Harnessing the power of AI for biodiversity forecasting

The challenges of our time are becoming increasingly complex, for example, how we address the biodiversity crisis and climate change are multifaceted problems. The post 2020 global biodiversity framework drafts have outlined ambitious conservation targets and strategies to prevent further species extinctions. Management strategies require an understanding of how biodiversity might respond to future scenarios and threats.

However, the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) expressed low confidence in abilities to forecast ecological change. While there have been large-scale efforts devoted to projecting climate change, modelling frameworks for projecting future biodiversity and ecological changes have been neglected.

Species responses to climate and land use change are typically modelled in a correlative manner, yet it is increasingly recognised that process-based models are essential for understanding the complex feedbacks between threats to species and their responses1. Data for parameterising such models are however frequently sparse or lacking for individual species2.

To facilitate the development of process-based models for forecasting, the integration of machine learning (ML) methods to address data limitation problems is needed.

This project will develop ML methodologies for predicting species' traits with uncertainty quantification. The development of computational approaches that synthesise and address scarcities in species data through generative ML models could transform research and how the current biodiversity crisis is managed. Potential methods could include generative deep learning models, e.g. transformers with variational autoencoders for multitasking, with data sourced from ever-growing trait repositories (e.g.

TRY plant database, MammalBase). These approaches will be used to generate data sets for different taxa that can be used in process-based species range models. This pipeline will allow for projections of how future land use change and/or climate change might affect the distribution of species and the identification of species that may be more vulnerable to changing landscapes.

As such, this project will address key knowledge and methodological gaps in biodiversity forecasting and ML and help to prioritise conservation efforts.