The successful reintroduction of African wild dogs (Lycaon pictus) to Gorongosa National Park, Mozambique

Large carnivores have experienced widespread extirpation and species are now threatened globally. The ecological impact of the loss of large carnivores has been prominent in Gorongosa National Park, Mozambique, after most were extirpated during the 1977–92 civil war. To remedy this, reintroductions are now being implemented in Gorongosa, initiating with endangered African wild dogs (Lycaon pictus), hereafter ‘wild dogs’. We describe the first transboundary translocation and reintroduction of founding packs of wild dogs to Gorongosa over a 28-month study period and evaluate the success of the reintroduction based on five key indicator categories. We also assess how wild dog space use and diet influenced their success. We found that pre-release, artificial pack formation in holding enclosures aided group cohesion and alpha pair establishment. Post-release, we also observed natural pack formations as a result of multiple dispersal events. Founder and naturally formed packs produced pups in two of the three breeding seasons and packs successfully recruited pups. Survival rate for all wild dogs was 73% and all mortality events were from natural causes. Consequently, the population grew significantly over the study period. All indicators of success were fully achieved and this study documents the first successful reintroduction of wild dogs into a large, unfenced landscape in Mozambique and only the second on the continent. Potential mechanisms underlying these early successes were the avoidance of habitats intensively used by lions, dietary partitioning with lion, avoidance of human settlements, and Gorongosa’s management strategy. We predict further population expansion in Gorongosa given that 68% of the park is still unused by wild dogs. This expansion could be stimulated by continued reintroductions over the short- to medium-term. Recovery of wild dogs in Gorongosa could aid in the re-establishment of a larger, connected population across the greater Gorongosa-Marromeu landscape.


Abstract:
Large carnivores are important components of healthy ecosystems. However, carnivores are threatened globally, surviving mostly across fragmented ranges because of widespread extirpations. The effect of the loss of large carnivores has been prominent in Gorongosa National Park, Mozambique given most of the large carnivore guild was lost during the 1977-92 civil war. To restore the guild of large carnivores to Gorongosa, reintroductions are now being implemented with a focus recently on endangered African wild dogs ( Lycaon pictus ). Here, we describe the transboundary translocation and reintroduction of two packs of wild dogs to Gorongosa over a 28month period and use data on artificial group formation, demography, diet, space use and management to evaluate the success of the reintroduction based on 16 indicators. We artificially formed packs in holding enclosures prior to release which aided in group cohesion. Post-release, we also observed natural pack dynamics with dispersal resulting in new pack formations. Packs bred in two of the three breeding seasons, recruited new individuals, and exhibited a 73% survival rate with no human-related mortalities. Wild dogs also avoided lions and effectively hunted preferred prey. Overall, we show that 15 of the 16 indicators were fully achieved (one partially achieved) and that the reintroduction was successful. This represents the first successful reintroduction of wild dogs into an open landscape. We expect further the population expansion based on extensive vacant areas (68% of Gorongosa unused by wild dogs) and suggest that this expansion could be further stimulated by continued reintroductions over the short-term. Such an expansion has the potential to establish a large and connected wild dog population across the greater Gorongosa-Marromeu landscape which has important regional implications for the species. We also suggest that our study can be used as framework for evaluating future wild dog reintroductions and applied to other carnivore reintroductions.    