In this session, we deep dive into the state-of-the-art of spatial biodiversity modeling on local to global scales, ranging from populations through communities to ecosystems. This includes a multitude of models that integrate in-situ biodiversity data with remote sensing, such as species distribution models, macroecological models, natural value segmentation, causal inference methods, and beyond. It covers a broad spectrum of data-driven modeling techniques, from time-tested statistical models to modern deep learning frameworks that can facilitate learning across species and environments.
We will examine how such models can fuse multi-source inputs into ready-to-use metrics, such as species richness, community turnover, and functional diversity. Discussions will cover best practices for evaluation and uncertainty quantification, strategies to address gaps in biodiversity data, and the roles of data management, benchmarking and explainable AI in building transparent, trustworthy models.
Combining scientific talks, panel discussions and audience engagement, the session aims to identify current limitations of and outline key priorities for improving the state-of-the-art in this field.
Co-Convener: Tobias Andermann, Jan Borgelt, C. Vanalli, Sara Si-Moussi, Pierre Bonnet, Florian Hartwig
An important part of ecology is understanding where species occur in relation to their environment, a question addressed through Species Distribution Modelling (SDM). These models provide spatially explicit estimates of species distributions, which are essential for understanding biodiversity at large scales. SDMs extend far beyond what direct observation can capture and help identify which sites should be prioritised for monitoring or conservation.
SDMs are typically supported by two types of data: Presence–Absence (PA) and Presence-Only (PO). PA data, collected through structured surveys, provide high-quality information but are costly and therefore sparse. PO data, by contrast, come from opportunistic observations and contain only positive detections, making them abundant but biased. These datasets present a clear trade-off between quality and quantity, and data-integration approaches seek to combine PA and PO information so that models can exploit the complementary strengths of both.
It is not yet clear how much integration can improve our models, particularly when the available PA training data are geographically or environmentally distant from the prediction region. This issue is especially relevant for under-sampled areas, where PO data may be more readily available. Previous studies have explored this problem using simulations, but these lack the complexity and noise of real ecosystems. Using real data is therefore essential for understanding how these methods behave in practice and for assessing their value in biodiversity modelling.
Our work studies this question using real-world datasets. We propose a partitioning framework that creates controlled spatial separation between PA training and prediction regions while respecting the environmental constraints of each dataset. Using this framework, we evaluate several state-of-the-art SDM approaches, including deep neural network models adapted to support integrated PA–PO training through specialised loss functions and data-fusion strategies.
Our preliminary findings suggest that integrating PA and PO data consistently improves model performance across a range of spatial separation scenarios. This indicates that both PA-rich and PO-rich contexts can benefit from incorporating the complementary data source, highlighting data integration as a robust and broadly effective strategy for enhancing SDM generalisation and improving biodiversity assessment at scale.
How to cite: Ubilla Pavez, P., Marcos, D., Botella, C., Joly, A., Benerradi, R., and Marti, R.: Data Integration for Species Distribution Models Under Spatial and Environmental Separation, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-666, https://doi.org/10.5194/wbf2026-666, 2026.
Biodiversity is declining worldwide, driven by anthropogenic pressures such as land sealing, air pollution, agricultural practices, forest clearing, and climate-change impacts. These stressors interact in complex ways, altering habitats, disrupting ecosystem functions, and challenging the long-term survival of many species. In response, policy frameworks such as the European Union’s Biodiversity Strategy for 2030 aim to protect 30% of terrestrial and aquatic ecosystems by the end of the decade. Meeting these goals requires transparent and scalable approaches for identifying ecologically valuable areas. However, protected-area designation across Europe remains inconsistent, with national frameworks differing and often lacking transferability. At the same time, the growth of geospatial datasets, citizen science observations, and environmental time-series data offers new opportunities for data-driven conservation planning. Advances in machine learning lead to new tools to analyse heterogeneous datasets and support decision-making in complex ecological contexts.
In this study, we use open-source environmental data together with citizen-science species observations collected between 2016 and 2024. These time-series data allow us to assess how environmental conditions, land use, and climate variability influence species occurrences over time. Using these inputs, we apply species distribution modelling to estimate habitat suitability and to map how suitable environments shift in response to external pressures. To support conservation planning, we apply the CAPTAIN reinforcement learning framework. Developed by Silvestro et al., this tool (2022) enables the optimisation of different conservation targets under ecological, spatial, and socio-economic constraints. It is based on the interaction between an agent and its environment, where the agent learns to take decisions that maximise a defined reward. The approach evaluates trade-offs, such as balancing biodiversity outcomes with economic considerations, and produces spatially explicit prioritisation maps that identify forest areas of consistently high ecological value, even under changing environmental conditions. The resulting framework is designed to be transferable and scalable. It supports nature-conservation planning in Germany and can be adapted to other regions, ecosystems, or policy objectives. By combining open data, citizen-science observations, and a machine learning-based optimisation, the approach contributes to national biodiversity strategies and aligns with broader international efforts such as the EU Biodiversity Strategy for 2030.
How to cite: Horn, K., Silvestro, D., Wallis, C., and Rudolph, A.: Integrating Open-Source Data and Advanced Machine Learning for Forest Conservation Prioritisation in Germany, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-712, https://doi.org/10.5194/wbf2026-712, 2026.
