Disjunct resident population of Melodious Lark Mirafra cheniana discovered in East Africa

We describe the discovery of a previously unknown population of Melodious Lark Mirafra cheniana from northern Tanzania and southern Kenya, some 2000 km north of what was previously thought to be its northern limit in central Zimbabwe. The identification of this population as M. cheniana is based primarily on analyses of mitochondrial DNA and song. We also clarify relationships within this genus by sequencing material of White-tailed Lark Mirafra albicauda for the first time; although its relationships with other species in the group remain unclear, our analyses clearly show it to be different from M. cheniana in the newly discovered population. The discovery of a population of M. cheniana in East Africa conforms to a known pattern for closely related lark species to have disjunct distributions in southern and eastern Africa.


Introduction
The family Alaudidae (larks) comprises c. 100 species, which occur widely across Eurasia and Africa, with one species each reaching Australia and North America plus Colombia (de Juana et al. 2004;Fjeldså et al. 2020;Gill et al. 2022). Larks are found in various open habitats, such as different types of grasslands, open scrubland, and deserts. The phylogeny has been reconstructed for c. 80% of lark species, revealing a striking lack of agreement between traditional morphology-based taxonomy and phylogeny ). Many of the species are renowned for having extremely similar appearances (de Juana et al. 2004) and therefore being difficult to identify in the field.
The Eurasian species are fairly well studied with respect to taxonomy, and several species complexes have been analysed using different approaches, often including morphological, vocal, other behavioural, genetic, ecological, and other data. These studies have resulted in a substantial increase in the number of recognized species (Alström 1998;Guillaumet et al. 2005Guillaumet et al. , 2006Guillaumet et al. , 2008Drovetski et al. 2014;Ghorbani et al. 2020aGhorbani et al. , 2020bAlström and Gombobaatar 2021;. The African species are much less well studied, despite the fact that 60% of the species of larks occur exclusively in Africa. The southern African Karoo Lark Calendulauda albescens and Long-billed Lark Certhilauda curvirostris (sensu lato) complexes are nowadays often treated as three instead of two species and five instead of one species, respectively (Clements et al. 2021;Gill et al. 2022), based mainly on mitochondrial DNA (mtDNA) data (Ryan et al. 1998;Ryan and Bloomer 1999). The circumscriptions of Blanford's Lark Calandrella blanfordi and Rufous-capped Lark Calandrella eremica from the Horn of Africa and southern Arabian Peninsula are fundamentally different following the publication of a molecular phylogeny by Stervander et al. (2016), and Stervander et al. (2020) described a new subspecies of Red-capped Lark Calandrella cinerea from Nigeria. Both Alström et al. (2013), Stervander et al. (2016), and Stervander et al. (2020) discovered unexpectedly deep mtDNA divergences within several species of larks, and suggested that further studies based on independent data were warranted. In contrast, Spottiswoode et al. (2013) provided morphological, vocal, and molecular evidence that the Horn of Africa Sidamo Lark Heteromirafra sidamoensis is synonymous with Archer's Lark Heteromirafra archeri.
East and northeast Africa are home to 12 species of larks in the genus Mirafra (Gill et al. 2022). Four of these are rather small with very similar plumages: Williams's Lark M. williamsi, Singing Bush Lark M. cantillans, Friedmann's Lark M. pulpa, and White-tailed Lark M. albicauda (Sinclair and Ryan 2010;Ryan 2020a, b;Ryan and Sharpe 2020;Bradley 2021). Two other Mirafra species from southern Africa are extremely similar to these four species in appearance: Melodious Lark M. cheniana occurs in eastern South Africa, with a few records from southern Botswana and an isolated population in central Zimbabwe (there may also be a relict population in northern Botswana; (Davies 2011)), and Monotonous Lark M. passerina, which is more widely distributed in southern Africa, mainly north and west of M. cheniana (Ryan 2020c, d). In addition, Horsfield's Bush Lark M. javanica from Southeast Asia to Australia is similar to and sometimes considered conspecific with M. cantillans. All of these except M. pulpa and M. albicauda have been analysed phylogenetically and found to form a clade .
