Topographic ruggedness and rainfall mediate geographic range contraction of a threatened marsupial predator

Species range contractions are increasingly common globally. The niche reduction hypothesis posits that geographic range contractions are often patterned across space owing to heterogeneity in threat impacts and tolerance. We applied the niche reduction hypothesis to the decline of a threatened marsupial predator across northern Australia, the northern quoll (Dasyurus hallucatus).


| INTRODUC TI ON
The contraction of species' geographic ranges is the result of population decline and local extinction, and a stark, prominent manifestation of adverse anthropogenic environmental change (Hobbs, Valentine, Standish, & Jackson, 2017). Range contractions are increasingly common across the globe (Thomas, Franco, & Hill, 2006)-for instance, 40% of 177 mammal species studied by Ceballos, Ehrlich, and Dirzo (2017) have experienced range contractions of >80%.
As range contractions increase extinction risk (Purvis, Gittleman, Cowlishaw, & Mace, 2000), the extent of a species' geographic range, as well as its population size, are considered key elements by the International Union for Conservation of Nature (IUCN) in determining a species' global conservation status (IUCN, 2006). Range contractions entrain a raft of changes that occur when species are in decline, such as changes to species' niches (Breiner, Guisan, Nobis, & Bergamini, 2017). Small changes in a species' geographic range size can correspond to substantial reductions in its niche breadth, and vice versa, due to local extinctions occurring in distinct environments (Breiner et al., 2017), leading to spatial patterning in range declines (McDonald, Luck, Dickman, Ward, & Crowther, 2015).
To integrate the concepts of range decline and niche reduction, Scheele, Foster, Banks, and Lindenmayer (2017) introduce the 'niche reduction hypothesis'. This hypothesis distinguishes between a species' 'historical niche'-defined as the realized niche of a species prior to decline-and the 'contemporary niche', a subset of the historical niche after a reduction in niche breadth due to novel threats (e.g. introduced species, habitat loss, disease). Three factors that shape the contemporary niche of species are (a) threat occurrence, which, when a threat operates variably across different populations or parts of a species' range, may cause species to persist only in environmental refugia, (b) a species' threat tolerance, the ability to persist despite threats, which can be increased within a subset of environmental conditions due to, for example, larger populations sizes, and (c) geographic barriers that exclude threats from a part of a species range (Scheele et al., 2017). Understanding how threats shape species' niches, and geographic ranges, is critical to conservation, as it can assist in identifying ecological refuges (Reside et al., 2019). Further, basing conservation on the contemporary niche can lead to a narrow understanding of a species' potential habitat, thereby limiting the range of options available for conservation, translocation and restoration (Scheele et al., 2017).
Here, we apply Scheele et al. (2017) niche reduction hypothesis by quantifying changes in the historical and contemporary range and niche of four populations of a marsupial predator-the northern quoll (Dasyurus hallucatus). The northern quoll is the smallest of four Dasyurus species found in Australia, ranging from 300 to 1,200 g, and males within the species often exhibit a semelparous life history, surviving for only a single breeding season (Oakwood & Cockburn, 2001). The geographic range of northern quolls spans much of northern Australia (Figure 1), but is suspected to have declined considerably since European colonization of Australia (1788), and particularly in the past 50 years (Braithwaite & Griffiths, 1994).
Consequently, the northern quoll is classified as 'endangered' by the IUCN, with a 'declining' population trend (IUCN, 2016). The main threats to northern quolls are predation by introduced animals such as feral cats, poisoning by the introduced cane toads (Rhinella marina), and habitat degradation caused by altered fire regimes and grazing by introduced herbivores (IUCN, 2016).
The northern quoll offers an excellent case study of range contraction and niche reduction in a declining species for three reasons.