Recently, deep learning approaches to species distribution models (SDMs) have increasingly focused on integrating information-rich modalities such as natural language and remote sensing, motivated by the hypothesis that capturing the non-linear relationships between these inputs and species occurrences should help compensate for limited biodiversity data in poorly monitored regions. However, while leveraging additional modalities has been shown to improve predictions in certain settings, we argue that these improvements remain highly dependent on the task formulation and dataset. We consider the SatBird dataset (Teng et al., 2023) as an illustrative example, showing how leveraging representations derived from satellite imagery does not consistently translate into performance improvements, especially in low-data regimes. We argue that multimodality shouldn't be treated as a generic stepping stone towards improving deep learning-based SDMs, as it can often boil down to the naive assumption that any additional information will be beneficial regardless of their ecological relevance. We also highlight that multimodal approaches in deep learning-based SDMs are predominantly reducible to the inclusion of more and more abiotic covariates, and discuss how such a strategy can amplify the risk of overfitting to sampling biases and amplifying spurious correlations. Finally, we show that leveraging relevant, context-dependent biotic information offers a particularly promising alternative research direction, considering as case studies our work with 1) BATIS (Villeneuve et al., 2026), a novel Bayesian framework that iteratively refines prior predictions from an uncertainty-aware SDM using limited local observations in data-scarce regions, and 2) CISO (Abdelwahed et al., 2025), a novel transformer-based approach that leverages well-documented species groups to improve predictions for data-limited taxa. Results with both BATIS and CISO suggest that universal solutions are unlikely to be sufficient to address current limitations in deep learning-based SDMs, and that further improvements in predictive performance are more likely to come from targeted approaches dedicated to specific data gaps and ecological contexts.
How to cite: Villeneuve, C., Teng, M., Akera, B., Radi Abdelwahed, H., Zbinden, R., Pollock, L., Larochelle, H., Tuia, D., and Rolnick, D.: Reevaluating Multimodal Approaches To Deep Species Distribution Models, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-813, https://doi.org/10.5194/wbf2026-813, 2026.
Monitoring species and habitat distributions across space and time is critical for biodiversity conservation, as it allows ecologists and decision-makers to assess ecosystem dynamics, detect emerging threats, and prioritize interventions to mitigate biodiversity loss.
Species and Habitat Distribution Modelling (SDM/HDM) enable this by identifying correlations between environmental variables and species or habitat occurrences. In particular, Deep Neural Networks for Habitat Distribution Modelling (Deep-HDM) has demonstrated strong scalability and cross-modal learning capabilities across images, tabular data, and text.
However, most Deep-HDM approaches rely on mono-scale only data and overlook the potential of extracting complementary information from ground-level images, which encode medium-grained ecological and structural landscape cues absent from remote sensing or tabular data alone. To address this gap, we propose a multi-scale, image-focused habitat classification pipeline that jointly leverages satellite/remote sensing observations and landscape photographs.
Our method uses pre-trained modality-specific visual encoders (e.g., GeoCLIP, SwinV2, ResNet) to generate initial representations, which are then refined using contrastive learning to spatially align features from geographically close samples. A downstream habitat classifier is then finally trained on this shared representation space, allowing to infer habitats from multiple possible input data types.
To carry out our experiments, we rely on three European vegetation and land-cover datasets: (i) LUCAS (EUROSTAT) landscape images of which a minority has level-2 EUNIS habitat labels (~70k samples, ~15k survey sites); (ii) EVA (European Vegetation Survey) containing presence-absence plant observations with level-1-to-3 EUNIS habitat labels (~500k survey sites); and (iii) EMBAL (European Commission's DG ENV) which includes images of transects, where a minority has level-2 EUNIS habitat labels (~75k samples, ~4k survey sites). The code will be open-sourced in the future, and details about accessing the datasets will be provided.
Results show that contrastive spatial pre-training improves Deep-HDM performance, particularly for fine-grained habitat identification. This demonstrates that learning shared representations over multi-scale input data strengthens habitat prediction compared to mono-scale baselines. Better habitat classification from multi-modal data can improve habitat monitoring, but also the spatial delineation of habitats across Europe, and thus help regional targeting of more pertinent measures under the Nature Restoration Regulation.
How to cite: Larcher, T., Joly, A., Salmon, J., Bonnet, P., and Van Der Velde, M.: Bridging Ground and Satellite Views with Contrastive Learning for Scalable Habitat Monitoring, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-760, https://doi.org/10.5194/wbf2026-760, 2026.
Projecting species abundance under future environmental conditions remains a critical challenge in biodiversity modelling. Exceedingly, abundance frameworks are retrospective or taxonomically limited, lacking integration with high-resolution environmental data and their corresponding future projections. Recent research highlights that forward-looking predictions linking abundance to environmental drivers are crucial for effective conservation planning under global change.
Prescribed by the Global Biodiversity Framework, the United Kingdom’s legally binding targets of halting and reversing species abundance decline present unique modelling opportunities and challenges. The UK benefits from exceptionally rich, long-term citizen science and statutory monitoring schemes across multiple taxonomic groups yet integrating these into unified abundance projections remains technically challenging. Combining these datasets with comprehensive environmental covariates enables the development of national-scale abundance projections. Here, we present a spatial-temporal framework that harmonises data on hundreds of species across taxonomic groups (e.g. birds, plants, butterflies, moths) with high-resolution environmental covariates covering climate, soils and land cover, and their scenario-based projections (RCPs, SSPs).
Using inlabru for Bayesian approximation, we develop spatially explicit models at 1 km resolution across the UK. Carefully formulated spatial-temporal random effects capture residual dependencies whilst remaining computationally efficient at national scale. This framework enables abundance predictions at a notable combination of spatial extent (national), resolution (1 km) and taxonomic breadth.
This framework explicitly links abundance to environmental covariates and their future projections, enabling predictions under novel climate and land use conditions rather than simple temporal extrapolation. We estimate a baseline abundance and project changes under alternative scenarios, quantifying uncertainty throughout. This hierarchical approach provides policy-ready outputs at multiple biological scales; from individual species through taxonomic groups to community-level measures.
Model outputs provide 1 km resolution gridded abundance estimates with associated uncertainties. The integration of comprehensive environmental datasets with robust spatial-temporal modelling at national scale represents significant advances in abundance projections. This scalable framework can be adapted by other nations working towards Global Biodiversity Framework targets; with such forward-looking modelling approaches essential for tracking progress and identifying interventions to halt and reverse biodiversity decline.
How to cite: Rush, C. and Wilkes, M.: A spatial-temporal framework for species abundance modelling and scenario assessment in the UK, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-899, https://doi.org/10.5194/wbf2026-899, 2026.
