Biology subjects

Allen-Perkins, A.

Publications and source records attributed to Allen-Perkins, A..

6 recordsLinked to original sources

bioRxiv2026

reserBUGS: A reservoir computing framework for probabilistic forecasting of ecological abundance time series

Forecasting species population dynamics is a central challenge in computational ecology, yet existing approaches rarely combine flexible nonlinear modelling, support for count-based ecological data, and systematic uncertainty quantification within a single, scalable framework. Here we introduce reserBUGS, an open-source Python framework for ecological forecasting based on reservoir computing, a recurrent neural network architecture in which only a simple readout layer is trained while a fixed high-dimensional dynamical system encodes temporal memory and nonlinear dependencies. reserBUGS integrates species abundance time series with environmental covariates retrieved automatically from global climate products, generates probabilistic ensemble forecasts, and provides tools for forecast evaluation and reliability assessment. We evaluated reserBUGS using insect abundance time series from available biodiversity monitoring datasets, comparing its performance against seven statistical and machine-learning baselines over one- to five-year forecast horizons. Reservoir-based models consistently outperformed alternatives in both predicting future abundance and capturing forecast uncertainty, with environmental predictors increasing the proportion of stable forecasts and contributing additional predictive value beyond historical abundance dynamics alone, particularly at 3-4-year forecast horizons. Probabilistic forecasts further enabled the identification of conditions associated with reduced predictive skill, providing a practical basis for communicating forecast confidence to end users. While default configurations already achieved competitive performance across a taxonomically and geographically diverse set of time series, hyperparameter optimisation revealed substantial room for performance gains through series-specific tuning. reserBUGS offers a computationally efficient and extensible framework for ecological forecasting that is well suited to the short, heterogeneous time series typical of biodiversity monitoring programmes. Its combination of flexible nonlinear modelling, probabilistic uncertainty quantification, and automated environmental data integration addresses key practical barriers to the adoption of modern forecasting methods in conservation and ecological research.

bioinformatics↗

bioRxiv2026

Seasonal variation mediates the importance of species attributes in plant-pollinator interactions

Predicting species interactions remains a major challenge, as multiple species attributes operate simultaneously and their relative importance may vary seasonally and with temporal resolution. Here, we assess how the relative contributions of abundance, trait matching, and phenology to plant-pollinator interactions vary through a flowering season and across temporal resolutions using interaction data from three European botanical gardens. We show that predictive models explain a substantial proportion of variation in visitation patterns, with floral and pollinator abundances consistently explaining most variation across the flowering season. Trait matching between pollinator body size and floral size also contributes to visitation patterns, playing a secondary but persistent role in shaping interactions, while phenology plays a relatively minor role at broader temporal resolutions but becomes more important at finer temporal scales. Additionally, null models accounting for spatio-temporal variation in floral and pollinator abundance reveal consistent patterns of pollinator preference and avoidance, indicating that abundance alone cannot explain the observed interaction patterns. Our results show that seasonal variation and temporal resolution differentially shape the importance of species attributes, highlighting the need for multi-variable approaches that account for temporal dynamics to accurately explain and predict ecological interactions.

ecology↗

bioRxiv2023

Structural asymmetry in biotic interactions as a tool to understand and forecast ecological persistence

A universal feature of ecological systems is that species do not interact with others with the same sign and strength. Yet, the consequences of this asymmetry in biotic interactions for the short- and long-term persistence of individual species and entire communities remains unclear. Here, we develop a set of metrics to evaluate how asymmetric interactions among species translate to asymmetries in their individual vulnerability to extinction under changing environmental conditions. These metrics, which solve previous limitations of how to independently quantify the size from the shape of the so-called feasibility domain, provide rigorous advances to understand simultaneously why some species and communities present more opportunities to persist than others. We further demonstrate that our shape-related metrics are useful to predict short-term changes in species relative abundances during seven years in a Mediterranean grassland. Our approach is designed to be applied to any ecological system regardless of the number of species and type of interactions. With it, we show that is possible to obtain both mechanistic and predictive information on ecological persistence for individual species and entire communities, paving the way for a stronger integration of theoretical and empirical research.

