AI-managed remote sensing
Remote cameras at San Domingo Arbory that upload images on their own, day and night (infrared after dark), and AI processing of drone footage and smartphone photos from the field: part of a broader effort to use computer vision models and AI agents to understand sensor data and images from the sites, feeding our digital twins.
About this project
What we’re tracking
- Species recorded each week by the remote cameras, day and night (infrared)
- Camera uptime and automatic uploads
- Vegetation on treated and untreated ground across the seasons, from drone footage and field photos
- How often a person has to correct what the AI reports
Remote sensing here means recording the land with cameras and sensors instead of notes alone: cameras that run on their own, drones flying over the sites, and the phone cameras we carry on site visits. Together they produce far more images and readings than anyone can sort by hand. AI-managed remote sensing is how we aim to keep up: AI makes a first pass through the images and sensor data, and a person checks what it finds.
This project is part of a broader effort to use computer vision models and AI agents to understand sensor data and images from the field: what lives where, how the vegetation responds to treatment, and what changes across the seasons.
What it covers
- Remote cameras. Solar-powered cameras above a backcountry road at San Domingo Arbory send their images over a cellular connection to the cloud automatically, so new frames arrive without anyone walking out to collect cards. They record day and night, switching to infrared after dark. Black bear (Ursus americanus) is among the night visitors. The setup began as a prototype.
- Drone footage. Repeat flights over the sites record the forest, the meadows and treated ground through the seasons. AI helps us work through hours of footage.
- Smartphone photos. Photos from site visits (a plant in flower, storm damage, a track in the mud) are part of the same record, and AI helps sort and describe them.
What we want to learn
- What lives where. Which species the cameras record, how often, and at what hour of the night, by week and by season. Every frame carries the camera’s date and time.
- How the vegetation responds to treatment. How thinned, burned and seeded ground grows back compared with untreated ground, from repeat drone footage and field photos.
- What changes across the seasons. Green-up, dry-down, storms and regrowth, followed on the same ground year after year.
- How far the AI can be trusted. How often a person has to correct what the AI reports, and whether that share falls as the models improve.
- How the equipment holds up. Camera uptime and upload success through winter storms and summer heat. A remote, solar-powered camera is only useful if it stays up.
Why San Domingo
San Domingo Arbory is about 450 acres of mixed forest, meadows and creek valley along San Domingo Creek outside Murphys, with fuel reduction, kiln burns and trail work underway. That work may change which animals use the land, and when, and how the vegetation grows back. A record kept by instruments that report on their own is the baseline to compare against, and it can be kept for years.
How it feeds the digital twins
Our digital twins of the sites are designed to take sensor streams and images like these, alongside drone surveys, satellite time series and the treatment plans drawn for San Domingo. With the camera, drone and photo records on the same timeline as fuel reduction and kiln burns, we’ll be able to ask whether wildlife activity and vegetation change as the work goes on, and how quickly. Connecting the camera record to the twins is one of the next steps. The cameras’ positions stay private there too.
How we handle images and sensor data
- We publish wildlife and land, not people. Frames that show people or vehicles stay private, and we use no facial recognition.
- Footage from property-security cameras is for security only and is never published.
- The cameras’ exact locations, and the locations of sensitive wildlife such as dens and nests, are never published. Drone flight logs with positions stay private.
- Camera records can be shared with research collaborators on request.
What comes next
- Computer-vision labels for the camera frames. Software that sorts frames by species, with a person confirming or correcting each label.
- The camera record in the digital twins, as one of their sensor layers.
- A longer record. The record becomes useful as it covers whole seasons, and later field notes will report what it shows across the year.
Running camera traps on private land, or building computer vision models for field images? Get in touch to compare setups.