Biology subjects

Purves, D.

Publications and source records attributed to Purves, D..

4 recordsLinked to original sources

bioRxiv2025

Large language models possess some ecologicalknowledge, but how much?

Large Language Models (LLMs) have shown remarkable capabilities in question answering across various domains, yet their effectiveness in ecological knowledge remains underexplored. Understanding their potential to recall and synthesize ecological information is crucial as AI tools become increasingly integrated into scientific workflows. Here, we assess the ecological knowledge of two LLMs, Gemini 1.5 Pro and GPT-4o, across a suite of ecologically focused tasks. These tasks evaluate an LLMs ability to predict species presence, generate range maps, list critically endangered species, classify threats, and estimate species traits. We introduce a new benchmark dataset to quantify LLM performance against expert-derived data. While the LLMs tested outperform naive baselines, achieving around 20 percent-age points higher accuracy in species presence prediction, they reach only a third of the mean F1 score for range map generation and improve threat classification by just around 10 points over random guessing. These results highlight both the promise and challenges of applying LLMs in ecology. Our findings suggest that domain-specific fine-tuning is necessary to improve eco-logical knowledge in LLMs. By providing a repeatable evaluation framework, our benchmark dataset will facilitate future research in this area, helping to refine AI applications for ecological science.

ecology↗

bioRxiv2020

The Madingley General Ecosystem Model predicts bushmeat yields, species extinction rates and ecosystem-level impacts of bushmeat harvesting

Traditional approaches to guiding decisions about harvesting bushmeat often employ single-species population dynamic models, which require species- and location-specific data, are missing ecological processes such as multi-trophic interactions, cannot represent multi-species harvesting, and cannot predict the broader ecosystem impacts of harvesting. In order to explore an alternative approach to devising sustainable harvesting strategies, we employ the Madingley General Ecosystem Model, which can simulate ecosystem dynamics in response to multi-species harvesting given nothing other than location-specific climate data. We used the model to examine yield, extinctions, and broader ecosystem impacts, for a range of harvesting intensities of duiker-sized ectothermic herbivores. Duiker antelope (such as Cephalophus callipygus and Cephalophus dorsalis) are the most heavily hunted species in sub-Saharan Africa, contributing 34%-95% of all bushmeat in the Congo Basin. Across a range of harvesting rates, the Madingley model gave estimates for optimal harvesting rate, and extinction rate, that were qualitatively and quantitatively similar to the estimates from single-species Beverton-Holt model. Predicted yields were somewhat greater (around 5 times, on average) for the Madingley model, which would be expected given that the Madingley simulates multi-species harvesting from an initially pristine ecosystem. This match increased the degree of confidence with which we could examine other predictions from the ecosystem model, as follows. At medium and high levels of harvesting of duiker-sized herbivores, there were statistically significant, but moderate, reductions in the densities of the targeted functional group; increases in small-bodied herbivores; decreases in large-bodied carnivores; and minimal ecosystem-level impacts overall. The results suggest that general ecosystem models such as the Madingley model could be used more widely to help estimate sustainable harvesting rates, bushmeat yields and broader ecosystem impacts across different locations and target species.

ecology↗

bioRxiv2019

Mechanistic macroecology: exploring the drivers of latitudinal variation in terrestrial body size in a General Ecosystem Model

Many mechanisms have been hypothesized to explain Bergmanns rule - the correlation of body size with latitude. However, it is not feasible to assess the contribution of hypothesised mechanisms by experimental manipulation or statistical correlation. Here, we evaluate two of the principal hypothesised mechanisms, related to thermoregulation and resource availability, using structured experiments in a mechanistic global ecosystem model. We simulated the broad structure of assemblages and ecosystems using the Madingley model, a mechanistic General Ecosystem Model (GEM). We compared emergent modelled biogeographic patterns in body mass to empirical patterns for mammals and birds. We then explored the relative contribution of thermoregulation and resource availability to body mass clines by manipulating the models environmental gradients. Madingley produces body size gradients that are in broad agreement with empirical estimates. Thermoregulation and resource availability were both important controls on body mass for endotherms, but only temperature for ectotherms. Our results suggest that seasonality explains animal body mass patterns through a complex set of mechanisms. Process-based GEMs generate broadly realistic biogeographic body mass patterns. Ecologists can use them in novel ways: to explore causality, or for generating and testing hypotheses for large-scale, emergent ecological patterns. At the same time, macroecological patterns are useful for evaluating mechanistic models. Iteratively developing GEMs, and evaluating them against macroecological patterns, could generate new insights into the complex causes of such patterns.

ecology↗

bioRxiv2019

Food Webs: Insights from a General Ecosystem Model

Food webs have been intensively studied throughout modern ecology, using empirical evidence, statistics and modelling tools to search for consistent patterns, underlying commonalities, and variations between food webs and ecosystems. However, with few exceptions, the modelling approaches have not been based on the emergent properties of complex simulated ecosystems. For the same reason, there have also been few studies of sampling the model, in which different levels of sampling effort are imposed in a controlled simulated environment to explore the effects of varying sampling intensity in order to relate those back to empirical observations. Here, we introduce the Madingley Model, a general ecosystem model based on ecological and biological first principles of interactions between individuals, as a potential tool for analyzing food webs. In doing so, we present the first insights of the analyses of emergent Madingley food web networks. We describe the basic structure of these networks and introduce a process to aggregate and sample the food webs produced by this model so that it can be compared to empirical food web studies. We show that food webs created by this model reasonably reproduce the properties of empirical food web networks. Furthermore, we provide insights about the effects of species aggregation and sampling on the observed structure of empirically documented food webs.

ecology↗