Data and code from: Thin corridors limit wildlife: Variance of tropical carnivore distribution and habitat use in a critical rainforest corridor

Ross, Patrick 1 ; Patel, Erik2 ; Ravelijaona, Rojo Nandrianina3; Raoloniana, Guy Irenel2; Dollar, Luke4 ; Parker, Patricia1 ; Rasolondravoavy, Charles2

Published Jul 23, 2026 on Dryad. https://doi.org/10.5061/dryad.prr4xgxqd

Abstract

Madagascar is acclaimed for its floral and faunal endemism and biodiversity. Among the island nation’s most emblematic fauna are its native mammalian carnivores; they are members of the threatened and endemic Eupleridae family. The Corridor of Marojejy, Anjanaharibe Sud Special Reserve, Tsaratanana (COMATSA) is a newly protected corridor system that faces deforestation and lacks detailed assessments of its native carnivore community. We deployed 44 motion-activated trail cameras to identify which terrestrial carnivores are found within the rainforest corridor landscape of COMATSA-Sud and Marojejy National Park, and to examine carnivore occupancy, relative activity, and the impacts of habitat edge. Our sampling array operated from October 20th, 2021, to February 10th, 2022 (113 days) and confirmed the presence of four species in Eupleridae: Cryptoprocta ferox, Fossa fossana, Galidia elegans, and Galidictis fasciata, as well as a non-native carnivore species: Canis familiaris. We show that carnivore occurrences significantly increase with distance to forest edges. We discuss the negative impacts of forest loss on native carnivores and highlight the threats posed by the non-native carnivore C. familiaris. Our findings spotlight faunal detections across the corridor and indicate that fewer native carnivores are found in the corridor compared to nearby protected areas, suggesting that deforestation correlates with a skewed distribution of carnivore detections. Based on our data, we provide management recommendations for the protection of the forest corridor and its endemic faunal community.

This data was generated using a camera trap array positioned along transects bisecting a critical forest corridor in Northeastern Madagascar (COMATSA-Sud). The study aimed to evaluate how habitat fragmentation influences the presence and behavior of both endemic and invasive mammalian species.

Trigger Mode: 3-shot bursts per motion detection event.

Operational Duration: Each camera remained active for a minimum of three months to ensure seasonal representation.

The following instructions apply only after the following steps are completed:

A. Camera traps have been recovered from the field grid.

B. All photos have been tagged with species, number of individuals, capturing camera trap, and camera station.

C. The tags have been written to the metadata using the free program Digikam.

submission Date to Dryad: 09.14.2024

"README_for_Patrick_Ross_Corridors.txt": Instructions for proper use of the COMATSA-Sud Corridor Camera Trapping data.

Readme File: for "thin corrThins limit wildlife: variance of tropical carnivore distribution and habitat use in a critical rainforest corridor"

Description of Data and File List

This submission contains data from camera trap surveys used to model single-season occupancy and analyze the effects of forest edges on carnivore occurrences.

File Inventory:

1. COMATSA_Station_data.csv: Contains location and environmental metadata for each camera trap station.

2.record_table_30min_deltaT_2022-03-14.csv: The raw detection record table containing species sightings, filtered for a 30-minute independence interval.

3.patsdata.csv (or relevant filename): The processed dataset used for the GLMER analysis of carnivore occurrences.

4.comatsaspeciestable.R: The R script used to process the data, generate detection histories, and run occupancy models

5.data_combined.xlsx: A multi-sheet workbook containing the summarized data and calculated indices used specifically for the tables and figures in the manuscript. This data file is also uploaded as separate unformatted CSV files: data_combined__Naïve_occupancy.csv; data_combined_Combined_Results.csv; data_combined_Comparisons.csv; data_combined_record_table_chopup.csv; data_combined_spp_by_station_no_sort.csv; data_combined_Standard_Dev_of_camera_distance.csv; data_combined_TS_calc.csv; data_combined_TS_se_calc_2893tn.csv

pats_data.R: For fitting and summarizing a basic Poisson GLMM

patsdata.R.Printout.docx/patsdata.R.Printout.txt: R installed and loaded the lme4 package and successfully fitted a Poisson GLMM to test the effect of edge distance on carnivore occurrences while accounting for species-level variation.

