Conference Agenda

As international conservation efforts aim to address biodiversity loss driven by climate change and anthropogenic disturbance, comprehensive information on species distributions which when assessed over a specific temporal scope represent that species distribution essential biodiversity variable (SD EBV) – has become a potential key resource for policymakers and stakeholders. Currently, much of this data is publicly accessible through expert range maps (e.g., IUCN Red List) and species occurrence records (e.g., GBIF). However, these kinds of data are fundamentally incomplete and subject to a variety of biases, thus completing such an information base that addresses all species of a taxon at spatial resolutions relevant to actionable conservation plans demands a highly scalable approach to species distribution modeling (SDM). One central pillar for temporally specific, spatially contiguous and global predictions is the integration of a range of remote sensing and remote sensing-supported climate products, collected at different spatial scales. In support of a range of information products and downstream indicator users, Map of Life has been advancing 1km2 global SDMs for a diverse array of taxa from vertebrates to invertebrates, common to rare species, and generalists to specialists, all with their associated challenges. In this presentation, we will share insights from our progress in producing, curating, and validating global SD EBVs, and also discuss the integration of novel data streams, such as trait-based data to apply target group background sampling, to enhance predictions. We will also address critical remaining challenges, including data gaps, computational limitations, and the complexities of generating SDMs for multiple taxa, and outline key next steps toward establishing a comprehensive, robust set of global SD EBVs that support global biodiversity measurement and conservation.

In the context of ever-increasing human impacts and accelerating climate warming, we need to better understand and predict species occurrences and abundances over space and time. In recent years, new types of Species Distribution Models (SDMs) based on deep learning (deep-SDMs) have been successfully applied for occurrence prediction from satellite remote sensing data. It has been shown that deep-SDMs outperform conventional SDMs for occurrence prediction and their architecture is very promising for solving abundance prediction challenges. However, deep-SDMs require millions of observations to be trained and cannot therefore be trained to predict abundance. Indeed, due to acquisition difficulties, abundance datasets are considerably smaller than presence-only datasets.

Here we overcome this limitation by using the transfer learning from occurrence deep-SDM to abundance deep-SDM with the underlying hypothesis that the neural network layers of a previously trained model with presence-only data can capture general information and patterns that can be reused for abundance predictions. As an example, we used coastal fish in the Mediterranean Sea. We assessed the extent to which deep-SDM trained on only 406 fish counts can predict the abundance of fish species by taking advantage of transfer learning from a deep-SDM trained on 62,000 presence-only.

We show that this approach significantly improves the abundance prediction performance of deep-SDM, with average gains of 35% (based on the D2 Absolute Log Error score). This allows deep-SDMs to be more efficient than conventional SDMs, with an average gain of 20%. These gains are mainly linked to a better prediction of the abundance of rare species. This ability of deep-SDM to extract relevant information predicting species abundance from presence-only data is a new and unexpected result. This finding paves the way towards a more general use of deep learning to predict species abundance and biodiversity patterns, especially for rare species.

While policymakers are committed to a 30% global protection target by 2030, including the ocean, our knowledge of marine species distributions remains limited compared to terrestrial species. This gap is a major barrier to science-based decision-making in the field of marine conservation.

The high cost of data acquisition is partly responsible for the situation. But a significant lever would be to use the sparse data we have more efficiently. Indeed, current species distribution models (SDMs), though robust, are simplistic in terms of environment-species interactions: they often rely only on long-term climate averages. This seriously hinders the discovery of dynamic processes, such as seasonal migrations or adaptation to environmental change, and therefore limits our knowledge of how marine species use space.

Recently, computational ecologists have successfully designed and tested new types of SDMs based on deep learning to model plant species distributions in the terrestrial realm. They treat the environmental landscape as a multi-layered image and use convolutional neural networks to extract relevant geographical features and predict associated species, with promising results on data-poor species using knowledge transfer.

Our work adapts this approach to the open ocean, in particular by taking into account the high variability of environmental conditions. In an initial trial, we predicted relative presence probabilities for 38 marine taxa at a global scale using 18 satellite-derived environmental variables, achieving 89% Top-3 accuracy. This work provided valuable insights on data curation, variable importance and hyperparameter fine-tuning.