The primary aim of this study was to document the first two-years of the reintroduction 105 of wild dogs to GNP. We focused on describing and analyzing the effectiveness of artificial 106 group formation in pre-release enclosures and their subsequent post-release demography, diet 107 and space use. Our secondary aim was to use these data to evaluate if the reintroduction had 108 been successful at a short-term scale [40] and to formalize the baseline against which to monitor 109 progress over future years. We defined broad goals of success that included artificial formation 110 of packs, natural pack formation dynamics, reproduction and recruitment, survival, diet, packs 111 establishing home-ranges, hunting, and a limited impact of humans on individual survival. We 112 outline a range of key biological, ecological and management indicators of short-term 113 reintroduction success (Table 1)  The Gorongosa-Marromeu landscape located in Central Mozambique (Fig 1) is 125 composed of a highly diverse mosaic of habitats [75,76]. This landscape is managed as a 126 contiguous 20,000 km 2 complex of national park, national reserve, buffer-zones, forest reserves 127 and hunting concessions or "coutadas" (Fig 1). Designated as a potential stronghold for lions 128 [77,78], the Gorongosa-Marromeu landscape, is also habitat for leopard (Panthera pardus), 129 spotted hyenas (Crocuta crocuta) and wild dogs which were all present pre-civil war [35] but 130 are currently ephemeral, absent or occurring at very low densities. 131 The unfenced, 4,067 km 2 GNP is situated in the core of a contiguous landscape of 132 protected areas known as "Gorongosa-Marromeu" (Fig 1; -18 Post-release monitoring 291 While we attempted to locate the packs daily, this was reliant area on accessibility 292 constrained by flooding, weather and personnel availability. We used the fitted GPS and VHF 293 collars (see Collars and GPS data section below) to locate the reintroduced founders. Once 294 located, wild dogs were observed from a 4x4 vehicle with 1-2 observers recording data. At 295 each sighting, we recorded (1) number, identity, sex and age of individuals, (2) health condition 296 of individuals and any injuries, (3) dominance (known from overmarking, mate-guarding and 297 scent marking), (4) pregnancies, (5) hunting success, and (7) any behavioural interactions with 298 conspecifics and lions. We also relied on opportunistic field ranger and tourist reports to 299 supplement our dataset. As each wild dog is identifiable from unique coat patterns, we utilized 300 against the photographic identification kits constructed during translocation phases. At the 302 worst resolution, each individual was seen and recorded at least once per month. Consequently, 303 we built individual and group-level life histories for wild dogs in GNP at a monthly resolution 304 from June 2018 until 30 September 2020, encompassing 28 consecutive months since the first 305 pack was released. We then used the demographic data to assess monthly (1) group 306 composition, (2) group size, (3) reproduction and recruitment, (4) inter-group movement 307 dynamics, (5) survival, (6) mortality, (7) GPS location and (8) kills. 308 Breeding 309 We used VHF collars, GPS collars and direct observations to determine the timing and 310 location of denning. Birth females were known from monitoring prior to denning based on 311 direct observations of alpha pairing, mating, visible signs of pregnancy, disappearance 312 underground, not joining pack hunting events and packs returning to the same site twice a day 313 (indicative denning behaviour). Thus, we were able to confidently assign birth to these females 314 while also recording the birth month and coordinates of dens. We estimated litter size by 315 placing camera traps at identified den sites (location determined from GPS collar clustering 316 and direct observations) to record images and videos. We rotated cameras on a weekly to 317 biweekly schedule as batteries permitted. This also allowed us to add photographs from pups 318 (i.e. non-founders) into our individual life history database and in our identification kits where 319 we assigned age and sex based off camera trap images and subsequent direct observations of 320 pups at dens. We generally recorded pups emerging from dens approximately 3-4 weeks after 321 females went underground. This is the age when pups first emerge from underground [68] and 322 we are confident that this represents actual litter size as per other wild dog studies [53, 69, 70, 323 85, 86]. We also counted the number of surviving pups for each litter to three, six, nine and 12 324 months as important age milestones to determine the proportion of pups raised. We defined 325 recruitment as the number of pups raised to 12 months old (i.e. yearling) as a proportion of the 326 population size (adults and yearlings) in the focal month. The proportion of pups surviving to 327 six, nine and 12 months does not include the six 2020 litters because we could not determine 328 survival to these ages at the end of this study. 