First, declines in northern quoll range have occurred relatively recently (i.e. from the mid to late 20th century to the present day), and so, an extensive catalogue of species records exists from within both F I G U R E 1 Northern quoll presence records across each of the four populations. Blue markers represent historic presence records, and maroon markers represent contemporary presence records (>2,000). Populations are represented by varying line patterns its historic and contemporary range. This is not the case for many of the Australian mammals that declined earlier following European colonization (Woinarski, Burbidge, & Harrison, 2015).
Second, in addition to range contractions, northern quolls are thought to have suffered a reduction in niche breadth, due to the disproportionate loss of local populations from particular environments, such as lowland savanna and the arid fringe of their range (Braithwaite & Griffiths, 1994;McKenzie, 1981;Oakwood, 2000).
Consistent with the niche reduction hypothesis, local extinctions are thought to have related to both threat occurrence and threat tolerance. For instance, topographically rugged areas (e.g. rocky outcrops) are considered to offer fixed refuges (sensu Reside et al., 2019) from predation, grazing and fire (Burnett, 1997), increasing the likelihood of persistence of northern quolls (Begg, 1981). In addition, local extinctions are hypothesized to have occurred more often in marginal habitats (i.e. low rainfall and lower population size) that predispose populations to a lower threat tolerance due to smaller population sizes (Burnett, 1997).
Third, range contractions of northern quolls likely differ across its four main populations (Queensland, Northern Territory, Kimberley and Pilbara), due to geographic barriers that have so far prevented one primary threat-the cane toad (R. marina)-from reaching parts of the northern quoll's geographic range. Cane toads were introduced to Queensland in 1935 and spread westwards across the northern third of Australia, reaching the Northern Territory during the early 1970s (Urban, Phillips, Skelly, & Shine, 2008), and the Kimberley in 2010 (Doody et al., 2018;although they have not yet colonized that region entirely), but have not yet reached the Pilbara (Appendix S3).
We quantify and compare environmental conditions within the historic and contemporary geographic range of the northern quoll across its four major populations, compare range declines with niche reductions (i.e. declines in niche volume), and compare environmental conditions within the contemporary niche to those within the historic niche. We predicted that range size and niche volume would be most reduced in Queensland and the Northern Territory, due to their long-term exposure to cane toads. Consistent with the niche reduction hypothesis, we predicted that range declines within each population would relate to environmental variables that buffer (due to reduced co-occurrence) or increase the threat tolerance of quolls. Specifically, contemporary ranges would trend towards more topographically rugged areas with higher rainfall and greater protection from fire when compared to historic ranges due to extinctions occurring more often in more open, topographically simple and marginal habitats.

| Study area
The region over which northern quolls are believed to have occurred is ~2,634,641 km 2 and stretches from south of Brisbane on Australia's east coast, along northern Australia, to the Pilbara region in Western Australia. This region includes four discrete populations of northern quolls: in Queensland, the Northern Territory (including several islands), the Kimberley (including several islands) and the Pilbara (Figure 1). We used previous estimates of northern quoll range to guide the placement of population boundaries (Braithwaite & Griffiths, 1994;Oakwood, Woinarski, & Burnett, 2016). Recent analysis suggests each of these populations function as distinct evolutionally significant units, and thus in the interests of genetic conservation, they should be considered separately (How, Spencer, & Schmitt, 2009

| Data collection
A database containing 6,391 northern quoll records was collected from national and state/territory fauna databases, museum records, past publications, trap records as well as mining and government agencies, dated between 1788 and January 2019. The majority of records were sourced from online databases the Atlas of Living Australia (ALA, 2019), NatureMap (NatureMap, 2019) and Wildnet (Wildnet, 2019) as well as a 2008 Australian government report investigating northern quoll decline in northern Australia (Woinarski et al., 2008). Duplicates were identified and removed. Records missing information or with high locational inaccuracy were also removed.
