A bird's‐eye view: Evaluating drone imagery for the detection and monitoring of endangered and invasive day geckos

Species monitoring can be strongly limited by terrain accessibility, impeding our understanding of species ecology and thus challenging their conservation. This is particularly true for living in the canopy, on cliffs or dense vegetation. Remote sensing imagery may fill this gap offering a cost-effective approach allowing to improve detection inaccessible areas. We investigated applicability drone-based Critically Endangered insular gecko (Phelsuma inexpectata) two invasive alien representing risk former (P. grandis P. laticauda). determined distance before inducing behavioural response caused drone's presence. All three study showed no escape behaviour presence until very close distances (mean inexpectata: 33.8 cm; grandis: 21.9 laticauda: 26.4 cm). then performed horizontal vertical approaches with drone, taking photos every meter starting at 10 m away from canopy edge determine an optimal while ensuring species-level identification. examined total 328 photos. found bimodality number detected geckos species, different individuals recorded between short intermediate distances. Therefore, we recommend 2–2.5 5 ideally facing sun low wind conditions. encourage application methodology Phelsuma spp., but also other similar size Le suivi des espèces est fortement limité par l'accessibilité au terrain, ce qui freine notre compréhension de leur écologie et Ceci s'applique particulièrement aux vivant dans la canopée, sur les falaises ou végétation dense. La télédétection ouvre possibilité combler ces lacunes tout en améliorant détectabilité endroits inaccessibles. Nous avons exploré faisabilité du drone d'un insulaire danger critique d'extinction deux autres invasifs représentent un risque pour dernier déterminé d'approche maximale avant d'observer une réaction à présence drone. Les trois étudiées n'ont montré aucune réponse très courtes cm moyenne). ensuite effectué approches horizontales verticales avec le prenant photo chaque mètre entre 1 bords canopée afin déterminer optimale identification d'espèce fiable. examiné trouvé bimodalité nombre détectés espèces, différents individus identifiés plans rapprochés intermédiaires. recommandons prise photos, respectivement idéalement dos soleil sous vent faible. encourageons l'application méthode toutes Phelsuma, mais aussi d'autres taille d'écologie similaire d'améliorer détection Population assessment endangered rare often multitude factors, including funding, observer experience, detectability, accessibility. Consequently, complex habitats dynamic habitat use are poorly understood conservation becomes challenging. The development novel methods cryptic threatened priority (Monks al., 2022). Insular endemic vulnerable (IAS). spread IAS generally facilitated lack surveillance efforts, preventing early detection, initial dispersal (Cuthbert rising economic costs biological invasions settlement preventive measures (Diagne 2021). Detection considerably improved methods. Recent technological advancements reduced electronic devices have contributed biodiversity monitoring. Novel methodologies such as computer assisted slow-speed road cruising (Jones 2022) camera-trapping (e.g., Deso 2022; Roesch 2021) already proven efficient. Drones been used estimate abundance, monitor individual animal disturbance related its presentations (Schad & Fischer, They were recently (Landeo-Yauri 2020; Varela-Jaramillo 2023), reptiles 2022), reptile tree (Aota Manapany day gecko, inexpectata Mertens 1966, Reunion Island. Its distribution restricted narrow strip along southern coastline. frequently uses screw pine Pandanus utilis, where it locally abundant (Bour 1995). throughout seasons, more frequent during winter (Choeur 2023). year-round remote protocol dedicated increase temporal resolution surveys, help understand species' ecology, ultimately management. Among few fragmented populations inexpectata, several reported sympatry that Madagascar giant Gray 1870 gold dust laticauda Boettger 1880 (Dubos, 2013; Porcel A colonization Manapany-les-Bains, stronghold has successfully controlled NGO Nature Océan Indien 2010 2012 not observed area ever since (M.A. Roesch, pers. obs.). Both thrive share resources, competition (Deso 2023; Hoarau 2021; raises concerns due larger size, imposing high predation smaller (Buckland 2014). established world (Dubos 2014, Fieldsend 2020, Krysko, 2019), strong invasion potential tropical islands spp. variety primary forests, shrub land, urban environment, agricultural areas (D'Cruze 2009; Dubos Beyond promoting uninvaded areas, impact native they established. Drone offers bird's-eye view otherwise difficult survey. It thus, contribute IAS. allow interactions better characterize dynamics invisible ground. investigates sensing-based aim probability (i.e., canopy), test whether accessible trunk eye level). (i) quantified approaching (ii) maximum (iii) count outperforms human visual easy access sites, (iv) variation relative time eventually propose standardized framework based imagery. Our research took place sites Island: (1) village Manapany-les-Bains (−21.37 S; 55.58 E; conducted November 22, site only present; (2) botanical garden Domaine Café Grillé 55.42 23, co-occur; (3) public park city Saint Benoît (−21.03 55.72 25, occupied grandis. In all surveys pines, which represent highly favorable species. mostly bright-green red spots back, distinguished through differing patterns. identified blue crescent-shaped pattern snout, white-black-white stripe behind eye, length 12 cm. characterized snout much (up 30 length). longitudinal stripes wider yellowish tail, 13 For analysis, juveniles adults when was (individual aging uncertain considered adult). Females basis endolymphatic calcium sacs sexual dimorphism inexpectata. DJI Phantom 4 Pro V2.0 equipped standard camera. camera 1-inch 20M pixel sensor 24 mm (35 format equivalent) lens, corresponding 84° field view. take-offs landings located secured open public, least geckos' habitat. tested would induce first could approached safely criteria: obstacle little cover precise geolocation maneuverability. stabilized image monitored height. Then, steadily flew horizontally toward individual. interrupted either reacted observing behaviour), about 20 propellers (for individual's safety material integrity). suggests did respond case behaviour, maintained position measured trunk. target depending orientation perception avian predator effect propellers' blow). aforementioned method evaluate ground (mostly volcanic rock beaches; Probst, 2007). As known having lost vigilance predators, expected longer than differ linear model (LM, assuming Gaussian distribution). removed data approaches, acquired variable explanatory variable. stronger because bird harrier Circus maillardi red-whiskered bulbul Pycnonotus jocosus (J.