Data Dictionary:

1.COMATSA_Station_data.csv

Column Name Description Units/Format

Station Unique identifier for each camera trap location Alphanumeric (e.g., Station_01)

utm_x Easting coordinate (UTM) Meters

utm_y Northing coordinate (UTM) Meters

Setup_Date The date the camera trap was deployed YYYY-MM-DD

Retrieval_Date The date the camera trap was retrieved YYYY-MM-DD

Forest_type Classification of the forest surrounding the station Categorical (e.g., Primary, Secondary)

Elevation.m. Height above sea level Meters (m)

2. record_table_30min_deltaT_2022-03-14.csv

Column Name Description Units/Format

Station Abbreviated station identifier e.g., S1, S2

Species Scientific specific epithet of the detected animal e.g., elegans, ferox, fossana

DateTimeOriginal Date of the individual detection event YYYY-MM-DD

Time Time of the individual detection event HH:MM:SS

delta.time.secs Time elapsed since the previous detection of any species at this station. Seconds

delta.time.mins Time elapsed since the previous detection. Minutes

delta.time.hours Time elapsed since the previous detection. Hours

delta.time.days Time elapsed since the previous detection. Days

FileName The original filename of the image containing the detection. .JPG / .PNG

metadata_Animal The specific species tag assigned to the image in metadata. Text

n_images The number of images included in the specific trigger/event. Integer

HierarchicalSubject The hierarchical metadata tag structure (from tagging software). Text

3. Analysis Data (patsdata)

Column Name Description Units/Format

Carnivore_Occurrences The count of native carnivore detections at a site Count (Integer)

Edge.Dist Linear distance from the station to the nearest forest edge Meters (m)

Edge.Dist.log Log-transformed distance to the forest edge Log-meters

No_Species Categorical identifier for the species used as a random effect Categorical

elegans Detection status of Eupleres elegans. 1 = Detected, 0 = Not Detected

fasciata Detection status of Galidia fasciata. 1 = Detected, 0 = Not Detected

ferox Detection status of Cryptoprocta ferox. 1 = Detected, 0 = Not Detected

fossana Detection status of Fossa fossana. 1 = Detected, 0 = Not Detected

4. comatsaspeciestable.r

The rcode used to extract metadata from the photos and generate all analyses.

5.data_combined.xlsx.

Combined Results: The primary output sheet. It contains finalized tables ready for presentation or publication, including occupancy estimates and Trap success rates per species.

Comparisons: A comparative analysis sheet. It aligns the results of this current study with data from previous studies I have conducted. These tables are used to identify trends or shifts in species presence using similar methodology within the same forest.

A working file used during the data cleaning phase. It is a subset of the original record_table_30min_deltaT file, filtered to include only the essential columns and rows required for specific downstream calculations.

Standard Dev of camera distance: Contains spatial statistics regarding the camera trap array.Mean Distance: 484m Standard Deviation: 71m Purpose: To document the spacing consistency between stations.

SPP by station no sort: A raw data table listing each species detected and the specific count of detections recorded at each station. This data remains unsorted to preserve the original station order.

TS calc: The calculation sheet for Trap Success (TS).Formula used: $TS = \frac{\text{Total Captures}}{\text{Total Trap Nights (TN)}}$This sheet converts raw detection counts into a standardized rate of activity.

TS se calc 2893tn: Manual calculation of the Standard Error (SE) for Trap Success per species. Calculations are based on a total of 2,893 trap nights. This provides the statistical variance for the TS metrics found in sheet 6

naive occNaivey: Calculates the Naive Occupancy for each carnivore species.Definition: The proportion of total sites where at least one capture of a species occurred.Formula: $\frac{\text{Number of sites with } \ge 1 \text{ detection}}{\text{Total number of sites}}$

Methodology Note

Temporal Independence: As seen in the provided R script, records were filtered using a 30-minute minDeltaTime to ensure independent observations.

Detection Histories: Occupancy models used an occasionLength of 7 days, with the first day based on the individual station's setup date.

The camera trap array was positioned inside western Marojejy NP and eastern COMATSA-Sud. The center of our research area was (14°28'2.48"S 49°32'56.39" E) (Figure 1). We chose this unsurveyed study site for its corridor structure and proximity to the previously surveyed rainforest of Anjanaharibe-Sud Special Reserve (Ross et al. 2020). The study site covers the center of the corridor, which, through its deforested shape (Figure 1), provides a funneling of habitat through which carnivore presence and possible movement could be examined. Eupleridae carnivores are known to prefer contiguous forests compared to degraded or fragmented areas (Gerber et al. 2012). The forest within the corridor features patches of deforestation with habitat degradation that is visible in satellite imagery (Figure 1).

