Journal list menu

Volume 57, Issue 5 p. 963-974
RESEARCH ARTICLE
Open Access

Drawn out of the shadows: Surveying secretive forest species with camera trap distance sampling

Mattia Bessone

Corresponding Author

Mattia Bessone

School of Biological and Environmental Sciences, Liverpool John Moores University, Liverpool, UK

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Correspondence

Mattia Bessone

Email: [email protected]

Search for more papers by this author
Hjalmar S. Kühl

Hjalmar S. Kühl

Department of Primatology, Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany

German Centre for Integrative Biodiversity Research (iDiv) Halle-Leipzig-Jena, Leipzig, Germany

Search for more papers by this author
Gottfried Hohmann

Gottfried Hohmann

Department of Primatology, Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany

Search for more papers by this author
Ilka Herbinger

Ilka Herbinger

Department of Africa and South America, WWF Germany, Berlin, Germany

Search for more papers by this author
Kouame Paul N'Goran

Kouame Paul N'Goran

WWF Regional Office for Africa—Yaoundé Hub, Yaoundé, Cameroon

Search for more papers by this author
Papy Asanzi

Papy Asanzi

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Pedro B. Da Costa

Pedro B. Da Costa

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Violette Dérozier

Violette Dérozier

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Ernest D. B. Fotsing

Ernest D. B. Fotsing

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Bernard Ikembelo Beka

Bernard Ikembelo Beka

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Mpongo D. Iyomi

Mpongo D. Iyomi

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Iyomi B. Iyatshi

Iyomi B. Iyatshi

Institut Congolais pour la Conservation de la Nature (ICCN), Kinshasa, Democratic Republic of the Congo

Search for more papers by this author
Pierre Kafando

Pierre Kafando

WWF in the Democratic Republic of the Congo, Kinshasa, Democratic Republic of the Congo

Search for more papers by this author
Mbangi A. Kambere

Mbangi A. Kambere

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Dissondet B. Moundzoho

Dissondet B. Moundzoho

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Musubaho L. K. Wanzalire

Musubaho L. K. Wanzalire

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Search for more papers by this author
Barbara Fruth

Barbara Fruth

School of Biological and Environmental Sciences, Liverpool John Moores University, Liverpool, UK

Faculty of Biology/Department of Neurobiology, Ludwig Maximilians University of Munich, Planegg-Martinsried, Germany

Centre for Research and Conservation, Royal Zoological Society of Antwerp, Antwerp, Belgium

Search for more papers by this author
First published: 30 March 2020
Citations: 55

Abstract

en

  1. With animal species disappearing at unprecedented rates, we need an efficient monitoring method providing reliable estimates of population density and abundance, critical for the assessment of population status and trend.
  2. We deployed 160 camera traps (CTs) systematically over 743 locations covering 17,127 km2 of evergreen lowland rainforest of Salonga National Park, block South, Democratic Republic of the Congo. We evaluated the applicability of CT distance sampling (CTDS) to species different in size and behaviour. To improve precision of estimates, we evaluated two methods estimating species' availability (‘A’) for detection by CTs.
  3. We recorded 16,700 video clips, revealing 43 different animal taxa. We estimated densities of 14 species differing in physical, behavioural and ecological traits, and extracted species-specific availability from available video footage using two methods (a) ‘ACa’ (Cappelle et al. [2019] Am. J. Primatol., 81, e22962) and (b) ‘ARo’ (Rowcliffe et al. [2014] Methods Ecol. Evol. 5, 1170). With sample sizes being large enough, we found minor differences between ACa and ARo in estimated densities. In contrast, low detectability and reactivity to the camera were main sources of bias. CTDS proved efficient for estimating density of homogenously rather than patchily distributed species.
  4. Synthesis and applications. Our application of camera trap distance sampling (CTDS) to a diverse vertebrate community demonstrates the enormous potential of this methodology for surveys of terrestrial wildlife, allowing rapid assessments of species' status and trends that can translate into effective conservation strategies. By providing the first estimates of understudied species such as the Congo peafowl, the giant ground pangolin and the cusimanses, CTDS may be used as a tool to revise these species' conservation status in the IUCN Red List of Threatened Species. Based on the constraints we encountered, we identify improvements to the current application, enhancing the general applicability of this method.

