AI-based Mapping of the Conservation Status of Orchid Assemblages at Global Scale

TL;DR

This study develops a global, kilometre-scale framework to map the conservation status of orchid assemblages by training a deep-SDM on about 1 million orchid occurrences from 14 thousand species. The model produces set-valued predictions S_lambda(x) and computes two key indicators, I_O (most critical status) and I_c (proportion of statuses), along with the Shannon index I_H to quantify assemblage threat and diversity, using both official IUCN statuses and automatically predicted ones via IUCNN. High-resolution global maps and zonal statistics reveal spatial patterns of threat, with pronounced hotspots in Madagascar, tropical Asia, and certain island regions, and a Sumatra case study shows how predicted statuses can reveal conservation gaps inside and outside protected areas. The approach demonstrates the potential of deep learning to generate scalable, policy-relevant conservation indicators for a major umbrella taxon, offering a tool to inform global and regional biodiversity action under post-2020 frameworks. The online interactive maps and multi-scale analyses provide a practical platform for researchers and decision-makers to identify priority areas and assess PA effectiveness, while acknowledging limitations in data quality and the need for ground-truthing and transparent species-level outputs.

Abstract

Although increasing threats on biodiversity are now widely recognised, there are no accurate global maps showing whether and where species assemblages are at risk. We hereby assess and map at kilometre resolution the conservation status of the iconic orchid family, and discuss the insights conveyed at multiple scales. We introduce a new Deep Species Distribution Model trained on 1M occurrences of 14K orchid species to predict their assemblages at global scale and at kilometre resolution. We propose two main indicators of the conservation status of the assemblages: (i) the proportion of threatened species, and (ii) the status of the most threatened species in the assemblage. We show and analyze the variation of these indicators at World scale and in relation to currently protected areas in Sumatra island. Global and interactive maps available online show the indicators of conservation status of orchid assemblages, with sharp spatial variations at all scales. The highest level of threat is found at Madagascar and the neighbouring islands. In Sumatra, we found good correspondence of protected areas with our indicators, but supplementing current IUCN assessments with status predictions results in alarming levels of species threat across the island. Recent advances in deep learning enable reliable mapping of the conservation status of species assemblages on a global scale. As an umbrella taxon, orchid family provides a reference for identifying vulnerable ecosystems worldwide, and prioritising conservation actions both at international and local levels.

Figures (14)

• Figure 1: Global comparison of the most critical IUCN status indicator according to three methods. (a) represents the IUCN information on our dataset: observations and available spatial data (polygons and points from https://www.iucnredlist.org/resources/spatial-data-download) taken together. Spatial data is available for only 167 IUCN-assessed orchids from our dataset, i.e. 1.2% of all species. (b) is the result of our species assemblage prediction model coloured by the most critical known IUCN status whereas (c) includes predicted IUCN status too in the indicator calculation. [Figure maps are under-sampled, see the website for full-resolution]
• Figure 2: Four indicators based on species assemblage predictions. (a) $\mathcal{I}_{\mathcal{H}}$ the Shannon index, (b) $\mathcal{I}_{\text{THREAT}}$ the weighted proportion of threatened species, (c) and (d) the weighted proportions of respectively CR species $\mathcal{I}_{\text{CR}}$ and EN species $\mathcal{I}_{\text{EN}}$. [Figure maps are under-sampled, see the website for full-resolution]
• Figure 3: Average proportion of species predicted as threatened by botanical country (WGSRPD level 3) versus average Shannon index. Countries are coloured in function of their continent (WGSRPD level 1) and top-15 countries of both variables are highlighted. Myanmar, Assam and Laos are the only three regions in the top-15 intersection whereas Pakistan and Cape Verde show especially high threatened species proportions with low diversity indices.
• Figure 4: Five indicators of species assemblage extinction risk applied on Sumatra island. Elevation is also provided and protected areas are hashed in blue (downloaded from https://www.protectedplanet.net/en). (a) elevation map, (b) Shannon index, (c) proportion of IUCN-assessed CR species in the predicted species assemblages. On the second line, species proportion of: (d) threatened species, (e) VU species only, and (f) CR species only (all statuses combined). [Maps in figures are under-sampled, see the website for full-resolution]
• Figure S1: Maps and statistics illustrating the post-filtering step with the geographic-prior. The support is a global regular grid with $0.5$ decimal degree resolution (59,823 points). (a) $|\hat{S}_\lambda|$, i.e. the species assemblage size before the filtering step. Northern latitudes -and especially northern Europe- present abnormally large species assemblages. This is a consequence of the generalisation / over-prediction trade-off described in Discussion. The prediction model is over-confident because of the extensive occurrence training data in northern European countries. (b) $|\hat{S}'_\lambda|$, i.e. the species assemblage size after the filtering step. The over-prediction bias at northern latitudes has been largely compensated. Empty predictions zones (red surrounded) have increased because of the geographic filtering, especially in the Sahara. (c) $|\hat{S}'_\lambda| - |\hat{S}_\lambda|$, i.e. the absolute size difference of the species assemblage before/after the filtering step. Regions having lost the highest number of species are northern European countries and the South Arabian Peninsula. (d) $\frac{|\hat{S}'_\lambda| - |\hat{S}_\lambda|}{|\hat{S}_\lambda|}$, i.e. the relative change in the species assemblage size before/after the filtering step. Regions mentioned in (c) are highlighted again. Saharan regions with empty predictions after geo-filtering do not appear to have lost high species number in (c). However, the clear yellow on map (d) indicates that these regions have lost all of the few species they were predicted to host. (e) Statistics on the absolute and relative size difference of the species assemblage before/after geo-filtering. $\Delta$species corresponds to map (c) and Relative change [%] corresponds to map (d).