Applications of machine learning techniques in ecology

Abstract

The size and diversity of ecological data are growing in exponential ways due to the modern advances in informatics applications, web services, and cloud systems that yield a great flux of information available to scientists, stakeholders, and the public. The great global challenges at the level of nature conservation, biodiversity loss due to anthropogenic effects, global changes, vector epidemiological monitoring, and sustainability are complex problems that require fast and accurate real-time analysis with suitable statistical tools. Machine Learning (ML) is a subfield of Artificial Intelligence (AI) that includes a heterogeneous set of theories and data-driven algorithms that allow computers to capture relationships and hidden patterns “not explicitly given by humans” better than traditional statistical methods. Here, we trained and evaluated using different approaches a set of ML algorithms, to identify environmental drivers that shape the realized niche of different species and to evaluate the effects of climate change in organisms from freshwater and marine ecosystems. Moreover, ML was used to study wings’ shape variation of sibling malaric vectors in a contest of epidemiological surveillance and to identify the influence of chemical and physical environmental features on the assemblage patterns of different freshwater zooplankton communities. A particular branch of ML that acquired importance in the last years, deep learning, was applied to ecoacustics, to demonstrate how deep learning captures different aspects of the marine environment using large marine Passive Acoustic Monitoring (PAM) data. We demonstrated how the flexibility of the ML algorithms address successfully different ecological problems across taxa and different environments. Finally, data sharing and free AI programs might improve the use of ML in ecology to speed up the process that leads to new scientific discoveries.