Résumé

fr

  1. Avec des espèces animales disparaissant à un rythme sans précédent, nous avons besoin d'une méthode de suivi efficace fournissant des estimations fiables de la densité et de l'abondance des populations, essentielles pour l'évaluation de l'état et des tendances des populations.
  2. Nous avons déployé 160 pièges photographiques (camera traps) systématiquement sur 743 emplacements couvrant 17,127 km2 de forêt pluviale au parc national de la Salonga, bloc sud, République Démocratique du Congo. Nous avons évalué l'applicabilité du camera trap distance sampling (CTDS) à des espèces de tailles et de comportements différents. Pour améliorer la précision des estimations, nous avons évalué deux méthodes d'estimation de la disponibilité des espèces (availability ‘A’) pour la détection par des camera traps.
  3. Nous avons enregistré 16,700 clips vidéo, révélant 43 taxons d'animaux différents et avons estimé les densités de 14 espèces différentes en termes de caractéristiques physiques, comportementales et écologiques, en calculant la disponibilité spécifique à l'espèce à partir des séquences vidéo disponibles en utilisant deux méthodes (1) 'ACa' (Cappelle et al. [2019] Am. J. Primatol., 81, e22962) et (2) 'ARo' (Rowcliffe et al. [2014] Methods Ecol. Evol., 5, 1170). Les tailles d'échantillon étant suffisamment grandes, nous avons trouvé des différences mineures entre ACa et ARo dans les densités estimées. En revanche, la faible détectabilité et la réactivité à la caméra étaient les principales sources de biais. Le CTDS s'est avéré efficace pour estimer la densité d'espèces réparties de façon homogène plutôt que celle des espèces inégalement réparties.
  4. Synthèse et applications. Notre application du camera trap distance sampling (CTDS) à une communauté diversifiée de vertébrés démontre l’énorme potentiel de cette méthodologie pour les études sur la faune terrestre, permettant des évaluations rapides de l’état et des tendances des populations d'espèces qui peuvent se traduire par des stratégies de conservation efficaces. En fournissant les premières estimations d'espèces sous-étudiées comme le paon du Congo, le pangolin géant et les mangoustes du genre Crossarchus, le CTDS peut être utilisé comme outil pour réviser l'état de conservation de ces espèces dans la liste rouge des espèces menacées de l'UICN. Sur la base des contraintes rencontrées, nous identifions des améliorations à l'application actuelle, renforçant l'applicabilité générale de cette méthode.

1 INTRODUCTION

The use of camera traps (CTs) to study wildlife has seen an exponential increase in the last decade (Burton et al., 2015), offering innovative approaches for obtaining species' distribution, density, abundance, behaviour and community structure in an economical and minimally invasive way (Rovero & Zimmermann, 2016). Animal density is an extremely informative parameter in community ecology, providing data for monitoring temporal trends in population status and comparing populations across sites (Nichols & Williams, 2006), crucial information for effective wildlife conservation.

The first density estimators based on CT footage were designed for large, individually recognizable species, using a capture–recapture (CR) framework (Karanth & Nichols, 1998). The method was applied to few mammals, mainly felids identified by their coat pattern (Jackson, Roe, Wangchuk, & Hunter, 2006; Karanth, 1995). However, defining the effective surveyed area was problematic and CR methods estimated population size within an area of unknown size, rather than density (Sollmann, Mohamed, & Kelly, 2013). Mark-reSight methods (MS; Rich et al., 2014) and spatially explicit capture–recapture methods (SECR; Efford, Borchers, & Byrom, 2009) were a big improvement. By estimating the area effectively sampled, density estimates became statistically valid (Sollmann, Linkie, Haidir, & Macdonald, 2014). However, requiring at least a proportion of individuals to be recognizable, they were not applicable to all species. Recently, the development of statistical estimators of animal density has overcome these limitations. SECR methods have been extended allowing density estimates of unmarked populations (Chandler & Royle, 2013). Here, sampling effort must be spatially intensive and estimates lack precision unless supplemented with auxiliary data such as genetic sampling or telemetry (Evans & Rittenhouse, 2018; Linden, Sirén, & Pekins, 2018; Sollmann et al., 2014). Random encounter models (REMs; Rowcliffe, Field, Turvey, & Carbone, 2008) were considered a promising development. REMs assumed a certain detection within an estimated area in front of the camera and, by using the gas model of Hutchinson and Waser (2007) to describe animal movement, required estimates of average animal speed for estimating animal density. Animal speed however, is hard to estimate accurately, and REMs broad applicability is still being tested (Chauvenet, Gill, Smith, Ward, & Massei, 2017; Sollmann et al., 2013). To address these issues, recent studies have used a modified version of REMs. Nakashima, Fukasawa, and Samejima (2017) replaced animal speed with the time animals remained in the camera field of view (obtained from recorded videos), whereas Campos-Candela, Palmer, Balle, and Alós (2017) in a simulation study and Moeller, Lukacs, and Horne (2018) in a study on elks Cervus canadensis, circumvented the need for animal average speed by collapsing sampling occasions into predetermined instantaneous moments where the surface covered by the camera field of view was known and 100% detectability assumed. Although these methods were a promising development for estimating density of unmarked species, they remain to be tested in various field situations and, as some (e.g. Campos Candela et al., 2018; Nakashima et al., 2017) are mathematically demanding, broad applicability without a user-friendly software seems unlikely.