One of the most important knowledge gaps for nature management is the species diversity in an area. In order to know the spatial distribution of richness of different species groups, observations of species are needed. This is challenging, because for many species groups it is resource-intensive to systematically map the occurrence of species over large areas. Instead, citizen science data has become more important, but as it is mostly collected opportunistically , there is a high geographical bias in the collection. If this is not taken into account, one gets the impression that the most species-rich areas are where there are the most people. This bias can be estimated and corrected if citizen science data is correctly combined with survey data, e.g. from national monitoring programs.
Integrated species distribution models (iSDM) are relatively new statistical tools that combine opportunistically collected data with systematically collected data with information about collection intensity. These models have great potential to benefit from the strengths of different types of data, while at the same time being able to take into account some of the challenges, but the models have not been used to a large extent in large-scale modeling of species diversity. Large-scale data presents both opportunities and problems. It allows us to leverage information across species, but the size of the data makes it infeasible to model every species together.
Here we will present our approach to this large-scale modelling, using iSDMs across species to produce fine-scale maps of communities over Norway, along with their associated uncertainties and biases. The model is based on a state space, using ideas from point processes, which has proved flexible for different data types and for which there are efficient and flexible fitting methods. We will outline the model and the methods we have used to overcome the problems of large data and the requirement for fine spatial scales. We will also outline how this is being extended to include time and topologies such as river networks.
How to cite: O'Hara, R., Mostert, P., Perrin, S., Adjei, K., Togunov, R., and Finstad, A.: Large-scale integration of data for distribution models (and beyond), World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-779, https://doi.org/10.5194/wbf2026-779, 2026.
Reliable biodiversity forecasts depend on knowing where species occur and how they interact with their environment. Yet most of the world’s species—particularly those that are rare, range-restricted, or threatened—remain severely under-sampled. These data gaps cascade through global biodiversity assessments, systematically excluding data-deficient species from conservation prioritization, climate vulnerability analyses, and scenario planning. As countries work toward the goals of the Kunming–Montreal Global Biodiversity Framework, the persistent invisibility of data-deficient species has become a major, but often unrecognized, barrier to evidence-based action.
We present a new predictive modeling framework that directly addresses this challenge by allowing species with few records to “borrow strength” from their closest relatives. Because related species tend to share aspects of their ecological niches, evolutionary history offers a powerful source of information where empirical data are lacking. By embedding phylogenetic relationships directly into a multi-species distribution model, we are able to generate robust environmental niche estimates and spatial predictions even for species with only 1–10 observations.
Using a continental-scale analysis of South American vertebrates, I show that phylogenetically informed models dramatically outperform traditional species distribution models under extreme data scarcity. The greatest gains occur for data-deficient species, whose predicted distributions and accuracy metrics improve substantially compared to standard state-of-the-art SDMs. As species become data-sufficient, model performance between approaches converges, highlighting that the primary value of phylogenetic information lies in rescuing the species we understand the least.
Crucially, this work lays the foundation for better-informed conservation decisions by quantifying how the exclusion of data-deficient species biases conservation analyses and alters our understanding of global change impacts, particularly under climate change scenarios and climate-vulnerability assessments. By making it possible to include under-sampled, understudied, and often highly threatened species in scenario planning and priority-setting exercises, this framework enables more inclusive and equitable conservation outcomes. Overall, this framework extends species distribution modeling to under-sampled taxa, reducing a major source of bias in ecological and conservation analyses.
How to cite: Sharma, S., Cohen, J., and Jetz, W.: Mapping All Species: Closing Biodiversity Data Gaps with Phylogenetically Informed Predictive Models , World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-903, https://doi.org/10.5194/wbf2026-903, 2026.
Achieving the Kunming-Montreal Global Biodiversity Framework’s 30x30 target requires accurate, high-resolution biodiversity maps. However, for plants, distribution data remains spatially sparse and taxonomically biased, limiting our ability to create representative maps of global plant diversity for conservation prioritisation. To address this, we developed a scalable workflow that uses optimised representative sampling of species to generate robust global plant diversity maps.
Using global plant taxonomic checklists as a baseline, our approach employs a genetic algorithm to maximise geographic and compositional representativeness of the selected species for modelling relative to established coarse-scale global diversity patterns. This ensures that a comparatively small number of species are able to capture broader biodiversity patterns. We then apply ensemble species distribution modelling techniques, such as MaxEnt-based binary predictor ensembles and multi-algorithm ensemble models, to model species distributions at fine scales. By iteratively modelling multiple optimised samples across a range of sample sizes, we are able to determine sample size thresholds beyond which additional species no longer affect predicted diversity hotspots, allowing us to define a minimum number of species needed for stable and reliable diversity mapping at different scales. We use the results to create a high-resolution (10km2) global plant diversity map that explicitly incorporates uncertainty by deriving predicted diversity intervals for each cell from the repeated samples.
The resulting high-resolution global plant diversity map explicitly accounts for sampling biases, geographic gaps, and predictive uncertainty. Importantly, we demonstrate that robust biodiversity prediction is achievable using only a fraction of all species when using a representative sample, offering a pragmatic solution where comprehensive modelling remains unfeasible. This methodology provides a data-efficient foundation for spatial conservation planning and priority-setting aligned with global targets such as 30x30.
Our framework is transferable across taxa and scales, offering a generalisable path forward for biodiversity assessment under data limitations. By strategically leveraging representative samples, this approach supports more inclusive and globally consistent conservation planning.
How to cite: Baldaszti, L., Brummitt, N., Moonlight, P. W., Pironon, S., and Särkinen, T.: The power of few: Using representative sampling to predict global plant biodiversity patterns, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-9, https://doi.org/10.5194/wbf2026-9, 2026.
Monitoring species distributions at large spatial and temporal scales is critical for understanding ecological dynamics and informing conservation strategies. Standardized survey protocols can produce presence-absence data that are particularly valuable, but they remain costly, time-consuming, and require botanical expertise, resulting in limited and sparse data. In contrast, opportunistic presence-only data from citizen science programs such as Pl@ntNet offer unprecedented spatial, temporal, and taxonomic coverage. However, they are affected by strong sampling, detection, and reporting biases that can obscure true species distributions and trends. To address these challenges, we propose a framework that integrates deep learning with site-occupancy models to estimate species distributions from presence-only data while explicitly accounting for these biases.
