Wild insects pollinate numerous agricultural crops, but the role of nocturnal pollinators, while increasingly acknowledged, remains poorly understood. We examined nocturnal insect communities and pollination in agroforests of robusta coffee (Coffea canephora) – a crop that exhibits floral traits suggestive of nocturnal pollination – in India’s Western Ghats mountains. Specifically, we (1) compared nocturnal insect communities of a shaded robusta coffee agroforest and a nearby secondary tropical rainforest using light screens, and (2) assessed nocturnal and diurnal pollination of coffee using floral exclosure experiments in the agroforest and in a former coffee agroforest located within the secondary rainforest. Nocturnal pollinators visiting light screens were 21% fewer in the agroforest than the rainforest, mainly due to reduced numbers of Lepidoptera, Coleoptera, and Diptera in the former. Lepidoptera and Coleoptera differed in genus richness and composition between habitats, with the agroforest having fewer Lepidoptera and more Coleoptera genera than the rainforest. Coffee pollination success was largely attributable to diurnal pollinators in both the agroforest and rainforest. While nocturnal pollination effects were absent in the agroforest, we found some evidence of nocturnal pollination in the secondary rainforest, where coffee flowers accessible to diurnal and nocturnal pollinators had higher pollination success (60%) than flowers accessible to diurnal pollinators alone (46%). In summary, the nocturnal insect community of coffee agroforestry, which is distinct from the rainforest community, contributes little to coffee pollination. However, a greater contribution of nocturnal pollination under less intensive coffee cultivation is a possibility that warrants further exploration.

Study site:
This study was conducted at a robusta coffee farm and a secondary tropical rainforest in the Sakleshpur Taluk, Karnataka State, located in the Western Ghats biodiversity hotspot, southern India. 

Insect sampling:
Nocturnal insects were sampled using light screens, which comprised a vertical white 3.24 m² square screen illuminated by four ultraviolet LEDs, two blue LEDs, and one green and white LED (Fig. 1a), following Brehm (2017). Light screens were set up at dusk (1800 h) and nocturnal insects were inventoried between 2330 and 0130 h. The timing of data collection was determined based on a pilot study in the rainforest comprising 12 screens monitored hourly from 1830 to 0630 h, in which insect accumulation was observed to peak during 2330–0130 h. We compared the nocturnal flying insect communities of the robusta coffee agroforest and secondary tropical rainforest during March–April 2022.  

Pollinator exclusion experiments:
The experiment comprised the following treatments: (1) complete insect exclusion (Negative control), (2) pollinator exclusion by day (0630–1830h) and open at night (1830–0630h; Night - accessible), (3) pollination exclusion by night and open in the day (Day - accessible), and (4) no exclusion (Positive control). 

Statistical analysis: 
All data analyses and visualizations were performed using R version 4.2.2 (R Core Team, 2022). We compared light screen encounter rates of nocturnal pollinators between the agroforest and rainforest using generalized linear models (GLM) from the R package MASS (Ripley et al., 2023).

We compared Lepidoptera and Coleoptera taxonomic diversity at the genus level between coffee agroforest and rainforest, standardized for coverage, using the R package iNEXT (Chao et al., 2014; T. C. Hsieh et al., 2022).

We used Non-Metric Dimensional Scaling (NMDS) to visually represent the dissimilarity of Lepidoptera and Coleoptera genus-level composition between agroforest and rainforest habitats. We performed a three-dimensional ordination based on the Bray-Curtis dissimilarity index using the metaMDS function in the ‘vegan’ R package (Oksanen et al., 2022).

For the pollination exclusion experiments, we used generalized linear mixed-effects models (GLMM) with a binomial error distribution to compare pollination success (a binary variable) across the different exclusion treatments and controls. We ran separate models for the 2022 experiment in the coffee agroforest and the 2023 one in abandoned coffee bushes within the rainforest. Each model comprised the exclusion treatments and controls as fixed effects and bush ID as a random effect. GLMMs were run using the ‘lme4’ R package (Bates et al., 2015).