All individuals were detected from the banks of two streams in Ankarafa Forest. The new species represents the only member of the B. rappiodes group endemic to Madagascar’s western coast, with the majority of other members known from the eastern rainforest belt. Despite its conspicuous call, it has not been detected from other surveys of northwest Madagascar and it is likely to be a local endemic to the peninsula. The ranges of two other amphibian species also appear restricted to Sahamalaza, and so the area seems to support a high level of endemicity. Although occurring inside a National Park, this species is highly threatened by the continuing decline in the quality and extent of its habitat. Due to these threats it is proposed that this species should be classified as Critically Endangered according to the IUCN Red List criteria.


Introduction
The genus Boophis is a monophyletic group of treefrogs belonging to the family Mantellidae. Endemic to Madagascar and the Comoros, the group comprises over 70 species, many of which have only recently been described (Köhler et al. 2007(Köhler et al. , 2008(Köhler et al. , 2011Wollenberg et al. 2008;Glaw et al. 2010;Vallan et al. 2010;Vences et al. 2010Vences et al. , 2012. The genus is classified into two subgenera, Boophis and Sahona, which largely correlate with breeding habit; the nominal Boophis being predominantly stream-breeders, and the Sahona pond-breeders . Of the two subgenera, Boophis is the most speciose and can be further divided into eight groups, although at least some of these appear to be paraphyletic and await taxonomic revision ). Among these is the rappiodes group -small, green treefrogs with red pigment patches and partially translucent skin -an appearance shared with the neotropical glass frogs of the family Centrolenidae (Vences and Glaw 2002;Glaw and Vences 2007;Köhler et al. 2007). As with most Malagasy amphibian diversity, the majority of members of the rappiodes group can be found in the eastern rainforests, and until now there have been no documented species from the drier forests of the western coast (Vences and Glaw 2002;Glaw and Vences 2007).
The Sahamalaza Peninsula in north-western Madagascar has undergone only two previous amphibian surveys (Andreone et al. 2001;Raselimanana 2008); the most recent of which led to the discovery of Boophis tsilomaro and Cophyla berara (Vences et al. 2005. However, one site on the peninsula, the Ankarafa Forest, had never been surveyed before. From this forest we describe a new species of Boophis, molecularly assigned to the rappiodes group, vocally distinct and genetically different to all other known species of this monophyletic group.

Study sites and survey periods
The Sahamalaza Peninsula is in the province of Mahajanga, northwest Madagascar (Fig. 1). Parts of the peninsula were declared the Sahamalaza -Iles Radama National Park in July 2007 and have been included in UNESCO's network of Biosphere Reserves since 2001 (Schwitzer et al. 2007). The peninsula covers around 26,000 ha and is characterised by a number of low hills of about 300-350 m a.s.l. intersected by a few largely seasonal streams (Andreone et al. 2001). Located between 13°52'S and 14°27'S, and 45°38'E and 47°46'E, it is defined by the Sahamalaza Bay to the east, the Mozambique Channel to the west and the Loza River to the south (Volampeno 2009).
The climate is sub-humid and has two distinct seasons: a cooler, drier season from May to November; and a hotter, wetter season from December to April. Monthly mean maximum temperature ranges from 28.5 ± 3.61 °C in July to 39.1 ± 2.11 °C in February; and monthly mean minimum temperature ranges from 13.2 ± 0.81 °C in October to 21.8 ± 0.81 °C in January (Volampeno et al. 2011). Rainfall is highest during January and February and the mean annual precipitation rate is around 1,600 mm (Schwitzer et al. 2007).
Sahamalaza supports a unique type of transitional forest that harbours plant species from both the wetter Sambirano domain and drier western domain (Birkinshaw 2004;Schwitzer et al. 2006). This forest is concentrated in two separate blocks: Anabohazo in the northeast (14°18.56'S, 47°54.89'E) and Ankarafa in the west (14°22.82'S, 47°45.46'E) (Schwitzer et al. 2007). Until recently a third block existed, called Analavory Forest (14°23.30'S, 47°56.15'E), but it experienced near complete destruction following an uncontrolled man-made fire in 2004 (Volampeno 2009). The forest blocks are composed of a matrix of smaller fragments isolated by savannah with high levels of human disturbance (Schwitzer et al. 2007).