Site-occupancy models allow disentangling species presence from observation processes, yielding more reliable estimates of species presence probabilities. Incorporating deep learning enables these models to be fitted efficiently and flexibly through stochastic gradient-based optimization, making it possible to analyze massive opportunistic datasets at scale.
We first validate the approach using realistic simulated datasets, comparing deep-learning-based inference with classical methods, including gradient-based maximum likelihood and Bayesian approaches, to demonstrate both computational scalability and reliability of the resulting predictions.
We then evaluate model performance on benchmark spatial species distribution modeling datasets using presence-only data. Results highlight the ability of deep-learning site-occupancy models to capture spatial variation in occurrence probabilities while mitigating reporting biases.
Finally, we explore the potential of our framework to leverage large citizen science datasets for assessing spatio-temporal changes in species distributions. In particular, we compare trends inferred from opportunistic observations – such as those from Pl@ntNet – using our approach, with trends derived from structured survey protocols to assess similarities, differences, and complementary information provided by these different types of data.
By combining site-occupancy modeling with deep learning on massive opportunistic datasets, our approach would bring new insights into large-scale species distributions and monitor changes over time. In addition, the flexibility of deep learning could allow for refined modeling of observer behaviors, detection patterns, enabling more accurate assessments of species distribution and trends from heterogeneous data sources.
How to cite: Benerradi, R., Botella, C., Servajean, M., and Joly, A.: Deep-Learning Site-Occupancy Models for Disentangling Biases in Species Distribution and Trend Assessment from Citizen Science Data, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-738, https://doi.org/10.5194/wbf2026-738, 2026.
Understanding how biodiversity patterns are shaped and accurately modeling them is critical for establishing effective conservation strategies. Species distribution models (SDMs) are a fundamental tool in this effort, linking species occurrences with environmental drivers to estimate their spatial distribution. While SDMs have traditionally been fitted on one or a few species at a time, recent deep learning-based approaches jointly model thousands of species, with the potential of leveraging shared environmental structure and co-occurrence patterns in the data. However, evaluating multi-species models is inherently non-trivial, and we show that existing metrics do not adequately capture model performance.
Most studies dealing with large sets of species summarize performance using a few averaged metrics. However, species distribution modeling is strongly affected by geographic and taxonomic biases in the occurrence data. A single metric blurs the extent to which these biases shape model performance. Moreover, widely used metrics, such as the area under the receiver operating characteristic curve (AUROC), inadvertently reflect these biases making them more difficult to interpret, especially when aggregated between species. These issues exist in any multi-species setting, but are amplified when scaling to thousands of species across broader geographic ranges and more heterogeneous biases.
To address these gaps, we propose a bias-aware evaluation framework for multi-species SDMs. We define several proxy scores for characterizing various per-species biases in the training data - including sampling imbalance, occurrence sparsity, and taxonomic neglect. These scores allow us to evaluate model performance across species groups with differing bias levels, revealing, for example, whether a model is robust to geographic bias, struggles with sparsely sampled taxa, or performs well only for well-sampled species. Along with these bias-informed metrics, we introduce a curated, fully non-anonymized global plant dataset combining GBIF citizen-science records with SPlotOpen vegetation plots. This dataset is explicitly designed to enable transparent, species-resolved performance evaluation.
Together, the dataset and bias evaluation scheme provide the framework needed to test the currently most pressing yet unsolved SDM challenges, including long-tailed distributions and spatiotemporal observation biases, at scales and complexities that are appropriate for modern-day, deep learning-based multi-species distribution models.
How to cite: Arens, E., van Tiel, N., Zbinden, R., Vanalli, C., Robert, D., Drees, L., Kellenberger, B., Pellissier, L., Wegner, J. D., and Tuia, D.: Bias Aware Benchmark for Species Distribution Modeling, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-772, https://doi.org/10.5194/wbf2026-772, 2026.
Many fundamental processes in ecology are nowadays modelled with ever-increasing complexity, also thanks to advancements in data science. Species distribution modelling is no exception to this trend, and in particular deep learning-based models have steadily been maturing in recent years, promising high prediction performance for many species at large scales. Yet, a fundamental desire in ecology is not just to predict, but understand, observed processes, both environmental and model-intrinsic. Deep learning models are often described as black boxes due to their complexity, and hence have a notorious reputation of being unsuitable for either. However, recent years have seen great advancements in both unravelling and more explicitly quantifying the decision process of deep neural networks.
In this work, we explore the potential of a deep learning-based species distribution model (SDM) that explains itself by design. The model achieves this via a learned top-K sampling scheme with attention mechanisms on the environmental covariates it receives. In detail, the model is forced to select a subset of user-definable size (K) of covariates that is as useful as possible for the prediction of species encounter likelihoods at each data point. Within this sampling scheme, covariates are either available or not (and not modulated as in regular attention mechanisms), and unlike in post-hoc explainability methods, no auxiliary model is required to explain the SDM's decision process. The result are per-covariate importance scores that are as trustworthy as possible.
We evaluate our model on a set of around 830,000 observations for 356 mammal species, sampled over North America, comparing prediction performances and investigating obtained covariate importances. We find that our sampling scheme does result in highly consistent covariate combinations across runs, and further see plausible correlations with the environmental configuration across the continent. We further investigate correspondence with post-hoc explainability methods and find improvable agreement, highlighting the challenges in explainability for machine learning models in general, and deep learning SDMs in particular.
How to cite: Kellenberger, B. and Jetz, W.: Self-explaining Deep Learning-based Species Distribution Models, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-921, https://doi.org/10.5194/wbf2026-921, 2026.