Camera trap distance sampling (CTDS; Howe, Buckland, Després-Einspenner, & Kühl, 2017) is another recently proposed method for density estimations of unmarked populations. It uses a distance sampling (DS) approach (Buckland et al., 2001), adjusting point transect DS to the use of CTs. Similar to Moeller et al. (2017) and Campos-Candela et al. (2017), CTDS makes use of predetermined instantaneous snapshot moments, but assumes 100% detection at 0 m only, accounting for imperfect detection by modelling detectability as a function of distance. In addition, CTs only detect animals when available, a problem when studying arboreal or subterranean species. Therefore, CTDS requires estimates of species-specific availability 'A', that is, the proportion of time a species is available for detection. So far, two methods have been used for estimating ‘A’: (a) ‘ACa’ (Cappelle, Després-Einspenner, Howe, Boesch, & Kühl, 2019) refers to the time of activity ‘Ti’ with Ti being defined as the number of 1-hr intervals with at least one video; (b) ‘ARo’ (Rowcliffe, Kays, Kranstauber, Carbone, & Jansen, 2014) estimates ‘A’ by fitting a circular kernel distributions to times of independent detections, with the peak of activity defined by the maximum value of the kernel distribution. Importantly, in both methods ‘A’ is extracted from the same videos used for estimating density. CTDS was applied to wild populations of Maxwell's duiker Philantomba maxwellii (Howe et al., 2017), and Western chimpanzees Pan troglodytes verus in Taï National Park, Côte d'Ivoire (Cappelle et al., 2019) returning unbiased estimates for the latter. In addition, by using a DS approach, CTDS takes advantage of a consolidated mathematical framework, open-source software and a vast community of users and developers, making the method easily accessible. Therefore, CTDS could be considered among the most promising methods to assess animal density, particularly suitable for habitats where species taking advantage of dense vegetation for their cryptic existence are rarely encountered.

Our planet's tropical rainforests, particularly the Amazonian and the Congo basins, provide these features. Disappearing with unprecedented speed, ecological information for the vast majority of terrestrial vertebrates is urgently required (IUCN, 2019). Central Africa's Congo basin provides 1,620,000 km2 of evergreen rainforests, with 1,000 bird and 400 mammal species currently known (Campbell, 2005). Its heart, the Cuvette Centrale, 800,000 km2 in size situated south of the Congo River, Democratic Republic of the Congo (DRC), has the continent's largest protected area of pristine African lowland rainforest: Salonga National Park (SNP), an IUCN World Heritage Site. Here, we estimate vertebrate density by applying CTDS to the large and remote South block of Salonga National Park, assessing applicability of the methodology in relation to species-specific properties such as (a) size; (b) activity patterns; (c) sociality; (d) abundance; (e) distribution and (f) reactivity to CTs. The latter referring to any responsive behaviour occurring because of the presence of an observer (i.e. the CTs) causing the animal to modify its travelling trajectory. This either by (a) moving away from the camera (avoidance), (b) approaching it (attraction), or (c) stopping, standing in front of the camera.

2 MATERIALS AND METHODS

2.1 Study area

Salonga National Park (36,000 km2), situated in the Cuvette Centrale, DRC (Figure 1a), consists of two blocks, North and South. We investigated block South (17,127 km2), composed of 99% of primary lowland mixed forest, 1% of savannahs, regenerating forest, cultivation, marshes and water bodies (Bessone et al., 2019).

Details are in the caption following the image
Location and survey design: (a) Democratic Republic of the Congo (DRC; light blue) with Congo river (blue line), Cuvette Centrale (green) and Salonga National Park (SNP; yellow); (b) SNP, block South with surveyed (black dots) and unsurveyed (white dots) line transects; (c) Camera trap locations along line transect

2.2 Data collection

Camera trap data were collected between September 2016 and May 2018 as part of a comprehensive biodiversity inventory (PNS-Survey©), conducted along 405 systematically placed sample units (i.e. line transects), generated from a random origin. The 1 km transects running east–west were evenly spaced by 6 km (Figure 1b). Two infrared CTs (Bushnell Trophy CamTM, Model 119776), with angle of view θ = 45° and inter-trigger lag time = 1 s, were set up at 250 and 750 m from the beginning of each transect (Figure 1c). To avoid disturbance caused by the passage of field teams, cameras were systematically positioned 50 m to the north or south of the transect line, oriented north between 70 and 90 cm above-ground. Given the size of SNP and the limited number of devices (n = 160), the study area was divided into 37 sub-areas covering 380 km2 on average (range = 72–1,188 km2, SD = 274.8), each surveyed once. CTs were relocated to a new sub-area after a minimum of 2 weeks (average = 38.4 days, range 14–78, SD = 12.4).

Of the 405 transects, 27 were not surveyed due to their proximity to major rivers, or armed poachers, resulting into 378 surveyed line transects (Figure 1b). Due to logistical constraints, one transect remained without, and four transects with only one CT each, resulting in 750 sampling locations. Time of installation, habitat type and GPS location were noted for each device. Cameras were active 24 hr/day and sensor sensitivity was set to ‘high’. For a discussion of potential limitations of our survey design, see Appendix S1.