Fieldwork took place over two separate periods: a three month survey between October 2011 and January 2012, and an additional survey between January and February 2013. This ensured the coverage of most of the wet season when rain is more abundant and the individuals are expected to be more active. Surveys were conducted in Ankarafa Forest, Anabohazo Forest and around the villages of Antafiabe and Betsimpoaka.

Survey methods
Surveys were conducted by opportunistic searching and directed towards vocalising males using headlamps and torches. Transects were made across a range of habitats and degradation levels, covering the four previously mentioned areas of the peninsula. Searches were repeated during the day and night to account for any diel differences in activity, taking place in the morning and evening. Searching took place approximately two metres either side of the transect and up to two metres in height. Most searches in Ankarafa were repeated at least once both in the dry and wet season (during the 2011-2012 period) following the same routes where possible. The sites in Anabohazo Forest, Antafiabe and Betsimpoaka village were sampled only once. Sites were sampled in a randomised order and all searches were conducted by the same two individuals to avoid systematic observer bias. The frogs' frequency of occurrence was estimated by dividing the total number of frogs encountered during each transect search by the respective length.
The vouchers were photographed to document life colouration and calls were recorded whenever possible. Acoustic description follows that outlined in the acoustic analysis methods below. The tissue samples (fourth digit of the left toe removed with scissors) were stored in 70% ethanol or 96% ethanol for genetic analysis. Location was logged using a handheld GPS receiver (Garmin eTrex Vista HCx; Garmin International Inc., Olathe, United States). Microhabitat was noted and vertical position from the ground measured using a tape measure.

DNA analysis
Tissue samples were available for four individuals. Total genomic DNA was extracted from the tissue samples using proteinase K digestion (10 mg/ml concentration) followed by a standard salt-extraction protocol (Bruford et al. 1992). We sequenced a fragment of ca. 550 bp of the 3' terminus of the mitochondrial 16S rRNA gene. For primers and cycling protocols see Crottini et al. (2011). The light strands were sequenced using an ABI3730XL by Macrogen Inc.
To assess genetic distinctness of the new species from all other Malagasy frogs and ascertain its belonging to the Boophis rappiodes group, sequences were compared using the BLAST algorithm with a database containing homologous sequences of reliably identified adult individuals of almost all Malagasy frog species (Vieites et al. 2009). Mean genetic distances matrix (uncorrected p-distance transformed into percent, using the complete deletion option) between and within individuals belonging to the type series of B. ankarafensis sp. n. (holotype and three paratypes) and of other species of the B. rappiodes group were computed using MEGA, version 6.06 (Tamura et al. 2013).

Acoustic monitoring and analysis
Most amphibian calls are species-specific and it is usually possible to identify syntopic calls to the species level. Sound recordings were taken to obtain detailed information on the acoustic parameters of the species and investigate any intraspecific variability and overnight temporal patterns in activity. Acoustic recordings were made continuously from dusk until dawn on sixty nights between October 2011 and January 2012. Data were collected from 37 different locations; the majority of these locations were within Ankarafa Forest (29), followed by Anabohazo Forest (7) and a single location on the Vavan'aneno River near the village of Antafiabe. Nineteen locations had recordings made on two or more nights, separated between 9 and 79 days.
Calls were recorded in the field using a Song Meter SM2 digital recorder (Wildlife Acoustics Inc, Concord, United States) at a 16-bit resolution and 16 kHz sampling rate using two side-mounted SMX-II microphones. The digital recorder was placed one to two metres above the ground or water by securing it to deadwood or a protruding branch using bungee cords. Continuous recordings split into sections of 120 minutes each were saved in the standard uncompressed .WAV format. Preceding analysis, these were split using a custom-written MATLAB (The Mathworks, Natick, USA, v7.14.0.739) script into minute long segments to allow for more efficient analysis. Spectrograms were viewed individually as a dual channel output using Avisoft SASlab Pro (Berlin, Germany, v5.2.06); frequency resolution of 512 FFT, a 100% frame rate, Hamming window and an intensity threshold of 50%.