In the interest of controlling for threat absences on islands, all island records were removed from our analysis. To minimize the effect of localized survey effort on analysis, all records collected within the same year in the same 1 km × 1 km grid cell were condensed to one. (2016), the historical data set included the contemporary observations (i.e. was 'nested'), so that it included all populations that have likely been present (but may not have been detected) since European colonization. This approach was justified given the relatively sparse sampling effort in the historical period. To test the effect that nesting contemporary records within historic records had on our results, we ran a series of analyses (methods below) on both nested and un-nested data sets and compared the results. In Queensland and the Northern Territory, we found nesting had little impact on range contraction, niche volume or habitat suitability predictor importance (Appendix S2). Conversely, in the Kimberley and the Pilbara, where historical sampling was less thorough, we found range size and niche volume increased through time in un-nested data sets-trends with no supporting evidence. Based on this comparison, we elected to only use nested data in further analysis.
We chose the year 2000 as the separation point between historic and contemporary data sets as it allowed us to have a large number of presences both before and after this point, facilitating the development of separate range estimates for each population during each of the two time periods (historic and contemporary). While separating the data into more than two time periods would have been preferable, the amount of presences within finer temporal windows did not allow for the development of reliable models. Further, while we note that local extinctions have occurred post-2000 (Oakwood, 2004), these are likely minor compared to the reductions that occurred in the 20th century when cane toads and invasive predators spread across a large proportion of the quoll's range (Braithwaite & Griffiths, 1994).

| Range contractions
We measured change in geographic range using α-hulls, a standard

| Niche change
We quantified niche volume (the space defined within the bounds of n independent environmental axes) for each population in each time period. To do this, we used niche hypervolumes generated by a one-class support vector machine method (SVM), as described by Blonder et al. (2019). Hypervolumes were defined by the bounds of the five environmental variables (Table 2) and were scaled prior to analysis (following Tingley, Vallinoto, Sequeira, & Kearney, 2014).
The SVM was used as it is insensitive to outliers and generates a smooth boundary around the data, yielding binary predictions of niche volume, facilitating comparisons between data subsets. Once all n-dimensional hypervolumes had been assembled, niche volume was calculated and compared within populations to assess niche reduction. Hypervolumes were calculated in the R package 'hypervolume' (Blonder, Lamanna, Violle, & Enquist, 2014).
We created a layer of habitat quality for northern quolls across study populations using ecological niche models (ENMs).
Ecological niche models combine presence records with environmental data to predict habitat that is most likely to support populations of a given species (Guisan & Thuiller, 2005). Ecological niche models were developed using the MaxEnt algorithm (Elith et al., 2011). MaxEnt default output consists of a map with every cell assigned a log value representing relative probability of occurrence, ranging from 0 to 1. Cell size for all ENMs in this study was 1 km 2 (1 km × 1 km). To quantify habitat quality within the range of northern quolls, we incorporated five ecogeographical layers based on evidence within the literature supporting their relevance to northern quoll ecology ( Table 2). As presence-only data are often subject to selection bias, we included a bias grid within MaxEnt models, indicating sample biases across the study area (Kramer-Schadt et al., 2013). We used a 'target group' background sampling approach to generate the bias layers (Phillips et al., 2009). We defined our target group as all critical weight range (CWR) mammals (including northern quoll) within the study area (following Molloy, Davis, Dunlop, & van Etten, 2017). These species were selected as survey methods used to detect them would also detect the northern quoll (following Molloy et al., 2017)-typically TA B L E 1 Study area size and climatic variability in terms of mean coldest and warmest quarter temperature, as well as annual precipitation  (Phillips, 2011).
To quantify changes in the environmental conditions within the contemporary range of quolls, we sampled each of our five environmental predictors 10,000 times within historic and contemporary α-hulls, across all populations. Finally, to quantify changes in habitat quality over time, we sampled the MaxEnt habitat quality layer within both historic and contemporary α-hulls. We used generalized linear models (GLMs) to compare the average values of the five environmental predictors between the two time periods, using the historic period as the reference category. We considered there to be significant difference between time periods when the 95% confidence interval of the regression coefficients did not overlap zero.