Larks have two centres of endemism within Africa: one in southern Africa, particularly in the Cape, Karoo, and Namib regions (Dean and Hockey 1989), and another in north-eastern Africa, particularly Ethiopia, Somalia, Kenya, and northern Tanzania (Fjeldså and de Klerk 2001). Some genera, such as Spizocorys, Calendulauda, and Heteromirafra, are largely or wholly confined to both regions but with different species represented in each (Archer and Godman 1961). This pattern of clustering of dry-country species into species hotspots in southern and eastern Africa was studied by Moreau (1966), who suggested that the high number of dry-country endemics in both regions is due to their long and stable history of dry climates, perhaps longer and more stable in south-western Africa. A similar pattern of interrupted distribution is found also in other groups of animals, such as the harvester termites in the genus Hodotermes and their mammalian predators the Bat-eared Fox Otocyon megalotis and the Aardwolf Proteles cristata. The similarity in the taxonomic composition in these two hotspots of endemism is likely to be due to the fact that they have not always been separated, as they are now, by a humid zone from central Tanzania to Zimbabwe (Moreau 1966). An "arid corridor", remnants of which are still apparent in western Tanzania, is likely to have linked the two repeatedly over time due to fluctuating climates, allowing exchange of animals between them.
In 1987, B.W.F. realised that what was generally assumed to be White-tailed Lark M. albicauda occurring in the Maasai Mara, Kenya and Serengeti, Tanzania (i.e., west of the Great Rift Valley) differed markedly (especially in song) from other populations of M. albicauda (Finch 2002(Finch , 2009). The identity of these birds, referred to below as "Mara Lark", remained a mystery until it was originally suggested in a personal email sent by A.S.K. in 2020 that they appeared to match the highly disjunct M. cheniana rather than any known East African species.
We here analyse the taxonomic status of the "Mara Lark" based on analyses of songs and mitochondrial DNA. We conclude that the "Mara Lark" represents a presumably geographically isolated population of M. cheniana, and comment on its distribution and conservation status. We also for the first time include M. albicauda in a phylogenetic analysis (including the holotype) and describe its poorly documented song.

Field observations
B.W.F. studied "Mara Lark" in the field in the Maasai Mara in Kenya on multiple occasions between 1987 and 2019, and recognised that an identical form was present in Serengeti in Tanzania over the same period. R.S.H. and A.S.K have studied the "Mara Lark" in the Maasai Mara in Kenya multiple times between 2010 and 2021. C.C. has sound recorded it in the Serengeti National Park, Tanzania in 2013 where it occurs sympatrically with M. albicauda. Rich Lindie videoed a singing bird in the Serengeti National Park, Tanzania in 2015 and Nik Borrow photographed a singing bird in the Serengeti National Park, Tanzania in 2018 (Mirafra albicauda ML20546688). All localities with observations of "Mara Lark" are shown in Fig. 1.
B.W.F. has studied M. albicauda on multiple occasions in Kenya, Tanzania, and Uganda; C.C. has observed and sound recorded this species at multiple locations in Uganda, at Eluanata Dam in the Maasai Steppe east of the Great Rift Valley in Tanzania and especially in the Serengeti, west of the Great Rift Valley in Tanzania where it co-occurs with "Mara Lark"; J.G. has observed it in the Serengeti on several occasions, and P.A. has observed and sound recorded it in Nairobi National Park, Kenya in May 2018. D.d.S, D.E., J.G., P.F.D., and P.A. have ample experience with M. cheniana from South Africa.