329 Population size, survival and mortality 330 We used three age class categories for wild dogs including pup (< one year old), 331 yearling (1 -2 years old) and adult (> 2 years old). We defined a pack as a group of wild dogs 332 with at least one adult male and one adult female, dispersal group as a cohort of a single-sex 333 and splits as a mixed sex cohort that permanently left their most recent pack. We generally 334 classified pack and population size as the sum of adults and yearlings unless otherwise stated 335 where the count also included pups. We determined two types of population density; one as the 336 population size per 100 km 2 and the other as the number of packs per 100 km 2 to allow for 337 individual and group-level comparisons with other populations. Through the field data 338 collection, we assigned alive or dead to each individual at the end of each month. From these 339 monthly data of individual fate, we were able to estimate survival as the number of individuals 340 alive at the end of the study relative to the total number of individuals recorded during the 28 341 months. This method included both founders and non-founders (i.e. pups). For individuals that 342 were not alive at the end of the study, we assigned mortality and the month of mortality as any 343 individual observed to have died and the carcass or collar recovered. If an individual was not 344 confirmed to have died and was not alive at the end of the study, we assigned it as disappeared 345 and estimated disappearance month as one month after last confirmed observation. For 346 confirmed mortalities, we coded the cause as human (road accident, indirect snaring, direct 347 shooting), natural (predation, intra-specific, hunting) or disease (rabies, distemper) as defined 348 by Woodroffe, Davies-Mostert (52). The sex and age class of dead and disappeared individuals 349 were also recorded. We also recorded changes in group association as evidenced by successful 350 dispersal (i.e. natural pack formation with opposite sex group), unsuccessful dispersal (have 351 yet to form new pack) and splits. 352

353
We recorded all successful hunts or located carcasses killed by wild dogs. At each kill 354 or carcass, we recorded the prey species and age class (where possible). We used the GNP 355 biennial herbivore census data to determine the amount of available prey biomass [87]. We 356 then applied Jacob's Index of selection [88] to assess prey selection by wild dogs. We separated 357 waterbuck biomass into two distinct categories related to their distribution in GNP: (i) 358 grassland/woodland waterbuck and (ii) floodplain waterbuck because of the large difference in 359 waterbuck biomass relative to each habitat type [87]. We also used lion kill data between 2012 360 and 2020 (available from the GNP Carnivore Recovery Unit) to assess lion prey selection using 361 Jacob's Index of selection which allowed us to contrast prey selection between wild dogs and 362 lions. 363 Tracking; 8x Animal TrackEm). Thus, multiple individuals per pack were collared 370 simultaneously. However, we only used GPS data from one individual per pack at any time to 371 reduce pseudoreplication. We also attempted not to select collar data from breeding females to 372 avoid small and non-representative territories during denning as breeding females tend to stay 373 near the den for large portions of the denning season [68]. Consequently, we included data from 374 nine of the 30 collared individuals (S1 Table). GPS collar fix schedule varied per collar 375 recording optimally six fixes per day. This full dataset contained 7,944 GPS fixes. We truncated 376 this dataset, to include a maximum of four fixes per group per day, representing two resting 377 periods and two movement periods, each separated by at least six hours. We considered these 378 temporal intervals as adequate to avoid spatio-temporal auto-correlation between consecutive 379 locations without losing relevant information on the ecology of the animals [68]. This truncated 380 dataset included 3,048 fixes with a mean number of 339 fixes per group (S1 Table). African Wildlife Tracking satellite collars (S1 Table). These collared lions account for 392 approximately 10% of the population and 68% of the extent of area used by lions in the park 393 (P. Bouley, unpublished data) and we acknowledge there are uncollared individuals/groups 394 within GNP over this study period. Lion collars were programmed to record fixes every 2-4 395 hours and we truncated the lion GPS dataset in the same way we did for the wild dogs to reduce 396 pseudoreplication and ensure sampling in the same time periods as the wild dog population to 397 enable meaningful analyses. 398 399 Territory estimation KUDs and territory size 401 We used the truncated spatial dataset for wild dogs and lions (S1 Table)  We used the getverticeshr function in adehabitatHR package in R and specified the WGS84 410 geographic projection to extract the area size (km 2 ) enclosed by the 95% and 50% isopleths 411 extracted from KUDs for each half of the study and for the entire study period for wild dogs. 412