2.3 Camera trap distance sampling

2.3.1 Measurements

Following Howe et al. (2017), we measured distances between the CT's lens (i.e. 0 m) and the midpoint of each detected animal (=radial distances) in each video at predetermined snapshot moments (=observations) by comparing animal locations to 1 m distance labels recorded during camera installation (from 1 to 12 m). Predetermined snapshot moments represent observations at specific times of day, starting with a snapshot at midnight 00 hr 00 min 00 s with an interval between snapshots ‘t’ set to 2 s, a value considered appropriate to obtain adequate sample sizes even for fast moving and rare species (Howe et al., 2017). Temporal effort is then determined by the value of t (the longer ‘t’, the lower the effort—see Appendix S2).

We expected that species-specific features could potentially affect CTDS estimates. Therefore, for each observation, we also recorded (a) individual maturity (immature/adult), (b) animal group size and (c) reactive behaviours (see Appendix S3).

2.3.2 Species-specific availability

We corrected for species-specific availability ‘A’ applying (a) ‘ACa’ (Cappelle et al., 2019) and (b) ‘ARo’ (Rowcliffe et al., 2014), calculating ARo by using the r package ‘activity’ (R Core Team, 2019; see Figure S1). In order to ensure independence of observations of times of detection, we used the number of capture events, defined as the first video recording the same individual/animal group, while subsequent videos of the same individual/group were discarded. A new event was recorded when a different individual/animal group entered the field of view.

2.3.3 Density estimation

Densities were estimated by applying the formula of Howe et al. (2017; see Appendix S2 for further details). All operational days, excluding days of camera installation and retrieval, were considered when calculating survey effort.

As reactivity to CTs is expected to induce bias (Buckland et al., 2001), we discarded all observations where animal behaviour indicated a reaction to CTs. Then, we left- and/or right-truncated each dataset after visual inspection of the histogram of observed radial distances (see Figure S2). We fitted the detection functions to the remaining radial distances and calculated species-specific density in distance 7.3 (Thomas et al., 2010), correcting for both ACa and ARo, and considering six CTDS models (half normal with 0 and 1 hermite polynomial adjustment terms; hazard rate with 0 and 1 cosine adjustments terms; uniform with 1 and 2 cosine adjustment terms).

In CTDS, violation of the assumption of independence of observations is expected. Violation does not affect point estimates of abundance (Buckland et al., 2001), but introduces ‘over-dispersion’, which is partially addressed by defining predetermined instantaneous snapshot moments (Howe et al., 2017; Moeller et al., 2017). In addition, the assumption can be relaxed by estimating variances using a nonparametric bootstrap, resampling points with replacement (Buckland et al., 2001), and by using model selection methods adjusted for over-dispersed data. Accordingly, we estimated variance from 999 bootstrap resamples, with replacement across camera locations, and selected between competing models comparing QAIC scores, following a two-step method (Howe, Buckland, Després-Einspenner, & Kühl, 2019).

2.4 Considered species

Recorded species were considered suitable for density estimation if (a) the number of independent capture events was ≥20; (b) the number of recorded radial distances was ≥80 (Buckland, Rexstad, Marques, & Oedekoven, 2015). To test general applicability of the method, we selected species showing differences in size, activity, abundance and distribution patterns. Information on the following species-specific traits was acquired from published literature: (a) body mass, a proxy of body size (Smith et al., 2003); (b) activity pattern (diurnal, nocturnal or crepuscular); (c) sociality (gregarious or solitary); (d) expected abundance; (e) expected distribution (homogeneous or heterogeneous) (2–5: Kingdon et al., 2013) and (f) conservation status (IUCN Red List of Endangered Species, 2019).

3 RESULTS

A total of 160 CTs were fully functional and active at 743 locations. Total effort was 27,045 camera days, returning 16,734 videos showing animals belonging to 43 different taxa (see Table S1). Average species richness per location was 4.7 (range 0–24, SD = 3.04). Of these 43 taxa, 29 provided adequate data for density estimation. Table 1 shows 14 out of these, selected due to their differences in biological (body mass), behavioural (activity pattern, sociality), ecological (abundance, distribution) and conservation (IUCN status) traits. Except for the endemic Congo peafowl, all chosen species were mammals.

TABLE 1. Species (Common name, scientific name) selected for method evaluation. Body mass (average in kg; Smith et al., 2003); Activity pattern (sD = strictly diurnal; sN = strictly nocturnal; mD = mainly diurnal; mN = mainly nocturnal; Cr = crepuscular); Sociality (G = gregarious; S = solitary); Approximate expected abundance (n/km2) available from literature; Distribution available from literature; IUCN status (IUCN, 2019). Reference: Kingdon et al. (2013)
ID Species Body mass (kg) Activity pattern Sociality Approximate expected abundance [n/km2] Distribution IUCN status
1

Congo peafowla

Afropavo congensis

1.4 sD G Unknown Homogeneous VU
2

Forest elephant

Loxodonta cyclotis

3,940.0 mD G 0.05 Heterogeneous EN
3

Bonobo

Pan paniscus

34.0 sD G 0.42 Homogeneous EN
4

Allen's swamp monkey

Allenopithecus nigroviridis

4.7 sD G 100 Heterogeneous LC
5

Honey badger

Mellivora capensis

8.0 mD S 0.03 Homogeneous LC
6

African golden cat

Caracal aurata

10.6 mD S 0.04–0.1b Homogeneous VU
7

Genets

Genetta sp.