Song
For sound analysis, we selected recordings of 8 M. albicauda individuals (4 individuals from Kenya, 3 from Tanzania, 1 from Uganda; 11 recordings in total), 4 "Mara Lark" individuals (all from Kenya, 4 recordings in total), 5 M. cheniana individuals (from throughout the range in South Africa, 5 recordings in total), and 4 M. cantillans individuals (3 individuals from Kenya and 1 from Somalia; 6 recordings in total). These include all of the species of larks that are potentially confusable with "Mara Lark", and which occur in Kenya and Tanzania (Sinclair and Ryan 2010). The data were from our own recordings and additionally from xenocanto (http:// www. xeno-canto. org) and the Macaulay Library (https:// www. macau layli brary. org/). The M. albicauda were identified based on our experience with songs of these birds from Kenya, Tanzania, and Uganda, as described in the literature (Britton 1980;Zimmerman and Turner 1996;Sinclair and Ryan 2010;Fanshawe 2002, 2020;Ryan 2020b); the birds identified as "Mara Lark" resembled M. cheniana from South Africa in song; and the M. cantillans were identified based on experience with songs of these birds from Kenya and Somalia. In several cases, the vocal identifications were confirmed by plumage characters (as described in the literature (Britton 1980;Zimmerman and Turner 1996;Sinclair and Ryan 2010;Fanshawe 2002, 2020;Ryan 2020a,bandc) and studied in museum collections). Our own recordings have also been uploaded to the Macaulay Library and to Xeno-canto. A list of the recordings studied is in Table S1. Additional recordings have been listened too, and multiple additional birds have been heard in the field.
The sampling frequency of all recordings was set to 48 kHz and sampling depth to 24 bits before analysis. Sonograms were created in Raven Pro 1.5 (Cornell Laboratory of Ornithology, Ithaca, USA), with the following settings: window size = 256 samples; window type = Hann; 3 dB filter bandwidth = 270 Hz; overlap = 50%; size = 2.67 ms; DFT size = 256 samples; spacing = 188 Hz. The contrast was kept fixed at 93, and brightness was adjusted manually for each recording to give good visual separation between the song and the background noise. Strophes consist of notes, and different strophes are separated from each other by silent pauses. The following variables were measured for each strophe: duration (s), maximum frequency (Hz), and minimum frequency (Hz). The duration of intervals between strophes was also measured (s). In the simple parts of "Mara Lark" and M. cheniana songs (see below), the number of blocks of repeated notes (phrase) inside a strophe was counted. All analyses of songs were done by the same person (S.C.).
We used linear discriminant analysis (LDA) to assess similarity and clustering in song structure in multivariate space between "Mara Lark", M. cheniana, M. albicauda, and M. cantillans based on the above measurements. In each analysis, we used correlation clustering to identify and remove strongly intercorrelated (at r > 0.6) variables. In the LDA analysis, all variables were scaled to zero mean and unit variance due to their greatly different scales of measurement. Confusion matrices were used to assess the discriminatory power of the models.
To determine the identity and frequency of mimicked species in "Mara Lark" and M. cheniana songs (see below), we counted the number of times within a recording that a certain species was mimicked. For the song mimicry analysis, we selected recordings of 7 "Mara Lark" individuals, 6 from Kenya (the same 4 individuals from the sound analysis cited above, plus another 2; 8 recordings in total) and 1 from Tanzania. We also selected recordings of 24 M. cheniana individuals (the same 5 individuals from the sound analysis cited above, plus another 19; 32 recordings in total) (Table S1). This analysis was performed by three individuals for South African M. cheniana (D.E., J.G. and D.d.S.) and by one person for "Mara Lark" (B.W.F.).

