Australian coastal migratory shorebirds species data for population trend assessment (including R code)

Full description

Database uses data on shorebird counts from around Australia. The majority of the records are from Birdlife’s Birdata database. We supplemented this data with bird surveys within the Coorong, from David Paton, from data available from the South Australian Government (Paton, Paton, and Bailey 2016). Data for some shorebird areas, namely eighty mile beach, Roebuck Bay, Werribee/Avalon, did not have count area level data for a number of recent years (2019-2022), but had aggregated summary data available. Within the database, observations of the number of individuals per species are organized in “count areas”, which are generally one high tide roost, or segments of beach. Count areas are situated within “Shorebird areas”, which are the maximum areas in which individual birds are likely to move during the non-breeding season (Clemens, Herrod, and Weston 2014). The database contains >380,000 records from 448 shorebird areas around the country. Data on individual species generation times was sourced from Birdlife’s Data Zone (‘BirdLife Data Zone’ 2022). For our analysis, we aggregated the data into independent count occasions for each shorebird area within each Australian summer, here termed “season”. To achieve this, we summarized the total number of each species observed at a shorebird area for each month during the summer months (October, November, December, January, and February). First, the database was subset to records that had complete fields for “shorebird area”, “point count ID”, “count”, and “date”. Data were then aggregated to find the max observations per species per count area per month. The max observations per count area per month were then summed per shorebird area. For shorebird areas with counts across multiple months, we used the top two counts per season as input for our data analysis. Finally, we only included shorebird areas that had at least 500 birds observed over the entire time series and had at least one count for at least half of the years in the entire time series (14 years of the 29 years in the time series). Structured, regular monitoring began in 1993, so we used data from 1993-2022. Modelling abundance and population trends - The objective was to estimate abundance and population trends of the targeted species at the national level, using the time-series data described above. Following the successful example of modelling population trends of shorebirds in Australia by Studds et al (2017), we also used N-mixture models, which estimate the abundance of each species at each shorebird area each year, while accounting for imperfect detection of individuals as well as among-area difference, temporal trends, and over-dispersion in abundance. The model allowed us to estimate: (i) the abundance of each species at each shorebird area each year, (ii) the total abundance of each species across all areas each year, and (iii) the nationwide population index of each species, which shows “average” changes in the species abundance across all shorebird areas. As N-mixture models tend to be highly complex with many parameters and thus require much information (i.e., data) for those parameters to be successfully estimated, we developed two types of N-mixture models with varying levels of complexity/assumption: (i) the model assuming that detection probabilities at a given area vary among months within each year, and (ii) the model assuming that detection probabilities at a given area are constant across months within each year. We first fitted model (i) above to all targeted species using the program JAGS (Hornik et al. 2003) through the R2jags package (Su, Yajima, and Edu 2022) in R version 4.2.1 (R Core Team 2015). Model convergence was checked with R-hat values and trace plots. If the model still did not converge, we next fitted model (ii) above and increased the number of iterations until the model converged. If both models did not converge, we fitted a simpler model, which had the same structure but without accounting for the imperfect detection of individuals (i.e., assuming that all individuals are detectable). Using model outputs, we then estimated the rate of change in abundance. For a given time frame (29 years, three generations, 1993-2013, 2013-2022) we calculated growth rates using generalized least squares regression to account for temporal autocorrelation. We then sampled 1000 growth rates from each regression result and calculated the mean and standard deviation based on these samples. We then took the difference in the samples and calculated the probability of the difference being larger than zero using the code: 100*length(difference_in_samples[difference_in_samples <0])/length(difference_in_samples) We used IUCN criteria A2, to assess how the species should be listed based on the estimated declines from our analysis. These thresholds are: 80% - critically endangered 50% - endangered, and 30% - vunerable From the IUCN IUCN Red List of Threatened Species “