2.0–1.9c sN S 0.8–4.5c Homogeneous LC
8

Cusimanses

Crossarchus sp.

0.7–1.5d mD G Unknown Homogeneous LC
9

Aardvark

Orycteropus afer

52.3 sN S 1–2 Homogeneous LC
10

Giant ground pangolin

Smutsia gigantea

33.0 sN S Unknown Unknown VU
11

Sitatunga

Tragelaphus spekii

78.0 mD S 92–180 Heterogeneous LC
12

Water chevrotain

Hyemoschus aquaticus

10.8 sN S 1.5–5 Homogeneous LC
13

Brush-tailed porcupine

Atherurus africanus

1.9 sN S 2.4–13 Homogeneous LC
14

Four-toed sengi

Petrodromus tetradactylus

0.2 Cr S 210 Homogeneous LC
  • a McGowan, Kirwan, & Sharpe (2019).
  • b Bahaa-el-din et al. (2016).
  • c Genetta servalina and G. maculata.
  • d Crossarchus alexandri and C. ansorgei.

Table 2 shows species-specific information obtained from CTs, including activity times, availability according to Cappelle et al. (2019) and Rowcliffe et al. (2014; examples provided in Figure 2), as well as truncation distance. Detectability positively correlated with body size, with small-sized species being undetected within the first 2 m from the camera (Figure 3).

TABLE 2. Species-specific information obtained from camera traps (CTs). Individual locations: Number of CT locations with detections >0; Capture events: Number of video-clips; Total Radial distances: Total of measured radial distances; Radial distances ‘Immatures’ (%): Number of radial distances of immature individuals (percentage of total); Radial distances ‘Reactivity’ (%): Number of radial distances of reactive individuals (percentage of total); Activity time Ti [hours]: number of activity hours per 24-hr-day used when calculating specific availability after Cappelle et al. (2019); Availability (ACa): specific availability calculated after Cappelle et al. (2019); Availability (ARo) (SE): specific availability (standard error) calculated over 24 hr after Rowcliffe et al. (2014); Truncation [m] left; right: meter(s) of left and right-truncation of radial distances
ID Species Individual locations Capture events Total Radial distances Radial distances ‘Immatures’ (%) Radial distances ‘Reactivity’ (%) Activity time Ti [hours]a Availability (ACa)a Availability (ARo) (SE)a Truncation [m] left; right
1 Congo peafowl 73 137 3,104 249 (8%) 240 (8%) 12 0.57 0.34 (0.04) 1; 8
2 Forest elephant 14 20 840 68 (8%) 676 (80%) 24 0.15 0.34 (0.10) 0; 12
3 Bonobo 66 100 5,604 1,376 (25%) 515 (11%) 14 0.45 0.34 (0.05) 0; 12
4 Allen's swamp monkey 12 49 1,067 81 (17%) 209 (19%) 12 0.31 0.39 (0.04) 3; 8
5 Honey badger 18 21 217 None 48 (22%) 11 0.38 0.28 (0.07) 0; 7
6 African golden cat 23 25 84 None 6 (7%) 13 0.34 0.45 (0.07) 0; 7
7 Genets 63 95 431 None None 11 0.61 0.43 (0.03) 2; 7
8 Cusimanses 24 34 1,301 72 (5%) None 13 0.44 0.44 (0.06) 2; 7
9 Aardvark 46 54 354 None 25 (7%) 12 0.45 0.31 (0.04) 0; 6
10 Giant ground pangolin 15 24 119 None None 11 0.50 0.40 (0.07) 1; 8
11 Sitatunga 10 23 417 101 (24%) None 14 0.41 0.64 (0.11) 0; 7
12 Water chevrotain 38 152 1,462 None None 12 0.61 0.34 (0.03) 1; 6
13 Brush-tailed porcupine 140 527 1,765 48 (3%) None 12 0.62 0.34 (0.02) 2; 8
14 Four-toed sengi 7 40 216 None None 13 0.31 0.28 (0.05) 2; 5
  • a See Appendix S2.
Details are in the caption following the image
Camera trap-derived daily activity patterns for (a) Capture events < 50 (Elephant); (b) Capture events > 50 (Aardvark); (c) Capture events > 100 (Water chevrotain). Solid lines show availability (ACa) as relative frequency of capture events for each hour interval according to Cappelle et al. (2019); Dashed blue lines show availability (ARo) as relative frequency of capture events fitted over 24 hr according to Rowcliffe et al. (2014)
Details are in the caption following the image
Species-specific truncation distances (m) for Left-truncation (left); and Right-truncation (right). Blue lines represent linear regression (left-truncation R2 = 0.48, p < 0.01; right-truncation R2 = 0.19, p = 0.07) with 95% confidence intervals (light blue areas). Black dot with number indicates species; numbers see ‘ID’ in Table 1

Of the 14 species, immature individuals were never detected in seven, made up less than 10% of observations in another four, and more than 10% in three species only (Table 2). Therefore, we decided to exclude observations of immature individuals, providing population estimates for adults only. Three different types of reaction to CTs (i.e. attraction, avoidance and olfactory) were recognized in seven species (Table 2); examples are provided in Figure 4.

