Below, we summarise our modelling procedure. For more details, see our report.
We modelled habitat suitability for threatened
(for more information see the
Saving Our Species Program).
Landscape-managed species are plants and animals that need broad landscape scale conservation projects. The objective for these species
is to maximise the viability of species and their habitat by strategically investing in priority locations and management actions and
working in partnership with stakeholders across NSW (OEH 2016). Some land-managed species might be widely distributed, and possess highly
mobile or dispersed, or be affected by landscape-scale threats. Thus, recovery for these species should address threats such as habitat
loss or degradation within a landscape. There are 98 landscape-managed species for which 81 had sufficient data for habitat suitability models
to be developed.
Site-managed species are threatened plants and animals that can be secured by conservation projects at specific locations within NSW.
For these species the objective is to maintain a 95% probability of having a viable population in the wild in 100 years from now, and ensure
that the species' status under the TSC Act does not decline (OEH 2016). Different conservation actions can be implemented for those species,
including weeding, controlling erosion or revegetation, and monitoring the results, among others. These actions allow the long term protection
of these species securing their survival. There are approximately 440 site-managed species, and we developed models for 238 species
(34 vertebrates and 204 plants).
Occurrence records for the species included in this study were obtained from (a) OEH Atlas; (b) Victoria's Biodiversity Atlas; and (c) the Australias
Virtual Herbarium (AVH) hub of the Atlas of Living Australia (ALA, www.ala.org.au).
Note that we did not fit models for species with less than 20 unique location records after data cleaning (i.e. where we defined unique as 1 km x 1 km grid cells).
While Global Climate Models (GCMs) project mean annual temperature to increase as the century progresses,
there remains uncertainty in the magnitude of this change. Similarly, across alternate GCMs there can be
considerable difference with respect to the magnitude and direction of precipitation changes. This means
that multiple climate futures need to be considered when trying to understand future impacts and develop
We used data derived from climate simulations performed as part of the
NSW and ACT Regional Climate
Modelling (NARCliM) project1. These data comprise climate surfaces projected by four GCMs (Table 1).
Importantly, these scenarios have undergone a rigorous selection process1. The models project futures
that we refer to as "Warmer/wetter", "Hotter/little change" in precipitation, "Hotter/wetter", and
"Warmer/drier", given the future state with respect to mean annual temperature and annual precipitation
of the baseline period (1990-2009). It is important to recognise that, at present, these futures are
Table 1. Climate futures used in this study. GCMs assumed the SRES A2 emissions scenario2.
|Climate Future||GCM||Represents a future that is:|
||Warmer and wetter than present, particularly in NE NSW, although alpine regions are projected to become drier.
||Has the greatest increase in temperature, of the four scenarios. Precipitation trend varies across the state (slightly wetter in the NE and coastal regions, slightly drier elsewhere).
|Warmer than MIROC, and wetter across most of the state, although areas in NW and SE of the state may be slightly drier.
||Warmer than present, and the driest of the four models.
Projections from each climate scenario were downscaled, then summarised to a standard set of 19
bioclimatic variables commonly used in species distribution models (SDMs). These data were generated
for each of the NARCliM time periods, representing baseline climate (1990-2009), near-future (2020-2039)
and distant future (2060-2079). We then interpolated these data for intervening decades, resulting in
climate data for 2000, 2010, 2020, 2030, 2040, 2050, 2060, and 2070. Finally, data were transformed
to a 1 x 1 km resolution.
From the 19 bioclimatic variables, we selected a subset for model calibration: (1) mean diurnal temperature
range; (2) temperature seasonality (the coefficient of variation of weekly mean temperature); (3) maximum
temperature of the warmest week; (4) minimum temperature of the coldest week; (5) precipitation of the
wettest week; (6) precipitation of the driest week; and (7) precipitation seasonality (the coefficient
of variation of weekly total precipitation). These represent climatic variables that influence
ecophysiological functions, and hence, species' distributions.
Static environmental data
To supplement bioclimate predictors, we also used data describing topsoil attributes, weathering intensity
and topographic complexity. These data were originally at a spatial resolution of ~100m, and were
resampled to 1 km.
Soil: These layers were based on spectral characteristics of soil samples from across the continent.
