21 – Determining trends in irruptive desert species^281described brief ly, followed by a modelling approach that quantifies the error
structure of the data and helps to identify drivers of the species’ dynamics. The
chapter concludes by discussing the lessons learned and provides recommendations
for how irruptive, threatened species might be monitored in future.
Monitoring program
Field methods
An important consideration in any monitoring program is to use methods that
perturb the system as little as possible. In this instance, twelve 1-ha grids of pitfall
traps were set up in 1990–91 on Ethabuka Station (now Ethabuka Reserve)
(23°4 6′S, 138°28′E) in the north-eastern Simpson Desert in western Queensland,
with three of these grids used for long-term monitoring. Between 1995 and 1997, a
further 11 sets of trapping grids were established, each with two to five 1 ha
monitoring plots, on Ethabuka and neighbouring properties. All 12 sites are
located within spinifex Triodia basedowii-dominated sand dunes, and mammals
have been trapped and released at each site every year until the present.
The original motivation for the sampling regime, the characteristics of the
sites, sampling methods and key results have been described in detail elsewhere,
with summaries in Dickman et al. (2014) and Greenville et al. (2016).
Here, capture data on two species of dasyurid marsupial – the brush-tailed
mulgara Dasycercus blythi (~100 g) and lesser hairy-footed dunnart Sminthopsis
youngsoni (~10 g), from nine of the 12 study sites – are summarised. These species
were selected for analysis because they show divergent population trajectories and
hence are instructive for present purposes. To simplify and standardise the capture
data, results are expressed as numbers of captures per 100 trap-nights (one trap-
night equals one pitfall trap open for one night) per year.
Analytical methods
Bayesian multivariate autoregressive state space (MARSS) models were used to
analyse the capture data. These models have several advantages for analysing this
kind of long-term data, in particular accounting for both process (state) and
observation variability (Ward et al. 2010). Process variability represents temporal
variability in population size due to environmental and demographic stochasticity,
whereas observation variability includes sampling error (e.g. temporal changes in
detectability or error resulting in only a sub-sample of the population being
counted) (Ward et al. 2010). The MARSS equations and detailed descriptions of the
modelling approach are provided by Greenville et al. (2016).
Several factors that might inf luence populations of the two species were
identified. For D. blythi, these were mean spinifex cover per site, annual rainfall
and total rodent population size, both in the prior year. For S. youngsoni, these