Violin plots were generated to visualized these data using the 'plotly' package in r (Sievert et al., 2017).

| RE SULTS
The total historic database consisted of 2025 historic records and 1,153 contemporary records (Appendix S1). Range size, as estimated Topographical ruggedness Calculated from the difference in elevation between a cell and the eight cells surrounding it (following Riley, 1999) Studies suggest quality northern quoll habitat is often associated with rocky rugged areas (Braithwaite & Griffiths, 1994;Burnett, 1997;Hernandez-Satin, 2016;Schmitt et al., 1989) Calculated according to Riley (1999)  Increased productivity as a result of high annual precipitation may boost the capacity of northern quolls to tolerate threats (Burnett, 1997;Hohnen et al., 2016;McKenzie et al., 2007) WorldClim (2019) Precipitation seasonality Average monthly variation in rainfall  expressed as a percentage ratio of the mean monthly precipitation total and the standard deviation of the monthly precipitation Northern quoll records in Queensland associated with higher levels of rainfall seasonality (Woinarski et al., 2008) WorldClim (2019) Distance to water Distance to permanent water measured in decimal degrees Studies suggest areas proximate to permanent water are more likely to provide high quality for northern quolls (Begg, 1981;Braithwaite & Griffiths, 1994;Burnett, 1997;Molloy et al., 2017) Derived from 1:100,000 watercourse mapping provided by Geoscience Australia (2018 declines occurred in the Pilbara (0.1%). While substantial niche reduction was also observed in Queensland (49.2%), it was a proportionally smaller reduction compared to range declines (Figure 3).

| D ISCUSS I ON
We applied the niche reduction hypothesis (Scheele et al., 2017) to quantify changes in the geographic range and ecological niche of a declining marsupial predator through time, across multiple populations.
Our results confirm a contraction in geographic range and niche volume for northern quolls in Queensland, the Northern Territory and the Kimberley (Braithwaite & Griffiths, 1994;Woinarski et al., 2008).
As predicted, range declines have been most severe in Queensland and the Northern Territory, less severe in the Kimberley, and negligible in the Pilbara. These findings are consistent with localized declines in each of these regions (Braithwaite & Griffiths, 1994;Burnett, 1997;McKenzie, 1981) and at least partly reflect the incremental spread of toxic cane toads across the species' range (Urban et al., 2008). Reductions in the northern quoll's niche volume were >20% in all populations other than the Pilbara, and, consistent with the niche reduction hypothesis, we found evidence of environmental variables mediating niche reduction, potentially due to threat occurrence, species' tolerances and geographic barriers.
Niche volume-the array of environmental conditions within a species' realized niche-is a predictor of extinction risk both now and over geological time-scales (Saupe et al., 2015). Reductions in niche volume were largely proportionate to range contractions, except in Queensland, where proportional range declines exceeded proportional reductions in niche volume. This result aligns with theory suggesting species range and niche should be correlated for the most part (Slatyer, Hirst, & Sexton, 2013); for example, a recent analysis of 148 species using simulated data found niche volume and range size were mostly similar, although considerable residual variation did exist (Breiner et al., 2017). This variation was also detected in our analysis with range decline exceeding niche decline by 29% in Queensland, reaffirming that while a consistent relationship between niche volume and range size holds for the most part, one is not always fully reciprocal of the other (Colwell & Rangel, 2009 (Oakwood, 2000). While less diverse in terms of topography and vegetation than coastal habitats, this habitat accounted for the majority of Queensland northern quoll historic range, and thus in disappearing from these areas, range size was reduced to a larger extent than niche volume.