Details are in the caption following the image
Reactivity to camera traps. Columns: Left (yellow): Attraction (honey badger); Middle (grey): Olfactory (elephant); Right (green): Avoidance (bonobo). Rows: Top: reactive observations included; Middle: reactive observation excluded; Bottom: final fitted model after right-truncation (honey badger); binning (elephant); as is (bonobo). Bars provide detection probability as function of distances (1.0 = 100%) prior to truncation (reactivity observations included). Dashed black lines show distances of truncation. Dashed blue lines detection probability as a function of distance (reactivity observations discarded)

Table 3 shows density estimates for each species after bootstrapping, corrected for species-specific activity patterns, providing selected model and coefficient of variation. ACa-corrected estimates consistently provided higher densities than ARo, with ACa-corrected and ARo-corrected estimates being similar with at least 100 capture events (Table 2 and Figure 5). Estimates' precision significantly improved with a higher number of individual locations with at least one capture event (log-transformed—R2 = 0.63; p < 0.01). A similar trend, with precision increasing with higher numbers of recorded radial distances or estimated densities, was not significant (see Figure S3). These results were confirmed by a multiple regression analysis, where precision was modelled as a function of the three aforementioned variables (see Table S2), suggesting precision was mainly driven by the number of individual locations and sample size.

TABLE 3. Species-specific density estimates. Selected model for density estimation showing (a) key function (HN = Half normal; HR = Hazard rate; UNI = Uniform); (b) series expansion (co = cosine; hp = hermite polynomial); (c) adjustment terms; Radial distances (=observations): Number of measurements after truncation and discarding of immatures and reactive individuals; Density (ACa) [ind/km2] (95% CIs): Density of mature individuals per km2 (in bold) corrected for availability using ACa (Cappelle et al., 2019) with 95% confidence intervals (CIs); Density (ARo) [ind/km2] (95% CIs): Density of mature individuals per km2 (in bold) corrected for availability using ARo (Rowcliffe et al., 2014) with 95% confidence intervals (CIs); C.V. ACa; ARo: Estimated coefficient of variation for (a) Density (ACa), (b) Density (ARo)
ID Species Selected model a; b; c Radial distances

Density (ACa) [ind/km2]

(95% CIs)

Density (ARo) [ind/km2]

(95% CIs)

C.V.

ACa; ARo

1 Congo peafowl UNI; co; 1 2,383 0.91 (0.66–1.27) 0.76 (0.55–1.06) 0.17; 0.17
2 Forest elephant HN; hp; 0 151 0.03 (0.01–0.07) 0.02 (0.01–0.03) 0.44; 0.45
3 Bonobo HR; co; 0 3,658 0.70 (0.32–1.53) 0.54 (0.24–1.21) 0.41; 0.43
4 Allen's swamp monkey HN; hp; 0 691 0.53 (0.24–1.14) 0.20 (0.09–0.43) 0.41; 0.40
5 Honey badger HN; hp; 0 121 0.05 (0.02–0.09) 0.03 (0.01–0.06) 0.39; 0.41
6 African golden cat HN; hp; 0 78 0.04 (0.02–0.07) 0.02 (0.01–0.03) 0.36; 0.36
7 Genets HN; hp; 0 374 0.27 (0.17–0.43) 0.18 (0.11–0.28) 0.24; 0.23
8 Cusimanses HN; hp; 0 1,104 1.16 (0.58–2.36) 0.62 (0.31–1.26) 0.37; 0.37
9 Aardvark HN; hp; 0 255 0.20 (0.10–0.34) 0.15 (0.09–0.26) 0.27; 0.28
10 Giant ground pangolin HN; hp; 0 112 0.05 (0.02–0.13) 0.03 (0.01–0.08) 0.50; 0.54
11 Sitatunga HN; hp; 0 253 0.12 (0.03–0.42) 0.07 (0.02–0.25) 0.71; 0.74
12 Water chevrotain HN; hp; 0 1,250 0.72 (0.38–1.35) 0.68 (0.37–1.27) 0.33; 0.32
13 Brush-tailed porcupine HN; hp; 0 1,624 0.71 (0.48–1.03) 0.64 (0.44–0.96) 0.20; 0.20
14 Four-toed sengi HR; co; 0 191 289.46 (2.67–31,294.00) 181.09 (2.16–15,177.00) 17.20; 12.71
Details are in the caption following the image
Relative species-specific concordance of density estimates corrected with ARo (Rowcliffe et al., 2014) and ACa (Cappelle et al., 2019) by number of capture events. Relative accordance (ARo/ACa): 1.0 = 100% accordance. Blue line represents linear regression (R2 = 0.62, p < 0.01) with 95% confidence intervals (light blue area). Black dot with number indicates species; numbers see ‘ID’ in Table 1

4 DISCUSSION

Our results are the first application of CTDS to a multi-species animal community and show the enormous potential of this method for biomonitoring. However, as in other CTs studies (Burton et al., 2015), biological, ecological and behavioural features affected specific detectability. In the following, we will discuss how six of these features could influence the applicability of CTDS: (a) body size/mass; (b) sociality; (c) activity pattern; (d) distribution; (e) abundance and (f) reactivity to the camera.

