Bringing BearID to Ecuador (a sabbatical story)

One of the many perks of working at Arm is the sabbatical program. This is described on the Life at Arm page as:

After four years of working with us, you can take an extended paid four-week sabbatical to really hit pause or pursue a passion project.

If you are connected to me, you are likely aware of me and my partner Mary’s involvement with the BearID Project . My first sabbatical seemed like a good time for us to try something new, yet related to bears! ??

One of our key collaborators is Dr. Russ Van Horn , Scientist at the San Diego Zoo Wildlife Alliance . Russ is a super-connector in the world of bear researchers. When we asked him for advice on how we could work with new bear species combined with interesting travel, he simply asked: what species do you want to work with and where do you want to go? Since we had collaborated on a paper, Multispecies facial detection for individual identification of wildlife: a case study across ursids, which included a section on Andean bear identification, we decided on that species. South America was the destination, but we left the specifics up to Russ’ suggestion for a collaborator.

In September of 2023, Russ introduced us to Dr. Becky Zug , Professor and Director of the Carnivore Lab at the Universidad San Francisco de Quito (USFQ). We hit it off with Becky from the start and we were excited to see how we could help. Not only is she studying Andean bears in Ecuador, she is involved with multiple camera trapping projects in various ecosystems, all relying on manual analysis. In addition to helping with Andean bear identification, could we provide tools to help with analyzing all the camera trap data, and maybe even look into identifying individuals of other keystone species, like pumas? This was sounding like a valuable way to spend a sabbatical in 2024!

Over the next months we planned a four-week itinerary of collaboration at USFQ combined with exploration of Ecuador. We settled on early June to early July of 2024 to take advantage of the quieter summer period at the university. Counting weekends and 2 holidays, we had 32 days in total. Our final itinerary included:

• 8 days working on the USFQ campus with Becky
• 8 days visiting affiliated research sites
• 13 days exploring Quito and the surrounding sites
• 3 travel days

Biodiversity Server

As we were preparing for the collaboration aspect of the trip, one thing became apparent, Becky (and the university in general) could really use a centralized repository for all their camera data! Representing the extensive biodiversity of Ecuador, researchers were barely scratching the surface of this data gold mine. From a software perspective, I had been looking into TRAPPER (from OPEN SCIENCE CONSERVATION FUND ), which seems to have all the major functionality and AI expandability we would need. However, Becky and her team were mainly using outdated PCs and laptops. A performant and scalable yet sustainable server is needed.

To this end, I reached out to Joe Speed , Head of Edge at Ampere less than a month before our trip. As an Arm partner, I knew Ampere-based servers would be a perfect fit for a USFQ-wide biodiversity server. Joe loved the idea and got straight to work on securing hardware. The week before our trip, the BearID Project received an important donation, in the form of an Ampere Computing Mt Collins 2U server fitted with 2 Ampere? Altra? Series Processors (that’s 160 cores)! With 24 SSD bays, room for 8TB of SDRAM, dual 1GbE networking, and multiple PCIe risers, this beast can easily scale to meet USFQs needs. After “owning” this server for a couple days of quick testing, we (re)donated it to the USFQ Carnivore Lab and shipped it to the university’s importer in Miami. We expected to meet again in Quito!

On the Ground in Ecuador

One of the first things we did after arriving in Quito was to visit the US Embassy. Thanks to the great connections of my Arm colleague, Stephen Ozoigbo , we had arranged a meeting with the Environment, Science, Technology, and Health Officer and some members of the local USAID team. We introduced them to the BearID Project and filled them in on the collaborations with USFQ and Ampere. Both teams expressed interest in what we were working towards. Hopefully we can build on this relationship in the future.

It was fantastic meeting Becky in person. For our first week together, since the server was still in transit, we focused on helping with basic analysis. Mary focused on creating scripts to automate laborious tasks like creating CSV files from lists of video content and fixing the date on videos from malfunctioning cameras. I set up and demonstrated how to use EcoAssist (from Addax Data Science ) to sort videos and images into blanks, animals, humans and videos.