Genetic analyses
A faecal sample of a "Mara Lark" was collected in Mara Naboisho Conservancy on 18 September 2020 immediately after it was deposited on a rock (Fig. 2c). This sample was dried and stored in an envelope at room temperature. DNA was isolated from the faecal sample using the QIAamp fast DNA Stool mini kit (QIAGEN, Valencia, CA, United States) according to the protocol: "Isolation of DNA from Stool for Human DNA Analysis" from the QIAamp fast DNA Stool mini handbook version February 2020. Two short overlapping fragments of the mitochondrial cytochrome b gene (Cytb) were amplified by means of Monoplex PCR and PCRs were run on a GeneAmp® PCR System 9700 (Online Resource 1 for primer sequences). The PCR product was visualized using the QIAxcel and purified using the QIAquick® PCR purification kit (QIAGEN) according to the protocol from this kit. For each of the PCR fragments, a sequencing PCR was performed on a pre-heated GeneAmp® PCR System 9700 using an input of purified PCR-product samples. PCR product was purified using the DyeEx® 2.0 Spin kit (QIAGEN) following the protocol for Dye-Terminator Removal. Purified sequencing PCR product was run on an AB3100 Genetic Analyzer. The two overlapping sequences were aligned and concatenated in BIOEDT 7.2.5. To exclude premature (numt induced) stop codons, open-reading frame finder (https:// www. ncbi. nlm. nih. gov/ orffi nder/) was used. All fragments aligned well against M. cheniana and M. albicauda, and there was no evidence of a chimeric sequence.
Four toepad samples of M. albicauda, including the holotype (ZMB 49.225) from Tanzania, were obtained (Table S2). These were extracted following the protocol described in Irestedt et al. (2006). Subsequently, Illumina library preparation followed the protocol published by Meyer and Kircher (2010), including pre-treatment with USER enzyme (New England BioLabs) to reduce post-mortem DNA damage patterns. In total, 11 samples with four sequencing libraries per sample (dual indexes) were then pooled in equimolar ratios and sent to SciLifeLab, Stockholm for paired-end sequencing using Illumina Next-Generation Sequencing (NGS) technology on one lane NovaSeq sequencing machines. Cytb sequences from two fresh samples of M. cheniana were obtained from unpublished wholegenome sequences (Alström et al. in prep.; Table S2). The raw reads were trimmed with Trimmomatic 0.36 (Bolger et al. 2014) by removing adapters identified as overrepresented sequences in FastQC 0.11.9 (Andrews 2010) and performing a 4-base wide sliding window trimming, cutting when the average quality per base dropped below quality score 15. Bases at the start and end of the read were cut if they were below a threshold quality score of 3. Reads shorter than 50 nucleotides were discarded. The trimmed reads of M. cheniana DZUG U5759 were mapped to an Eremophila alpestris mitochondrial reference sequence in GenBank (NC 948,479) in Geneious 9.1.2 (www. genei ous. com) with the Geneious mapper, medium-low sensitivity, iterated 10 times. The resulting consensus sequence was edited for quality and used as a template for a second round of mapping of M. cheniana DZUG U5759, using the same Geneious settings. The trimmed reads of M. cheniana DZUG U5760 and the four M. albicauda samples were mapped to the same template. The Cytb sequences were then extracted from the final respective consensus sequences supported by a sequencing depth of at least 10X. Cytb sequences of all other potential close relatives (except M. pulpa, for which no sequence data are available; corresponding to clade B1b in Fig. 1 in Alström et al. 2013), including one additional M. cheniana, as well as a few outgroup species (corresponding to clades B1a and B1d in Fig. 1 in Alström et al. 2013) were downloaded from GenBank (Table S2).
Sequences were aligned in SeaView 5.0 (Gouy et al. 2010). The best-fit model for phylogenetic analysis was determined by the Bayesian Information Criterion in jModel-test2 (Darriba et al. 2012). Phylogenetic analysis of the Cytb matrix was run under the best-fit model (HKY + G), a Yule speciation model, a strict molecular clock (with rate 1.0), and default priors in BEAST 1. 10.4 (Suchard et al. 2018). 20 million generations were run, with trees sampled every 10,000 trees. Convergence to the stationary distribution of the single chains was inspected in Tracer v. 1.7.1 (Rambaut et al. 2014) using a minimum threshold for the effective sample size (> 200). The joint likelihood and other parameter values reported large effective sample sizes (> 1,000), and the trace plot had the shape of a "dense, straight, furry caterpillar". Good mixing of the MCMC and reproducibility were established by multiple runs from independent starting points. The first 20% of the sampled trees were discarded as burn-in, well after stationarity had been reached, and the posterior probabilities were calculated from the remaining samples. Trees were summarized using TreeAnnotator v. 1.7 (in the BEAST package), choosing "Maximum clade credibility tree" and "Mean heights" and displayed in FigTree 1.4.4 (Rambaut 2016). Both jModeltest and BEAST were run on the CIPRES portal (Miller et al. 2010).
