Deep learning reveals the individual-level effects of artificial light at night on a wild insect

Artificial light at night (ALAN) is a widespread anthropogenic phenomenon with varied physiological, behavioural, and ecosystem-level effects. Its impacts have been studied extensively at the population level, however less is known about the individual changes that underpin these larger trends. We use a networked video system combined with GryllAI--a deep learning-based system for continuous individual monitoring--to explore this in the field cricket, Gryllus campestris. Applying field-realistic artificial light (10-25lx) or a control treatment to burrows, we continuously track the activity of 144 nymphs across >38,000h of video footage, recording life history outcomes for each individual. Results indicate that ALAN exposure does not influence daily activity timing, total activity duration, predation risk, or nymphal development duration. However, changes in fine-scale behaviour were apparent, with ALAN causing crickets to enter and exit burrows less frequently, especially at night; and spent greater proportion of time outside burrows during daytime. These behavioural adjustments were not evident from broad scale activity trends that could be observed manually. Consequently, our findings suggest that aggregate measures of activity may fail to capture the full scope of ALAN-mediated impacts in nature, and that automated monitoring techniques offer a promising means of addressing this.