Field note
Apr 2026
Sisters' Islands, after the bleach: what the reef looked like to our model
An eight-week diary from the southern reefs — what hyperspectral foundation models see when a coral begins to fade.
An environmental AI research institute · SingaporeEnvironmental AI · Singapore
Foundation models for a living planet.
We build AI foundation models to help us understand, protect, and restore the natural world — with Singapore as our home and the tropics as our laboratory.
Research pillars
Six landscapes. One island.
Each programme answers a question Singapore's policymakers, scientists, and communities are already asking — with models that ship into the field, not just into papers.
01
Marine intelligence
Coral resilience for the Southern Islands
Hyperspectral imaging and vision foundation models trained on more than a decade of reef surveys — surfacing the early signs of bleaching, mapping cover and species across Singapore's Southern reefs.
- Reef sites monitored
- 63
- Hectares mapped
- 1,240
02
Climate equity
Urban heat that moves with you
A neighbourhood-scale heat model trained on 9,400 IoT sensors and Sentinel-3 thermals — finding cooler walking routes for the elderly and surfacing where the next park, awning or louvre belongs.
- Streets mapped
- 100%
- °C avg cooling delivered
- −2.1
03
Biodiversity
Listening to the rainforest
Bioacoustic ML models distinguish 312 native species across Bukit Timah and the Central Catchment — tracking pangolin, hornbill and slow loris populations without disturbing them.
- Species identified
- 312
- Recording hours / week
- 4,800
04
Blue carbon
Mangroves & the carbon coastline
LiDAR + diffusion models reconstruct above- and below-ground biomass of every mangrove in Singapore, validating blue-carbon credits with field-grade certainty.
- ha mapped
- 812
- tCO₂e verified
- 146k
05
Water quality
Clean water for a tropical island
Foundation models read every reservoir, canal and stream — spotting algal blooms, sediment plumes and pollution sources so that the water we drink, swim in and live around stays clean.
- Catchments observed
- 17
- Sample resolution
- 10 m
06
Atmosphere
Air pollution at the postal-code scale
Multimodal models fuse satellite imagery, low-cost sensors and traffic to map air quality street by street — giving every household clarity instead of a city-wide average.
- Postal codes covered
- 8,200
- Pollutants tracked
- 9
Field impact · 2025–2026
Numbers that walk out of the lab and into the city.
We measure ourselves by what changes outside the lab — cleaner air, cleaner water, cooler streets, reefs that spawn.
- 146,800
- Tonnes CO₂e mapped & verified
- 63
- Reefs under continuous AI watch
- −2.1°C
- Avg. neighbourhood cooling
- 312
- Native species acoustically tracked
- 8,200
- Postal codes mapped for clean air & water
- 100%
- Open methods, datasets & weights
Journal
Field notes from the island.
Paper
Mar 2026
WaterMark-SG: a foundation model for tropical water quality
How a single multimodal model reads every reservoir, canal and stream on the island to surface algal blooms, sediment plumes and contamination — and why that matters for the water we drink.
Open dataset
Feb 2026
Bukit Timah 24/7: 4,800 weekly hours of rainforest acoustics
A four-year, fully labelled bioacoustic corpus of 312 species — released under CC-BY for researchers everywhere.
The decade ahead
Singapore is small enough to get this right — ambitious enough to lead.
We are building the most loved environmental research institute in the tropics. The next decade is being decided now.
hello@singapore-research.com
28 Orchard Boulevard, Singapore 248649