Spatially explicit power analysis reveals challenges for a long-term threatened species monitoring program in Australia

Menon, Vishnu 1 ; Southwell, Darren2; Robley, Alan3; Rees, Matthew4; Wilkinson, David1; Giljohann, Katherine4; Pascoe, Jack1; Wintle, Brendan1; Hradsky, Bronwyn1

Research facility: The University of Melbourne

Published Mar 16, 2026 on Dryad. https://doi.org/10.5061/dryad.80gb5mm4h

Data files

Mar 16, 2026 version files 13.63 MB

code-Dryad.zip

19.50 KB
data.clean.zip

13.60 MB
README.md

9.30 KB

Abstract

Long-term monitoring programs are crucial to assess trends in biodiversity, and so make informed decisions for conservation and resource management. However, disregarding the statistical power of a monitoring program can lead to incorrect conclusions about species population trends, potentially resulting in ineffective management and misdirected resource allocation. In Australia, predation by introduced red foxes (Vulpes vulpes) and feral cats (Felis catus) remains a major cause of native faunal decline and extinction. Australia spends more than $16 million yearly in controlling foxes for biodiversity conservation, primarily through landscape-scale poison baiting. Using a long-term fox baiting and threatened species monitoring program in south-eastern Australia, we collated data from 2,132 camera-trap deployments to: (1) explore drivers of the distribution of threatened native mammals and introduced predators, (2) conduct a spatially explicit power analysis to assess the program's ability to detect trends in native and introduced species occupancy for the next 10 years, and (3) provide recommendations for improving monitoring efforts through alternative scenarios. We found that threatened native mammals were more likely to occupy areas with high densities of fox baits, whereas foxes were less likely to occupy these areas; however, these areas were quite localised within baited regions. The power of the existing monitoring design was sensitive to the magnitude of change in occupancy, but robust to approximately 15% changes in the number of survey sites. The monitoring program showed adequate power (> 0.8) to detect its original aims: increases in threatened native mammal occupancy and decreases in fox occupancy in baited areas. Hence, the lack of a strong signal of increasing native mammal occupancy in the last eight years likely indicates that the system has reached a stable state under current management, rather than poor statistical power. This may potentially be the case in many long-term predator management programs. If removing some sites from an existing monitoring design does not considerably vary power, managers could consider diverting these resources to, for example, improving understanding of species-habitat relationships or intensifying predator management efforts.

Dataset DOI: 10.5061/dryad.80gb5mm4h

Description of the data and file structure

We use a long-term threatened species monitoring and predator management program conducted over 18 years in south-east Australia as a case-study to: (1) explore the drivers affecting the distribution of threatened native mammals and introduced predators, (2) conduct a spatially explicit power analysis to evaluate the statistical power of the monitoring program to detect changes in species occupancy over time, and (3) develop recommendations for improving monitoring effort by assessing alternate realistic monitoring scenarios. The study was conducted in Glenelg in SE Australia where Glenelg Ark Program was set up in 2005 to facilitate the recovery of native mammals from predation by foxes.

We collated data from multiple camera-trap studies in the Glenelg region: 240 sites that are monitored yearly as part of Glenelg Ark (total of 1624 deployments, data sourced from 2013-2019), 425 sites that were surveyed once (Rees et al. 2024) and 83 sites surveyed once (Menon et al. 2025). This totalled 2,132 camera-trap deployments between 2013 to 2020.

Files and variables

File: code-Dryad.zip

Description: The code includes (in order):

(1) Create occupancy-detection models using R package 'ubms'. Each site-year combination is considered as a separate site. Here we input :
- weekly species detection histories for two invasive predators (stacked_weekly_PA_glenelg_combined_fox.csv, stacked_weekly_PA_glenelg_combined_cat.csv) and two threatened native mammals (stacked_weekly_PA_glenelg_combined_sbb.csv, stacked_weekly_PA_glenelg_combined_lnp.csv). This includes presence-absence (0/1) data for 2,132 camera deployments between 2013-2020; and
- Values for scaled environmental covariates used in the occupancy-detection models for each site-year (env_cov_allsites_glenelg_scaled_updated.csv).
(2) Using the occupancy-detection models created for each species in step 1, create occupancy and detection raster for each species by predicting it onto a scaled environmental raster stack which includes all environmental covariates (input raster: env_stack_GA_scaled_final.tif). These occupancy and detection rasters are required to run power analysis in the following steps. Occupancy raster for each species is presented in Figure 3 in the manuscript, detection raster is provided as a figure in the supplementary materials.