Our finding that the Pilbara population has experienced less range contraction and niche reduction compared to other regions is consistent with expert opinion and the literature-the Pilbara is generally regarded as a stronghold for the northern quoll, due largely to the (current) absence of cane toads and abundance of topographically complex rocky outcrops (Cramer et al., 2016). However, we recognize that a smaller range contraction in the Pilbara may also However, our findings are supported to some extent by genetic data from the Pilbara, which shows no signs of substantial range contraction (Spencer et al., 2013).
By comparing the average values of environmental variables that shape northern quoll distributions within the historic and contemporary niches, we assessed a series of hypotheses regarding the F I G U R E 3 Change in northern quoll range (α-hull), and niche volume (hypervolume) plotted against one another. Queensland (purple), Northern Territory (blue), Kimberley (green), Pilbara (red). The solid line shows where a 1:1 correspondence between range decline and niche decline would fall types of environments that would favour northern quoll persistence.
We found that the contemporary niche quolls is characterized by more topographically rugged areas, supporting the hypothesis that complex landscapes increase the resistance of northern quolls to threats (or because in such areas the threats are absent or of lower intensity) by acting as fixed refuges (Reside et al., 2019). We found support for this hypothesis in all populations other than the Pilbara, with the greatest increase observed in the Queensland population.
The increase in ruggedness within the Queensland contemporary range likely reflects the loss of populations that previously occupied highly flammable, topographically simple landscapes, thought to have been more exposed to the threats of livestock grazing, altered fire regimes (more frequent and intense fires) and invasive predators (Hernandez-Satin, 2016;Oakwood, 2000). Similarly, central rock rats (Zyzomys pedunculatus) have contracted from more simple to more complex rugged habitat where the impact of their primary threat, the feral cat, is reduced (McDonald, Stewart, & Dickman, 2018). Rocky habitats are also important to northern quolls because they provide highly suitable areas within which females are able to establish maternal denning sites (Oakwood, 1997)-critical features in providing developing offspring with protection from predators (Oakwood, 2000). The importance of suitable denning habitat is further compounded when considering the unusually short, semelparous life history of northern quolls, which makes population persistence particularly reliant on recruitment (Moro, Dunlop, & Williams, 2019), and thus the effectiveness of dens in providing young with protection. While there is no evidence that rugged habitats reduce the direct impacts of cane toads, it has been suggested that, by limiting the impacts of grazing, fire and invasive predators, rocky habitats may indirectly buffer against toads by increasing population sizes that can offset mortality due to toads (Burnett, 1997). And then this finding is consistent with previous work showing that northern quolls are found at higher densities and that individuals live longer, in rocky habitats (Begg, 1981;Schmitt et al., 1989). A similar mechanism might explain why northern quoll contemporary range is characterized by a decreasing distance to permanent water when compared to the historic range, as northern quolls have been found in better reproductive condition closer to creek lines-a habitat associated with free-standing water (Braithwaite & Griffiths, 1994). The authors of that study suggest quolls closer to water are likely to engage in increased rates of reproduction, presumably because of an increased availability of resources, leading to increased recruitment and overall population size (Braithwaite & Griffiths, 1994). This finding is also consistent with Reside et al. (2019), who identified water sources as important fixed refuges for threatened species.
Some of the largest changes in environmental values between the historic and contemporary niche were for annual rainfall, which was also the most important variable driving quoll distributions in three of the four regions. The contemporary niche is characterized by much higher average annual rainfall compared to the historic niche, particularly in the two regions that have experienced the greatest range declines (Queensland and the Northern Territory).