From each night-long recording where B. ankarafensis sp. n. was detected, five representative vocalisations for each identified call type were chosen. Calls with the highest recording quality were selected and where possible equally distributed over the entire recording period. This reduced the risk of selecting the same individual repeatedly (pseudoreplication) by maximising the time period between each of the selected vocalisations. From a subset of five trill notes, acoustic parameters were measured for ten sequential broadband pulses and ten sequential narrowband pulses per note.
From each of these notes spectral and temporal characteristics were measured, and the minimum, maximum and average values (with SD) calculated (Avisoft SA-Slab Pro; Berlin, Germany; v5.2.06). To remove interspecific and abiotic noise outside of the B. ankarafensis sp. n. bandwidth, and increase detection thresholds, sound files were band-pass filtered with a finite impulse response filter (Hamming) between 3.2 and 6 kHz using Avisoft SASlab Pro (Berlin, Germany, v5.2.06). The spectral measurements taken were peak frequency (= frequency of maximum energy) and bandwidth. These were calculated using averaged power spectra with peak interpolation over three data points (Hamming window, FFT width 512 points; bandwidth threshold -10dB; peak detection threshold -20 dB, hysteresis 10 dB). For the trill note (type 1) these were taken from each of the two pulse types present and also a section of 40-60 sequential pulses; measurements were taken from the click note (type 2) in its entirety. Durations were taken for the two notes types in their entirety and also for individual pulses. The pulse rate of the trill note was calculated by taking the inverse of the average time differences between pulses, calculated through pulse train analysis. All measurements are stated as the mean ± standard deviation followed in parentheses by the maximum and minimum and number of analysed units (n). Calls were then compared to an existing database of frog vocalisations Rosa et al. 2011): comparisons were limited to notes recorded at similar ambient temperatures as, to a limited extent, some call parameters can be temperature dependent .
Minute by minute changes in activity were recorded for all detected species from dusk (sunset) until dawn (sunrise) for 60 nights. The activity index per minute was 1 for one caller and 2 for more than one. The time period between dusk and dawn was split into percentiles for each night; this controlled for the slight variations in night length experienced across the sampling period, enabling the summation of data across multiple nights for each analogous percentile. The number of minute periods containing calling activity within each percentile was calculated and the values for each analogous percentile across all nights summed. The totals for each percentile were then divided by the total activity to determine the proportional change in activity.
Holotype Diagnosis. A treefrog assigned to the genus Boophis based on absence of femoral glands in males, the presence of an intercalary element between the ultimate and penultimate phalanges of fingers and toes (verified by external examination), presence of nuptial pads in males, general morphological resemblance to other Boophis species, and molecular evidence. Assigned to the Boophis rappiodes group based on small size (adult males 22.9-24.0 mm and one female 28.5 mm SVL), absence of lateral fringes along lower arm and tarsus, greenish and slightly translucent dorsal colouration and translucent venter (inner organs can be clearly seen through the skin in live specimens). Boophis ankarafensis sp. n. is distinguished from B. erythrodactylus by lack of reddish colour of fingertips, and from B. erythrodactylus and B. tasymena by lack of regular pattern of red dorsal spots. Distinguished from B. viridis by its smaller size (SVL up to 31 mm in males and 35 mm in females of B. viridis), iris colouration (B. viridis shows a distinctive brown inner-iris area and blue outeriris area), and the presence (vs. absence) of yellowish dorsolateral stripes. Compared to B. rappiodes, the new species is distinguished by a more extensive darker pattern on the dorsum, which is most evident in living or freshly preserved specimens; while the dorsal pattern in B. rappiodes is intensely red, and remains red in preservative before it eventually fades, the pattern in B. ankarafensis sp. n. is reddish-brown in life and becomes persistently dark brown in preservative, often covering almost the entire dorsum. B. ankarafensis sp. n. is most similar in morphology to Boophis bottae and no apparent morphological feature distinguishes them. However, the new species can be distinguished by molecular analysis and by its advertisement calls (see below).
