Movement ecology of an endangered mesopredator in a mining landscape

Background Efficient movement and energy expenditure are vital for animal survival. Human disturbance can alter animal movement due to changes in resource availability and threats. Some animals can exploit anthropogenic disturbances for more efficient movement, while others face restricted or inefficient movement due to fragmentation of high-resource habitats, and risks associated with disturbed habitats. Mining, a major anthropogenic disturbance, removes natural habitats, introduces new landscape features, and alters resource distribution in the landscape. This study investigates the effect of mining on the movement of an endangered mesopredator, the northern quoll (Dasyurus hallucatus). Using GPS collars and accelerometers, we investigate their habitat selection and energy expenditure in an active mining landscape, to determine the effects of this disturbance on northern quolls. Methods We fit northern quolls with GPS collars and accelerometers during breeding and non-breeding season at an active mine site in the Pilbara region of Western Australia. We investigated broad-scale movement by calculating the movement ranges of quolls using utilisation distributions at the 95% isopleth, and compared habitat types and environmental characteristics within observed movement ranges to the available landscape. We investigated fine-scale movement by quolls with integrated step selection functions, assessing the relative selection strength for each habitat covariate. Finally, we used piecewise structural equation modelling to analyse the influence of each habitat covariate on northern quoll energy expenditure. Results At the broad scale, northern quolls predominantly used rugged, rocky habitats, and used mining habitats in proportion to their availability. However, at the fine scale, habitat use varied between breeding and non-breeding seasons. During the breeding season, quolls notably avoided mining habitats, whereas in the non-breeding season, they frequented mining habitats equally to rocky and riparian habitats, albeit at a higher energetic cost. Conclusion Mining impacts northern quolls by fragmenting favoured rocky habitats, increasing energy expenditure, and potentially impacting breeding dispersal. While mining habitats might offer limited resource opportunities in the non-breeding season, conservation efforts during active mining, including the creation of movement corridors and progressive habitat restoration would likely be useful. However, prioritising the preservation of natural rocky and riparian habitats in mining landscapes is vital for northern quoll conservation. Supplementary Information The online version contains supplementary material available at 10.1186/s40462-023-00439-5.

Appendix 1: Methods used to create environmental covariate layers for habitat selection analyses.
To create the map of habitat classifications, we applied semi-supervised classification of a normalised difference vegetation index (NDVI) layer [1][2][3], derived from Sentinel-2 imagery [4].The NDVI layer for the habitat map was captured at a scale of 10 m and we used a consistent layer (captured in October 2021) for both tracking periods.Woodie Woodie had no fires between tracking periods and habitat features remained relatively consistent between time periods.NDVI is calculated using the near infrared (NIR) and red (RED) bands: For Sentinel 2 data the NIR band is band 8 and the RED band is band 4.
We used the semi-automatic classification (SCP) plugin [5], in QGIS to classify habitat classifications from NDVI imagery.Classified habitats included spinifex grassland, riparian habitat (which was mostly associated with heavily vegetated creek lines), disturbed land, and water.To define all disturbed land accurately, we then overlaid a disturbance raster layer (converted from a vector) provided by ConsMin which reflected all mining disturbance in the landscape and was used for reporting at Environmental Protection Authority (EPA) standards.
Finally, because rocky habitat is crucial for northern quolls in the Pilbara [6][7][8], we overlaid a northern quoll potential natural denning habitat raster layer (converted from a vector) which was digitised manually by Western Wildlife by outlining all visible rocky features from highresolution aerial imagery.These features included rocky outcrops, gorges, and rocky mesas, as these are the areas which provide denning habitat for northern quolls in the Pilbara [9,10].
This resulted in a habitat raster of the following habitat types: spinifex grassland, riparian habitat (dense vegetation associated with creek lines), water, rocky habitat, and mining disturbed land.During breeding season, quolls were often tracked to dens within rocky waste dumps and mine pits, therefore, we split mining disturbed land into two types: 1) mine pits and waste dumps, and 2) other disturbed land (e.g., roads, buildings, and large cleared areas).
The final map was cross-examined with the corresponding satellite imagery to ensure the accuracy of habitat feature classification [11].
The topographic ruggedness index (TRI) is defined as the difference in elevation between a cell and the eight cells surrounding it [12].To create a TRI map for our landscape, we sourced a high-resolution radiometric terrain-corrected digital elevation model (12.5 m scale) [13], and used the 'Terrain Ruggedness Index' function in QGIS to calculate TRI for each cell [14].
To create maps representing distance to disturbance and distance to potential denning habitat, we used the same disturbance and habitat vector layers provided by ConsMin that we used to create the habitat classification map.We converted these vector layers to distance rasters using the "rasterize" function in the 'raster' package in R. Each resulting cell of the respective rasters (10 m scale) reflected the distance from disturbed land or potential natural denning habitat.
To determine differences in environmental characteristics among habitat classifications, we extracted the mean NDVI value and the median topographic ruggedness for each habitat.To define the area from which to sample from, we combined the observed and available movement ranges (values were compared separately for observed and available ranges, but were very similar so they were combined for simplicity), using QGIS [14].We clipped all habitat types to the combine observed and available area and extracted the NDVI and topographic ruggedness values using the 'extract' function in the "raster" package in R [15].

