A deep learning framework for enhancing habitat identification based on species composition

Prepared by César Leblanc, Pierre Bonnet, Maximilien Servajean, Milan Chytrý, Svetlana Aćić, Olivier Argagnon, Ariel Bergamini, Idoia Biurrun, Gianmaria Bonari, Juan A. Campos, Andraž Čarni, Renata Ćusterevska, Michele De Sanctis, Jürgen Dengler, Emmanuel Garbolino, Valentin Golub, Ute Jandt, Florian Jansen, Maria Lebedeva, Jonathan Lenoir, Jesper Erenskjold Moeslund, Aaron Pérez-Haase, Remigiusz Pielech, Jozef Šibík, Zvjezdana Stančić, Angela Stanisci, Grzegorz Swacha, Domas Uogintas, Kiril Vassilev, Thomas Wohlgemuth, and Alexis Joly

The hdm-framework is a generic, free, and open-source framework using artificial intelligence and combining plant species composition and environmental data to enhance vegetation classification. It outperforms the traditional expert systems on habitat type identification. Schema from the paper.

Understanding and protecting natural areas is essential for preserving biodiversity. One way to do this is by accurately identifying and classifying different types of habitats, because it can help scientists and conservationists know where to focus their efforts. This study focused on improving how we identify and classify habitats across Europe, home to many unique ecosystems.

The main goal of this research was to explore if advanced computer techniques, specifically deep learning, could be used to better classify European habitats. Deep learning is a type of artificial intelligence that can learn from large amounts of data. In this study, we used deep learning to analyze around 1,000,000 vegetation surveys. These surveys included information about all occurring plant species from different areas and their importance value (i.e., how dominant or abundant each species is). Additionally, they contain information on the structure of the vegetation, location, time, and, in most cases, at least some environmental factors measured or estimated in the plot (e.g., slope, land use type, soil parameters).

We tested various computer models to see which one could best identify the type of habitat based on the plant data. We also experimented with different ways of processing and using the data to get the most accurate results. After careful validation and testing by human experts, we developed a generic, free and open-source framework. This tool can classify habitats much more accurately than traditional methods used by researchers and practitioners. Traditional approaches rely on human judgment, while deep learning can handle much larger datasets and detect patterns that may not be obvious, even to experts.

Overview of hdm-framework. The panels display the sequence of tasks performed during each of the four main stages (dataset preparation, parameters evaluation, model training and habitat prediction). Schema taken from the original paper.

One of the key findings was that the computer models could accurately identify habitats even without environmental data, or detailed information about how much of each plant was present. This suggests that flora alone is a strong marker of ecosystems and that the presence of certain key species is a strong indicator of the habitat type. The model we developed achieved an accuracy rate of 88.74% across Europe and more than double the accuracy of previous methods on an independent test set.

Why does this matter? Accurate habitat classification is crucial for conservation efforts. By using our new deep learning framework, users with various backgrounds can better understand where different types of habitats are located and how they are changing over time. This can help in making more informed decisions about where to protect or restore habitats, ultimately contributing to the preservation of biodiversity in Europe.

The tool developed in this study is shared and maintained through a GitHub repository: https://github.com/cesar-leblanc/hdm-framework. It is easy to use and allows anyone to start identifying habitats. If you’re neither a deep learning nor a vegetation expert but want to predict the habitat type of a given area using species compositions, this framework is for you. In conclusion, this study shows that advanced computer techniques like deep learning can greatly improve our ability to classify and understand natural habitats. This new approach is not only more accurate but also more efficient, making it a valuable tool for anyone involved in habitat conservation.

This is a plain language summary of the paper by César Leblanc and colleagues, published in Applied Vegetation Science (https://doi.org/10.1111/avsc.12802). This post was prepared by César Leblanc.