Biology subjects

Dormann, C. F.

Publications and source records attributed to Dormann, C. F..

8 recordsLinked to original sources

bioRxiv2026

Evaluating Transferability and Robustness of Process-Guided Neural Networks in Forest Carbon Flux Modelling

Making robust and generalizable predictions within ecological systems such as forests remains challenging due to limited data availability and the slow pace of environmental change. To address this, we integrate a semi-empirical environmental process model (PRELES) to support deep learning approaches, specifically artificial neural networks (ANNs). We replicate and extend previous work on process-guided neural networks (PGNN) by introducing new model types and conducting a comprehensive hyperparameter optimisation within systematic nested cross-validation analyses in both data-thinning and extrapolative scenarios. Results show that both data-driven ANNs and PGNNs consistently outperform the stand-alone process model, while PGNNs provide additional advantages over ANNs in data-sparse settings and under transfer scenarios to unseen, changing climatic conditions. We further estimate the generalisation error for data-driven models as a function of the amount of training data, allowing for guidance on model suitability under different data availability. A variable importance analysis using accumulated local effects reveals that both PGNNs and ANNs learn simple, physically plausible relationships, whereas PRELES exhibits a strong bias toward boreal conditions and limited ability to predict unseen, climatically divergent sites. HighlightsO_LIProcess-guided, and plain neural networks outperform a calibrated process-based model (PRELES) in predicting forest ecosystem carbon fluxes. C_LIO_LIProcess-guided neural networks provide advantages over naive neural networks in sparse-data settings and show greater robustness under transferable scenarios with unseen changing climatic conditions. C_LIO_LIVariable-importance analyses using accumulated local effects show that both process-guided and naive neural networks learn simple yet physically plausible relationships between meteorological drivers and target responses, whereas the process model (PRELES) exhibits a better fit toward boreal conditions and limited ability to predict unseen, climatically divergent sites. C_LI

ecology↗

bioRxiv2025

Climate-driven specialisation in plant-pollinator networks peaks outside the tropics

Pollination is a key ecological process sustaining biodiversity and food security, yet global patterns of plant-pollinator specialisation have remained unresolved. Using the largest global dataset of quantitative networks (>3,400 networks, >110,000 interactions), we show that the latitudinal specialisation gradient (LSG) exists, but it is non-linear, hemispherically asymmetric, and strongly taxon-dependent. Network-level and pollinator specialisation were lowest in the tropics and peaked at northern mid-latitudes, whereas plants tended to become more specialised toward higher latitudes. Climate consistently outperformed latitude, species richness, and environmental productivity as a predictor of these patterns. Specialisation declined with increasing temperature, rose with moderate rainfall before declining at the wettest sites, and increased with temperature seasonality, but plants and pollinators responded differently to these drivers. Functional groups diverged strongly: ectothermic insects were most specialised in cooler, seasonal climates, while birds showed weaker links to latitude but reduced specialisation in wetter regions. These findings demonstrate that climate, rather than latitude or species richness, structures global variation in specialisation. Because warmer and less seasonal climates promote generalisation, climate change is likely to disrupt the most specialised pollination systems, unevenly across taxa and regions, with important consequences for biodiversity and ecosystem stability.

ecology↗

bioRxiv2025

From smartphones to satellites: Uniting crowdsourced biodiversity monitoring and Earth observation to fill the gaps in global plant trait mapping

Plant functional traits are fundamental to ecosystem dynamics and Earth system processes, but their global characterization is limited by the availability of field surveys and trait measurements. Recent expansions in biodiversity data aggregation, including large collections of vegetation surveys, citizen science observations, and trait measurements, offer new opportunities to overcome these constraints. Here we demonstrate that combining these diverse data sources with high-resolution Earth observation data enables accurate modeling of key plant traits at up to 1 km resolution. Our approach achieves high predictive power, reaching correlations up to 0.63 (15 of 31 traits exceeding 0.50) and improved spatial transferability, effectively bridging gaps in under-sampled regions. By capturing a broad range of traits with high spatial coverage, these maps can enhance our understanding of plant community properties and ecosystem functioning globally, and can serve as useful tools in modeling global biogeochemical processes and informing worldwide conservation efforts. Ultimately, our framework highlights the power and necessity of crowdsourced biodiversity data in high-resolution plant trait modeling. We anticipate that advancements in biodiversity data collection and remote sensing capabilities will further refine global trait mapping, fostering a dynamic trait-based understanding of the biosphere.

ecology↗

bioRxiv2022

Process-guidance improves predictive performance of neural networks for carbon turnover in ecosystems

Despite deep-learning being state-of-the-art for data-driven model predictions, it has not yet found frequent application in ecology. Given the low sample size typical in many environmental research fields, the default choice for the modelling of ecosystems and its functions remain process-based models. The process understanding coded in these models complements the sparse data and neural networks can detect hidden dynamics even in noisy data. Embedding the process model in the neural network adds information to learn from, improving interpretability and predictive performance of the combined model towards the data-only neural networks and the mechanism-only process model. At the example of carbon fluxes in forest ecosystems, we compare different approaches of guiding a neural network towards process model theory. Evaluation of the results under four classical prediction scenarios supports decision-making on the appropriate choice of a process-guided neural network. O_TEXTBOXSignificance StatementDeep-learning is the state-of-the-art for data-driven model predictions. Given the low sample size typical in many environmental research fields, these approaches can rarely be applied. When data are complemented by process understanding, as coded in physical or empirical models, both predictions and generalisations can be substantially improved over both data-only neural networks and mechanism-only process models. Comparing different approaches of such process-guidance helps decide on how to best combine process models and neural networks. C_TEXTBOX

ecology↗

bioRxiv2021

A novel perspective on the meaning of nestedness with conceptual and methodological solutions

