Benefits of modelling abundance for rare species conservation: a case study with multiple birds across one million hectares

Aim: Many management programs that are based on the needs of rare or threatened species are ineffective because they fail to collect enough data to reliably estimate abundance and map distributions for their target species. Information that does exist for rare species is often based on presence-only data, because it is difficult to collect sufficient data on abundance for such species. We targeted ten rare bird species that were excluded from a recent study due to insufficient data. For these species, we aimed to (a) collect sufficient abundance data, (b) identify important locations and (c) estimate population sizes.

Location: A large reserve system (~1M-ha) in south-eastern Australia.

Methods: We undertook intensive field surveys, using repeat area searches of 660 independent 25-ha sites, totalling 2,640 hrs of surveys (2-hr surveys; two surveys per site). We used N-mixture models to estimate abundance whilst accounting for imperfect detection.

Results: This survey effort returned enough high-quality data on nine rare bird species to identify important locations and estimate their population sizes. To illustrate potential applications of mapped important locations, we used our results to assess the likely impact of a planned burn program in part of the study region. We identified planned burns that are likely to have a significant impact on important locations for rare species that may not have otherwise been identified.

Populations were generally larger than previously estimated using expert opinion. For example, our population estimate for the threatened Red-lored Whistler (Pachycephala rufogularis) was ~16 times larger than the previous estimate.

Main Conclusions: Our results show (a) the benefits of using abundance to identify important locations for rare species (b) the value of developing bespoke survey methods for estimating abundance of rare species with low detectability and (c) a pathway for the application of mapped important locations in conservation land management.

Methods
This DOI comprises all the data required to run the N-Mixture models for each of the target species. It is also formatted for this purpose, with separate files for Bird observations, site covariates (e.g. fire-age, elevation) and observation covariates (e.g. wind, time of day). Details of data collection method are below: 

Site selection 

We randomly sampled 660 sites (25 ha each; 410 x 610 m), stratified according to fire-age (years since fire) and fire type (planned burn or wildfire). Through stratification we attempted to balance the dataset by maximising the number of sites in uncommon fire-age classes and in planned burns, which were scarce compared to wildfires. All sites were separated by > 1 km. Sites were arranged in ‘sets’ of three so that a single surveyor could complete one set per day. At least one site per set was within 1 km of the nearest track.

Survey method

From May to October 2022, we conducted 1,346 surveys (2,692 hrs covering 33,650 ha). To achieve this survey effort, we conducted 10 trips (9 days each, 4-10 people per trip). In total there were 57 surveyors, 54 of whom were volunteers. To account for inter-observer variability, we rated the experience level of all surveyors and incorporated this into analyses.

Each surveyor surveyed one set of three sites per day. Surveys started 40 minutes before dawn (± 15 mins), resulting in three distinct survey periods labelled ‘dawn’, ‘mid-morning’ and ‘noon’. Sites were surveyed 2.04 times on average (once: 45 sites; twice: 545 sites; three times: 70 sites). Repeat surveys of the same site were conducted on consecutive days. The order in which sites were visited was reversed for each repeat survey day.

Surveys consisted of a single person conducting a 2 hr area search, walking 1,500-2,500 m per survey and recording counts (abundance) of each target species. To manage surveying such large sites, surveys were broken up into six consecutive 20 minute survey bouts, covering adjacent ~4 ha cells (~200 x 200 m). The data from the six survey bouts was combined at the end of each survey to form the site survey data. Each site was also given a 5 m buffer on all sides, so that birds recorded exactly on the boundary were counted as occurring within the site, so for analysis, sites were 410 x 610 m i.e., ~25 ha. Survey bouts began when the surveyor entered the cell. The surveyor then conducted a 7 minute meandering area search on the way to the centre point (~200 m). At all times the surveyor was free to wander throughout the cell to search for birds and confirm bird species identities and abundances. At the centre-point of each cell, the surveyor conducted playback for eight of the ten target species (20 seconds per species, 20 second gap between species). The Crested Bellbird and Black-eared Miner were excluded from playback because it is ineffective for the Crested Bellbird, and playback of the Black-eared Miner can negatively affect detectability of other species due to its interspecific aggression (MF Clarke pers. comm.). The playback process at the centre point of each cell took ~6 minutes. This included time required to take site photos, record wind speed, take general notes and identify any birds detected. The surveyor then conducted another 7 minute meandering area search on the way to the cell boundary where that survey bout ended (also ~200 m).

Detecting birds through their calls is the most common form of detection for the target species. After initial detection by call, surveyors attempted to sight birds to confirm numbers. If a bird was heard in the 25 ha site but in an adjacent cell (i.e., not the cell currently being surveyed), the surveyor still recorded that individual as in the 25 ha site. Surveyors took great care to avoid double-counting birds in site surveys, using information relating to bird species mobility, direction and time of previous detection and movement of the surveyor in that time. We did not use external speakers to amplify playback because we wanted to minimise the risk of ‘calling birds in’ from outside the site, leading to inflated estimates of bird abundance (Kéry & Royle 2015). We continued surveys in light rain. In moderate-heavy rain we paused surveys and waited for them to pass. In moderate-heavy and consistent rain we cancelled surveys for that day.

Covariates

We used abundance covariates (i.e. representing the environment at sites) and detection covariates (i.e. representing survey conditions). We used eight abundance covariates that were hypothesised to affect the occurrence and abundance of the target species. We only used abundance covariates that had associated spatial data because we aimed to extrapolate model predictions to estimate abundance across the entire study area. We transformed the spatial data for each abundance covariate to generate a single value for each 25 ha site (e.g. using the mean value for the site).

We used detection covariates to account for factors affecting detectability of species during surveys. Detection covariates were: Observer Skill (ordinal: Beginner, Intermediate, Expert; classified by the lead author), Time Of Day (ordinal: Dawn, Mid-Morning or Noon), Season (continuous numeric: Month of the Year: 5-10 corresponding to May-October) and Wind Speed (continuous numeric: scored on a qualitative scale but calibrated between observers during a workshop at the start of each trip: None = 0, Slight = 1, Breezy = 2, Gusty = 3, Strong Winds = 4, Gale = 5).