Insect communities under skyglow: diffuse night-time illuminance induces spatio-temporal shifts in movement and predation.

Artificial light at night (ALAN) is predicted to have far-reaching consequences for natural ecosystems given its influence on organismal physiology and behaviour, species interactions and community composition. Movement and predation are fundamental ecological processes that are of critical importance to ecosystem functioning. The natural movements and foraging behaviours of nocturnal invertebrates may be particularly sensitive to the presence of ALAN. However, we still lack evidence of how these processes respond to ALAN within a community context. We assembled insect communities to quantify their movement activity and predation rates during simulated Moon cycles across a gradient of diffuse night-time illuminance including the full range of observed skyglow intensities. Using radio frequency identification, we tracked the movements of insects within a fragmented grassland Ecotron experiment. We additionally quantified predation rates using prey dummies. Our results reveal that even low-intensity skyglow causes a temporal shift in movement activity from day to night, and a spatial shift towards open habitats at night. Changes in movement activity are associated with indirect shifts in predation rates. Spatio-temporal shifts in movement and predation have important implications for ecological networks and ecosystem functioning, highlighting the disruptive potential of ALAN for global biodiversity and the provision of ecosystem services. This article is part of the theme issue 'Light pollution in complex ecological systems'.

Microhabitat conditions remedy heat stress effects on insect activity.

Anthropogenic global warming has major implications for mobile terrestrial insects, including long-term effects from constant warming, for example, on species distribution patterns, and short-term effects from heat extremes that induce immediate physiological responses. To cope with heat extremes, they either have to reduce their activity or move to preferable microhabitats. The availability of favorable microhabitat conditions is strongly promoted by the spatial heterogeneity of habitats, which is often reduced by anthropogenic land transformation. Thus, it is decisive to understand the combined effects of these global change drivers on insect activity. Here, we assessed the movement activity of six insect species (from three orders) in response to heat stress using a unique tracking approach via radio frequency identification. We tracked 465 individuals at the iDiv Ecotron across a temperature gradient up to 38.7°C. In addition, we varied microhabitat conditions by adding leaf litter from four different tree species to the experimental units, either spatially separated or well mixed. Our results show opposing effects of heat extremes on insect activity depending on the microhabitat conditions. The insect community significantly decreased its activity in the mixed litter scenario, while we found a strong positive effect on activity in the separated litter scenario. We hypothesize that the simultaneous availability of thermal refugia as well as resources provided by the mixed litter scenario allows animals to reduce their activity and save energy in response to heat stress. Contrary, the spatial separation of beneficial microclimatic conditions and resources forces animals to increase their activity to fulfill their energetic needs. Thus, our study highlights the importance of habitat heterogeneity on smaller scales, because it may buffer the consequences of extreme temperatures of insect performance and survival under global change.

The travel speeds of large animals are limited by their heat-dissipation capacities.

Movement is critical to animal survival and, thus, biodiversity in fragmented landscapes. Increasing fragmentation in the Anthropocene necessitates predictions about the movement capacities of the multitude of species that inhabit natural ecosystems. This requires mechanistic, trait-based animal locomotion models, which are sufficiently general as well as biologically realistic. While larger animals should generally be able to travel greater distances, reported trends in their maximum speeds across a range of body sizes suggest limited movement capacities among the largest species. Here, we show that this also applies to travel speeds and that this arises because of their limited heat-dissipation capacities. We derive a model considering how fundamental biophysical constraints of animal body mass associated with energy utilisation (i.e., larger animals have a lower metabolic energy cost of locomotion) and heat-dissipation (i.e., larger animals require more time to dissipate metabolic heat) limit aerobic travel speeds. Using an extensive empirical dataset of animal travel speeds (532 species), we show that this allometric heat-dissipation model best captures the hump-shaped trends in travel speed with body mass for flying, running, and swimming animals. This implies that the inability to dissipate metabolic heat leads to the saturation and eventual decrease in travel speed with increasing body mass as larger animals must reduce their realised travel speeds in order to avoid hyperthermia during extended locomotion bouts. As a result, the highest travel speeds are achieved by animals of intermediate body mass, suggesting that the largest species are more limited in their movement capacities than previously anticipated. Consequently, we provide a mechanistic understanding of animal travel speed that can be generalised across species, even when the details of an individual species' biology are unknown, to facilitate more realistic predictions of biodiversity dynamics in fragmented landscapes.

Inferring the location of neurons within an artificial network from their activity.

Inferring the connectivity of biological neural networks from neural activation data is an open problem. We propose that the analogous problem in artificial neural networks is more amenable to study and may illuminate the biological case. Here, we study the specific problem of assigning artificial neurons to locations in a network of known architecture, specifically the LeNet image classifier. We evaluate a supervised learning approach based on features derived from the eigenvectors of the activation correlation matrix. Experiments highlighted that for an image dataset to be effective for accurate localisation, it should fully activate the network and contain minimal confounding correlations. No single image dataset was found that resulted in perfect assignment, however perfect assignment was achieved using a concatenation of features from multiple image datasets.