EcoAssist can be run on a laptop, reducing the amount of time spent manually analyzing videos, especially the blanks. I reached out to the developer, Peter van Lunteren , about adding in species classifiers relevant to Ecuadoran ecosystems. I connected Peter with Kyra Swanson , Research Coordinator at SDZWA, who has been working on models for similar ecosystems in Peru. By the end of our trip, an Amazon model from SDZWA was fully integrated and work was starting on one for the Andes.

To really take full advantage of the Ampere Server, we wanted to establish a relationship with the Computer Science department at USFQ. Becky set up a meeting with Daniel Riofrio, Coordinator of Computer Science Engineering. Daniel’s department had collaborated with the biology department in the past and was interested in doing so again. He was excited to hear about the Ampere Server, and thought his department could help expand its capabilities and work to develop undergraduate projects to support the biodiversity server concept.

One of the puma projects Becky is working on is in collaboration with the Mashpi Lodge , situated in the Mashpi Reserve at the transition from the lower montane rainforest and the cloud forest. Through the relationship and thanks to Research & Biology Director, Mateo Roldán R. , we were able to visit the reserve and connect with the team there. In addition to the puma project with USFQ, researchers are still discovering new species, such as the Mashpi Glass Frog (discovered in 2014).

Through the university we also visited the Tiputini Biodiversity Station (TBS), a research facility in Amazonian Ecuador affiliated with USFQ. In addition to experiencing the Amazon, we could learn about camera trapping projects there as well. Before our trip, we had a video conference call with Becky and TBS Director Gonzalo Rivas-Torres. Gonzalo was excited about the biodiversity server heading to USFQ, and wanted the station to participate. He and Ciara Wirth, Investigation Coordinator, hinted at decades of valuable camera trap data and many ongoing and future projects. We managed to visit the station for 5 days. Between climbing the canopy tower, guided walks in the rainforest and night floats on the Tiputini River to see caiman, we managed to introduce EcoAssist and the Amazon model to Santiago Arroyo, Field Supervisor, who immediately downloaded and started working with the application.?

By the time we returned to USFQ for our final week, the Ampere server had arrived and been racked in their server room. Working with the USFQ IT team, we got the server set up and connected to the university system. Once we have VPN access to the server, we can start setting it up and start planning projects with Becky, Daniel and the others.?

In the meantime, we continued to develop helper scripts and made sure we had all the data we need for Andean bear identification. We also copied some data to help test the species model for the Andes ecosystem. We’re hoping this will also work well for the cloud forests in the lower elevations.

Hasta la vista, Ecuador

I’m sitting at the airport in Quito waiting for our flight back as I work on these final words. Earlier today I received the credentials for the USFQ VPN, so we can get started on setting up the server once we’re back State-side.

Here’s what we’ve got planned for the coming months:

• Install a large SSD on the Ampere server for the core data
• Install and configure TRAPPER on the Ampere server
• Integrate the relevant species models into TRAPPER-AI (Andes, Amazon, etc.)
• Discuss how USFQ CS department can get involved with the server
• Work on Andean bear identification!

We better get on it, there’s only 2 years, 3 months until I’m eligible for my next sabbatical - and a whole new adventure.

Keep an eye on this space for updates.

Prologue

The highlight of our trip happened while we were on a trip across the Andes with our guide, Santiago Molina . Santi is a bear expert and has collaborated with Becky on numerous projects. He took us near an area where they had done an Andean bear camera trap survey years before. Having heard there had been a bear in the area a few days earlier, Santi was pointing out signs of recent feedings (torn up bromeliads). Amazingly, we found the bear and experienced an incredible 1.5 hours taking photos ands videos while she was eating.

Even more incredible, we were later able to identify the individual (using manual methods) as one of the bears from the earlier survey! See Oso_14_A (bottom row, second from left):

We hope our efforts will enable more data connections like this, across distance and time!