To evaluate how the "Mara Lark" differed from M. cheniana and M. albicauda in the 387-bp Cytb sequence that was available for the "Mara Lark", as well as how other closely related lark taxa differ in the same region, an alignment was created for this stretch for the "Mara Lark", three M. cheniana, four M. albicauda, a single M. cantillans marginata (from Tanzania), a single M. javanica williamsoni (from Australia), three Horned Lark Eremophila alpestris alpestris (from Newfoundland, Canada), and four Horned Lark Eremophila alpestris brandti (from Gobi, Mongolia) (Table S2).

Morphology
Morphological comparisons between the "Mara Lark" and M. cheniana from South Africa and M. albicauda from East Africa were based on c.15 photos of the former and field observations (latter of course never side-by-side); nearly all the photos are available at Macaulay Library (https:// www. macau layli brary. org), and a selection is shown in Fig. 2. It was not considered meaningful to try to quantify the differences, which mainly concerned minor differences in colour, and which were also affected by photos being taken under different light conditions, of birds at different stage of wear, etc. Measurements of specimens of M. cheniana and M. albicauda were taken in the Natural History Museum, Tring, UK by P.F.D., but no "Mara Larks" were available for comparison.
The song of M. albicauda ( Fig. 3a-b) is usually composed of well-defined strophes that are separated by silent pauses of rather regular length; when given in flight (see below), the strophes average shorter (0.67-5.05 s, mean 2.75 ± 0.90 s, n = 37 strophes from 3 individuals) and the pauses shorter (0.78-3.42 s, mean 2.34 ± 0.79 s, n = 32 pauses from 3 individuals) than when given from a perch (strophes: 0.11-27.07 s, mean 4.40 ± 3.80 s, n = 95 strophes from 5 individuals; pauses: 0.30-18.61 s, mean 2.64 ± 3.36 s, n = 89 pauses from 5 individuals) (Table S3). The strophes consist of short, scratchy notes, which are sometimes interspersed with short, soft notes. Individual notes can be very different from each other in terms of frequency and shape, and usually, all notes in a strophe are unique as seen on a sonogram, although a few similar-sounding notes are often given in a sequence; occasionally, a single note may be repeated a few times before switching to another one. The notes are delivered at a rather slow, steady, chattering speed, usually with a slightly hesitant beginning; sometimes, the strophes are preceded by single notes that are well separated from the rest of the strophe. The notes are frequently given on somewhat rising and falling scales, producing an undulating rhythm, with an abrupt ending by a rather short, rather low-pitched note. In our sample, none of the strophes was repeated identically, and accordingly, each male has a very large repertoire of strophes. Unlike in "Mara Lark" and Melodious Lark, Fig. 3 Songs of White-tailed no apparent mimicry was included in the analysed songs of this species. Whilst the song is usually delivered in flight, it is more rarely given from the top of a low bush. In display flight, it traces a very large ellipse and may climb until lost to the naked eye. Hovering in one spot is only momentary after each repeated song phrase. The complete song-flight rarely lasts more than 15 min, but covers a large territory.