Collars and GPS data
For lions, we only created KUDs for the entire study period and the extracted the 95% and 50% 413 isopleths. The 95% isopleths represented the total territory area (excluding 5% of outliers) and 414 the 50% isopleth represented the core territory area of intensive use. 415

416
To determine if the wild dog population increased its range over time, we merged the 417 extent of the group-level 95% isopleths for the first 14 months of the study period and for the 418 second 14 months of the study period. We also estimated the proportion of the park used by 419 wild dogs using the merged 95% isopleths for each group covering the entire study period. We 420 also developed monthly 95% minimum convex polygons (MCPs) for packs using the mcp 421 function from the adehabitatHR package v0.4.15 [95]. We did this to represent the total extent 422 of the area covered by wild dogs without worrying about use. 423 Overlap area 424 We assessed spatial interactions between neighbouring wild dog dyads using overlap 425 as an index for territoriality [96,97]. We estimating overlap area (km 2 ) between a pack's range 426 for each neighbouring dyadic association to determine what proportion of each focal range 427 overlapped with a neighbour's. We estimated this for both the 95% and 50% isopleths extracted 428 from the KUDs created over the whole study period. We used the merged 95% and 50% 429 isopleths for wild dogs and the merged 95% and 50% isopleths for lions to estimate the amount 430 of overlap between these two carnivores. We did this by relating the proportion of overlap in 431 area size between the 95% and 50% wild dog territories to that of the 95% and 50% lion 432

territories. 433
Habitat use 434 We determined the number of fixes per wild dog and lion group from the truncated 435 dataset that fell within each of the two broad vegetation categories (woodland or floodplain). 436 From this, we determined the proportion of fixes for each carnivore population that fell within 437 each vegetation type. We were thus able to contrast species-specific habitat use. We excluded 438 fixes that fell outside the official protected area boundary of the park for the habitat use 439 analysis. We also determined the number of wild dog dens in each of the two broad vegetation 440 types. Some female wild dogs denned in the same site as other females in the pack (S2 Table)  441 and we did not include these duplicates in this analysis (i.e. n = 23 unique den locations). We did not observe signs of aggression during their entire period in the enclosure and 504 we observed close association between one female (Matenga) and one male (Nhamaguena) that 505 suggested the formation of an alpha pair (indicator 1B, Table 1). We did not observe mating 506 during this group's enclosure period. In contrast to the Gorongosa pack, post-release the 15 507 individuals only stayed together for four days before the first of two dispersal events occurred 508 (see Natural pack formation section below). Despite the dispersals, by the end of this study 509 period, one female and seven males of the original founders have remained together in the 510 Pwadzi pack (indicator 1A, Table 1). 511 512 Natural pack formation 513 We documented 12 of the 29 founder individuals change their group association (Fig  514   2; S3 Table). This was in the form of dispersals (n = 4 events) (indicator 2A, Table 1) and a 515 split (n = 1 event) from artificially formed packs that resulted in the formation of two new 516 packs (Fig 2; S3 Table) (indicator 2B, Table 1). We observed one subordinate female from the 517 Pwadzi pack to leave the pack when she started denning in May 2020 (Fig 2; S2 and S3 Tables). 518 While this is technically a dispersal, this female is currently only accompanied by her 519 remaining pups and it remains to be seen if she rejoins Pwadzi pack (S2 and S3 Tables) 520 We recorded 11 pregnancies from seven different females but we only confirmed nine 530 litters. For the two unconfirmed litters, the pups never emerged from underground dens. We 531 recorded these 11 pregnancies in each of the denning seasons in 2018, 2019 and 2020, but only 532 recorded litters emerging in the 2019 and 2020 seasons (indicator 3A, Table 1). Mean litter size 533 for the nine confirmed litters was 9.11 ± SE1.06 (range 5 -15). Birth was highly seasonal, with 534 all litters recorded between May and July (χ 2 (11) = 51.07, p < 0.001). Multiple maternities per 535 pack was standard for this population with seven of the nine confirmed litters being born when 536 more than one female per pack gave birth (S3 Table).