The most likely explanation for this shift is that rainfall increases primary productivity (Pianka, 2017), and food resources (Dickman, Mahon, Masters, & Gibson, 1999), leading to improved body condition, increased offspring survivorship and population size (Meserve, Gutiérrez, Yunger, Contreras, & Jaksic, 1996), thereby offsetting increased mortality due to threats. Support for this hypothesis in northern Australia is strong, with native mammals typically persisting longer in areas that receive higher annual rainfall (Fisher et al., 2014;Start, Burbidge, McDowell, & McKenzie, 2012). Northern quolls also persist more in higher rainfall areas (Radford et al., 2014;Woinarski et al., 2008;Ziembicki et al., 2013), even though threats such as grazing, predation and intense fire are present across both the arid and mesic extents of their range. Hohnen et al. (2016) suggested one reason northern quoll populations may be more stable in high rainfall areas is because genetic connectivity is greater. They supported this hypothesis by showing quolls in wetter habitats were more closely related than quolls in drier habitats (Hohnen et al., 2016). Here, it is important to note that our analysis did not account for changes in rainfall patterns over time, a factor which has the potential to drive shifts in species range (Davis & Shaw, 2001) and niche (Broennimann et al., 2007). While the inclusion of such analysis was outside the scope of this study, we strongly suggest this topic be addressed in future studies as it has been in other species of Permutation importance quoll (Fancourt et al., 2015), particularly in the context of the earth's rapidly changing climate (Urban, 2015).
We also observed a substantial increase in rainfall seasonality within the contemporary niche in Queensland. As with the previ- ous two examples, we suggest that this shift most likely reflects increased population size and output related to the timing in quoll breeding activity, where offspring dispersal (November-February; Oakwood, 2000) coincides with a boom in resource availability brought by summer monsoonal systems in northern Australia (Oakwood, 2000). By synchronizing their reproductive time line with highly seasonal rainfall patterns, northern quolls are likely able to increase offspring survivorship, allowing populations to be more likely to persist despite increased mortality due to threats (e.g. cane toads, invasive predators).
By comparing habitat quality derived from MaxEnt models within the historic and contemporary niche, we found that the contemporary geographic range of northern quolls comprises a subset of the historical niche that is of higher predicted habitat quality, highlighting the importance of high-quality habitats for providing refuge from stressors. Here, it is important to recognize that refuges, like niches (Guisan & Thuiller, 2005), are multivariate and can be defined by the bounds of more than one environmental parameter (Keppel et al., 2012). The most effective refuges should occur on the environmental continuum where resource availability and buffering from threats permit the greatest chance of persistence (Keppel et al., 2012). In Queensland and the Northern Territory, where shifts to high-quality habitat were greatest, habitat quality was best predicted by annual precipitation, distance to water and precipitation seasonality. This result conforms with research showing mammal attrition in Australia has been greatest in resource-poor arid and semi-arid landscapes (Johnson, 2006).
Topographical ruggedness was also an important predictor for habitat quality in Queensland and the Kimberley, suggesting a combination of complex and high-resource habitats may in some cases act as more effective refuge habitat than habitat that is simply high in resources.
While effective refuge habitat is clearly critical to the persistence of species in the short term, it can also be important at an evolutionary scale as refugium-a space within a species niche that can support populations over evolutionary time-scales (Keppel et al., 2012;Reside et al., 2019). In the case of the northern quoll, habitat that limits but not fully eliminates exposure to key threats such as cane toads and feral predators may facilitate the development of both behavioural and phenotypic traits F I G U R E 5 Violin plots comparing 10,000 random samples of habitat variables within the historic (blue) and contemporary (maroon) range of northern quolls: (a) annual precipitation, (b) precipitation seasonality, (c) elevation, (d) distance to water, (e) topographical ruggedness and (f) habitat quality derived from the MaxEnt model through behavioural learning or evolution that could potentially allow re-expansion into their historic niche. Previous studies have found the timeframe for such changes to occur in response to threats (including cane toads) can be surprisingly short (<70 years; Hudgens & Garcelon, 2011;Phillips & Shine, 2006). Further, recent studies have found that not only can northern quolls be trained to avoid cane toads in a single generation, but that the trained quolls offspring also avoided eating toads, suggesting quolls have the capacity to rapidly learn behaviour and distribute the information to their young (Cremona, Spencer, Shine, & Webb, 2017;Webb, Legge, Tuft, Cremona, & Austin, 2015). These results suggest investment in protecting refuges (and populations within them) over a relatively short period of time has the potential to yield important conservation outcomes.
