Scientists have devised a revolutionary new method to harness the cutting-edge capabilities AI to map and model the natural environment in great detail.
Charlie Kirkwood, University of Exeter’s expert in modeling Earth’s natural features, has developed a sophisticated new method of modeling them in greater detail.
The new technique can recognize complex features and aspects of terrain that are not possible with traditional methods. This allows for enhanced-quality environmental maps.
The new system could also open the door to new discoveries about the relationships in the natural environment that may help address some of the larger climate and environment problems of the 21st century.st century.
The study has been published in a leading journal Mathematical GeosciencesAs part of a special issue in geostatistics, machine learning.
It is time-consuming, costly, and takes a lot of effort to model and map the environment. Cost limits the number and quality of observations that can easily be obtained. This means that the creation of comprehensive, spatially-continuous maps is dependent on filling in any gaps between these observations.
Scientists have a wide range of information sources available to fill these observation gaps. Conventional modelling methods rely on users to manually create predictive features from these data. This could include generating slope angles and curvatures using terrain elevation data, in order to help explain the spatial distribution.
Scientists believe that there are many nuanced relationships in the natural environment that models using traditional manual feature-engineering approaches might miss.
The study reveals a new AI approach that is radically different. It uses environmental information extraction as an optimization problem. This allows it to recognise and use relationships that may otherwise go unnoticed or unutilised by humans using traditional modelling methods.
This not only improves map quality but also opens up the possibility of discovering new relationships in nature by AI. At the same time, it eliminates a lot of trial and error in the modelling process.
Charlie Kirkwood, a University of Exeter postgraduate student, said: “Our models must be able to provide useful information for decision-making. This means that they need to be as precise as possible and trustworthy.
“Our AI approach is built within a Bayesian statistic framework that allows us quantify these uncertainties and provide a range o uncertainty measures including credible intervals, exceedance probabilities, and other bespoke products that will feed directly to decision making processes. All this is done while harnessing all available information more efficiently than traditional approaches allow. This can be seen in the detail of map.
The new approach was demonstrated with stream sediment calcium concentration observations from the British Geological Survey’s Geochemical Baseline Survey of the Environment, (G-BASE).
The distribution of calcium in the environment is an important factor in soil fertility. This is primarily controlled by geology (with different rock types containing different levels of calcium) and hydrological processes at ground level.
Calcium presents a difficult use case for the AI approach. It must learn to recognize and utilise features that relate to both bedrock geology (e.g. Different terrain textures and slope breaksSurface hydrology (e.g. Drainage, river channels
The method, the scientists say, has produced a spectacularly detailed and accurate map which, despite depicting just one element — calcium, reveals the geology of Britain in arguably a new level of detail thanks to the information-extracting power of the new AI approach. The team believes that this work represents a new era in environmental mapping, thanks to the combined expertise, research skills and data resources of its collaborators, the University of Exeter, Met Office and British Geological Survey.
Professor Gavin Shaddick of the University of Exeter stated that “This is a remarkable example of Environmental Intelligence. This is the use of AI in solving challenges in environmental sciences. This work is a model in combining technical knowledge of AI/machine learning with expertise in geosciences in order to produce a new methodology that directly addresses key questions in mapping environmental data. These methodological innovations could be used to produce detailed maps for a wide range of environmental hazards. They also have the potential to provide rich information for decision makers and scientists.
Garry Baker (Interim Chief Digital Officer, British Geological Survey) added: “This paper shows how environmental information such the BGS geochemical databases can be reassessed through new approaches (AI Spatial Interpolation). It demonstrates the value of ongoing environmental research and how it can draw on the extensive datasets that are available to everyone through the National Geoscience Data Centre, NERC, UKRI data repositories.
Dr Kirstine Dale, Principal Fellow for Data Science at the Met Office and Co-Director for the Joint Centre for Excellence in Environmental Intelligence, commented on the importance of this work. “This is an important example how data science can transform our understanding about the natural world. It highlights the benefits of collaboration across disciplines. In this instance, mathematicians, computer scientists, and weather specialists enrich our understanding of nature in a way that no one discipline can.