When plants die, neighbours escape competition. Living conspecifics could disproportionately benefit because they are freed from negative intraspecific processes; however, if the negative effects of past neighbours persist, other species may be advantaged and diversity maintained through ‘legacy effects’ of dead trees. Using large mapped forest plots, we examine whether antecedent communities shape forest dynamics and how long dead-tree legacies persist. We find that per-capita interaction strength of dead neighbours exceeds that of living trees, and that density-dependent effects from dead trees can persist for up to 10 years after death. Over 75% of species show negative responses to dead-tree neighbours, compared with 67% to living neighbours. Preliminary evidence suggests consistent legacy effects across tropical and temperate forests, with no clear mediating gradients. The drivers of tree legacy effects remain a mystery.

Conformal prediction methods are statistical tools designed to quantify uncertainty and generate predictive sets with guaranteed coverage probabilities. The two works I will present, introduce refinements to these methods for classification tasks, one specifically tailored for long-tailed classification and, the other one, for scenarios where multiple observations (multi-inputs) of a single instance are available at prediction time. Our approach is particularly motivated by applications in citizen science such as Pl@ntnet, where the distribution of species has this behavior and where multiple images of the same plant or animal are captured by individuals at test time. These works have been done in collaboration with Joseph Salmon (UM), Tiffany Ding (UC Berkeley) and Mohamed Hebiri (Université Gustave Eiffel).

Dr Baraloto présentera des résultats originaux sur les patrons et les processus dans les forêts de l’Amazonie, le long des gradients altitudinaux du Bouclier guyanais et du Costa Rica, ainsi que sur la récupération des forêts perturbées. Il souhaite profiter de son passage pour explorer de possibles collaborations entre son équipe à Miami et les équipes françaises sur l’étude des communautés d’arbres néotropicaux

My stay at the UMR AMAP aims to assess the influence of the spatial variability of topo-climatic, edaphic and anthropogenic factors on the distribution of the Mamora cork oak forest “understory” species in northwestern Morocco. We use a JSDM to model species cover-abundance from 161 floristic relevés inventoried according to Braun-Blanquet's sigmatist phytosociological method during 1999-2004. We employ Multivariate Ordinal Probit Models, which are a natural extension of the JSDMs traditionally used to model presence/absence data. The flexibility of these models is based on the existence of a latent Gaussian field of dimension J that depends on environmental conditions. The findings show that the spatial heterogeneity in precipitation, temperature, and overgrazing rate differently control the cover-abundance distribution of the Mamora understory species. The estimated influences and residual correlations among species have implications for the effect of droughts and increasing overgrazing on species abundance and composition in the study area.

Assessing the true stability of slow-dynamic ecosystems, such as forests, remains a major challenge in ecology. Beyond advancing our understanding of ecosystem functioning, this issue has critical implications for predicting their long-term trajectories and improving management strategies. Here, we integrate theoretical modeling with field and remote sensing data to evaluate the stability of a potential alternative stable state within Central African tropical moist forests: the Marantaceae forest. As forests face increasing disturbances, the model accounts for the combined effects of fire and forestry on the relative stability of forest types. Long-term simulations predict notable shifts in the Central African forest mosaic under certain scenarios, informing sustainable management approaches.

Differentiable programming encompasses methods for computing the gradient of a numerical model’s output. As a cornerstone of modern scientific computing, it presents immense opportunities to enhance our understanding and prediction of natural systems. In this talk, I will introduce the fundamental concepts of differentiable programming, present the available toolbox for practitioners, share lessons learned from a user perspective, and showcase three research projects from my portfolio that leverage differentiable programming to advance ecological research: predicting spatial biodiversity patterns, modeling ecological community dynamics, and optimizing biodiversity conservation planning. Through this talk, I aim to illustrate how differentiable programming can accelerate the integration of scientific models and data, a crucial step toward building interpretable models with robust extrapolation capabilities.