We placed 44 motion-activated camera traps across 28 camera stations, 12 single camera stations and 16 double camera stations. We set up camera trap stations across four transects, each 4km long, 16km in total. The elevational range of camera trap stations varied from 942m ASL to 1286m ASL. No lure or bait was used at any station, and all camera traps were placed 15cm above the forest floor, which is an ideal height for our targeted taxa (terrestrial carnivores). Camera stations were in the forests of COMATSA-Sud n=12 and Marojejy National Park n=16 (Figure 1). We used double camera stations in an attempt to photograph both sides of an individual animal so that we could use markings or unique traits to identify individuals that the stations photographically captured. This was done so that the movement of individual carnivores across the landscape could be examined. We were ultimately unable to identify individual carnivores effectively and were unable to assess their movement or ranging. The camera trap model used was the Stealth Cam PXP3NG. All camera stations were spaced on average (±SD) 483m ± 70m apart, thus allowing for spatial independence, as a 483 m diameter is equal to or greater than the average home range for the majority of Eupleridae carnivores (Farris et al. 2014a). The cameras were placed on October 20th, 2021, and were all taken down by February 10th, 2022 (113 days). The camera traps operated during the project period without any gaps in capturing,g and none suffered memory-related faults. Our sampling array totaled 2893 trap nights. We chose the location of the camera station array based on the narrowing of the forest corridor to photographically capture carnivores and their individual movement through the narrowed corridor structure (Figure 1). All cameras were set to high sensitivity and 3-shot bursts for each independent trigger. We recorded the location and elevation of each station (UTM) using handheld Garmin GPS units. After the initial setup and check of all camera stations, we checked the cameras after the first 14 days, and then every 30 days for the remainder of the study period.

Data Analysis

We categorically sorted and labeled every image using the open-source program Digikam (www.digikam.org). Once the images were sorted, we created a capture history file using the package "CamtrapR” within the program RStudio (R version 3.6). We defined a unique capture event in our analysis as a photograph of an organism at a camera station within 30 minutes. We used this time period to ensure temporal independence among capture events (Di Bitetti et al. 2006). Using these unique capture events across all the camera stations, we assembled a new capture history using these unique capture events for each carnivore species, where 0 indicates non-detection, and 1 indicates detection for each 24 h period. We defined a trap night as a 24 h period in which one camera station was functioning properly with no malfunctions. For each carnivore, we calculated trap success (TS) as the total number of captures divided by total trap nights multiplied by 100. The ratio of TS is used as a measure of relative activity for all detected carnivores to compare their activity and presence across the entire surveyed landscape (Kelly et al. 2008). To evaluate species occurrence while also accounting for imperfect detection, we used single-season, single-species occupancy modeling implemented in the ‘Unmarked’ package in R v.3.6 (MacKenzie et al. 2002, Fiske & Chandler 2011, R Core Team 2014). Given the limited spatial scale of the camera trap array (900ha), our 4-month collection period, and the limited habitat degradation sampling, we chose not to include covariate model estimates in our species occupancy estimates as to not overstate our findings. We believe that a longer-running survey would be required to assess covariates such as habitat degradation and distance to edge. We encourage a multi-season sampling period that would be sufficient to model these covariates. We calculated species occupancy (ψ) and detection probability (p) for all native and non-native carnivores. We assumed that occupancy did not change during the survey period and that detection of each species at each location was independent (Anderson & Burnham 2002). We calculated the effects that distance to forest edge would have on carnivore occurrences with a Generalized Linear Effects Model (GLMER) using the ‘lme4’ package in R v3.6 (Bates et al. 2015). The GLMER model fits a Poisson distribution and features “Distance to Forest edge” as a fixed effect and “Species” as a random effect, while the number of carnivore occurrences is the response variable. To examine potential carnivore preferences that correspond to longitude, such as favoring one side of the road that divides the study area or preferring one protected area over the other, we compared native carnivore occurrences to the longitude of the capturing camera station. We calculated regression analyses for distance to forest edge by G. elegans and native carnivore occurrences by UTM (West to East) with the Analysis ToolPak in Microsoft Excel (Microsoft 2022).

Data and code from: Thin corridors limit wildlife: Variance of tropical carnivore distribution and habitat use in a critical rainforest corridor

Data files

Abstract

README: Data and code from: Thin corridors limit wildlife: Variance of tropical carnivore distribution and habitat use in a critical rainforest corridor

Methods

Usage notes