This approach provides ways to enhance our knowledge of data-poor marine regions or species, and to deepen our understanding of the dynamic impact of environmental conditions. It paves the way for a diversity of use cases ranging from other marine environments to predictions of future species distributions. This allows more comprehensive biodiversity mapping, which is a significant step towards well-designed ecosystem protection measures.

Species distributions are one of the fundamental units in biogeography and conservation and a key Essential Biodiversity Variable. Species distribution models (SDMs) are popular tools that characterize a species distribution by statistically relating occurrence data with environmental variables, remote sensing products, and other habitat variables. SDMs have become increasingly sophisticated, but they are unsuitable for data-deficient species; 30% of known species have insufficient data to characterize geographic distributions appropriately. Since many analyses rely on robust SDM outputs, data-deficient species are often left out, biasing scientific and conservation efforts.

Recently, ecology has seen unprecedented growth in the amount and variety of data collected, including occurrence data, phylogenetic data, and fine-resolution remote sensing products. We integrate these data sources and present a novel modeling framework that extends SDMs to allow data-deficient species to borrow strength from data-rich species. Specifically, we demonstrate how evolutionary history, WorldClim, EarthENV, and MODIS products can inform the distributions of data-deficient species. We apply our modeling framework to the tropical clade of South American hummingbirds, where SDMs are often hampered by a need for more data, even in well-studied taxons such as birds.

The results of our analysis include up to 40% improvement in the Area Under the Curve for over 75% of the species. Species that showed little to no improvement lacked a recently diverged sister species, indicating that this method works best when species’ pairs are recently diverged. We quantify the improvement of our model and produce novel richness maps. We suggest these maps are our best current understanding of South American species’ distributions. This work represents a concrete way forward for SDMs to integrate phylogenetic information and remote sensing products. By improving data-deficient distribution estimates, we will develop more robust species distribution-related EBVs and better understand how our biodiversity is distributed across geographic space.

Preserving species' genetic diversity is crucial for maintaining ecosystem functions and services in the face of global change. However, to preserve it effectively, we first need to monitor genetic diversity efficiently. While DNA sequencing remains the gold standard for assessing genetic diversity, it is often too expensive and time-consuming for routine and large-scale monitoring.

The Genes from Space project offers a complementary approach: using Earth Observations (EO) for large-scale monitoring of species' genetic diversity. Changes in species' genetic diversity often reflect local population declines driven by habitat changes that can be readily detected via EO—such as deforestation, ecosystems response to shifting climate and other biotic and abiotic conditions. Therefore, tracking habitat changes with EO allows for large-scale and routine monitoring of genetic diversity using EO-based indicators.

In collaboration with the BON in a Box platform, we have developed a tool that enables monitoring of genetic diversity across a range of ecosystems using EO data. The tool is versatile, allowing users to integrate their own species distribution or habitat change data, or to retrieve such data from publicly available databases (e.g., GBIF, Global Forest Watch, or global land cover maps). Based on these inputs, the tool produces genetic diversity indicators consistent with the global biodiversity conventions.

This free and open-source tool is designed to accommodate a range of users. For example, biodiversity conservation practitioners can access the tool via a user-friendly website interface to generate indices for specific ecosystems. For advanced users with programming expertise, the tool can be run locally, and they are encouraged to contribute to its development by adding new workflows.

The Genes from Space monitoring tool provides a scalable, accessible solution for monitoring genetic diversity across large spatial scales, serving as an early-warning system to direct and optimize in situ and DNA-based assessments where needed.

We present a framework to monitor biodiversity by calculating ecological niche models over time with a temporal series of remote sensing products. We have implemented this methodology in the Natural Park of Montesinho (Northeast Portugal) through the MontObEO research project. The framework estimates species vulnerability by analysing trends (Mann-Kendall test) over time (2001-2023) of the habitat suitability index from a set of ecological niche models (Maxent) calculated with a time series of remote sensing variables (MODIS sensor). Positive trends are associated with increases in habitat suitability; negative trends with decreases in habitat suitability. All procedures (e.g. gathering the satellite data, calculating the MaxEnt models, and analysing the habitat suitability trends) are performed in Google Earth Engine (GEE). We considered five taxonomic groups: vascular flora, amphibians, reptiles, birds, and mammals. We analysed habitat suitability trends for each species, taxonomic group, functional group, and potential species richness over time. We created an R package and a GEE App, where users can use our framework easily and efficiently. We built a spectral library for some vascular flora key species in Montesinho. Our framework is an effective monitoring methodology as it can be adapted to any study area at different spatial and temporal resolutions. This work is funded by Centro de Investigação em Ciências Geo-Espaciais, reference UIDB/00190/2020, funded by COMPETE 2020 and FCT, Portugal.

