Les animations scientifiques et techniques constituent un temps fort du programme FORESTT. Elles seront programmées annuellement et vous pouvez y contribuer en suggérant des thèmes de recherche à l’adresse equipe@pepr-forestt.fr.
La première animation se tiendra le 22 septembre de 14h-17h, et portera sur les modèles de distribution d’espèces.
Le principe :
- Des capsules vidéo préenregistrées – en format 5 ou 20 minutes, en français ou en anglais, déposées sur la chaîne YouTube du PEPR d’ici le 29 août, pour être mises à disposition de la communauté scientifique pour visionnage et commentaires.
- Un webinaire de conversation scientifique, le 22 septembre après-midi, pour échanger sur ces présentations et répondre aux questions.
Vous pouvez participer activement à l’élaboration de ce webinaire, en rejoignant l’équipe des contributeurs ! Pour cela proposez, lors de votre inscription, un sujet en lien avec la thématique. Les contributeurs se verront proposer un échange avec les organisateurs entre le 16 et le 30 juin pour préciser le cadre de leur présentation.
Cette animation est bien entendu ouverte aux doctorant.e.s, ingénieur.e.s et technicien.ne.s de la sphère académique et des parties prenantes, s’agissant de discuter aussi des enjeux plus techniques sur le continuum de l’acquisition de données à la création d’outils d’aide à la décision.
Les inscriptions (contribution vidéo ou participation à la conversation scientifique du 22/09) sont ouvertes jusqu’au 16 juin.
Diffusion du programme définitif du webinaire (liste des orateurs, le titre et résumé de leurs présentations) le 30 juin.
D’avance merci pour vos contributions à venir ! Nous sommes bien sûr disponibles pour répondre à vos questions. Pour cela, il vous suffit de nous contacter à l'adresse equipe@pepr-forestt.fr.
Animateurs : Marta Benito Garzon, Isabelle Chuine, Christophe Plomion
Avec dores et déjà la participation de Jonathan LENOIR, Xavier MORIN, Isabelle CHUINE, Nicolas DELPIERRE, Victor VAN DER MEERSH, Marta BENITO GARZON, Thibault CAPBLANCQ
Pour vous inscrire : ICI
Advancements in Species Distribution Models: Addressing Challenges and Applications in Forest Management
Species distribution models have rapidly evolved from their origins. Early SDMs relied on the statistical correlation between species and local climate variables. While these models provided initial insights, they were often overly simplistic, leading to considerable uncertainties in their interpretations. From a biological perspective, early SDMs lacked critical biological processes occurring at different scales such as local adaptation, phenotypic plasticity, phenology, reproductive cycles, dispersal mechanisms, microclimatic processes, lagging dynamics and biotic interactions. From a mathematical perspective, these models were hindered by limited empirical data for calibration, poor transferability across different climatic conditions, inherent validation issues, and inherent uncertainties in the future climatic scenarios themselves. Advancements in empirical data collection, particularly for commercially important forest species, have allowed us to address several biological limitations of SDMs. Data on physiology, phenology, microclimatic conditions, time series of past environmental conditions, species interactions, adaptation mechanisms, plasticity, and genomics now allow for the inclusion of these critical biological processes in model frameworks. Likewise, mathematical and computational progress has improved SDMs by incorporating formal uncertainty analysis, robustness testing, and advanced tools like artificial intelligence and machine learning. These developments have enhanced the accuracy and applicability of SDMs in various ecological and management contexts. For forest trees, one of the most important applications of SDMs is the choice of forest reproductive material better adapted to future climatic conditions. However, the application of SDMs to forest management requires a critical evaluation of the models’ strengths, weaknesses, uncertainties, and practical relevance. This workshop will provide a comprehensive review and comparison of current SDMs, assessing their utility in forest management, and discussing pathways for scientific improvement and practical application. The goal is to foster collaboration between scientists and forest managers, ensuring that SDMs are properly addressing the challenges imposed by climate change on forests and could rapidly provide new tools for decision making.
