Artificial Intelligence Revolutionizing Biodiversity Conservation: Groundbreaking Study Showcases the Power of Recurrent Neural Networks
- February 5th, 2025
- Biodiversity
- Print Article
A recent study published in Ecological Indicators explores the potential of artificial intelligence (AI) to significantly enhance biodiversity conservation efforts. The study, led by PhD student Sébastien Lhoumeau of the University of the Azores and co-supervised by the Atlantic International Research Centre, in collaboration with researchers from these institutions, namely João Pinelo and Paulo A.V. Borges, examines the potential of Recurrent Neural Networks (RNNs) in predicting arthropod population dynamics based on long-term ecological data.
Implications for Conservation Science
As biodiversity loss accelerates due to climate change and habitat destruction, advanced AI models offer promising solutions for conservationists seeking to mitigate these impacts. The ability of RNNs to analyse time-series data without pre-set assumptions makes them particularly well-suited for ecological applications, where environmental conditions and species interactions are highly dynamic.
"As a call to action, our study demonstrates the practical use of RNN in a specific context and encourages researchers to explore its potential in diverse ecosystems. It can advance biodiversity conservation research and decision-making."
Breakthrough in Ecological Forecasting
The present study utilised data gathered between 2012 and 2023 from a long-term biodiversity monitoring programme conducted in the native forests of Terceira Island, Azores, to compare the predictive capability of recurrent neural networks (RNNs) with traditional forecasting methodologies, encompassing Seasonal Autoregressive Integrated Moving Average (SARIMA) and local polynomial regression (LOESS). The research focused on 39 arthropod species, using seasonal abundance data collected via SLAM traps to model ecological trends over time. Furthermore, the trained IA model was used to forecast arthropod populations under several scenarios, thereby showcasing the usefulness of its application in the conservation of biodiversity.
Key Findings
- RNN models significantly outperformed LOESS in both goodness-of-fit and predictive accuracy, demonstrating their ability to detect and process complex ecological patterns.
- While SARIMA models performed well in direct data prediction, RNNs exhibited superior training fit, allowing for more accurate scenario simulations.
- Simulations of extinction and invasion scenarios provided valuable insights into ecosystem resilience and vulnerability, emphasizing the potential impact of invasive species on native arthropod communities.
- The study highlights the potential of AI-driven models to support conservation decision-making by offering robust and flexible ecological forecasting tools.
About the Research
The study, titled "Artificial Intelligence for Biodiversity: Exploring the Potential of Recurrent Neural Networks in Forecasting Arthropod Dynamics Based on Time Series," was conducted at the cE3c-Centre for Ecology, Evolution and Environmental Changes, within the Azorean Biodiversity Group and the CHANGE – Global Change and Sustainability Institute at the University of the Azores, co-supervised by the Atlantic International Research Centre.
The research is published in Ecological Indicators and is available as an open-access article under the CC BY license.
For more details, access the study here: https://doi.org/10.1016/j.ecolind.2025.113119
Sébastien Lhoumeau has produced a video to provide an explanation of the study. The video can be accessed via the following link: https://youtu.be/ei_qlOzTJ8U
Media Contact
Sébastien Lhoumeau, PhD student
Email: seb.lhoumeau@gmail.com
https://www.sciencedirect.com/science/article/pii/S1470160X25000482?viaihub