Regular counts of walruses (Odobenus rosmarus) across their pan-Arctic range are necessary to determine accurate population trends, and in turn understand how current rapid changes in their habitat, such as sea ice loss, are impacting them. However, surveying a region as vast and remote as the Arctic with vessels or aircraft is a formidable logistical challenge, limiting the frequency and spatial coverage of field surveys. An alternative methodology involving very high-resolution (VHR) satellite imagery has proven to be a useful tool to detect walruses, but the feasibility of accurately counting individuals has not been addressed. Here, we compare walrus counts obtained from a VHR WorldView-3 satellite image, with a simultaneous ground count obtained using a remotely piloted aircraft system (RPAS). We estimated the accuracy of the walrus counts depending on 1) the spatial resolution of the VHR satellite imagery, providing the same WorldView-3 image to assessors at three different spatial resolutions (i.e., 50, 30, and 15 cm per pixel) and 2) the level of expertise of the assessors (experts vs a mixed level of experience – representative of citizen scientists). This latter aspect of the study is important to the efficiency and outcomes of the global assessment programme because there are citizen science campaigns inviting the public to count walruses in VHR satellite imagery. There were 73 walruses in our RPAS “control” image. Our results show that walruses were under-counted in VHR satellite imagery at all spatial resolutions, and across all levels of assessor expertise. Counts from the VHR satellite imagery with 30 cm spatial resolution were the most accurate, and least variable across levels of expertise. This was a successful first attempt at validating VHR counts with near-simultaneous, in situ, data. But further assessments are required for walrus aggregations with different densities and configurations, on different substrates.

Monitoring vulnerable wandering albatross (Diomedea exulans) populations presents significant challenges due to their remote nesting locations, making traditional ground or aerial surveys costly, infrequent, and often incomplete. However, with advancements in geospatial remote-sensing technologies, citizen science is emerging as a valuable tool for generating accurate, georeferenced wildlife data. In this study, we conducted the first citizen science campaign aimed at counting wandering albatrosses in South Georgia, utilising 31 cm resolution satellite imagery. The campaign spanned 24 breeding areas with imagery captured between 2015 and 2022. Volunteers were tasked with identifying presumed albatrosses in 150 m x 150 m image chips (with 5 m overlap), each reviewed by a minimum of seven unique users. Over the course of the campaign, 639 citizen scientists classified a total of 11,839 image chips, covering 154 km². Our results show a strong positive correlation (r = 0.98, df = 16, P < 0.001) between adjusted ground counts and satellite-based estimates, with deviations ranging from 4.5% to 30.9% for colonies containing more than 100 breeding pairs. This study demonstrates the accuracy and effectiveness of crowdsourcing as a reliable method for long-term monitoring of wandering albatross populations and highlights the potential to expand this approach to other seabird species and breeding sites, offering a scalable solution for wildlife monitoring in remote regions.

Geomagnetic navigation as an animal migratory strategy has been studied across several taxa, but how animals, mainly long-distance migrants (i.e. birds), use the geomagnetic field on their journeys is still relatively unknown. The Earth’s magnetic field varies across both space and time. The variability across temporal scales, which are relevant for animal navigation (at seconds to days), mostly comes from solar activity and may affect the potential animal choice of direction during navigation. To date, ecologists have not been able to study how these short-term geomagnetic field variations affect navigation because of the lack of reliable geomagnetic data (Deutschlander 2014). This, however, is important since it has been demonstrated that animals can sense minor geomagnetic field differences (Beason 1987; Semm and Beason 1990). This talk will introduce a tool called MagGeo that will help ecologists obtain detailed geomagnetic data at the location and time of the passing animal. I will describe the technical process of implementing a spatio-temporal data fusion method (Benitez-Paez 2021) to link wildlife tracking data with geomagnetic data provided by Swarm Constellation (from the European Space Agency - ESA). Our tool, MagGeo, is available as free and open-source software (FOSS) and uses a set of Jupyter notebooks to let users interact with the process.