Event Details

DSI Wednesday Lunch Time Talks - all are welcome! Calendar Entry: 26th May @ 12.30 Speaker: Victoria Janes-Basset , Senior Research Associate, LEC

Title: Data science approaches for soil carbon mapping: comparing and evaluating statistical and machine learning methods

Soils are the largest terrestrial store of carbon, storing more carbon than the atmosphere and the biosphere combined. Soil carbon plays a key role in the delivery of a wide range of ecosystem services including climate regulation, food production, water quality and regulation. A need exists to quantify soil carbon stocks at regional and national scales to guide long-term sustainable management of this natural asset.

Using national data sets and statistical approaches, spatial estimates of soil carbon across the UK have been generated. However, the level of uncertainty within these estimations is not fully recognised due to the uncertainty associated with the methodological choice and its application in generating these projections. Additionally, there is no standard method of validation across these datasets, complicating comparison. There are a range of novel data science methods including advanced statistical and machine learning techniques that can be applied, but currently little experience of how useful they can be in assessing and making sense of complex (multivariate) environmental data exists. Much like with process-based models, there is a need to understand which data science methodology is most suitable for a given research question.

As part of the Ensemble project, we’ve been using the Countryside Survey soil carbon data in combination with other nationally available datasets to generate spatial maps of soil carbon at a national scale using several data science approaches (e.g., generalised additive models, Gaussian processes, random forests, neural networks, boosted regression trees). By allowing these models to select from the same input data, we will provide a direct comparison of each method in terms of application, prediction power, uncertainty quantification and process understanding. Our aim is to develop an understanding of the suitability of data science methods for spatial mapping questions, whilst feeding into future sampling strategies and data needs to further understanding and reduce uncertainty in national soil carbon estimates.

Teams link