Nous pouvons inclure des approches similaires pour d'autres organismes lorsqu'ils sont liés aux forêts dans le cadre d’une approche unifiée. Cela peut faire partie de la partie du séminaire consacrée aux interactions biotiques (Joint SDM)
Programme
An overview of DeepMaxent method for species distribution modeling
Maxime Ryckewaert, Diego Marcos, Christophe Botella, Maximilien Servajean, Pierre Bonnet, Alexis Joly
The rapid growth of citizen science initiatives has greatly expanded biodiversity databases, particularly presence-only (PO) data. Maxent is one of the most widely adopted point process methods for species distribution modeling (SDM). Maxent estimates species distributions by maximizing the entropy of a probability distribution between locations, based on predefined transformations of environmental variables, known as features. In this presentation, we introduce DeepMaxent, which relies on neural networks to automatically learn shared features between species while preserving Maxent's fundamental principle of maximum entropy. We are evaluating DeepMaxent on two different datasets: the NCEAS dataset, covering six distinct regions and five biological groups, and GeoPlant, a large-scale dataset of European plant observations.
Concept and challenges associated with using genomic data to refine species distribution models
Thibaut Capblancq
The current global climatic crisis has begun affecting the entire biosphere, posing a serious threat to the health and persistence of climate-sensitive populations and species.
At the same time, recent technological advancements give us access to massive quantities of data pertinent to biodiversity conservation (e.g., high-resolution DNA sequencing and climate models) and new sophisticated computational tools that can take advantage of these data to identify conservation risks and opportunities under a changing climate. A key question is how to harness the predictive power of statistical algorithms to generate actionable predictions about which populations will be most heavily impacted by climate change, and how these losses could be mitigated. To address this question, scientists have been developing computational approaches that pair large genomic datasets with high-resolution climate information to estimate the gene-climate relationship and identify populations most vulnerable under a changing climate. During this talk, I will describe how genomic offset can help informing management plans aiming at mitigating the negative impact of climate change on natural populations.
Accounting for microclimate processes matters for modelling forest biodiversity redistribution
Jonathan Lenoir, Stef Haesen, Koenraad Van Meerbeek
Species distribution models (SDMs) are traditionally based on ambient-air temperature (i.e., macroclimate) but fail to capture apparent temperature near the ground (i.e., microclimate). Yet, microclimate matters for the distribution of many organisms, especially in the understory of forest ecosystems. Very recent progress has been made to interpolate forest microclimate at very fine spatial resolution. Yet, very few SDMs’ studies have used microclimatic data to train species distribution at large spatial extent. Using a forest microclimatic grid at 25 m resolution across Europe, we compared conventional macroclimate-based SDMs with models corrected for forest microclimate buffering. We show that microclimate-based SDMs outperformed macroclimate-based SDMs. We also found that macroclimate-based SDMs introduced a systematic bias in modelled species response curves, resulting in erroneous range shift predictions. We emphasize the role of microclimate when fitting SDMs for forest biodiversity, especially so for tree seedlings as a mean to anticipate, and even mitigate, the potential negative impact of climate change on tree recruitment.
How tree demographic performance and competition vary within tree species’ ranges
Georges Kunstler, Anne Baranger, Jullien Barrere, Maxime Jaunatre, Bjoern Reineking
Currently, the tools available to forest managers for anticipating climate change impacts remain largely based on a static view of forests. Linking National Forest Inventories with functional traits and competition models reveals how growth, survival, recruitment, and competition vary within species.
Evaluation de la qualité de la prédiction des modèles de répartition corrélatifs
Xavier Morin, Valentin Journé, Jean-Yves Barnagaud, Pierre-André Crochet
L’immense majorité des prédictions de changement de répartition d’espèces (pas seulement les arbres) ont été réalisés avec des modèles corrélatifs (SDMc) liant environnement et répartition. Le succès de cette approche est principalement lié à la relative facilité d’utilisation de ces outils. Cependant, les fortes concordances généralement trouvées entre la répartition actuellement observée et prédite par ces modèles est aussi un argument fréquemment mis en avant pour promouvoir cette approche. Or il a pu être montré que ces fortes concordances peuvent être essentiellement dues aux caractéristiques de l’approche et des données, et ne reflétant que partiellement une réalité biologique. Je détaillerai ainsi cet aspect, en m’appuyant que l’étude que nous avons réalisée il y a quelques années (Journé et al. 2020).
Présentation du modèle mécaniste PhorEau
Xavier Morin, Tanguy Postic, Isabelle Chuine, Nicolas Martin
Je présenterai rapidement un nouveau modèle mécaniste (‘PhorEau’, Postic et al. 2025) pour les écosystèmes forestiers, permettant de réaliser des projections de répartition d’espèces mais aussi de communautés.
