Expert Vetted Distribution Models and Biodiversity Hotspot Maps of Terrestrial and Freshwater Taxa of Conservation Concern in Northern Australia

Abstract

Abstract: This collection is comprised of raster layers of present day distributions of terrestrial and freshwater taxa that are of conservation concern in Northern Australia, which includes the Australian Conservation Management Zones �North Australian Tropical Savanna� and �North Eastern Australia Tropical Rainforests� and drainage basins intersecting these regions boundaries. Distributions were derived either from species distribution models (also referred to as �ecological niche models�, �habitat suitability models�, �habitat models� or �bioclimatic envelope models�) created with the program Maxent or from buffered occurrence records cut to suitable areas (for data deficient taxa) and were vetted by experts and modified accordingly. The additional biodiversity hotspot maps (i.e. areas with relatively high concentrations of taxa of conservation concern) included in this collection show the number of taxa within each relevant taxonomic, functional or threatened species status group (e.g. all endangered species or all vulnerable species) of organisms present across the study region. The collection also includes a supplementary spreadsheet (�NESP vetting and taxa information�) with further information on each taxon, such as relevant ecological traits, conservation listings and model details (e.g. thresholds used for Maxent raw outputs, model quality as assessed by experts, AUC/ model fit statistics, cautionary notes on any remaining issues or uncertainties). The collection was created for the purpose of informing conservation decision-making in Northern Australia and was funded by the Australian National Environmental Science Program (NESP) as part of the Northern Australia Environmental Resources Hub.Purpose: The collection was created for the project �Prioritising threatened species and threatening processes across northern Australia�, which was funded by the Australian National Environmental Science Program (NESP) as part of the Northern Australia Environmental Resources Hub. It aimed to provide relevant stakeholders with high quality, high resolution representations of the distribution of relevant taxa to aid with conservation decisions within Northern Australia. Taxa included in the models were considered to be of conservation concern because they were listed under state and federal threatened species legislation, have restricted spatial distributions, or are of other priority concern as indicated by our direct communication with environmental departments of the WA, NT and QLD governments.� The spatial layers were furthermore designed to be used in conjunction with spatiotemporal models of threats potentially affecting these taxa in northern Australia, which are outputs from the second stage of the same project. Distribution layers and threat layers can be combined to estimate spatiotemporal variation in vulnerability (sensitivity vs. exposure) of individual taxa or groups of taxa to particular threats and assist in targeted investment of management actions to conserve these priority species.Lineage Statement:Selection of taxa and sources of distribution dataTaxa were included in this project if they were (i) listed as near threatened, vulnerable, endangered, and/or critically endangered under Australia�s Environment Protection and Biodiversity Conservation Act (EPBC) and/or under Northern Territory, Queensland, and Western Australia legislation and/or on the IUCN red list. Additional range restricted priority taxa were included on advice of the relevant state and territory environmental departments and based on their listing on the relevant action plans for Australian mammals, birds and fishes. The listings of each taxon at the time of data collation are shown in the additional information spreadsheet �NESP vetting and taxa information� included in this data collection.Distribution models were based on occurrence records extracted from several data bases, including the Atlas of Living Australia (ALA), state and territory government data bases managed by the Western Austr lia Parks and Wildlife Service, the Queensland Department of Environment and Heritage Protection, the Northern Territory Department of Environment and Natural Resources, the Victoria Department of Environment, Land, Water and Planning, and several databases managed by universities, museums or individuals (Frank Koehler from the Australian Museum, Ashley Field and Mel Greenfield from the Australian Tropical Herbarium/ James Cook University, Ian Cowie from the Northern Territory Herbarium, David Westcott from CSIRO, Arthur Georges from the University of Canberra, Mark Kennard from Griffith University, and Mark Hamann from James Cook University).Occurrence records were cleaned (i.e. dubious records excluded) and filtered to the appropriate precision and date range. Any non-Australian records were excluded (however, this only affected a