The song of "Mara Lark" is easily distinguishable from that of M. albicauda by being faster, more varied (often including mimicry) and more continuously flowing, with short, irregular pauses (most different compared to M. albicauda in flight). It is highly varied, and switches back and forth between monotonous spells of short rattling series of simple notes (Fig. 3e-f) and spells of much more complex notes, including mimicry (Fig. 3 i-j). The simple parts usually consist of rattling series (phrases) (1-6, mean 4.51 ± 0.70, in n = 217 phrases from 4 individuals) of identical or marginally different short, dry, scratchy notes that are separated by short silent pauses (Table S3). Usually, two, sometimes three, phrases with different notes are given in rapid succession, separated by very brief intervals, and these "pairs" of phrases are separated by slightly longer silent pauses (0.52-7.13 s, mean 0.92 ± 1.01 s, n = 102 pauses from 4 individuals) (Table S3); the pitch usually differs slightly between the phrases in these "pairs" (first one being slightly higher pitched than the second or vice versa). The more varied parts are much more complex and melodious, and consist of a medley of highly varied notes of different pitch, duration, and quality, including their own improvised notes, masterful mimicry (see below), and notes that are considered to be modifications in tone and tempo of mimicked songs. The notes are usually arranged in series of various lengths that are separated by short pauses, but can also be rather continuously flowing; series of rapid repetitions of single notes are common. Similar-sounding, almost identical, series are often given a few times in succession. As in M. albicauda, the song can be given both in flight and from a perch such as the top of a bush, and there is no apparent difference in the frequency of the two song types depending on whether the song is given in flight or when perched. Unlike the display flight of M. albicauda, the bird remains over a small territory and hovers for long periods, tracing small circles and is rarely high enough not to be discerned by the naked eye. Such restricted range song-flights may exceed an hour in duration. The single recording of a bird by C.C. in 2013 from the Serengeti in Tanzania (ML274636) was not included in the vocal analyses because of inferior quality, but vocally confirms the presence of the "Mara Lark" as being in the Serengeti. Finally, "Mara Lark" singing males form loose associations in the same general area while M. albicauda in Kenya are always found singing in very low numbers with no local aggregations at any time of the year.
The song of M. cheniana from South Africa (Fig. 3k-l) is very similar to that of "Mara Lark" in its structure as both populations alternate between series of simple notes and more complex parts which include a high number of mimicked species. Because of the enormous variation both in a single recording and among recordings of different individuals, it is very hard to detect and quantify differences, if any, between M. cheniana and "Mara Lark". However, the simple parts (Fig. 3g-h) usually comprise either rattling series of up to five identical notes (1-5, mean 3.07 ± 0.84 in n = 415 phrases from 5 individuals) with rather uniform pauses between the series (i.e., not "paired series") or more complex series of notes (Table S3). In the more complex parts of the song, identical series of notes that are repeated a few times are not uncommon, unlike in our sample of "Mara Lark", though this might be an effect of our smaller sample size of the "Mara Lark". As in the above, the song may be given from a perch or in flight.
The songs of each of M. albicauda, "Mara" and Melodious Larks differ clearly from the song of M. cantillans, which occurs in the same geographical area as the two former species, although almost invariably separated by habitat. The song of M. cantillans from African populations (Fig. 3c-d) consists of short strophes (0.55-3.96 s, mean 2.16 ± 0.67 s, n = 64 unique strophes from 4 individuals) that are well separated by silent pauses (0.64-6.52 s, mean 2.48 ± 1.24 s, n = 94 pauses from 4 individuals) (Table S3). They usually begin somewhat falteringly with a series of a single note type and then change to a fast rattle consisting of one or two series of notes. The repertoire size seems to be much smaller than in the above-mentioned taxa. Unlike in "Mara Lark" and Melodious Lark, no apparent mimicry was included in the analysed songs of this species.
In the LDA, "Mara Lark" and M. cheniana are broadly overlapping along LD1 (which explains 91.9% of the variance), whereas they are fully separated along LD2 (explaining 6.4% of the variance) (Fig. 4). In contrast, M. albicauda and M. cantillans overlap on both axes, while they are widely separated from "Mara Lark" and M. cheniana along LD1. All songs were correctly classified, except one M. albicauda, which was classified as M. cantillans (Table S4).