4.1 Body size/mass

Our results confirm previous studies (Rowcliffe, Carbone, Jansen, Kays, & Kranstauber, 2011; Sollmann et al., 2013; Tobler, Carrillo-Percastegui, Pitman, Mares, & Powell, 2008), showing that body-size positively correlates with detectability (Table 2; Figure 3). In such cases, we left-truncated our data when estimating density, a method known to effectively address low detectability at short distances (Buckland et al., 2001). However, left- truncation implies loss of data needed for achieving accurate estimates, especially for rare species. Therefore, we suggest that deploying cameras at a height of 50 cm above-ground would increase detection rates of small-sized species. Although surprisingly found to be detected imperfectly within the first 2 m from the CT despite an average shoulder height of 95 cm (Figure 4), left-truncation was not applied to the bonobo, for reasons explained later (‘6. Reactivity to camera’). As expected, elephants were detected at short distances. However, given their size, both body length and width were considered when measuring radial distances, and fit was improved by binning data in 2-m intervals (Figure 4).

4.2 Sociality

When applying DS to gregarious species, detection rates may be inflated, as detecting the first animal in a group increases the probability of detecting others (Treves, Mwima, Plumptre, & Isoke, 2010). However, we found no clear evidence for overestimated density in gregarious species (see Table S3) and obtained satisfactory coefficients of variation (<25%) for both brush-tailed porcupine (solitary) and Congo peafowl (gregarious), with low precision equally affecting solitary and gregarious species when capture events were rare (Table 3).

4.3 Activity patterns

Although differences in the specific availability calculated according to Cappelle et al. (2019), and Rowcliffe et al. (2014) were minor with large sample sizes, ARo presented major advantages: (a) calculations provided standard errors of estimated availabilities (that could be included in the estimation of total variance of density) and (b) values appeared to be less influenced by the peak of observation and stochasticity (Figure 2; Table 3). Both methods rely on the assumption that at peak time, 100% of the population is available for detection (Rowcliffe et al., 2014), with asynchronous activity patterns of individuals within species leading to an overestimation of activity time, hence underestimated densities. ARo is consistently calculated over 24 hr, whereas ACa refers to the hour intervals of observed activity, considering activity hours with one capture event only (Figure 2). In ACa, activity intervals may remain undetected due to low sample size, potentially causing underestimation of survey effort. For example, we observed elephants from 20 capture events only, revealing activity in 11 of 24-hr intervals (Figure 2). Elephant activity however, is reported to occur throughout a 24-hr-day (Kely et al., 2019), suggesting we missed part of the species' activity time. While available knowledge could be used to interpret results according to species-specific ecology and behaviour, with ARo activity time fitted over 24 hr, additional sources of variation were avoided (Figure 2). In sum, sample sizes larger than 100 capture events allowed accurate and consistent estimates (Figure 5). Limited numbers of capture events however lead to underestimated ACa, inflating density values (Tables 2 and 3). Therefore, we recommend longer CT deployment, allowing a minimum of 100 capture events per species and the use ARo (Rowcliffe et al., 2014) for the calculation of specific availability. Unless supported by large enough sample sizes, comparing specific availability and density estimates across different studies is precarious and should be performed with care.

4.4 Distribution

We expected high variability in encounter rates for heterogeneously distributed species, leading to imprecise density estimates due to spatial variation (Buckland et al., 2001). In fact, all the species we expected being heterogeneously distributed (Table 1), showed a coefficient of variation >40%, due to low sample size and observations obtained from very few locations (Table 3—but see also Figure S3). Future research should aim at increased spatial effort with synchronous camera deployment to reduce potential bias and strengthen precision (Buckland et al., 2015). When this is not possible, a stratified random design might increase estimate precision (Foster & Harmsen, 2012).

4.5 Abundance

Consequently, while precision was more satisfactory for abundant species, it was also good (C.V. below 35%) for rare, but widespread species such as the genets and the aardvark (Table 3—but see Figure S3), with precision being mainly a function of sample size (i.e. number of radial distances) and heterogeneous distribution (i.e. number of individual locations with at least one capture event—see Table S2). When estimating density of abundant species, a limiting factor was the time required for distance measurements from video-clips. To reduce time of analysis, (a) snapshot interval t could be increased, or (b) observations may be restricted to peak of activity only (Howe et al., 2017). For validation, we compared the full with the reduced methods for all species, and obtained consistent densities despite increased snapshot intervals and peak of activity observations only (see Table S4). However, in the case of rarer species, the application of these methods would further reduce the number of exploitable radial distances, making safe estimates impossible.