Soil1 describes the distribution of highly weathered soils, soil2 the distribution of
soils with large amounts of organic matter, and soil3 the distribution of low relief landscapes with
soils containing abundant smectite (clay) minerals3.
Weathering intensity: This layer characterises the regolith, which has a major influence on geomorphic
and hydrologic processes4.
Topographic complexity: We used two layers characterising topographic complexity. The Topographic Wetness
Index5 (TWI) estimates the relative wetness within a catchment, while the Topographic
Position Index6 (TPI) categorises grid cells as belonging to the upper, middle and lower
parts of the landscape.
Habitat suitability models
We modelled habitat suitability with Maxent version 3.3.3k7,8, a machine learning approach
to habitat suitability modelling known for its high performance9. Maxent produces a grid,
where the value of each grid cell may range between 0 to 1. Values can be interpreted as a relative
index of habitat suitability with respect to the included predictors. Locations with higher values are
deemed to have greater suitability for the modelled species7,10. More details on the modelling
procedure can be found in our report to OEH.
For each species, we calibrated models using three sets of the predictor variables described above.
Of the three models run per species, the model resulting in the highest predictive power was selected
to project onto future climate scenarios. When generating projections of future habitat suitability,
soil predictors were assumed to remain static.
Current and future habitat suitability
Using Maxent, habitat suitability for each species was estimated for the 'current' period (based on 2000),
as well as for the future climates for each decade from 2010 to 2070. These maps can be viewed on our website,
by searching for a given species, selecting one of the four climate scenarios from the layers icon at the top right,
and playing the animation across the time periods.
Maps with cell values ranging from
0 (unsuitable) to 1 (highly suitable) were then converted to binary layers indicating suitable/unsuitable
habitat. Thresholded maps can be viewed in the website by selecting the Threshold option.
For each decade after 2000, the habitat suitability surfaces for each species were summarised to a single layer indicating the number
of scenarios in which a given grid cell was classified as suitable. On the website, these maps can be visualised using the Consensus
button under the climate scenarios.
The website can also be used to identify populations that may be at more or less risk from climate change. Suitable regions that currently
contain populations, and are projected to remain suitable in the future, can be classified as internal refugia. Conversely, regions from
which populations are currently absent but that become or remain suitable in the future can be classified as external refugia. Populations
likely to be least at risk from climate change are those with suitable habitat under all of the scenarios for a given time period (and
can be identified by selecting the Consensus option on the website). In our reports, we refer to these regions as areas of consensus.
Identifying multi-species refugia for threatened species
Localities that are likely to continue to have suitable climate for multiple threatened species represent areas that are particularly
valuable for conservation. To identify these higher value refugia, we combined maps of each species' internal refugia for each climate
scenario, thereby calculating the number of species for which each grid cell was suitable. In addition, we repeated this process for maps
of areas of consensus, calculating the number of species for which each grid cell was suitable under all climate scenarios. These results
can be viewed for landscape- and site-managed species separately.
1. Evans, J. P. et al. Design of a regional climate modelling projection ensemble experiment--NARCliM. Geoscientific Model Development 7, 621-629 (2014).
2. Nakicenovic, N. et al. Special report on emissions scenarios (SRES), a special report of Working Group III of the Intergovernmental Panel on Climate Change. (Cambridge University Press, 2000).
3. Viscarra Rossel, R. & Chen, C. Digitally mapping the information content of visible-near infrared spectra of surficial Australian soils. Remote Sensing of Environment 115, 1443-1455 (2011).
4. Wilford, J. A weathering intensity index for the Australian continent using airborne gamma-ray spectrometry and digital terrain analysis. Geoderma 183-184, 124-142 (2012/8).
5. CSIRO. Topographic Wetness Index (3" resolution) derived from 1 second DEM-H version 1.0. (2012).
6. CSIRO. Topographic position index (3" resolution) derived from 1 second DEM-S version 0.1. (2012).
7. Phillips, S. J., Anderson, R. P. & Schapire, R. E. Maximum entropy modeling of species geographic distributions. Ecological Modelling 190, 231-259 (2006).
8. Elith, J. & Leathwick, J. R. Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution and Systematics 40, 677-697 (2009).
9. Elith, J. et al. Novel methods improve prediction of species' distributions from occurrence data. Ecography 29, 129-151 (2006).
10. Phillips, S. J. & Dudík, M. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31, 161-175 (2008).