Song mimicry in "Mara Lark" and Mirafra cheniana
As stated above, both "Mara Lark" and M. cheniana from South Africa include much mimicry in their songs. Table 1 lists the species that have been identified as mimicked in the songs of "Mara Lark" and M. cheniana from South Africa. The birds that are imitated have apparently been learned from the local avifauna, and accordingly, there are some geographical differences between "Mara Lark" and M. cheniana from South Africa. In the selected recordings, a total of 80 different species belonging to 22 different families are imitated. The higher number of mimicked species by M. cheniana from South Africa is probably due to a larger number of available recordings for this population. The most frequently mimicked species by both populations belong to the families Cisticolidae, Alaudidae, and Hirundinidae; in particular, three species are mimicked consistently: Desert Cisticola Cisticola aridulus, Quailfinch Ortygospiza atricollis, and Grassland (African) Pipit Anthus cinnamomeus. Songs of Rufous-naped Lark Mirafra africana and Redcapped Lark Calandrella cinerea are incorporated by both "Mara Lark" and South African M. cheniana. In both "Mara Lark" and M. cheniana, mimicked songs are (i) often interspersed with their own songs and calls, (ii) characterized by frequent changes of the mimicked species, and (iii) interrupted by improvisation, unidentifiable song series, constituted by modified borrowed strophes.

Mitochondrial DNA
The Cytb tree (Fig. 5) identified the single "Mara Lark" as sister to the three M. cheniana from South Africa with posterior probability 1.00. The four M. albicauda also formed a strongly supported clade. The latter was sister to the M. javanica + M. cantillans clade, but with poor statistical support (posterior probability 0.55).
The 387 bp sequence of the "Mara Lark" differed from the three South African M. cheniana at 5 positions (two individuals) or 7 positions (third individual). In comparison, three of the M. albicauda had identical sequences to each other in the same 387 bp sequence, whereas the fourth individual differed from the others by a single bp, and the single M. javanica williamsoni (Australia) and M. cantillans marginata (Tanzania) differed from each other at 6 positions in the same stretch of DNA. The three E. alpestris alpestris (Newfoundland, Canada) and four E. alpestris brandti (Gobi, Mongolia) had two fixed single-bp substitutions in the same 387-bp stretch of Cytb, with four additional, autapomorphic single-bp substitutions among these individuals (Online Resource 1).  Morphology "Mara Lark" and M. cheniana are nearly identical in plumage, but "Mara Larks" appear to show more variation in plumage colour. Some worn "Mara Larks" show very white underparts and rufous lesser coverts, and these traits have not been recorded in M. cheniana in southern Africa. Mirafra albicauda is extremely similar, but is generally less warmly coloured overall, especially on the crown, nape, upperparts, and breast/flanks, and tends to show darker and more uniformly coloured centres to the tertials. The colour differences are most apparent in fresh plumage and can be negligible when worn. As we have not seen any specimens (alive or dead) of birds in the hand of "Mara Lark", we cannot say whether there are differences in morphometrics.

Habitat
The habitat of "Mara Lark" and M. cheniana in South Africa is basically similar. It is characterized by extensive open, moderately grazed, often dominated by Themeda triandra, grasslands with minimal or no bush cover. In the west part of its South Africa distribution range, M. cheniana can also be found in drier grasslands. "Mara Larks" are found at 1500-1800 m above mean sea level over both vertisols (black cotton clay) and planosols. "Mara Larks" appear to be resident, but during long wet periods, we have observed local irruptions into areas that recovered from recent overgrazing. Mirafra albicauda favours both lightly bushed, scrubby grasslands and more open grasslands, often over vertisols, and is very rarely seen in the open (even less frequently than "Mara Lark" and M. cheniana).

