Effective conservation relies on accurate species distribution assessment and monitoring methodologies. Recent studies have shown the promises of airborne eDNA metabarcoding to characterize terrestrial vertebrate communities, opening new opportunities for biodiversity monitoring in remote and highly diverse ecosystems such as in the tropics. However, understanding factors influencing species detection is essential to fully harness the potential of this new approach. Here, we used occupancy modeling on airborne eDNA metabarcoding surveys of tropical bat roosts in Belize. We sampled five roosts over a 2-week period, three days and three nights per roost, with four to six air filters per roost. Samples were extracted and PCR amplified in triplicates with two markers for sequencing. We found a high probability of occupancy (70-90%), congruent with known presence and ecology of the species as assessed by traditional surveys including photography, netting, and trapping. However, for most species the availability probability (conditional probability that the target DNA occurs in a sample, given that the species is present at the site level) was lower than the detection probability (the conditional probability that the DNA is detected in a PCR replicate, given that the DNA is present at the sample level), suggesting that i) species likely to occupy a site may not always be detected in samples collected, leading to potential underestimation of occupancy and distribution; and ii) the probability of detecting species at the PCR replicate level can still be high despite low availability of eDNA in the samples. Altogether these results emphasize the importance of deploying multiple samplers per roost. Three PCR replicates were enough to detect most common species, and detection probability increased with the number of reads, reaching 100% at low sequencing depth (≤ 1,000 reads) for common species when using the 16S marker. Finally, day vs night sampling influenced the availability probability for all species except Desmodus rotundus, indicating a localized and temporally dynamic airborne eDNA signal within roosts which can provide evidence of behaviour and suggests that optimal sampling time may need to be adjusted depending on the target species. Our results highlight key methodological aspects which maximize the detection of bat species using airborne eDNA metabarcoding and show that it can characterize bat communities at a fine temporal scale, providing valuable insights for future conservation planning and management.