Monitoring abundance trends for species is essential for conservation reporting under frameworks like the Global Biodiversity Framework, yet existing methods are reliant on ecological time series data which often contain gaps for populations across space and for long time periods. We addressed these data limitations by introducing a novel Bayesian multi-population autoregressive state-space (MARSS) modelling framework that estimates total abundance trends of species, along with uncertainty due to population deviations and sampling methods. We incorporated spatial synchrony in the modeling framework to estimate trends during unsampled time periods for populations using information from nearby sampled populations, informed by migration networks. Simulation experiments revealed that spatial synchrony improved population trend estimates, but only marginally, and could only be reliably estimated for species with high numbers of monitored populations. We applied this modeling framework to two broad case studies on pinniped species and sea turtles because of their high conservation concern and diverse trajectories after widespread industrial exploitation. We created a database of 555 time series of population abundances for 29 of 34 extant pinniped species and 304 time series for 7 of 7 extant sea turtle species. On average, pinnipeds and sea turtles had strong increases in abundance between 1980 and 2010, followed by a slowing of population growth rate to near zero from 2010 to 2020, though there is considerable variability across species and conspecific populations. We identified prominent spatial synchrony in species with comprehensive data coverage (e.g., California sea lion, green sea turtle), emphasizing the benefits of strengthening monitoring efforts for species with declining or uncertain trends. Observation errors of sampling methods (e.g., aerial, beach, boat surveys) varied widely, with implications for optimizing monitoring efforts given precision-cost tradeoffs. Our spatial Bayesian MARSS framework is broadly applicable to species with spatially structured populations and improves species assessments by leveraging spatial dynamics, identifying indicator populations, and better quantifying process and observation uncertainty for conservation status determination.
How to cite: Wang, T., Hilborn, R., and Ward, E.: A spatial Bayesian state-space framework for estimating species abundance trends: global case studies in marine megafauna, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-417, https://doi.org/10.5194/wbf2026-417, 2026.
The Hutchinsonian shortfall, i.e. lack of knowledge about the tolerance of species to abiotic conditions, represents one major drawback besetting our understanding of biodiversity and its response to a changing environment. The environmental responses and tolerances of species are commonly inferred with two distinct approaches: i) Trait performance studies, which use lab experiments to model species’ functional responses, often overlooking real-world complexity, and ii) Species Distribution Models (SDM), which infer environmental responses from occurrence data, yet offering limited causal insight. Here, we argue that the described strategies bring complementary information that can be integrated to better estimate species responses and more accurately map species distributions.
We select insect species whose ecology is strongly impacted by climate and are either important for ecosystem health (pollinators and biocontrol species) or disease vectors that threaten public health (Aedes mosquitoes). For each species, we retrieve from published studies the ecological model that estimates the trait-derived probability of occurrence as a function of temperature. Concurrently, we use occurrence records with average temperature data during the species activity season to train a neural network architecture (multi-layer perceptron) and estimate the SDM-based thermal response curves. The performance of the two approaches are compared on an independent test set, together with their respective thermal responses and identified thermal optimal and extremes. Among the two approaches, trait-based ecological models underperform deep learning SDMs in mapping distributions, possibly because the latter have been trained on the similar task of predicting species distribution. Interestingly, the thermal response curve of species occurrence seems consistently overestimated by the ecological model, suggesting that, when thermal functions generated in the lab are applied under real-world conditions, they might need correction with a shift towards lower temperatures. Last, we explore possible hybrid approaches, such as trait-guided deep learning or model ensembling, that can leverage both the mechanistic understanding of species ecology and the power of deep learning SDMs trained on the vast amount of community-based observational data.
Ultimately, we show how integrating experimental ecology with observational biogeography can lead to more accurate and ecologically grounded predictions for species distributions and their environmental responses to climatic changes.
How to cite: Vanalli, C., van Tiel, N., Zbinden, R., and Tuia, D.: Mapping species thermal suitability by integrating mechanistic ecological theory and deep learning , World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-107, https://doi.org/10.5194/wbf2026-107, 2026.
Connected landscapes are fundamental for species persistence under global change, as recognised by the International Convention on Biological Diversity. Connectivity enables species range shifts in response to climate change and maintains gene flow across fragmented populations, enhancing adaptive potential and long-term persistence.
Network theory offers powerful approaches to analyse landscape connectivity and prioritise conservation interventions, but current approaches to inform landscape connectivity networks face important limitations. Methods such as least-cost path analysis and circuit theory often lack ecological realism, neglecting dispersal behaviour and population dynamics, thus failing to represent . Moreover, most frameworks assume static landscapes, overlooking natural and human-driven change. These limitations risk misrepresenting connectivity and under- or over-estimating habitat availability and isolation, which highlights the need for approaches that integrate both ecological realism and temporal dynamics.
Individual-based models (IBMs) offer a promising, yet unutilized, way to inform landscape networks by simulating dispersal and demography with biological realism. Their process-based nature generates temporal outputs of functional connectivity, which can be used to inform multi-layer networks. The network representation offers the generality, scalability and comparability of connectivity analysis, which IBMs lack. While multi-layer networks show promise for representing spatio-temporal connectivity, incorporating landscape heterogeneity (space) and dynamics (time) into the network representation of the connected landscapes, their application in landscape ecology is in its infancy.
We developed a workflow that integrates the individual-based modelling platform RangeShifter with a multi-layer network theory framework for spatio-temporal connectivity analysis. RangeShifter integrates complex population dynamics and dispersal behaviours, includes explicit genetics, and simulates scenarios on spatially landscapes. Using multi-layer networks, the framework captures functional connectivity of one or multiple species across dynamic landscapes and enables connectivity analyses with diverse network metrics. We demonstrated the potential of this framework by comparing the effectiveness of alternative conservation actions, including ones derived from our framework based on different multi-layer connectivity metrics, in facilitating range expansion and patch occupancy for virtual species.
By linking IBMs with spatio-temporal network analyses, this workflow provides a tool to advance connectivity research for conservation planning in an era of rapid environmental change.
How to cite: Bunnenberg, C., da Silva Baptista, M., Henry, R., Rubido, N., Zurell, D., and Bocedi, G.: Integrating individual-based modelling and multilayer networks to advance landscape connectivity analyses, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-280, https://doi.org/10.5194/wbf2026-280, 2026.