Discussion
Our results based on analyses of songs and mitochondrial DNA and studies of photographs taken in the field convincingly show that the lark population in south-western Kenya ("Mara Lark") that has always been treated in the literature up to the end of 2020 as M. albicauda (Britton 1980;Zimmerman and Turner 1996;Stevenson and Fanshawe 2002; Ryan 2020a) is certainly not M. albicauda, but instead matches M. cheniana, which has previously been considered a Southern African endemic. That this population differs from M. albicauda has been known for some time based on perceived differences in song (Finch 2002(Finch , 2009  Numbers indicate the total number of times within our selected song recordings that a particular species was mimicked ("Mara Lark" n = 9 recordings, South Africa Melodious Lark M. Cheniana n = 32 recordings). Mimicked species present in Kenya but not in South Africa are highlighted in red, species found in South Africa but not in Kenya are highlighted in blue, and species shared by both regions are printed in black The recognition that the "Mara Lark" is in fact M. cheniana instead of M. albicauda constitutes a marked range extension of M. cheniana from South Africa and the isolated population in central Zimbabwe. The southernmost of the point localities of "Mara Lark" shown in Fig. 1 lies almost exactly 2,000 km north of the isolated population in central Zimbabwe, and 2,500 north of the northernmost population in South Africa. Whilst the "Mara Lark" was known by B.W.F. to be in the Serengeti in Tanzania almost from the time it was first recognised as an unusual form, there is now the evidence backing the Tanzanian presence from photographs, a sound recording and a video from the Serengeti National Park, all confirming that M. cheniana occurs also in Tanzania and is at least partially sympatric with M. albicauda. It seems possible that further explorations will show that M. cheniana is even more widely distributed. Although Stevenson & Fanshawe (2020) show only "Mara Lark" to be present in the Serengeti, multiple sound recordings by C.C. show that M. albicauda is common there and thus occurs sympatrically with "Mara Lark"; further research is needed to determine the fine-scale habitat differences between these two species in the Serengeti.
More data on the "Mara Lark", such as specimen and more comprehensive genetic data, will be needed to evaluate whether this East African population is sufficiently differentiated from the South African and Zimbabwe populations of M. cheniana to require a subspecific name. The differentiation of the "Mara Lark" from South African birds in the single available, extremely short mitochondrial sequence was similar to that between M. cantillans marginata from Tanzania and M. javanica williamsoni from Australia, which are usually treated as separate species (e.g., Clements et al. 2021, Gill et al. 2022, and larger than that between E. alpestris alpestris from Newfoundland, Canada and E. alpestris brandti from Gobi, Mongolia, which are markedly different from each other in plumage (Beason 2020). This indicates that the isolation between the "Mara Lark" and South African M. cheniana might be substantial. However, more genetic/genomic data and larger sample sizes are required to evaluate this. The differences in song found in the LDA are difficult to interpret because of the small sample sizes and great variation in both populations, which is also affected by which species are mimicked.
Three other lark complexes, Rudd's Lark Heteromirafra ruddi-Archer's Lark H. archeri, Spike-heeled Lark Chersomanes albofasciata-Beesley's Lark C. beesleyi and Fawncoloured Lark Calendulauda africanoides-Foxy Lark C. alopex, have similarly disjunct distributions between southern and eastern Africa as "Mara Lark"-M. cheniana (e.g., Gill et al. 2022), but are probably more anciently diverged, Fig. 5 Mitochondrial cytochrome b tree including the "Mara Lark" and all other potential close relatives (except Friedman's Lark Mirafra pulpa, for which no sequence data are available; corresponding to clade B1b in Fig. 1 in Alström et al. 2013), inferred using Bayesian Inference in BEAST under the HKY + G substitution model, a Yule speciation model, and a strict molecular clock with rate 1.0. Posterior probabilities are indicated at the nodes approximately 6.5, 4, and 2 million years ago, respectively Stervander et al. 2020). This may reflect different timings of movements along the "arid corridor" of eastern Africa as it has expanded and contracted over time.
The differences in song between the "Mara Lark" and southern African M. cheniana are partly based on the available subjects for mimicry. Peacock (2012) states that the three prime species mimicked in South Africa are Little Swift Apus affinis, Grassland (African) Pipit Anthus cinnamomeus, and Quailfinch Ortygospiza atricollis, which are frequently mimicked in our recordings from Kenya.

Conservation
Whilst the known distribution of the "Mara Lark" is geographically small, we believe that most of the population lies within formally gazetted protected areas and communityowned conservancies. More work is needed to ascertain the size of this population, but based on field observations, it is believed to be common in appropriate habitat.