Description of the holotype. MRSN A6973, adult male in a good state of preservation. SVL 24.0 mm (see Table 1 for more detailed morphometric measurements). The body is slender with the head much wider than the body. Snout rounded in dorsal view, slightly truncate in lateral view, nostrils directed laterally, slightly nearer to tip of snout than to eye; canthus rostralis and loreal region both slightly concave; tympanum distinct, rounded, 40% of eye diameter; supratympanic fold not recognizable; tongue ovoid, distinctly bifid, posteriorly half free; vomerine odontophores distinct; positioned posteromedian to choanae; choanae small, rounded. Arms slender, subarticular tubercles single, round; metacarpal tubercles unrecognizable; fingers with weak webbing; webbing formula 1(1), 2i(1.75), 2e(0.75), 3i(2.5), 3e(1.75), 4(1); relative length of fingers 1 < 2 < 4 < 3 (finger 2 distinctly shorter than finger 4); finger disks moderately enlarged; small unpigmented nuptial pads faintly recognizable on inner side of first finger. Hindlimbs slender; tibiotarsal articulation reaches nostril when hindlimb is adpressed along body; lateral metatarsalia separated by webbing; inner metatarsal tubercle distinct, no outer metatarsal tubercle; webbing formula between toes 1(0), 2i(0.75), 2e(0), 3i(0.75), 3e(0), 4i(1), 4e(1.5), 5(0.25); relative length of toes 1 < 2 < 5 = 3 < 4; toe disks slightly enlarged. Skin on dorsal surfaces and ventrally on throat smooth; ventral skin on belly and around cloacal opening glandular. The ground dorsal colour, including limbs, is light green. The webbing, finger and toe disks are greenyellow in colour. Speckles of reddish-brown and yellow pigment can be seen covering the dorsum and limbs. Thin yellow dorsolateral stripes run from behind the eye to the forelimb and then fade towards the mid-body. On the head a reddish-brown pigment forms a band between the eyes and covers the supra-ocular area, interspersed with yellow speckling. This reddish-brown pigment also forms a faint rostral stripe between the eye and nose tip. The pupil is horizontal, with a beige iris containing darker brown patches and reticulations; the iris periphery is blue. The specimen has a white venter with some translucence exposing the inner organs, and a bluish throat. At day-time the green on the dorsum became pale and the red markings fade out to a pale brownish red colour (Fig. 2D). In preservative it is in good condition and comparatively more hydrated. The colouration, at about 4 months from the capture is vivid: the dorsum is pigmented and the dark bar between the eyes is contrasting and evident, as well as the pigmentation around the nostrils. The specimen has the last phalanx of the 4th toe of the left foot missing for genetic analysis.
Description of paratypes and variation. The paratypes (MRSN A6974-6976) closely match the holotype but with slightly different patterning of pigment patches. Finer regular black spots were observed on the dorsum and limbs (possibly single melanophores); a feature similarly observed in the related B. bottae (Vences and Glaw 2002). Based on other individuals photographed in nature, specimens matched the holotype and paratypes. However, pigment patterns vary in their distribution and intensity. The males MRSN A6975 and MRSN A6976 are in mediocre state of conservation and dehydrated due to fixation and preservation in ethanol. The colouration, after about 18 months from the capture, faded from green to whitish-yellowish except for pigmentation on the upper eyelids and for the bar between the eyes. The dorsum and upper parts of legs are finely pigmented. Eyes are blackish, with whitish pupillae. The 5th toe of the left foot was removed and fixed in 96% ethanol for molecular analyses. The female specimen (SVL 28.5 mm) collected more recently (MRSN A6974) is in better condition, similar to the holotype. This specimen also has the last phalanx of the 4th toe of the left foot missing for molecular identification. See Table 1 for detailed morphometric measurements. Mitochondrial variation and differentiation. The molecular data confirms the attribution of B. ankarafensis sp. n. to the B. rappiodes group . The four analysed specimens of B. ankarafensis are genetically uniform and did not show any intraspecific divergence, while the analysed specimens of B. bottae, B. rappiodes, B. tasymena and B. viridis are more heterogeneous and show intraspecific uncorrected divergence ranging from 1.2% (B. viridis) to 2.2% (B. rappiodes), in the 16S rRNA gene analysed fragment ( Table 2). The species showing the highest values of intraspecific divergences is B. rappiodes (2.2%). This value of intraspecific divergence is computed based on the analyses of sequences coming from Andasibe and Ranomafana, two sites that are about 300 km far apart. The population of B. ankarafensis is about 400 km apart from the closest population of B. bottae (at Betampona) and their genetic distance is of 4.9%. As shown in Table 2, available values of intraspecific genetic divergence values never reach the minimum value of genetic divergence observed between different species of the B. rappiodes group (comparison B. ankarafensis/B. bottae).
