Effective Methods for Robust Population Estimates of a Nocturnal Predator: The Chuditch (Dasyurus geoffroii) Across the South-West, Western Australia

Melissa C Taylor

Population monitoring for cryptic species can be challenging. Low detection rates result in unreliable estimates of population size, hampering effective species management. Population size estimates for the cryptic and threatened chuditch (Dasyurus geoffroii) have been imprecise and highly variable, resulting in an uncertain conservation status for this species. Improved reliability (precision) in population size estimates was identified as a key action in the chuditch recovery plan. To achieve a targeted approach required: testing and optimising a fit-for-purpose, species-specific survey method to improve population size estimates and thereby species management outcomes. To address the recovery plan’s key action, I tested camera traps with spatially-explicit capture-recapture (SECR) statistics as these are expected to be more effective with cryptic species and produce more reliable density estimates compared to the cage trapping and non-spatial capture-recapture statistics that have been used to date. SECR models were run in a maximum likelihood estimation framework within Rstudio because this methodology calls for less modelling expertise and is therefore more accessible for use by wildlife researchers and managers. Development of methodology for SECR population estimates using cameras was approached in four stages, each stage informing the next. Firstly, model choice and setup at each camera site were optimised to maximise detection rates: an essential determinant of reliable mark-recapture estimates. Secondly, the spatial deployment of cameras across a site was optimised using simulations, SECR modelling and field testing. Thirdly, the method giving the most precise estimates was field-tested across multiple sites and, finally, SECR estimates were compared to spatial mark-resight estimates (SMR) to explore opportunities for including unidentified individuals in the dataset. Camera trap model and setup were initially investigated to determine which model and setup combination resulted in the highest Detection probability (i.e., was a chuditch detected by a camera given it was there), Proportional identification probability (i.e., probability of an individual chuditch being identified given detection) and Identification probability (i.e., probability of an individual chuditch being identified across all known detection events). Two camera models (Reconyx HC600/PC900 and Swift 3C) in four setups (single cameras at multiple heights and paired cameras at ground level) were tested in the presence of a fish oil lure in Julimar State Forest, south-western Australia. The three probabilities were compared using general linear mixed models (GLMM). Paired cameras at 30 cm performed better than the other setups for all three probabilities. Paired Swift 3Cs had a significantly higher Detection probability (back-transformed GLMM coefficient: 6.46 times higher probability than the baseline of a single Reconyx at 30 cm), while paired Reconyx had a higher Proportional identification probability, though this was not significant (1.29 times higher probability than single Reconyx at 30 cm). Identification probability was significantly higher when paired cameras at a height of 30 cm were used, regardless of model (Paired Swift 3.41, and paired Reconyx 2.41 times higher probability than single Reconyx at 30 cm), with models not significantly different from each other. Paired Swift 3C cameras with a fish oil lure were then used to optimise inter-trap spacing and trial two spatial layouts to determine the one that would result in the most reliable density estimates. After using existing data to get SECR parameter estimates and a conservative spacing of 500 m, I then ran an initial field trial using a grid. The data from this pilot field trial were used to determine location-specific parameters and an optimal inter-trap spacing using SECR simulations. These gave an optimal inter-trap spacing of approximately 1 km (relative standard error [RSE] = 13 %, relative bias = 0.05 and relative-mean-square-error = 0.13) which was then used to run two consecutive layout trials (grid and road array). These two layouts provided similar density estimates (1.29/km2 and 1.28/km2 respectively), though the grid was more precise in its estimate (SE; 0.25 and 0.43, RSE; 19% and 33% respectively). There was a large decline in the number of detections that identified individuals between the grid and road array, with the cause behind this unclear. A systematic grid approach allows for a more consistent and robust approach than the road array that is highly dependent on location-specific road availability. This, coupled with the lower SE and RSE values, resulted in the selection of the grid as the recommended layout. The 1 km grid was trialled in four locations in south-western Australia to determine the broadscale suitability of the method. These were: Julimar State Forest, Dryandra National Park (main block), Batalling State Forest and Tone-Perup Nature Reserve (Moopinup Block). Locations were chosen for their varying habitat types and expected population densities. The method successfully derived density estimates using SECR for all four locations (Julimar: 1.21/km2, Dryandra 0.45/km2, Batalling 0.35/km2 and Moopinup 0.69/km2); however, precision varied with location (RSE values and confidence interval coverage). Julimar was the only location that resulted in a reasonable precision (RSE < 20 %), though Batalling was close (RSE 23 %). To increase the precision of their density estimates, the other locations require further refinement of the method. Dryandra in particular had a density estimate that was the least precise given the small dataset when used in a multi-session model (n = 22, 1.47/km2). SECR parameters for all four locations were used to run SECR simulations and propose potential adjustments for different locations based on chuditch encounter rate and population density. For locations with lower encounter rates, increasing survey duration had the most marked effect on estimate precision, but in locations with higher encounter rates population density was the primary influence in estimate precision. SECR includes only identified individuals, so some deployments had large amounts of inviable data (~50 %). To address this, the data from the spacing optimisation, layout trial and broadscale test were rerun using spatial mark-resight (SMR) modelling. This modelling allows for ‘unmarked’ and ‘marked but unidentified’ detections to be included, therefore potentially increasing the reliability and precision of estimates. The SMR estimates produced from including unidentified detection data were compared to the SECR results. SMR estimates were more precise than their SECR counterparts with smaller SE and RSE values, varying confidence intervals, and higher density estimates, which were most noticeable when the proportion of detections identified was low (≤ 68 %). As chuditch are a naturally marked species where each individual could be ‘marked,’ there was no method of distinguishing individuals known in the dataset from unknown individuals and therefore all unresolved detections were classed as ‘unmarked’. This likely inflated the number of unmarked individuals in the population, overestimating density. While SECR is also likely biased, and may underestimate population size when unresolved individuals are excluded from analyses, underestimation was deemed a more acceptable outcome given current levels of identification success. I invoke this for camera studies on the threatened chuditch, based on the precautionary principle, to ensure declines are not missed or actioned on too late. Previously, most population density and abundance estimates for the chuditch have been derived from cage trap success rates using non-optimised spacing. These cage trap surveys are subject to low capture rates, and estimates are therefore highly variable. Camera traps increase encounter rates compared to cage traps, partially due to being multi-capture devices. Investigation into a fit-for-purpose method resulted in the development of a camera trap deployment consisting of minimum of 42 lured camera pairs in a grid at a spacing of 1 km with a minimum deployment of 30 days (up to 90 when encounter rates are low). This method will provide more reliable estimates of population density across the chuditch’s range, and improved population size estimates. These improvements will lead to greater confidence in detection and response to declines in a timely fashion, resulting in improved management outcomes, better assessment of reintroduction and translocation outcomes, and of management intervention. They will also allow for the conservation status of the species to be re-assessed and ensure that the appropriate level of protection can be enacted. Not only will this approach aid ongoing surveying and management of chuditch, the multi-phase optimisation approach used here can be adopted by conservation managers for a variety of species to improve survey effectiveness. The more reliable estimates derived from the method developed herein can be used by other ecological studies of the abundance and distribution of species (e.g., habitat preference, landscape use, invasive predator interactions).

Effective Methods for Robust Population Estimates of a Nocturnal Predator: The Chuditch (Dasyurus geoffroii) Across the South-West, Western Australia

Files and links (1)

Abstract

Details

UN Sustainable Development Goals (SDGs)

Metrics