These data are from a camera trapping project that occurred during 2024 in Guanacaste, Costa Rica. We examined the interactions between tree and animal communities during secondary forest succession in seasonally dry tropical forests in two conservation areas. We installed camera traps for one year in three well characterized chronosquences that included forest stands ~5 to > 100 years old.  All camera trap images were deposited in Wildlife Insights, a platform that uses Artificial Intelligence to aid in categorizing and identifying species in camera trap images, under the project name "Edadades".  Here we report the summaries of those images, focusing on the images that included unique animal sightings. The dataset includes a summary of the estimated forest age in years, characteristics of the camera trap images, and summaries of the plot-based number of unique sightings and sighting rates for the following groups of animals: deer, agoutis, tapirs, carnivores, peccaries, curassows, coatis, felids (cats), and just only ocelots. There is an additional data table with the full taxonomic and common names of the animals. Due to considerations that some of these taxa are of conservation concern, geographic coordinates of the plots are not provided, but can be made available upon request. 

STUDY SITES

We worked in 20 0.1 ha forest monitoring plots located in the Guanacaste Province of Costa Rica in two national parks located within two conservation areas, Parque Nacional Santa Rosa in Área de Conservación Guanacaste and Parque Nacional Palo Verde in Área de Conservación Tempisque. Both national parks have seasonally dry forest in variation stages of regeneration following agriculture or grazing (Calvo et al, 2009), with a five to six month dry season with minimal precipitation from December to mid-May. Parque Nacional Palo Verde (10.35° N, 85.35°W, 200 m.a.s.l.) has a 30 year mean annual precipitation of 1444 mm mostly concentrated between the months of May to November. The annual mean temperature is 27°C with a higher average during the dry season. Parque Nacional Santa Rosa, (10.84° N, 85.62° W, 300 m.a.s.l.), has a mean annual temperature of 25°C and a 30 year mean annual precipitation range from 880-3030mm.

CAMERA TRAPS

Within each plot, we deployed a single Browning Dark OPS EXTREME Model BTC-6HDX motion sensor camera. The cameras have ~ 24m maximum detection range for the motion sensor that triggers the camera, however, dense vegetation may limit detection ranges to shorter distances. Cameras were strapped to a medium sized tree (defined as those with diameters at ~1 m above the forest floor between 10 to 40 cm) 1 meter above the ground, and directed parallel to the forest floor. When triggered, the cameras were programmed to capture 4 images consecutively. Plots were visited every month to download data, clear the memory storage card, and replace batteries as needed. Each camera had a memory capacity of 10,000 photos, which was only reached 3 times total for all plots and months, which were accounted for in the total number of trapping days per camera. 

        All 20 cameras were deployed during the first week in January, 2024 and  monitored monthly until January, 2025. As images were collected, they were uploaded to Wildlife Insights. However, before accepting  an image with a species name, it was verified by a member of our team. After one year, all images were categorized and exported in a dataset that excluded any images that were categorized as “blank” or with only humans.

DATA FILTERING AND PROCESSESING

To ensure the camera trap captured clear images of animals, our cameras were set to capture 4 images consecutively within a 1 second window. Additionally, animals often were captured in multiple of these bursts due to lingering the the camera frame or remaining in the same area. To control for both the 4 image bursts and same animal replicates, we used R to filter our raw dataset to create “hour windows” to control for this recurrence of the same animal. An hour window was selected after manually viewing images and frequently seeing what appeared to be the same animal in a longer than 30 minute window. An “hour window” compressed any data points that were from the same plot, had the same species of animal, and were on the same day within the same hour block. These potential multiple data points were compressed into one data point, which included a count of how many original images were within this “hour window”. These observations controlled for the burst and the hour block are called “unique observations”. 

For each plot/ camera deployment, the first day of deployment was recorded, as well as the final day of deployment. The final day of deployment was determined by the last photo captured per camera in 2024. The final day was determined as the last photo taken instead of when the cameras were removed from the plots to control for if a camera had failed before it was removed from the field. Each month, each camera’s photos were downloaded and the SD card was cleared. On 5 occasions across all plots and months, the SD card was full and had stopped capturing photos before the photos were downloaded. These gaps were calculated by recording the date of the last photo of that month, to when the SD card was cleared. To control for gaps in captures as well as slightly varying lengths of deployment, the total number of active trap days was calculated per plot. This number was then used as a denominator against numbers of animal observations to create observation rates per trap day. 

WildlifeInsights AI-based species identifications and human confirmation of all images were not always able to unambiguously classify the animal in an image to species. 

In the file labeled "full_scientific_names.csv" cells containing missing information are flagged as "NA". For examples, in some images, AI and people could identify a felid, but not the genus or species. The cell values for genus and species in this case are listed as "NA".

In the file labeled "unique_animal_hourly", observations for which we could not classify an animal to a given taxonomic level (columns labeled class, order, family, genus, species) are flagged as "NA".  If both AI and humans flagged an animal in an image but could not classify it beyond "animal", the column labeled "common_name" has the value "animal" and the rest of the taxonomic descriptions are assigned the values "NA".  In cases where both AI and humans were not certain whether the image contained an animal, the "common_name" column was assigned a value of "Unknown".