The genetic distance between B. ankarafensis and the five other species of the B. rappiodes group ranges between 4.9% (comparison between B. ankarafensis and B. bottae) and 11.4% (comparison between B. ankarafensis and B. erythrodactylus). Among the species of the analysed species group the smallest genetic distance is observed between the newly described species and B. bottae (4.9%) and the highest value between B. erythrodactylus and B. viridis (13.0%). More details on genetic distances between species of the B. rappiodes group are provided in Table 2.
Advertisement call and acoustic monitoring. Vocalisations of Boophis ankarafensis sp. n. were recorded on thirteen occasions between October 2011 and Janu- Table 2. Genetic divergence in the analysed 16S rRNA mitochondrial gene fragment of the Boophis rappiodes group (p-distance transformed into percent using the complete deletion option). Pairwise distances calculated for intra-(in bold) and inter-specific genetic divergence. n.c. (not calculated). B. bottae B. rappiodes B. erythrodactylus B. tasymena B.  ary 2012. No males were found in calling activity between January and February 2013 (later in the wet season). Calls were detected from just 8 of the 37 locations monitored on the peninsula (Fig. 4). All eight of these locations were along the banks of a single stream in Ankarafa Forest. Vocalisations were detected within 3978 minutes of acoustic recordings and averaged 306 ± 181 minutes per night (n = 13). A total of 59.8% of these minutes contained vocalisations emitted by a single male caller, with the remainder emitted as a chorus of two or more callers. Total activity equalled 5576 minutes and average per night activity amounted to 429 ± 273 minutes (n = 13). Levels of calling fluctuated throughout the night (Fig. 5); there was a peak in activity just after dusk, followed by a slow decline until dawn, when the fewest vocalisations were detected (y = 0.019e -0.015 x; R² = 0.582). The acoustic repertoire consists of two note types: a multi-pulsed trill (type 1; Fig. 6A) and a 1-3 pulsed click (type 2; Fig. 6B). Only 11 of the 13 occasions produced recordings of a sufficient audio quality to perform spectral analysis. Measurements were taken from five representative notes of each call type per night, thus a total of 55 trill notes and 55 click notes were measured (Table 3). From a subset of five trill notes, acoustic parameters were measured for ten sequential broadband pulses and ten sequential narrowband pulses per note. The environmental temperature of the recorded trill notes ranged between 25.2 and 30.3 °C (27.4 ± 1.08 °C, n = 55) and of the click notes between 22.9 and 29.7 °C (26.7 ± 1.41 °C, n = 55).