Table S1:
The parameters used during kernel density estimation for the measurement of northern quoll movement ranges.Parameters show the buffer size, grid output, and the UTM zone for each individual (ID) when using the ad hoc method (had hoc), referred to as 'reference scaled' in the package "rhr" [16].

Figure S1 :
Figure S1: Northern quoll movement ranges during breeding season used in analysis.Individual ID is in the top left of each plot and each grid square is 5 km wide.Latitude is on the Y axis and Longitude is on the X axis.

Figure S2 :
Figure S2: Northern quoll movement ranges during non-breeding season used in analysis.Individual ID is in the top left of each plot and each grid square is 2 km wide.Latitude is on the Y axis and Longitude is on the X axis.

Figure S3 :
Figure S3: Northern quoll steps during breeding season used in analysis after data cleaning and removal of bursts with less than three steps.Individual ID is in the top left of each plot and each grid square is 5 km wide.Red lines signify steps.

Figure S4 :
Figure S4: Northern quoll steps during non-breeding season used in analysis after data cleaning and removal of bursts with less than three steps.Individual ID is in the top left of each plot and each grid square is 2 km wide.Blue lines signify steps.

Figure S5 :
Figure S5: Relevant range coefficients for step length and mean VeDBA related to the influence of temperature and season (specifically, breeding season).Red dashed arrows represent a negative relationship and blue solid arrows represent a positive relationship.Arrow width shows the size of the effect, with wider arrows representing a larger effect.An asterisk signifies that the relationship is significant (p = <0.05)and the conditional R 2 value for step length and mean VeDBA is listed for each model, outlining the variance explained by the predictor variables.Te represents the total effect coefficient of season on mean VeDBA, both directly and mediated through step length.Icons adapted from Microsoft PowerPoint.

Table S2 :
iSSF models compared to determine the habitat selection of northern quolls in a mining landscape.Models compared include variables: habitat = habitat classification (spinifex sandplain, dense vegetation, rocky habitat, mine pits and waste dumps, other disturbed land), sl = step length, ta = turn angle, TRI = topographic ruggedness, HD = distance from potential denning habitat, DD = distance from disturbance, and age = age.The global model for both seasons is highlighted in bold.

Table S3 :
Outputs of Bayesian zero-inflated regression models for broad scale use of each habitat type, and linear mixed-effects models for broad movement ranges and their median topographic ruggedness, mean distance from disturbance, and mean distance from potential denning habitat.Significant relationships are highlighted in bold and denote differences in used movement ranges from the intercept.The intercept is the available movement range.

Table S4 :
The outputs of iSSFs for models which had substantial support from AICc ranking (TableS3).Models are separated by their model number for each season.Response variables which were significantly different from the intercept (p = <0.05)are highlighted in bold.The intercept for categorical habitat classifications was rocky habitat.

Table S5 :
The mean, standard deviation (SD), minimum (Min), and maximum (Max) Normalised Difference Vegetation Index (NDVI) values for each habitat classification in the observed and available landscapes.As well as the median, interquartile range (IQR), minimum, and maximum Topographic Ruggedness Index (TRI) in the observed and available landscapes.