4.6 Reactivity to the camera

Reactivity to the camera is known to bias density estimates, as it violates the assumptions of DS (Buckland et al., 2001). Attraction to CTs, providing a high number of observations close to the camera has been previously addressed with left-truncation (Cappelle et al., 2019). However, to minimize induced bias, we consistently excluded all snapshots showing evident reactivity to the camera from analysis, not only for species attracted to CTs, such as the honey badger (Figure 4), but also for the elephants, showing strong olfactory reactivity by insistently smelling the area in front of the camera (regardless of the distance interval). In elephants, when using all observations including the 80% showing reactivity, estimates were inflated up to two orders of magnitude (Figure 4). Avoidance is less frequent, but Kalan et al. (2019) reported it for the bonobo. We confirm bonobos' avoidance of the camera, resulting in fewer observations within the first 2 m (Figure 5). However, bonobos were not undetected, but rather observed further away, and left-truncation was not applicable. Lack of detection close to the camera can be levelled out by excess of detection further away (Buckland et al., 2015), but densities were inflated by 15% because bonobo neophobia seems to be coupled with curiosity from a secure distance. Therefore, we discarded all snapshots showing reactive behaviours. This study suggests reactivity to CTs being the most impacting form of bias in CTDS. Not accounting for reactivity could result in largely inflated density estimates, and future studies should carefully examine the videos to detect reactive behaviours. To reduce visual and olfactory reactivity, we recommend to either deploy CTs for at least 1 month prior to the survey, allowing animals to habituate to cameras; or record reference distance labels after the survey, reducing the time of CT set-up, and by that ‘contamination’ with human odour. If neither is possible, methods not influenced by reactivity to CTs (e.g. SECR), is to be favoured over CTDS.

5 CONCLUSIONS

Camera trap distance sampling is an excellent survey method providing standardized and comparable information on wildlife density and abundance, particularly important for threatened species. Because of its highly diverse vertebrate community, SNP block South represents an excellent test-field, showing CTDS applicability to one of the remotest and least known rainforest areas of the globe. Density values for the Congo peafowl, the giant ground pangolin, and the cusimanses presented here, are the first ever obtained, and are of critical conservation importance providing the basis for IUCN Red Lists species assessments. Despite limitations in comparability due to methodological differences and a site-specific ecological set-up, eight out of 11 densities obtained fell within published ranges (see Table S5). However, our estimates' accuracy remains to be confirmed: longitudinal assessments of density using standardized methods such as those detailed here will validate our results and shed light on the status of these cryptic species. Continuous monitoring and population trend evaluation are crucial information for wildlife conservation. Allowing simultaneous surveys of large portions of the terrestrial vertebrate community, rather than single species, the information CTDS can provide is of pivotal importance for the development of conservation plans of multi-species' communities. It may allow to reveal the delicacy of location specific ecological equilibria, crucial for the conservation of the integrity of the few remnant intact habitats of our planet.

ACKNOWLEDGEMENTS

This work was funded by Liverpool John Moores University and by the Kreditanstalt für Wiederaufbau (KfW Group) on behalf of the German Government and WWF Germany. We are grateful to all people that assisted data collection: R. Souza, M. Bofeko, R. Booto, B. Katembo, J. Kukumanga, S. Mwanduku, P. Musenzi, R. Ratsina, P. Bondo, G. Mbambe, W. Bokungu, P. Bokembi, E. Way, P. Ingange, J. Iyondo, B. Nkoy, A. Bomondjo. We thank S. Abedi, J. Eriksson, M. Etike, A. Kabamba, G. Kasiala, R. Keller, L. Lokumu, E. Lopongo, A. Louat, S. Matungila, M. Mbende, O. Nelson, for facilitating fieldwork; S. Mbongo, I. Di Dato, G. Kayembe, C. Kabamba, L. Nsenga, M. and T. de Sousa, for logistic support. We are grateful to the comments of S. Buckland, S. Focardi, E. Howe, F. Maisels, F. Michalski, L. Thomas and an anonymous reviewer. Special thanks go to B. Grothe, C. Rörentrop, U. Zendel, A. Werner, R. McElreath, M. Tomasello, K. George, N. Jones, R. Leatherbarrow, A. Tattersall, and P. Wheeler. In DRC, we thank the Institut Congolais pour la Conservation de la Nature, the Ministère de la Recherche Scientifique and the Unité de Gestion du Parc National de la Salonga, for permitting research in Salonga National Park.

    AUTHORS' CONTRIBUTIONS

    H.S.K., G.H., I.H., K.P.N. and B.F. conceived the study; M.B., H.S.K., G.H., K.P.N., V.D. and B.F. organized data collection; P.A., P.B.D.C., E.D.B.F., B.I.B., M.D.I., M.A.K., D.B.M. and M.L.K.W. collected the data; I.B.I., P.K., G.H. and B.F. provided facilities and logistical support; M.B. and G.H. processed and analysed video footage; M.B. and B.F. wrote the manuscript with contributions from all co-authors.

    DATA AVAILABILITY STATEMENT

    Data available via the LJMU Data Repository https://doi.org/10.24377/LJMU.d.00000052 (Bessone et al., 2020).