European grasslands are among the world’s most species-rich ecosystems, yet they are increasingly shaped by two contrasting land-use trajectories: agricultural intensification and abandonment. Using 161,070 vegetation plots from the European Vegetation Archive covering all EU Member States and the United Kingdom, we quantified how landscape, environmental conditions, and nitrogen inputs jointly determine grassland plant diversity at continental scale. Applying a spatially explicit Bayesian modelling framework, we show that surrounding land use exerts a strong and consistent influence on local diversity. Landscapes dominated by low-intensity grasslands and heterogeneous farmland support approximately 8% higher plant diversity, whereas arable, urban, or woody-encroached landscapes reduce it. Diversity peaks at mid-elevation, in mildly acidic soils, and under moderately wet conditions, reflecting long-recognised gradients and providing a consistent, continent-wide estimate of their magnitude. Nitrogen inputs have a pronounced negative effect, with enrichment favouring nitrophilous, fast-growing species that outcompete slower-growing taxa. Spatial projection further indicates that, undercurrent nitrogen loads, grassland diversity is markedly reduced relative to a zero-input baseline, with national-level losses exceeding 10% in all EU Member States and reaching 29% in the Netherlands. These results highlight the considerable potential for biodiversity recovery if nitrogen inputs were reduced, particularly in intensively farmed regions. To complement this field-based analysis, we examined contemporary land-cover transitions using remotely sensed data from the Google-Dynamic-World dataset. Time-series classification allowed us to quantify the dominant processes driving ongoing grassland change across Europe. We found that succession linked to land abandonment, compounded by improving climatic conditions at higher elevations, is the prevailing process in montane and subalpine regions, where grasslands are increasingly transitioning towards woody vegetation. By contrast, intensification, conversion to cropland, and soil sealing are the most common pathways in flat, accessible lowlands, reflecting ongoing pressures from agriculture, infrastructure expansion, and urban development. This dual-pattern underscores how socio-economic drivers, accessibility, and biophysical gradients jointly structure grassland dynamics. Together, these two complementary lines of evidence show that sustaining Europe’s grassland biodiversity requires coordinated action across scales. Effective conservation and restoration will depend on integrating field-level mitigation with landscape-scale policies that internalise agricultural externalities and balance productivity, abandonment, and recovery within multifunctional rural systems.
How to cite: Marcantonio, M., Hagyo, A., Koeble, R., Musavi, T., See, L., Skoien, J., Strona, G., Urraca-Valle, R., van der Velde, M., and contributors, E.: The grasslands they’re a-changin’: land-use trajectories and the value of low-intensity agriculture, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-750, https://doi.org/10.5194/wbf2026-750, 2026.
Understanding the dynamics of migratory systems under global change is essential for effective land use and conservation planning. MIGRAZE is an individual-based ecological model designed to simulate the interactions between ungulate migration and food availability and land use in semi-arid regions. Here, the focus is on wildebeest. The model reflects the Serengeti–Masai Mara ecosystem, where roughly 1.3 million wildebeest migrate annually in response to grass biomass and nitrogen gradients. During the wet season, they feed on the nutrient-rich grasses in the southern Serengeti, before migrating to the wetter Masai Mara as the dry season begins. MIGRAZE integrates this behaviour by simulating movement of super-individuals using the stochastic movement simulator to make the step selection. The direction of the movement is chosen based on land cover type, grass biomass and grass nutrient content in their perceptual range. Vegetation dynamics incorporate rainfall patterns, the accumulation of dry matter and consumption by animals to simulate green grass biomass during the rainy season. We modelled the wildebeest migration in the Serengeti-Masai Mara ecosystem for two decades from 1999 to 2019. The study was validated using tracking data from 67 individuals over the same time period. In order to better identify the mechanisms underlying the large-scale movement patterns we tested two behavioural components. First, we provided the super-individuals with knowledge of the full landscape, which resulted in more directed movements toward their seasonal ranges. Second, we incorporated past environmental conditions by adding the normalized difference vegetation index (NDVI) as memory. This generated stronger movement toward the Masai Mara in the dry season but hindered return migration to the southern Serengeti, as NDVI does not capture grass quality. Overall, our results suggest that broad migratory destinations are shaped by memory and knowledge, whereas immediate movement decisions and the timing and arrival depend on current environmental conditions. This knowledge is of crucial importance when modelling future scenarios, and will assist in anticipating how wildebeest movements might respond to increasing anthropogenic pressures, such as climate and land use change. This, in turn, can inform conservation planning.
How to cite: Back, H., Allgayer, R., Bocedi, G., Ferretto, A., and Arneth, A.: MIGRAZE – Modelling Wildebeest Migration Patterns in the Serengeti-Masai Mara Ecosystem, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-881, https://doi.org/10.5194/wbf2026-881, 2026.
Dynamic global vegetation models (DGVMs) are increasingly capable of representing key forest structural attributes, such as canopy layers, deadwood, and size distributions (Eckes‐Shephard et al., 2025). Because these structural characteristics are frequently used as proxies for biodiversity, this improved modelling capacity offers new opportunities to project how forest biodiversity may respond to climate change and management.
However, despite their widespread use, the empirical relationship between forest structure and biodiversity remains unclear. Structural features often correlate positively with some taxa but show weak or even negative associations with others. The last major synthesis (Gao et al. 2015) highlighted this inconsistency, demonstrating that structural metrics differ greatly in their predictive power across species groups, forest types, and regions. Since then, many new studies have emerged, offering an opportunity to reassess which structural indicators are most reliable and in which context. We will therefore update the meta-analysis of Gao et al. (2015) with one decade of new studies, to identify robust taxon-specific structural indicators of biodiversity for European forests. We will then use the Dynamic Global Vegetation Model LPJ-GUESS (Lindeskog et al., 2021) to simulate the trajectories of these indicators under alternative climate and forest management scenarios. Through these projections, we will be able to determine which management strategies are likely to enhance or diminish biodiversity across different taxa, supporting a more comprehensive assessment of future biodiversity risks and opportunities.
Eckes‐Shephard, A. H., Argles, A. P., Brzeziecki, B., Cox, P. M., De Kauwe, M. G., Esquivel‐Muelbert, A., ... & Pugh, T. A. (2025). Demography, dynamics and data: building confidence for simulating changes in the world's forests. New Phytologist.