small number of taxa since most were Australian endemics). Cleaning was done according to feedback elicited from experts, information from the literature, and any information on trustworthiness available in the source data bases. Where possible, only records from 1975 or more recent and with precision of 250m or better were used. However, if either of these two filters (precision and date) reduced the number of unique records substantially (by more than half or to less than 20 records), criteria were relaxed (to 1000m or any precision; to all historic records). Records were further reduced to the single most recent and highest precision record per grid cell (all environmental predictor layers used for modelling were rasters with cell size of 0.0025 decimal degrees/~250m; see modelling methods below). Criteria were also relaxed for taxa with spatial bias in either precision or date range of records (e.g. if all records from a certain area had a precision of NA) and for some taxa that have undergone recent range contractions, if these contractions were likely to be related to spatial variation in intensity of threatening processes and not to changes in habitat suitability as described by our set of predictor variables. The latter enabled us to better capture the full set of conditions that define suitable habitat for a taxon; these areas of suitable habitat were subsequently cut to still occupied parts of species ranges according to expert knowledge (see modelling methods below).Environmental predictor variables�Environmental predictor variables used in any of the distribution models were chosen from a larger set of candidate environmental variables depending on each taxon�s ecological traits (such as freshwater dependence, or broad taxonomic affiliation), which was further reduced in a variable selection process. The base set of variables included environmental layers adapted to our required extent, resolution and coordinate system from various sources and included layers based on the National Catchment and Stream Environment Database version 1.1.5 (J.L. Stein, M.F. Hutchinson and J.A. Stein, Fenner School of Environment and Society, Australian National University),� the National Vegetation Information System (NVIS; http://www.environment.gov.au/land/native-vegetation/national-vegetation-information-system), climate layers created using ANUCLIM (http://fennerschool.anu.edu.au/research/products/anuclim-vrsn-61), the 9-sec digital elevation model available from Geoscience Australia (https://ecat.ga.gov.au/geonetwork/srv/eng/search#!a05f7892-d78f-7506-e044-00144fdd4fa6), AusCover foliage projective/forest cover (http://auscover.org.au/purl/landsat-persistent-green-2000-2010) and vegetation height and structure (http://auscover.org.au/purl/icesat-vegetation-structure), dominant lithology (as described in the Bureau of Meteorology Geofabric product groundwater cartography at ftp://ftp.bom.gov.au/anon/home/geofabric/), distance to water based water a combination of features described by Bureau of Meteorology Geofabric products at ftp://ftp.bom.gov.au/anon/home/geofabric/ and by Geoscience Australia �surface hydrology� products (https://data.gov.au/), distance to coast based on Australia�s oast line as described by IBRA7 products (http://www.environment.gov.au/land/nrs/science/ibra), soil properties (depth to regolith, pH, etc.) and soil type (calculated based on combination of sand, silt and clay in top 30cm of soil) as described by the CSIRO �Soil and Landscape Grid National Soil Attribute Maps� at 3-sec resolution (accessed through CSIRO data portal at https://data.csiro.au/dap/home?execution=e1s1), CSIRO soil classification (http://www.asris.csiro.au/themes/NationalGrids.html), fraction photosynthetic active vegetation (�Australia, MODIS-fPAR time series (2000-2014), 9 arcsec (~250 m)� accessed through the BCCVL at� http://www.bccvl.org.au/), and a weathering index (Wilford, J. 2012. A weathering intensity index for the Australian continent using airborne gamma-ray spectrometry and digital terrain analysis, Geoderma 183�184 (2012) 124�142, DOI: https://doi.org/10.1016/j.geoderma.2010.12.022). The full collection of layers from which the subset was selected, and detailed methods on the creation of each layer can be accessed via the Tropical Data Hub (�Pintor, A.; VanDerWal, J.; Graham, E. (2018). High resolution environmental layers for species distribution modeling. James Cook University. http://dx.doi.org/10.4225/28/5a9f2e04d6316).