Artificial Intelligence for Biodiversity: Exploring the Potential of Recurrent Neural Networks in Forecasting Arthropod Dynamics Based on Time Series

Abstract

In the current biodiversity crisis, the increasing demand for effective conservation tools aligns with significant advancements in artificial intelligence (AI). There is the need for the development of more robust and accurate forecasting methods, ultimately enhancing our understanding of ecological dynamics and supporting the formulation of effective conservation strategies. This research conducted a comparative analysis of Local Polynomial Regression (LOESS), Seasonal Autoregressive Integrated Moving Average (SARIMA), and Recurrent Neural Network (RNN) models for time-series prediction. Using a unique Long-Term Monitoring Program for island forest arthropods (2012-2023), wherein we selected the 39 most prevalent species collected using SLAM (Sea Land Air Malaise) traps within a native forest fragment on Terceira Island in the Azores archipelago. The results indicate that RNN outperformed LOESS in terms of both goodness of fit and overall accuracy. Although RNN did not surpass classical SARIMA in data prediction, it demonstrated superior goodness-of-fit on the training dataset. Furthermore, we investigated extinction and invasion scenarios within the Terceira arthropod assemblage, providing insight into broader implications and avenues for future research. This study discusses the utility and limitations of RNN models in biodiversity conservation through various scenarios. It contributes to the ongoing discourse at the convergence of conservation, ecology, and artificial intelligence (AI), highlighting advancements and innovative solutions crucial for the effective implementation of conservation strategies.

Keywords: Conservation scenario, Ecological modelling, Azores arthropods, machine learning, Long-term ecological data, Model comparison