Predictive text mining for global biodiversity indicators and models.

Biodiversity indicators and models are important for policy decisions, scientific understanding and public engagement. Although advances in remote-sensing are revolutionising the production of indicators of ecosystem structure and function, biodiversity indicators still rely on bottom-up aggregation and synthesis of data from many local surveys. Current ways of doing this are very time-consuming because finding and extracting them from the burgeoning literature is still a slow, manual process.

The PREDICTS (based at the NHM) and LPI (based at the ZSL) databases are global in coverage and their associated biodiversity indicators are internationally recognsed. However, both datasets have key spatial and thematic gaps. These gaps undermine attempts to produce indicators that are representative of biodiversity, rather than reflecting the data's geographic, taxonomic or ecological biases.

This project will use new developments in text mining and machine learning to greatly increase the rate of data flow into the above datasets, with the potential for preferentially targeting data concerning taxa and geographic regions which are currently under-represented. Subsequently, recalculating the associated indicators using the enhanced databases will provide a better overview of the state of the natural world and faciliate a comparison with the values obtained from the current, manually collated data. Given the broad ecological coverage of the two databases, further avenues of research will also be possible.