ecology↗

bioRxiv2022

The non-random assembly of functional motifs in plant-pollinator networks

Ecological processes leave distinct structural imprints on the species interactions that shape the topology of mutualistic networks. Detecting those relationships is not trivial since they go beyond pairwise interactions, but may get blurred when considering full network descriptors. Recent work has shown that the network meso-scale can capture this important information. The meso-scale describes network subgraphs representing patterns of interactions between a small number of species (i.e., motifs) that constitute the building blocks of the whole network. Despite the possible implications of network motifs to better capture species interactions, they remain overlooked in natural plant-pollinator networks. By exploring 60 empirical plant-pollinator networks from 18 different studies with wide geographical coverage we show that some motifs are consistently under- or over-represented worldwide, suggesting that the building blocks of plant-pollinator networks are not random. Furthermore, we find that distinct motif positions describing species ecological roles (e.g., generalisation and number of indirect interactions) are occupied by different plant and floral visitor groups on both trophic levels. Bees appear less frequently in specialised motif positions with high number of indirect interactions, while the rest of floral visitor groups are infrequent in generalised motif positions with low number of indirect interactions. All plant groups tend to be over-represented on specialised motif positions, except tall plant species with separated sexes (e.g., woody dioecious or monoecious species), which are more frequent on motif positions with low number of indirect interactions. Interestingly, the realized combinations of different species groups within a motif can not be retrieved from their joint probability distributions, indicating that group combinations are not random either. Our result highlights the non-random structure of the meso-scale on plant-pollinators networks and the association of different plant and floral visitor groups with certain motifs that involve different ecological roles at a macro-ecological scale.

ecology↗

bioRxiv2021

The non-random structure of multi-trophic ecological interactions maximizes species coexistence within ecologically realistic constraints

Theory posits that the persistence of species in ecological communities is shaped by their interactions within and across trophic guilds. However, we lack empirical evaluations of how the structure, strength, and sign of biotic interactions drive the potential to coexist in diverse multi-trophic communities. Here we model community feasibility domains, a theoretically-informed measure of multi-species coexistence probability, from grassland communities comprising more than 45 species on average from three trophic guilds (plants, pollinators, and herbivores). Contrary to our hypothesis, increasing community complexity, measured either as the number of guilds or community richness, did not decrease community feasibility. Rather, we observed that high degrees of species self-regulation and niche partitioning allow maintaining larger levels of community feasibility and higher species persistence in more diverse communities. Our results show that biotic interactions within and across guilds are not random in nature and both structures significantly contribute to maintaining multi-trophic diversity.

ecology↗

bioRxiv2021

Community level individual-based plant-pollinator networks unveil pollen flow dynamics and plant reproductive success

Interaction networks are a widely used tool to understand the dynamics of plant-pollinator ecological communities. However, while most mutualistic networks have been defined at the species level, ecological processes such as pollination take place at different scales, including the individual or patch levels. Yet, current approaches studying fine-grain sub-specific plant-pollinator networks only account for interactions among nodes belonging to a single plant species due to the conceptual and mathematical limitations of modeling simultaneously several plant species each composed of several nodes. Here, we introduce a multilayer diffusion network framework that allows modeling simple diffusion processes between nodes pertaining to the same or different layers (i.e. species). It is designed to depict from the network structure the potential conspecific and heterospecific pollen flows among plant individuals or patches. This potential pollen flow is modeled as a transport-like system, in which pollen grain movements are represented as random-walkers that diffuse on an ensemble of bipartite layers of conspecific plants and their shared pollinators. We exemplify this physical conceptualization using a dataset of nine fine-grain sub-specific plant-pollinator networks from a Mediterranean grassland of annual plants, where plant nodes represent groups of conspecifics within patches of 1m2. The diffusion networks show pollinators effectively connecting sets of patches of the same and different plant species, forming a modular structure. Interestingly, different properties of the network structure, such as the conspecific pollen arrival probability and the number of conspecific subgraphs in which plants are embedded, are critical for the seed production of different plant species. We provide a simple but robust set of metrics to calculate potential pollen flow and scale down network ecology to functioning properties at the individual or patch level, where most ecological processes take place, hence moving forward the description and interpretation of species-rich communities across scales.

ecology↗