B. ankarafensis
The trill note measured 4.29 ± 1.29 s (1.614-8.091 s, n = 55) and is composed of alternating broadband and narrowband pulses. These notes had a broadband pulse rate of 15.0 ± 1.19 pulses/s (11.7-17.3 pulses/s, n = 25) and peak frequency of 4.14 ± 0.131 kHz (3.69-4.48 kHz, n = 55). Broadband pulses measured 3.47 ± 0.485 ms (2.50-5.06 ms, n = 50) whereas narrowband pulses were slightly longer at 17.9 ± 2.56 ms (11.2-25.1 m; n = 50). The broadband pulse had a peak frequency of 4.19 ± 0.122 kHz (3.91-4.34 kHz; n = 50), whereas the narrowband pulse measured 4.15 ± 0.108 kHz (4.00-4.31 kHz, n = 50). Bandwidths were 0.288 ± 0.169 kHz (0.120-0.780 kHz, n = 50) for the broadband pulse compared to just 0.155 ± 0.036 kHz (0.129-0.280 kHz, n = 50) for the narrowband pulse when measured at a -10 dB threshold. The majority of click notes consisted of two pulses, although singular and triple pulses were also observed. Click notes had a total duration of 52.9 ± 5.20 ms (40.0-68.7 ms; n = 55) and peak frequency of 3.95 ± 0.162 kHz (3.57-4.30 kHz; n = 50). Bandwidth of the click note measured 0.538 ± 0.234 kHz (0.120-0.143 kHz) at a -10 dB threshold. The advertisement call of B. ankarafensis sp. n. sounds similar to that of the morphologically similar B. bottae; with both species possessing a trill note and a click note in their acoustic repertoires Rosa et al. 2011). However, a number of differences can be used to distinguish them. In our comparison with B. bottae we selected only those notes recorded at a similar ambient temperature, and we have also distinguished the two pulse types of the trill by their relative differences in bandwidth, rather than duration as used by Vences and Glaw (2002); the broadband pulse and narrowband pulse of B. ankarafensis sp. n. are equivalent to the shorter and longer duration pulses of B. bottae respectively.
The trill note (type 1) of B. ankarafensis has a faster broadband pulse rate than the trill of B. bottae (13.4-13.5 pulses/s, 25.2 °C versus 8.12-11.6 pulses/s, 23.0 °C). The click note (type 2) of B. ankarafensis sp. n. usually contains just two pulses, and only rarely consists of three, in contrast the click notes of B. bottae are usually three-pulsed. Although only slightly divergent, it is discernable that the spectral frequency of the click note is lower in B. ankarafensis sp. n than in B. bottae (3.75-3.84 kHz, 22.9 °C versus 4.23-4.42 kHz, 23 °C). The trill note of B. ankarafensis shows a slightly lower broadband pulse peak frequency (4.10-4.24 kHz, 25.2 °C versus 4.30-4.64 kHz, 23 °C) but a similar narrowband pulse peak frequency 23 °C) in comparison to B. bottae. Of the 56 encounters, 48 frogs were male and 8 female. Males of this species were found calling from vegetation approximately 0.5 to 2 m high. Vocalising males were often within close proximity to one another, positioned on different leaves of the same plant. All but one of the eight females were found in axillary amplexus with males ( Fig. 3D-E); mating pairs were positioned on leaves overhanging the stream bank, and in one instance on a rock. Calling and breeding activity were most intense during the first few months of the rainy season (Oct-Dec), decreasing later in the season (Jan-Feb). A lone female was found in a tree during the day approximately 3 m high and 30 m from a stream.
Distribution, conservation and proposed IUCN status. The suggested conservation status of this species was assessed using the criteria and guidelines of the IUCN Red List (IUCN Standards and Petitions Subcommittee 2011). This species is endemic to Madagascar and so far only recorded in a small forest fragment (Ankarafa Forest). Individuals were found along the vegetated banks of two streams, with forested sec- tions of these streams extending for approximately 2 km of the more northerly stream (Fig. 7B), and just 1 km for the stream south of this (Fig. 7A). No individuals were found upstream of these two locations, where the streams are replaced by Raphia-dominated swampland or downstream where the stream-banks are cleared of vegetation. An average occurrence of 3 specimens per 200 m of river was determined from those transect searches where the specimens were detected (Table 4). This species was not detected from Anabohazo Forest, which aside from Ankarafa represents the only other large area of intact forest remaining on the peninsula. A survey of the Vavan'aneno River between Antafiabe and Ambinda villages, and a survey of Betsimpoaka village also failed to detect it.