This code also includes how to extract model coefficients for each species from the occupancy-detection model which are presented as results in Figure 4 in the manuscript.
(3) The aim of the power analysis is to evaluate the statistical power of the Glenelg Ark monitoring program to detecting changes in threatened native mammals and introduced predators for the next 10 years. The four monitoring scenarios are described in the section 'Monitoring Scenarios' in Methods. The scenarios primarily looks at trends for native mammals or invasive predators (see manuscript). This step creates a native mammal occupancy and detection raster by combining the rasters created for southern brown bandicoot and long-nosed potoroo in Step 2. Same for invasive predators.
(4) We fitted an additional model with only survey year and treatment (categorical baited and unbaited) affecting occupancy (and site as random effect) and constant detectability to test whether they matched the observed naive occupancy trends from Glenelg Ark monitoring. Inputs are the same as in step 1. Results for this are provided in Appendix 1: Figure S3.
(5) Run power analysis for different scenarios outlined in Table 1 in the manuscript. The species (e.g. native mammals, invasive predators), trend (increasing, decreasing), number of monitoring sites in each scenario (e.g. 200, 240, 280) can be specified at the start of the script. Running power analysis which incorporates Bayesian occupancy-detection models created in 'ubms' is very computationally heavy. We ran this for different scenarios parallelly in University of Melbourne's high processing computing system (HPC) called 'Spartan'. All the necessary input files are provided in 'data.clean/input_poweranalysis'.

We ran 1000 simulations for each scenario by conducting 5 repetitions in 200 cores (number of simulations = repetitions * cores) using HPC. Power was then averaged across the simulations.
(6) Provides a script for extracting outputs from power analysis conducted in step 5 for one of the four scenarios if you are to run power analysis in HPC with parallel computing. The code provided is for the scenario 1 (see Table 1 in the manuscript). Do the same for each scenario presented in Table 1 in the manuscript.

File: data.clean.zip

Description: The input dataset provided includes:

Sub-folder 'stacked_PA' which provides weekly species detection histories for two invasive predators (stacked_weekly_PA_glenelg_combined_fox.csv, stacked_weekly_PA_glenelg_combined_cat.csv) and two threatened native mammals (stacked_weekly_PA_glenelg_combined_sbb.csv, stacked_weekly_PA_glenelg_combined_lnp.csv). This includes presence-absence (0/1) data for 2,132 camera deployments between 2013-2020. This is used in Step 1 and Step 4 in the code.
Values for scaled environmental covariates used in the occupancy-detection models for each site-year (env_cov_allsites_glenelg_scaled_updated.csv). This data frame is used in Step 1 and 4 in the code. The csv includes scaled covariate value for each site:
- station_year: sites surveyed across multiple years are considered as separate using a site x year combination. This is included as a random effect in the occupancy-detection models
- longitude: coordinates of site longitude
- latitude: coordinates of site latitude
- data source: the data used in the study is from three different studies conducted in the Glenelg Region. The values in data_source includes matt (Rees et al. 2024; doi: 10.1007/s10530-023-03200-6), vishnu (Menon et al. 2025; doi:10.1016/j.biocon.2024.110847) and glenelg_ark (doi:10.13140/RG.2.2.17033.83047)
- bait.station.density: fox bait density (baits per sq. km in a 2.3 km radius as that is a fox's home-range radius),
- dist.nonnative.veg: distance to non-native vegetation (metres),
- dist.roads: distance to roads (metres),
- dist.water: distance to water (metres),
- elevation: elevation (metres),
- twi: total wetness index,
- ndvi: normalised difference vegetation index,
- tri: Terrain Ruggedness Index
- tsf: time since fire (years),
- evc: Ecological Vegetation Classes (categorical variable, values 1 to 8 refer to different vegetation classes which can found be found in env_stacked_GA_scaled._final.tif)
- lure: lure type used in front of the camera (categorical variable: peanut butter or tuna oil).
- treatment: treatment or non-treatment (categorical variable) refers to whether a site was within the fox-baited or control area.
  
  Summary of all the environmental covariates used are presented in Appendix S1: Section S1.
Raster env_stack_GA_scaled_final.tif which is used in Step 3 in the code where you predict the occupany-detection model to the environmental raster to create respective occupancy and detection rasters.
Sub-folder 'input_poweranalysis' which has the input required for power analysis in Step 5 in the code.
- Occupancy and detection raster for each scenario outlined in Table 1. For example, if a scenario looks at statistical power to detect increases in native mammal occupancy, then the respective input raster files would be native_occ_stack_weekly_clean.tif and native_det_stack_weekly_clean.tif.
- Raster GA_treatment_raster_clean.tif which outlines the fox-baited and unbaited area in the study landscape.
- sites_monitoring_glenelg.csv which provides coordinates for the 240 sites in the ongoing Glenelg Ark monitoring program.
- Power analysis in our study allows up to four different detection methods to be incorporated. As all surveys were conducted using camera-traps, we only use one detection method in our analysis in Step 5 in the code, and these are the input files highlighting the number of detection methods used: Species_list_native.csv, Species_list_predator.csv and Species_list_cat.csv for the respective scenarios.

Code/software

All analyses were conducted in R. One section of the analysis required higher computing for which we used a high processing computer 'Spartan' from University of Melbourne.

Access information

Other publicly accessible locations of the data:

Data was derived from the following sources:

Managing Multiple Threats: Evaluating the Efficacy of Broad-Scale Introduced Predator Management in Improving Native Mammal Resilience to Fire. Biological Conservation, 301:110847. doi:10.1016/j.biocon.2024.110847
Fox Control and Fire Influences the Occurrence of Invasive Predators and Threatened Native Prey. Biological Invasions 26(3):685–703. doi:10.1007/s10530-023-03200-6.