Présentation du modèle PHENOFIT
Isabelle Chuine, Florent Mouillot, François de Coligny, Xavier Morin, Frédérik Saltré, Anne Duputié, Emmanuel Gritti, Bérangère Leys, Victor Van der Meersch, Iris Le Roncé
Je présenterai le modèle PHENOFIT, modèle de fonctionnement de l'arbre, initialement basé sur la phénologie, la résistance au gel et au stress hydrique. La première version du modèle a surtout été utilisée pour étudier la niche écologique des arbres forestiers (hors de la zone tropicale) et prédire leur répartition géographique, passée, présente et future. La nouvelle version (non publiée) incorpore le modèle SIERRA (Mouillot et al. 2001) qui permet notamment de tenir compte de la compétition intra et interspécifique, d’estimer l’état hydrique des arbres, et d’allouer la productivité à différents compartiments de l’arbre. Elle incorpore également un modèle de fécondité qui décrit le cycle de reproduction depuis l’initiation florale à la maturation des fruits et comment les différentes étapes sont affectées par les conditions environnementales.
Integrating intra-specific trait variation into species distribution models to assess tree sensitivity to climate change
Marta Benito Garzon
Accounting for intra-specific trait variation is essential for understanding the sensitivity of populations and species to climate change. ΔTraitSDMs (Delta Trait Species Distribution Models) provide a novel framework to analyze intraspecific variation in one or multiple traits, based on the rationale that trait co-variation and fitness can vary substantially across geographic gradients and novel climates. These models integrate trait data collected from common garden experiments encompassing a wide range of environmental conditions for adult trees, as well as translocation experiments focusing on early life stages. Overall, predictions from ΔTraitSDMs present a less alarming outlook on species distributions under future climates compared to traditional models and highlight differences in climate sensitivity across the tree life cycle.
Fiabilité et incertitudes des projections des SDMs
Victor Van der Meersch, Isabelle Chuine
By combining multiple modeling approaches, high-performance computing, paleoevidence, and the most recent projections of past and future climate changes, we propose to investigate the reliability and uncertainty of the projections of a large spectrum of models, from correlative models to process-explicit models and their hybrid counterparts. In particular, we discuss the key features necessary for building models transferable to future climatic conditions, and further propose a way forward to foster the use of process-explicit models which are underrepresented in climate adaptation and impact assessment studies.
Being desiccation-sensitive in a warming climate: Insights into the early stages development of three recalcitrant oak species
Marion Carme, Eduardo Vicente, Marta Benito-Garzon
Early tree stages are crucial for forests sustainability but understudied. We tested warming effects on them in recalcitrant-seeded oak species. Traits responses were mostly determined by population differences, reflecting how each species adapted to the main climatic constraint of its native range.
Spatial prediction of Abies alba early phenotypes: Effects of cold stratification and climate change scenarios.
Eduardo Vicente, Marion Carme, Marta Benito Garzón
We present spatial predictions of Abies alba germination phenology and early traits like growth, SLA and chlorophyll content. Using population-based transfer distance models, we show that cold stratification and climate change scenarios strongly influence germination timing and seedling performance.
Joint Species Distribution Models (JSDM): What Perspectives on Addressing Between-Species Interactions and Unobserved Variables?
Arthur F. Rossignol & François Morneau
JSDMs (joint species distribution models) are an important improvement of classical SDMs in that they explicitly account for between-species associations via a covariance matrix. The latent variable framework has enabled to significantly reduce the dimension of JSDMs (which increases with the number of species). Latent variables can be seen as missing covariates that can be reconstructed thanks to the information shared between species. While JSDMs do not explicitly model biotic interactions, it is hoped that residual associations can be used to infer some patterns of interspecific interactions and unobserved covariates. These JSDMs are currently the subject of intense research, and their ability to capture biotic interactions remains widely debated.
NMI: a method for measuring the distance of future bioclimatic conditions in Europe from the tree species’ realized niche margin
Tristan Ubaldi, Jonathan Lenoir
Alongside species distribution models, ordination methods have also been developed to quantify the environmental niche of species and estimate niche differences, such as niche shifts or niche overlap between disjunct populations, or across different time periods and spaces. Principal component analysis (PCA) is presented as a robust ordination technique for representing species niches in an n-dimensional environmental space, along independent ecological gradients. Recently, Broennimann et al. (2021) introduced a new metric, the niche margin index (NMI), to measure the distance of future bioclimatic conditions in the receiving study region from the realized niche margin of the target tree species. Despite the promising applications of the NMI for assessing the success or failure of assisted tree migration, we are not aware of any study adapting the NMI metric in this context of climate emergency for the forestry sector. In this presentation, we illustrate the use of NMI to identify potential areas that will be climatically sustainable until 2100 for these species.
