Fire is a powerful tool for conservation management at a landscape scale, but a rigorous evidence base is often lacking for understanding its impacts on biodiversity in different biomes. Fire-induced changes to habitat openness have been identified as an underlying driver of responses of faunal communities, including ants. However, most studies of the impacts of fire on ant communities consider only epigaeic (foraging on the soil surface) species, which may not reflect the responses of species inhabiting other vertical strata. Here, we examine how responses of ant communities vary among vertical strata in a highly fire-prone biome. We use a long-term field experiment to quantify the effects of fire on the abundance, richness, and composition of ant assemblages of four vertical strata (subterranean, leaf litter, epigaeic, and arboreal) in an Australian tropical savanna. We first document the extent to which each stratum harbours distinct assemblages. We then assess how the assemblage of each stratum responds to three fire-related predictors: fire frequency, fire activity, and vegetation cover. Each stratum harboured a distinct ant assemblage and showed different responses to fire. Leaf-litter and epigaeic ants were most sensitive to fire because it directly affects their microhabitats, but they showed contrasting negative and positive responses, respectively. Subterranean ants were the least sensitive because of the insulating effects of soil. Our results show that co-occurring species of the same taxonomic group differ in the strength and direction of their response to fire depending on the stratum they inhabit. As such, effective fire management for biodiversity conservation requires consideration of species in all vertical strata.

This dataset consists of ant sampling done in 2021-2022, using pitfall traps (n = 855), arboreal traps (n = 855), subterranean traps (n = 1664), and leaf litter extractions (n = 168). These traps were deployed over three sampling events, in 19 plots at the Territory Wildlife Park fire experiment, near Darwin, Australia. The experiment consists of 18 1-ha plots in a randomised block design, with six burning treatments in each of the three blocks (Figure S1). The treatments commenced in 2004 and are: burnt early (first week of June) in the dry season each one (E1), two (E2), three (E3), and five (E5) years; burnt late (first week of October) in the dry season every two years (L2); and unburnt (U). We also include fire activity data, which consists of the cumulated fire intensity experienced by a plot during the experiment, and woody cover data obtained with ground-based lidar scans.