Gao, T., Nielsen, A. B., & Hedblom, M. (2015). Reviewing the strength of evidence of biodiversity indicators for forest ecosystems in Europe. Ecological Indicators, 57, 420-434.
Lindeskog, M., Lagergren, F., Smith, B., & Rammig, A. (2021). Accounting for forest management in the estimation of forest carbon balance using the dynamic vegetation model LPJ-GUESS (v4. 0, r9333): Implementation and evaluation of simulations for Europe. Geoscientific Model Development Discussions, 2021, 1-42.
How to cite: Ferretto, A. and Arneth, A.: Forest structure and biodiversity: revisiting the evidence and assessing future trajectories with the process-based model LPJ-GUESS, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-892, https://doi.org/10.5194/wbf2026-892, 2026.
Predicting the effects of anthropogenic global change on biodiversity requires models that capture the mechanisms underpinning species’ responses, such as demography, dispersal, evoluiton and species’ interactions. RangeShifter is a spatially explicit individual-based modelling platform that integrates these processes, making it well-suited for investigating biodiversity responses to multiple drivers and management strategies under changing environmental conditions. First published in 2014, it has been continuously developed over the years, including an R package interface since 2021.
We illustrate three recent RangeShifter applications, highlighting some of its newly developed capability:
- Demography-environment relationships improve mechanistic understanding of range dynamics under climate change: Using a Bayesian framework, demography-environment relationships, dispersal, and other demographic parameters were inferred from long-term, country-wide population monitoring in Switzerland. The inversely calibrated model was then used to attribute observed population changes to climate effects using counterfactual scenarios, providing insight into how demographic rates scale with environmental conditions and how climate change has shaped past population dynamics.
- Potential translocation strategies for the European Bison and consequences for its genetic connectivity: Neutral genetic simulations in RangeShifter were used to evaluate functional connectivity between populations, beyond simple range-shifting potential, and testing for the role of current translocation efforts to maintain genetic diversity. This approach informs long-term conservation planning by identifying strategies to maintain self-sustaining spatially structured populations and ensure genetic health over time.
- Impact of recurrent anthropogenic disturbance on the distribution and dynamics of genetic diversity: In a theoretical study, RangeShifter was used to explore how repeated and frequent anthropogenic disturbances, such those occurring in agricultural landscapes, influence the spatial structure and temporal dynamics of genetic diversity. These simulations reveal genetic rescue and sink dynamics from undisturbed to disturbed habitats that shape maintenance and structure of genetic diversity depending on species’ life histories, disturbance and landscape characteristics.
Together, these studies highlight RangeShifter’s versatility for addressing diverse empirical and theoretical questions and ecological scenarios, from species’ responses to environmental change and their genetic consequences to large-scale conservation strategies.
How to cite: Bocedi, G., Wolf, J., Bunnenberg, C., Pannetier, T., Henry, R., Travis, J., and Zurell, D.: Investigating species responses to environmental changes with RangeShifter: a process-based eco-evolutionary modelling platform. , World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-510, https://doi.org/10.5194/wbf2026-510, 2026.
Under the threat of environmental changes, including climate and land-use change, the need to predict impacts on biodiversity, species distributions, and management decisions is becoming increasingly important. Process-based models are valuable tools for predicting ecological responses to environmental change and for understanding the underlying dynamics. RangeShifter is a spatially explicit, individual-based modelling framework designed to simulate eco-evolutionary processes and assess species’ responses to climate and/or land-use change. The model consists of three main modules: a demography module, simulating population dynamics; a dispersal module, encompassing emigration, transfer, and settlement processes; and a genetics module, enabling adaptation and tracking the populations’ genetic health. Each module can be parameterized from simple to complex representations, offering users a high degree of flexibility to adapt the model to their specific research questions.
First introduced in 2014 and available since 2021 also as an R package, RangeShifter has been applied to a wide range of species, from small insects to large mammals, and in various ecological contexts, including predicting range shifts, informing conservation planning, optimizing reintroduction management strategies, and addressing theoretical questions such as the consequences of habitat fragmentation on movement patterns and the evolution of dispersal traits. The latest major update introduces four additional features now available to all users.
We present these new features implemented in the most recent development phase of RangeShifter. The framework now provides an expanded genetic module enabling forward-time population genetics simulations, allowing users to explore evolutionary dynamics and genetic diversity in more detail. Importantly, the framework is no longer anymore to simulating single species but can now simulate multiple interacting species, thus scaling up to community-level processes. In addition, the updated version can incorporate demographic rates that vary in space and time, and supports the explicit simulation of translocations as a management strategy.
These enhancements expand the platform’s range of applications and provide users with greater flexibility in model design and conceptualization. Continued development of modelling frameworks such as RangeShifter opens new avenues for mechanistic, spatially explicit studies of species and community responses to environmental change.
How to cite: Wolff, J., Pannetier, T., Henry, R., Travis, J., Zurell, D., and Bocedi, G.: RangeShifter: Enhanced Tools for Eco-evolutionary Modelling of Species Range Dynamics, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-524, https://doi.org/10.5194/wbf2026-524, 2026.
The global imperative to address the biodiversity crisis requires transparent and trustworthy spatial data products to measure ecological state, assess trends, and inform policy. A critical data gap persists in sensitive Afromontane ecosystems of the Global South, such as the Ruo River in Malawi, a key biodiversity area facing intensifying pressure from irrigation and hydropower. An urgent priority is establishing a scientifically rigorous baseline by quantitatively linking hydrological variability (river flow) to biological response. This study contributes directly to the session’s objectives by employing a quantitative longitudinal design and time-tested statistical models to generate meaningful biodiversity indicators. The approach integrated newly generated in situ biodiversity data aquatic macroinvertebrates, classified using SASS version 5 with measured river flow rates and physicochemical parameters across a natural spatial gradient(upper stream, middle stream, and lower stream).. Crucially, the sampling design accounted for microhabitat diversity by sampling three diverse biotopes (gravel, sand, and mud; stone; and vegetation). Analysis relied on ANOVA to assess spatial and biotope heterogeneity and Generalized Linear Models (GLMs), specifically Poisson Regression, to explicitly model the non-linear relationship between flow parameters and key macroinvertebrate diversity indicators.