If suitable habitat is considered to be all areas of Ankarafa Forest (likely an overestimate) then this area totals less than 5 km 2 , giving an EOO (extent of occurrence) of less than 100 km². If plots with a scale of 2 km 2 are used to estimate AOO (area of occupancy), then this species occurs within 4 km 2 of habitat, resulting in an AOO of less than 10 km 2 . Therefore the Critically Endangered thresholds for extent of occurrence and area of occupancy are both met (EOO < 100 km² and AOO < 10 km²) (CR B1+2). The most serious threat to the species is habitat destruction through tavy practice (slash and burn agriculture), small-scale logging and the uncontrolled burning of neighbouring  grasslands; a large out of control fire could easily affect the two subpopulations as they are separated by a distance of less than 2 km. Therefore, all individuals can be considered to occur within a single location only (CR B1a+2a). Given this on-going destruction of suitable habitat, population declines can be expected to continue unless some remedial action is taken (CR B1b(i,ii,iii,iv,v + B2b(i,ii,iii,iv,v)). Thus the species should qualify as Critically Endangered under criterion B (CR B1ab (i,ii,iii,iv,v)+2ab(i,ii,iii,iv,v) (Vences and Glaw 2002). The closest relative of B. ankarafensis is B. bottae, known from over 400 km to the east of Sahamalaza. It can be possible that in the past B. bottae has spread along the southern slopes of Madagascar's northern massifs where it may have speciated and finally reached the transitional forest of the Sahamalaza Peninsula. Thus other populations of B. ankarafensis may be found here, although if they exist they will be isolated and fragmented due to habitat destruction. Other amphibian surveys in northwest Madagascar, e.g. in Sahamalaza, Manongarivo, Tsaratanana and Benavony, have failed to detect it, despite it possessing a conspicuous and distinctive advertisement call, Glaw 2007, Raselimanana 2008) and so the species may be locally endemic to the peninsula. With this description, a total of three amphibian species are now known solely from Sahamalaza, with the species B. tsilomaro and Cophyla berara discovered during a previous survey in 2000 (Andreone et al. 2001;Vences et al. 2005. Thus, at just 26,000 ha, the peninsula appears to support a high level of amphibian endemicity, although further surveys of northwest Madagascar should be undertaken to search for additional populations.
All breeding behaviour of B. ankarafensis sp. n. was observed during the wet season along the banks of fast-flowing streams, indicating that its spawn is most likely laid in or adjacent to bodies of lotic water; a feature in common with most members of the Boophis subgenus . The streams in Ankarafa Forest flow throughout the year, in comparison all streams in Anabohazo Forest are seasonal. The species' absence from this latter locality may be linked to the lack of perennial streams, for example its larval stage may not have enough time to complete development.
The frog was only found within intact forest and appears sensitive to anthropogenic disturbance. Intact forest is rare across the peninsula and aside from Anabohazo, where the species was not found, Ankarafa represents the largest area of remaining forest. Isolated populations of B. ankarafensis sp. n. may survive in residual pockets of gallery forest elsewhere on the peninsula, but Ankarafa is likely to harbour the largest and most sustainable population of this species. Their habitat in Ankarafa is fragmented and the two streams it is known from are separated by savannah (Fig. 8A). As B. ankarafensis sp. n. is arboreal, this will most likely limit gene flow between the two populations, potentially reducing its long-term viability.
Despite its protected status, Ankarafa is experiencing widespread deforestation (Seiler et al. 2012), furthermore much of this destruction is concentrated on the streamside forests which this species relies upon (pers. obs.). Boophis ankarafensis sp. n. is highly threatened and we propose that it should be classified as Critically Endangered (CR) according to the IUCN Red List criteria. The other apparent endemics face similar threats; C. berara is already listed as CR , while B. tsilomaro most likely qualifies for CR status . A halt to all forest destruction and agricultural practices within the park must occur immediately to stop any further decline of Sahamalaza's endemic amphibian fauna, or risk their possible extinction within the near future.