Joint Species Distribution models: Zero-inflated binary tree Pólya-splitting Regression approach
Mortier Frédéric, Jean Peyhardi, Dominique Lamonica
Understanding the impact of climate change on tropical rainforest ecosystems is crucial to promote efficient conservation strategies. The classical approach remains the use of species-specific distribution model. However, in species-rich ecosystems with many rare species, such an approach is doomed to failure. Moreover, univariate approaches ignore species dependencies. However, biodiversity is not merely the sum of species but the result of multiple interactions. Modeling multivariate count data allowing for flexible dependencies as well as zero inflation and overdispersion is challenging. In this presentation, we develop a new family of models called the zero-inflated binary tree Pólya-splitting models. This family allows the decomposition of a multivariate count data into a successive sub-model along a known binary partition tree. In the first part, I will present the general form of the zero-inflated binary tree Pólya-splitting model, studying the properties of the specified model in terms of marginal and conditional properties (distribution and moment). The second part presents the extension to the regression context. Finally, we finish presenting results on a real case study.
COLIBRI : principaux défis à relever pour améliorer l’offre et l’utilisation des outils d’aide à la décision
C. Perrier, M. Legay, A. Piboule, Th. Brusten, F. Lefèvre, N. Picard , L. Bulteau
Les forestiers ont besoin de pouvoir se projeter en tenant compte des incertitudes associées à ces projections, pour prendre des décisions en climat changeant. C'est à cette fin qu'ont déjà été développés un certain nombre d'outils, notamment ClimEssences et Bioclimsol, qui sont aujourd'hui intégrés aux processus de décision et font l'objet de formations auprès des décideurs et de cadrages stratégiques à leur utilisation. Des besoins restent cependant encore insatisfaits par rapport aux modèles sous-jacents à ces outils. Plus généralement, la démarche collective COLIBRI donne un cadre aux réflexions pour améliorer globalement l'aide à la décision appuyée par les connaissances scientifiques et l’expertise de terrain. Elle met en évidence les défis à relever collectivement dans les prochaines années pour mieux répondre aux besoins des décideurs en matière d’accompagnement à la décision et garantir parallèlement le bon usage des outils existants en veillant notamment à une meilleure compréhension des spécificités et limites de la connaissance sur laquelle ils s’appuient.
TuneSDM App
Nemer David
TuneSDM App est une application Shiny modulaire pour automatiser le réglage des modèles d’apprentissage automatique pour les SDMs. Développée en R avec le framework tidymodels, elle permet l’optimisation automatique des hyperparamètres (recherche bayésienne/grille) et l’évaluation des performances.
Axée sur l’accessibilité et la reproductibilité, l’application propose une interface intuitive pour configurer les workflows, charger les jeux de données et comparer les performances des modèles.
Elle s’adresse aussi bien aux utilisateurs expérimentés qu’aux écologues souhaitant intégrer des approches de machine learning sans expertise approfondie en programmation.
GeoPl@ntNet : A platform for exploring essential biodiversity variables
Lukas Picek, César Leblanc, Alexis Joly, Pierre Bonnet, Rémi Palard, Maximilien Servajean
This talk will describe GeoPl@ntNet, an interactive web application designed to make Essential Biodiversity Variables accessible and understandable to everyone through dynamic maps and fact sheets. Its core purpose is to allow users to explore high-resolution AI-generated maps of species distributions, habitat types, and biodiversity indicators across Europe. These maps, developed through a cascading pipeline involving convolutional neural networks and large language models, provide an intuitive yet information-rich interface to better understand biodiversity, with resolutions as precise as 50×50 meters. The website also enables exploration of specific regions, allowing users to select areas of interest on the map (e.g., urban green spaces, protected areas, or riverbanks) to view local species and their coverage. Additionally, GeoPl@ntNet generates comprehensive reports for selected regions, including insights into the number of protected species, invasive species, and endemic species.
