ABSTRACT
Landscape-scale monitoring is used to track changes in the status of fauna populations, including identifying when and where change is occurring over large areas. Yet few programs at this scale exist despite calls for long-term monitoring, especially after the ‘black summer’ fires of 2019/20. Landscape monitoring often sits in the ‘too hard and too expensive basket’. To encourage land managers to maintain existing programs or establish new ones, it is important to demonstrate approaches that can cost-effectively monitor fauna and assess the level of investment required to adequately assess trends. We used passive acoustic and ultrasonic monitoring with artificial intelligence to semi-automate the detection of vocalisations from Koalas, Yellow-bellied Gliders, Powerful Owls and Sooty Owls and echolocation calls from two threatened bat species (Eastern Falsistrelle and Little Bent-wing Bat) as part of a pilot study for forest monitoring at a landscape-scale. Analyses that accounted for imperfect detection were used to: (1) generate detection probability and initial occupancy estimates for selected priority fauna species, and (2) undertake power analyses to inform sampling effort, and duration required to detect a commonly applied trend (30% decline over 10 years). Cost-effectiveness of sampling with remote sensors and traditional methods was also examined. Several thousand detections of priority species were recorded, with the number of detections varying by species and season. This translated to differences in detection probability and occupancy for each species. Detection probability was sensitive to the sampling intensity (number of surveys/sensors) deployed at a given site and this varied between seasons, with >95% probability of detection achieved for most species when multiple subplots were sampled for a 2-week period. Power analyses revealed trade-offs between sampling effort (number of sites) and duration (number of visits) required to detect a trend. Only a modest investment was needed to monitor widespread species that are moderately detectable using remote sensors with artificial intelligence and this investment was 2-11 times lower than monitoring using traditional methods for the priority species examined. For very rare species that are not widespread, localised monitoring and/or targeted research may be more suitable. Nevertheless, landscape-scale monitoring for many forest fauna is achievable using cost-effective remote sampling, artificial intelligence and robust analytics that account for imperfect detection. Now, more than ever, we urge support for and implementation of landscape monitoring for biodiversity.
