Modelling Species Distributions with Deep Learning to Predict Plant Extinction Risk and Assess Climate Change Impacts

TL;DR

A novel method for classifying the IUCN status of species benefiting from the generalisation power of species distribution models based on deep learning is evaluated, which matches state-of-the-art classification performance while relying on flexible SDM-based features that capture species' environmental preferences.

Abstract

The post-2020 global biodiversity framework needs ambitious, research-based targets. Estimating the accelerated extinction risk due to climate change is critical. The International Union for Conservation of Nature (IUCN) measures the extinction risk of species. Automatic methods have been developed to provide information on the IUCN status of under-assessed taxa. However, these compensatory methods are based on current species characteristics, mainly geographical, which precludes their use in future projections. Here, we evaluate a novel method for classifying the IUCN status of species benefiting from the generalisation power of species distribution models based on deep learning. Our method matches state-of-the-art classification performance while relying on flexible SDM-based features that capture species' environmental preferences. Cross-validation yields average accuracies of 0.61 for status classification and 0.78 for binary classification. Climate change will reshape future species distributions. Under the species-environment equilibrium hypothesis, SDM projections approximate plausible future outcomes. Two extremes of species dispersal capacity are considered: unlimited or null. The projected species distributions are translated into features feeding our IUCN classification method. Finally, trends in threatened species are analysed over time and i) by continent and as a function of average ii) latitude or iii) altitude. The proportion of threatened species is increasing globally, with critical rates in Africa, Asia and South America. Furthermore, the proportion of threatened species is predicted to peak around the two Tropics, at the Equator, in the lowlands and at altitudes of 800-1,500 m.

Figures (18)

• Figure 1: Method scheme. The first column represents the current time frame (training and evaluation), while the right column represents a future scenario (projection). Step 1 consists in associating a given set of support points (true observations or dispersal scenario) with environmental covariates using deep-SDM inference. Species niches are indicated by dashed circles. In Step 2, the predicted features are summarised by taking their mean, standard deviation and sum, and the result is concatenated. This operation allows the information that was at the point level to be condensed to the species level. Finally, the Step 3 is the mapping between the species summary feature and its conservation status. After training and validation in the present with real observations, the random forest classifier is trained within dispersal scenarios to ensure coherence with future projections. Classification can be either binary, as shown, or at the IUCN status level.
• Figure 2: A. Distribution of the 889 IUCN extinction risk status from our dataset. B. Classification performance comparison between the state-of-the-art IUCNN method and ours zizka_iucnn_2021. The 10-fold cross-validation results in an accuracy distribution represented by boxplots. The first row shows the binary classification and the second row the status classification. Micro- and macro-average accuracies are identical, as the binary status distribution is balanced. The IUCNN method achieves an average accuracy of 0.81 for binary classification and our deep SDM method achieves 0.78. However, for status classification, our method gives a micro-average accuracy of 0.61 and a macro-average accuracy of 0.43 and the IUCNN method 0.60 and 0.41 respectively.
• Figure 3: Binary status proportions per continent and time period. All species are included and their number per continent is given in the subtitles. Error bars account for differences between the two dispersal scenarios.
• Figure 4: Species count histograms as a function of average latitude and time period. All species are included and colours indicate the binary extinction risk status. Bins cover four degrees of latitude.
• Figure 5: Species count histograms as a function of their average altitude and over time. All species are included and colours indicate the binary extinction risk status. Bins cover 150 metre elevation ranges.