(1) = 4.02, p = 0.04). 554 We observed 30 of the 111 individuals to have died or disappeared from the population 555 (27% of the population missing or dead). The majority of these were pups (90%) and it is likely 556 that these were all mortalities although only one pup carcass was found. Of the three confirmed 557 adult mortalities we documented, one male was killed by lions, a second suspected killed by 558 lions (based on post-mortem and observed signs of trauma on front leg-bones, although 559 inconclusive given the time that had elapsed) and one female was found dead in a den. One 560 male disappeared during the study and we could not confirm if he had died or dispersed. 561 Overall, we found no human-induced mortality on this wild dog population (indicator 9B, 562 The wild dog population in GNP has grown since their reintroduction in June 2018 (Fig  566   3). Over the 28 month study period, we identified 111 unique individuals in the population (29 567 founders, 82 non-founders) with population size peaking in June 2020 at 102 individuals (49 568 adults and yearlings, 53 pups) in four packs and three dispersal groups ( Table 2)  We observed 103 kills from nine prey species to have been killed by wild dogs. The 587 majority of kills comprised bushbuck (40%) and waterbuck (30%) but wild dogs only 588 positively selected bushbuck while avoiding all other species (Fig. 4, S4 Table) (indicator 6A, 589 Table 1). We also recorded 118 lion kills but with marked differences to those of the 102 wild 590 dog kills regarding species and age class (Fig. 4, S4 Table). For example, lions did not predate 591 upon bushbuck whereas bushbuck makes up 40% of the wild dogs diet and is their preferred 592 prey (Fig. 4, S4 Table) (indicator 8A, Table 1). Additionally, wild dogs rarely predated on 593 warthog, while warthog comprises 58% of observed lion diet (Fig. 4, S4 Table). increased over the 28 month study period (F(1,26) = 75.11, p < 0.001, R 2 = 0.74) (indicator 7A, 613 Table 1) and does not appear to have reached an asymptote (Fig 5). of the park and appeared to avoid high human density areas (Fig 6) (indicator 9A, Table 1). 622 Wild dogs incorporated the release enclosure at some stage in their early release as part of their 623 territories, but all groups eventually established adjacent core ranges excluding the enclosure, 624 with the exception of the first pack released in 2018 (Fig 6). Territory size for the four packs 625 was 356 ± 62 km 2 (mean ± SE) and core territory size was 71 ± 14 km 2 (mean ± SE). Area used 626 for the two dispersal groups was 613 ± 92 km 2 (mean ± SE) and their core areas were 118 ± 12 627 km 2 (mean ± SE). The mean ratio of core to total territory size was 0.20 ± 0.01 km (mean ± 628 SE, range 0.18 -0.22) for all six groups of wild dogs in GNP (Fig 6) (indicator 7B, Table 1). 629 Wild dogs in GNP overlapped with conspecifics and thus did not maintain exclusive territories. 630 The mean proportion of overlap between pack territories was 0.21 ± 0.02 km 2 (mean ± SE) 631 while overlap in core territories was 0.07 ± 0.03 km 2 (mean ± SE). The two dispersal groups 632 overlap by 48% in the total territory area and 22% in the core. coalitions (n = 4) and dispersers (n = 5) (Fig 8). Overlap between wild dogs and lions was higher in the total territory areas (56% overlap in the 95% isopleth; Fig 8A) while core territory 665 areas were more exclusive between wild dogs and lions (17% overlap in the 50% isopleths; Fig  666   8B) (indicator 8A, Table 1) Table 1). Additionally, local vets and on-the-ground monitoring teams have 682 provided oversight of the recovering population and a wealth of data that has been essential in 683 the identification of challenges and towards evaluating successes/failures (indicator 10A, Table  684 1). 685 Overall, the reintroduction of wild dogs to GNP has been highly successful. Reserve -Creel and Creel (68); Hluhluwe-iMfolozi Park -Marneweck (50)), breeding 699 occurred in all packs, pups were recruited into the population, no human-induced mortalities 700 occurred and wild dogs effectively avoided lions through territory establishment and habitat 701 use. Consequently, we conclude that over this short-term period since the first reintroduction 702 (i.e. 28 months), this population of wild dogs has been successfully reintroduced and is still 703 within the expansion phase (continued increase in space use, variable population size etc.). This 704 highlights how quickly wild dogs that were naïve to GNP adapted to the local conditions [24]. 705 Density-dependent regulation of vital rates and the emergent population size does not appear 706 to be influencing this population yet, although anecdotally, competition at dens and between 707 pups seemed to have limited survival of the 2020 litters. 708 Wild dogs in GNP displayed two notable patterns that we suggest aided in rapid 709 population expansion. Firstly, we observed multiple maternities per pack more often than 710 alpha-only litters. Normally, breeding is monopolized by an alpha pair whereas shared 711 maternity is very rare [50, 102, 103]. While we acknowledge this is a small sample size of 712 multiple maternities (five females in two packs, across two years), without subordinate females 713 combination of both. Next-steps will be to develop population projections for GNP that account 739 for uncertainties relating to demographic stochasticity (i.e. chance fates for individuals) and 740 environmental stochasticity (i.e. large environmental fluctuations) to guide GNP management 741 on what the optimal population size is that warrants no further reintroduction effort. This, in 742 combination with carrying capacity models are key next steps to help guide medium-term 743 objective for the reintroduced wild dog population in GNP to occupy as much suitable habitat