An important caveat of our study was that we did not explicitly account for threats such as cane toads, feral predators or fire, in any of our models. While including threats as predictors would likely aid in determination of their relative influence in shaping the northern quoll's contemporary range and niche, the scale and resolution of spatial layers required for this analysis do not currently exist and creating them was beyond the scope of our study. It may be possible to explicitly test the effect of threats on quoll populations in future studies by limiting the scale of analysis to a regional level and linking changes in northern quoll occupancy with threat distributions (Hernandez-Satin, 2016). For example, by measuring declines in range and niche in relation to fire, it may be possible to elucidate the value of logs and tree hollows-important refuge features for many northern boreal and arboreal mammal species (Goldingay, 2012)-as quoll habitat.

| Management implications
Previous research suggests cane toads have immediate and lasting detrimental impacts on northern quoll populations (Shine, 2010). It is therefore likely that cane toads are a key culprit responsible for larger range and niche contractions in Queensland and the Northern Territory (Burnett, 1997;Woinarski et al., 2008), given that cane toads have only very recently arrived in Western Australia and are yet to reach the Pilbara (Pizzatto, Both, Brown, & Shine, 2017). Cane toads are expected to colonize the remainder of the Kimberley within the next 5-10 years (Doody et al., 2018) and the Pilbara by 2037-2046 (Southwell, Tingley, Bode, Nicholson, & Phillips, 2017).
In the light of these predictions, several strategies have been developed to halt or slow the spread of cane toads from the Kimberley to the Pilbara (Phillips, Shine, & Tingley, 2016;Tingley et al., 2013).
These include closing off water bodies (Tingley et al., 2013), as well as introducing toads with 'less dispersive' genes to the front of the invasion line to impede the progress of toads carrying 'highly dispersive' genes (Phillips, 2016;Phillips et al., 2016). We suggest these strategies be given full consideration.
As with the northern quoll, previous studies show topographically rugged and high rainfall habitats also provide important ref- uge for a range of other threatened species of mammal (Southgate, Paltridge, Masters, & Carthew, 2007) and as such should form areas of focus for management (Reside et al., 2019). A first step in protecting refuge habitats is recognizing where they are located, a process that has been enhanced by remote sensing technologies (Allan et al., 2018). Once a refuge is located, a management priority should be to ameliorate any existing or emerging stressors to ensure they remain functional in facilitating species persistence (Reside et al., 2019). In the case of northern quolls, these may include limiting extensive grazing and burning, as well as implementing feral predator control programmes (Hill & Ward, 2010), and potentially applying cane toad aversion techniques (O'Donnell, Webb, & Shine, 2010). Although we excluded islands from considerations in our analyses, the marked contractions in range and niche for mainland quolls reinforce the importance of island populations and the need to maintain biosecurity for these islands to reduce the likelihood of toad or predator invasion. Finally, effective management of refuge systems requires high-quality ecological data collected over biologically meaningful timeframes (Lindenmayer et al., 2012). Therefore, the implementation of sustained and effective monitoring programmes inside and outside refuge habitats is highly recommended.

ACK N OWLED G EM ENTS
We thank all who contributed to record databases that were used in this study as well as authors of Woinarski et al. (2008) for collating many of the Queensland records. We also thank Deanna Duffy, Rachel Whitsed and Simon Mcdonald for their assistance with manipulating spatial data, and Debra Noy for assistence with administration.

DATA AVA I L A B I L I T Y S TAT E M E N T
Study data are publicly available via data sharing platform Dryad.

B I OS K E TCH
Harry Moore is a PhD student interested in how predators interact with biotic and abiotic factors within their environment, and how these shape their spatial ecology and behaviour. The researchers involved in this publication have broad expertise in mammal conservation, applied ecology, field ecology and ecological modelling.