The results yield crucial, ready-to-use metrics for decision-making. The statistical models established a significant spatial gradient and confirmed that flow rate and biotope-specific conditions are strong predictors of community structure. A clear community turnover was observed, with sensitive families (e.g., EPT groups) dominating less-disturbed reaches, and tolerant taxa increasing in areas impacted by reduced flow. By focusing on transparent GLMs, the study facilitated clear uncertainty quantification and enhanced model transparency for the derived indicators (richness, EPT scores). Furthermore, the project produced a new macroinvertebrate database for the Ruo River, advancing regional data management and benchmarking.
In conclusion, this research provides a rigorous example of generating high-fidelity spatial indicators from fundamental data. The established flow-ecology relationships form a vital biodiversity baseline and directly inform the urgent need for a locally calibrated biomonitoring index, supporting national decision-making and prioritizing the crucial role of transparent, validated models in global conservation strategy and development.
How to cite: Matemba, M. S.: Spatially Explicit Flow-Ecology Modeling to Inform Biomonitoring in Data-Scarce Afromontane Ecosystems, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-668, https://doi.org/10.5194/wbf2026-668, 2026.
Maintaining high biodiversity levels is of key importance in keeping ecosystems healthy and resilient, especially in landscapes undergoing intensive anthropogenic disturbances. Habitat destruction leads to severe biodiversity decline, and its spatial configuration affects its impact. Knowing the exact impact of habitat loss and fragmentation on species richness is valuable information. Predicting the effects of different habitat loss scenarios on biodiversity will facilitate making better, well-informed management decisions, but such predictions are not yet available.
We predict species-area relationships in tropical forest landscapes undergoing habitat loss and fragmentation. To do so, we use the information theoretical framework of Maximum Entropy Theory in Ecology (METE). METE uses state variables derived from macroscale parameters (i.e. total number of individuals, total number of species, and total metabolic energy in the selected area) to provide the most uninformed distribution of the species-area relationship. While METE makes accurate predictions in undisturbed systems, METE predictions in landscapes undergoing habitat loss and fragmentation are highly imprecise. This is due to the underlying assumption that individuals are spread homogeneously across space, which fails in such environments.
Using an individual-based model, we explored how METE’s state variables should be adjusted to provide a good species-area relationship prediction in disturbed environments. These environments were simulated using different combinations of habitat loss and habitat fragmentation levels. Our study revealed that METE’s state variable S0(total number of species in the area) depends on habitat loss and fragmentation levels.
We demonstrate the accuracy of our predictions using empirical data. By extracting two landscape characteristics – percentage forest cover and degree of habitat clustering – from satellite images, we can adjust the state variables of METE accordingly and predict species richness. We compared the empirically derived and METE predicted species-abundance distributions, and show that our method indeed works. Our adjusted METE predictions can demonstrate the severity of fragmentation’s impact on biodiversity, relative to what could be expected in a pristine, continuous environment. With this new method, we can estimate the immediate effects of habitat loss on biodiversity loss, as well as its long-term effects.
How to cite: de Jager, M., van Wordragen, F., Slegers, F., and Pos, E.: Predicting Species-Area Relationships in Fragmented Forests, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-50, https://doi.org/10.5194/wbf2026-50, 2026.
Marine plankton are incredibly diverse and play a central role in regulating ocean ecosystems. They drive primary production, sustain higher trophic levels, and export anthropogenic carbon to the deep ocean. Yet, different in situ observation types and diversity metrics capture distinct ecological processes, and the quantitative links between plankton diversity and ecosystem functioning remain only partially understood. Here, we use a state-of-the-art habitat modeling framework (CEPHALOPOD) to provide a unified assessment of global plankton diversity across occurrence, abundance, biomass and metagenomic observations. Our analysis is based on the regression output of more than 330,000 species-level in situ observations against environmental climatologies, enabling a scale-consistent comparison across data types, diversity assembling methods, and seasons.
We show that a pronounced latitudinal diversity gradient emerges across all estimates, but important divergences occur in high latitudes and coastal upwelling systems. Qualitative diversity estimates such as those constructed from occurrences and species richness, highlight productive or heterogeneous environments where many taxa co-occur despite strong bloom dynamics. By contrast, quantitative diversity estimates based on abundances, biomass, or metagenomic reads provide a complementary view linked to community evenness and persistent environmental stability, particularly within tropical oligotrophic gyres. Variance partitioning further reveals that these differences arise from distinct environmental drivers: productivity predominantly shapes occurrence-based diversity, whereas carbonate chemistry and temperature more strongly influence quantitative and metagenomic diversity.
Finally, we demonstrate that plankton diversity patterns, across all axes of variance considered, serve as robust indicators of key ecosystem properties. Qualitative diversity closely tracks global net primary production, reflecting its sensitivity to productivity-driven species turnover. In contrast, quantitative and metagenomic diversity correlate strongly with carbon export efficiency, consistent with their ecological link community evenness and trophic transfer. All diversity estimates also correlate with global megafauna diversity, revealing a coherent structuring of marine communities from microbes to upper trophic levels. Overall, this diversity intercomparison unifies complementary perspectives on plankton community structure and ecosystem functioning, providing a scalable framework to quantify the drivers of marine biodiversity globally.
How to cite: Schickele, A., Clerc, C., Benedetti, F., Sonnet, V., and Vogt, M.: Marine plankton diversity across in situ observation types, assembling methods, and seasons., World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-384, https://doi.org/10.5194/wbf2026-384, 2026.
How to cite: van der Veen, B., Coste, C., Lee, A., Solbu, E., Henriksen, M., Hovstad, K., Peres-Neto, P., and O'Hara, R.: A New Approach to Measuring Biodiversity, World Biodiversity Forum 2026, Davos, Switzerland, 14–19 Jun 2026, WBF2026-657, https://doi.org/10.5194/wbf2026-657, 2026.