O_LINestedness is a common pattern in metacommunities and interaction networks, whose causes are still subject to investigation. Nestedness inference is challenging because, beyond calculating an index, we need to compare observed and expected values using null models. There are many different null models and the choice between them strongly affects test outcomes. Furthermore, there is no theoretical basis to guide this choice. C_LIO_LIHere, we propose a different look at the meaning of nestedness that improves our understanding of its causes and directly links null models to hypotheses. The nestedness of a matrix is the combination of marginal sum inequality and high overlap. The higher the overlap, the more predictable the cell values by marginal sums. Therefore, the only way a process can increase nestedness is by promoting marginal sum inequalities, without concomitantly introducing preferences. We propose a combination of null models to test for different topological hypotheses. In these null models, interactions are randomized following a probability matrix. The probability matrix of the equiprobable model excludes marginal sum inequalities and consequently all nestedness-generating mechanisms, providing a distribution of expected values for nestedness significance tests. The probability matrix of the proportional model, by conserving marginal sums and maximizing overlap, conserves nestedness-generating mechanisms and excludes nestedness-disrupting mechanisms, thereby yielding the expected nestedness values for fully nested matrices. C_LIO_LIWe advance a new protocol to test for nestedness in binary and weighted matrices, in which those two null models represent different topological hypotheses. Our protocol helps to tell apart cases in which nestedness is different from expected by chance and cases in which nestedness represents the best topological archetype. We also evaluate the efficiency of applying this protocol using some common nestedness indices. Analyses based on weighted data proved to be highly efficient and informative, while binary analyses induced severe distortions. Finally, we empirically illustrate our approach using a plant-pollinator network. C_LIO_LIThrough a shift of perspective, our approach reconciliates contradictions in null model analysis and provides a guideline for nestedness interpretation. C_LI

ecology↗

bioRxiv2020

European agroforestry is no universal remedy for biodiversity: a time-cumulative meta-analysis

BackgroundAgroforestry is a production system combining trees with crops or livestock. It has the potential to increase biodiversity in relation to single-use systems, such as pastures or conventional agriculture, by providing a higher habitat heterogeneity. In a literature review and subsequent meta-analysis, we investigated the relationship between biodiversity and agroforestry and critically appraise the underlying evidence of the results. ResultsBiodiversity in agroforestry was higher than in conventional agriculture, but could not outcompete pastures, forests and abandoned agroforestry systems. There was no overall biodiversity benefit in agroforestry systems. Data were available for plants, birds, bats and arthropods. Arthropods and birds were the two taxonomic groups profiting from agroforestry systems. A time-cumulative meta-analysis shows that there was no general benefit of biodiversity at any point in the past besides in early 2015. Time-cumulative meta-analysis can unravel missing robustness of meta-analytical results if conclusions alternate between significant to non-significant summary effect sizes over time. ConclusionAgroforestry increases biodiversity only in silvoarable systems compared with conventional agriculture. But even this result is based on a small magnitude and single-study effect sizes were heterogeneous with sometimes opposing conclusions. The latter suggests the importance of other usually unmeasured variables, such as landscape parameters or land-use history, influencing biodiversity in agroforestry systems.

ecology↗

bioRxiv2019

Neither global nor consistent: a technical comment on the tree diversity-productivity analysis of Liang et al. (2016)

The publication of Liang et al. (2016, Science) seems to demonstrate very clearly that increasing tree species richness substantially increases forest productivity. To combine data from very different ecoregions, the authors constructed a relative measure of tree species richness. This relative richness however confounds plot-level tree species richness and the polar-tropical gradient of tree species richness. We re-analysed their orginal data, computing a regional measure of tree species richness and addressing several other issues in their analysis. We find that there is virtually no effect of relative tree species richness on productivity when computing species richness at the local scale. Also, different ecoregions have very different relationships between tree species richness and productivity. Thus, neither the \"global\" consistency nor the actual effect can be confirmed.

ecology↗

bioRxiv2018

A new model explaining the origin of different topologies in interaction networks

The architecture of interaction networks has been extensively studied in the past decades, and different topologies have been observed in natural systems. Despite several phenomenological explanations proposed, we still understand little of the mechanisms that generate those topologies. Here we present a mechanistic model based on the integrative hypothesis of specialization, which aims at explaining the emergence of topology and specialization in consumer-resource networks. By following three first-principles and adjusting five parameters, our model was able to generate synthetic weighted networks that show the main patterns of topology and specialization observed in nature. Our results prove that topology emergence is possible without network-level selection. In our simulations, the intensity of trade-offs in the performance of each consumer species on different resource species is the main factor driving network topology. We predict that interaction networks with low species diversity and low dissimilarity between resources should have a nested topology, although more diverse networks with large dissimilarity should have a compound topology. Additionally, our results highlight scale as a key factor. Our model generates predictions consistent with ecological and evolutionary theories and real-world observations. Therefore, it supports the IHS as a useful conceptual framework to study the architecture of interaction networks.

ecology↗