-M. Probst built second LM predictor. differences adult juvenile geckos, added maturity two-level factor (adult vs. juvenile). approaches. Horizontal enabled us compare performance counts, allowed prove efficient canopy. level, is, 3 6 above (depending height) oriented opposite direction applicable. reaching m. During each approach, point survey (human counts) observers (JC, ND, XP) per site. counted visible up 8, 6, 4, 2 duration min distance. height embedded barometer GPS, started repeated operation four times 8:00 a.m. 2:00 p.m. At shortest distances, camera's enable encompass whole tree, multiple same entirety resulted unbalanced design models replicates distances; see below), implies estimates certain shorted compared Images carefully afterwards (GD, XP), effort complexity. generalized additive mixed (GAMM; R package mgcv version 1.8-42; Wood, 2011) Poisson spline examine images. both analysis combined. Models included categorical fixed effect, sampling session one approach) random account exposure day). individually, accounting (fixed effect), (random effect). sites. human-based counts (Poisson GAMM distance, adjustment variables), merging adding versus factorial effect. there maximize sessions described above. (GAM; family), accounted abundance (three-level Eventually, assessed identification—at observed—using family) predict (spline effect) unidentified analyses under 4.1.3 (R Core Team, 26 n = 11; 8; 7), 19 7 juveniles. Interestingly, overall (Figure 1). significantly zero ± SE 5.4 p 0.02), mean 4.7 0.70; 5.0 0.22). significant difference (Table expected, vertically (+37.3 statistical (age class juveniles). produced 0 unit SD 0.70 1.20 given session, site, group; Figure 2) With 9 2.33 2.37). laticauda, 0.57 1.00) units drones, 10.40 4.49) counts. grandis, 0.09 0.29) drones 0.93 1.03) higher rate statistically (Human intercept 2.22 0.55, < 0.0001). differed long m). within angle view, location intervals, sex. modalities combined 3). highest rates 2.5 5.5 increased plateau near m, further 4). laticauda. decreased linearly errors narrower presumably result stable morning became around 11:00 a.m., after 12:00 5). Species-level promising avenue presence, images detect many remained undetected 2c). Approach unexpectedly short, even shorter previously New Zealand's lizards, average approximately 59 jeweled Naultinus gemmeus 107 grand skinks Oligosoma grande short-distance high-resolution accordance Varela-Jamarillo al. (2023), less disturbed suggesting non-invasive. sensitive approaching. possibly anti-avian among predators Overall, encounters. exotic ones. might unexpected oceanic island anti-predator behaviours (Blumstein Daniel, 2005), continental (Andreone weaker Malagasy explained adaptation habitats, (or mostly) exposed responses. Further life history needed behaviour. Regardless, suggest applied borbonica Island), systems, broadly arboreal Zealand (Lettink Monks, 2016) Anolis Caribbean Pacific When considering using who constrained (contrarily drone) choice trees good visibility trunks (which constitute important supports thermoregulation geckos). However, complementary leaves combination human- therefore useful, detectability species-specific daily migration, seasonal perch height; Augros 2018; Choeur On hand, outperform reduce rates. assume benefit become clearer shrublands distributed coastline, inhabiting steep slopes, cliffs. comprehensive spatial gaps sample accessibility 2010), filled approach. Future these remnant natural unprospected identify new populations. Similarly likely disperse vegetation, current knowledge (the Grillé), co-occur. surroundings predicted hosting most suitable climate future change emphasizes need pursue order plan intervention if needed. Drone-based should carried out upper part conditions favorable. applicable, orientated avoid backlight thermoregulation. (but models, degrees freedom observable median (not median, zero-inflation data), certainly sufficient S1). noticed (based sexing). respectively (both areas). large (2–2.5 m), taken Photographs offer fair trade-off (encompassing vegetation) sufficiently individuals, provides nearly-independent replicate accurate (and potentially individual) provide too hence fewer present study. These recommendations stand medium specifications those (1-inch lens), adjusted characteristics. do horizontally, opportunity well complementing vertically. opens possibility develop automated expenses increasing monitoring, artificial intelligence-based detection. deep learning techniques lizards carolinensis; Aota developed context. proactive programs, chances reduction reliable geckos. case, able perform individual-level possible mid-range lenses photo-identification Galàpagos marine iguanas; 2023) (Monk integrate studies take into imperfect aimed estimating population size. improvement broader range dynamic, summer, expect winter. year aspect needs explored methodology. will researchers operators coverage cost-effectiveness efforts. any endangered, ones world. Nicolas conceptualized manuscript, analyses, wrote draft. Gregory administered project. Dubos, Xavier Porcel, Romain Pinel designed strategies, provided material, piloted authors collected data, reviewed, edited manuscript. grateful Joanne Monk her constructive comments like thank staff support. Annelise Tran suggestions Lucas Grosolia his assistance field. flights abiding French laws regulations permissions obtained prior private properties. declare conflict interest. support findings openly available Figshare https://doi.org/10.6084/m9.figshare.22581595.v1. S1. Raw jittered Boxes third quartiles, bar represents points outliers Please note: publisher responsible content functionality supporting information supplied authors. Any queries (other missing content) directed author article.