More examples

Monitoring systems at scale

The observational data available to researchers is growing substantially and is vast and varied, helping us understand physical processes on scales ranging from centimetres to kilometres. AI-powered autonomous robotic platforms provide complementary solutions to monitoring activities requiring human presence. AI tools don’t fully replace the need for human-led collection and validation of observational data but can be used to unify datasets from those various monitoring platforms in order to provide a holistic picture.

Since the middle of 2019, researchers at the Turing and British Antarctic Survey (BAS) have been harnessing powerful AI algorithms to extract information from vast atmospheric, oceanic and sea ice datasets from the polar regions. With tens of millions of data points from our satellite-derived datasets, we are training our AI algorithms to predict future sea ice, with the ability to learn physical relationships between climate variables over both space and time. Additional AI applications in ASG research that have significantly improved environmental monitoring capacity include the project for monitoring iceberg populations in the Southern Ocean from space. This project applies ML techniques and synthetic aperture radar (SAR) satellite imagery to identify icebergs and track their disintegration in the Amundsen Sea, Antarctica. Approaches from this work will improve the identification of icebergs in both the open ocean and within the sea ice pack.

Closely related is Seals from Space, another project led by BAS for automating Antarctic ecosystem monitoring via high-resolution satellite imagery, tracking seals as potential indicators for the Antarctic ecosystem’s health. An AI-enabled system for classifying sea ice and mapping seals can be used to transform satellite images into numbers representing observed ice and seals. Using these environmental features, researchers can address ecological questions concerning seals’ preferred habitat and how their surroundings are changing over time.

The DeepSensor toolkit leverages cutting-edge advances in probabilistic AI modelling to intelligently fuse data from gridded satellite imagery and point-based surface sensors. In particular, DeepSensor enables new solutions in two problem areas: (1) increasing the spatial granularity of surface climate variables, generalising across spatially sparse and dense environmental sensor networks, and (2) optimising sensor placement to reduce uncertainties in climate variables and providing robust information to environmental monitoring groups. As the toolkit is being developed, the project team are focusing on three distinct use cases: temperature over Antarctica; soil moisture across the UK; and temperature in London.

Another relevant example for environmental research comes from Ecosystem of Digital Twins — an ASG theme with a broader scope to incorporate digital twins (multiple digital representations of physical infrastructure) at the component, asset, system and process levels. Researchers in the “Digital Twins of Fleets” project propose population-level analysis for modelling engineering systems by combining abundant data from systems in operation for a long time with sparse data from systems established more recently (Bull et al, 2022). Their approach encodes domain expertise to constrain the ML model via assumptions (and prior distributions), allowing the methodology to automatically share information between similar assets. This study has been optimised on interpretable fleet models of different in situ data, where ‘fleet’ refers to a population of assets that constitute engineering infrastructure such as civil structures of roads and bridges or trains in a transportation network. Showcasing examples from survival analysis of an operational truck fleet and wind-power predictions for an operational wind farm, they demonstrate the wide applicability in practical infrastructure monitoring.