The AI & Climate Sustainability Sandbox brings forward-thinking companies together with researchers to tackle real climate and sustainability challenges through practical AI development.
Delivered by The Co-Centre for Climate Biodiversity + Water in partnership with Technology Ireland ICT Skillnet, the programme is designed to move quickly from problem definition to early-stage technical solutions, with prototype outputs and clear next steps.
What follows are two case studies from recent Sandbox participants, showing the types of industry challenges that can be addressed within an intensive sprint environment.
Case Study 1 – ODOS Tech: AI-driven land-use mapping to strengthen agricultural sustainability analytics
The challenge ODOS Tech needed to solve
ODOS Tech supports agricultural suppliers who must meet carbon-reduction and biodiversity targets using high-resolution land-use and land-cover mapping.
Their existing mapping system was not reaching the 95% accuracy required for operational reliability. This created a dependency on time-consuming manual correction, which also introduced regional inconsistency and limited scalability.
What was delivered during the Sandbox
During the intensive Sandbox sprint, the team developed an end-to-end AI segmentation approach designed to move ODOS Tech closer to the 95% target.
The work included fine-tuning a pretrained Earth-observation foundation model (AIFS-Land) on ODOS Tech’s six-class agricultural dataset, as well as a class-balancing pipeline to improve learning on rare but important land-use types such as cropland. A proof-of-concept also explored semi-automated labelling using Segment Anything Model (SAM) to reduce manual effort while improving boundary accuracy.
Impact and outcomes
The enhanced model achieved an mIoU of 88%, showing a meaningful performance improvement and stronger land-use discrimination versus the baseline.
The analysis also highlighted data quality gaps, including cases where the model appeared to correctly identify forest regions that were missing in the original labels. The project set out a practical roadmap to reduce manual correction workloads and expand mapping capability into new regions.
Case Study 2 – Zinto Labs: AI for marine preservation through early detection of invasive seaweed
The challenge Zinto Labs needed to solve
The rapid spread of a known invasive seaweed is creating serious environmental and economic impacts along coastlines.
Manual monitoring is too slow to respond effectively, and once the species takes hold it becomes extremely difficult to remove. The core need was a faster, predictive approach that could anticipate risk areas so authorities can act earlier.
What was delivered during the Sandbox
Across an intensive sprint, the team developed the invasive seaweed Temporal (ROST) Early Detection Model as a multi-layered solution.
This included habitat suitability analysis using environmental variables such as sea temperature, salinity, and pH with the Bio-Oracle dataset, plus a CNN trained on Copernicus Sentinel satellite imagery to classify algae presence. The approach also proposed a Graph Neural Network architecture to capture spatio-temporal spread pathways, including factors like ocean currents and marine traffic.
Impact and outcomes
The project demonstrated a shift from reactive management to proactive, predictive strategy. The models identified high-suitability hotspots along the European Atlantic coast and highlighted potential future spread zones in South America and Australia. The work supports more targeted monitoring and provides a reusable framework that can be adapted to other invasive species using open-source data and advanced machine learning methods.
Explore the AI & Climate Sustainability Sandbox
If your organisation has a climate or sustainability challenge that could benefit from rapid AI prototyping, the Sandbox is designed to help you test approaches quickly and leave with tangible outputs and a clear roadmap.
Using AI to tackle climate and sustainability challenges.