� Species distribution modelling & vettingFor any taxon with 10 or less remaining records, simple distribution models were created by buffering occurrences and cutting buffers to areas with conditions similar to those observed at actual occurrence locations (for example, freshwater associated taxa were cut to areas not further from streams than observed at any of the occurrence records). Buffer size was specified according to a taxon�s extent of occurrence and number of records to allow for greater buffers for wide ranging taxa (likely greater dispersal capacities and sample gaps between known occurrences) and smaller buffers for more well sampled taxa within their extent of occurrence (less uncertainty about sample gaps between known occurrences). Experts were then consulted to choose the final buffer radius from three: our recommended buffer distance (based on the reasoning described above), half our recommendation, or twice our recommendation. For each taxon with over 50 records, we ran 10-fold cross-validated Maxent models (Steven J. Phillips, Miroslav Dud�k, Robert E. Schapire. Maximum Entropy Modeling of Species Geographic Distributions; Version 3.3.3, 2010) and performed variable selection: variable performance was ranked according to permutation importance, the best variables with a permutation importance of > 1% were selected, and this list further shortened to retain up to 10 variables for taxa with under 100 records and one variable per 10 occurrence records for taxa with over 100 records to balance the risk of overfitting models with providing the sufficient detail on habitat characteristics. The remaining variables were used to run a final Maxent model. For taxa with 11 to 50 records, Maxent models were created, but variable selection was based on summaries of best performing variables for other taxa with similar traits (from the set with more than 50 records discussed above).Each model went through an expert vetting process where experts on the different taxonomic or functional groups decided on a threshold, excluded dubious records, clarified taxonomic changes and inaccuracies, and identified areas that were predicted as suitable but are known or expected to be unoccupied. Final models were run with the vetted data and outputs were cut and thresholded accordingly. Occurrence thresholds were selected by experts from the following options; the Maxent threshold suggested options of (T1) �Equal training sensitivity and specificity� or (T2) �Equate entropy of thresholded and original distributions�; (T3) the most severe of the two thresholds T1 or T2 multiplied by 2 (if less than 0.35) or divided by 2 (if higher than 0.35 which was rarely the case) - this option aimed to provide an �extreme� alternative to the traditional thresholds because it may be that predictions for ranges of poorly sampled species are too strict (lack of representation of full range of occupied conditions) or for highly restricted species too generous (underestimates of specialists� dependence on exact conditions); (T4) a constant strict threshold of 0.5; (T5) an individual threshold nominated by the expert if all other thresholds were deemed inappropriate. Detailed information on which type of threshold was used for each taxon, which areas were cut are specified in the spreadsheet �NESP vetting and taxa information�, which is part of this collection.�These thresholds in the spreadsheet apply to the original Maxent outputs and NOT to the final distribution maps. Final model versions were rescaled to within the range of suitabilities included in the cut and thresholded output (for example if the raw suitability output ranged from 0 to 1 and the expert selected threshold was 0.4, the resulting range of suitabilities from 0.4-1 was rescaled to the full range of within realized habitat� suitability of 0-1). This was done to make final models for different taxa comparable. One reason this was necessary was that some threatened subspecies are under-sampled and underrepresented in whole-species models. For example for species with many East Coast records and few records for one isolated subspecies in the Kimberley, Maxent will treat latter as outliers and underpredict suitability in the Kimberley. However, such isolated subspecies are often the ones of greatest conservation concern. Because cresating useful models for conservation decisoons was the primary goal of this project, the rescaled final versions for each of the modelled subspecies was merged to the final version of the rescaled overall species model to fill in any such gaps.�All models ran through three vetting rounds: initial models were vetted by experts and corrections applied, models were then rerun and re-vetted by co-investigators (to make sure that original expert comments were still represented accurately in the new versions and that any potential previous oversights were corrected); lastly a small set of models with remaining inaccuracies was rerun with final modifications and quality checked again.Model outputs and suggested appropriate useAvailable files for each taxon include the following (binary suitable/unsuitable maps� g-l exist for all taxa, continuous suitability outputs a-f only for Maxent based models, not for buffer based models):a) 10km resolution distribution map (rescaled suitability index 0 to 1/ expert vetted unoccupied areas set to 0) � �taxon_currentG10km.asc�;�b) 250m resolution distribution map (rescaled suitability index 0 to 1/ expert vetted unoccupied areas set to 0) � �taxon_currentG.asc;�c) 250m resolution distribution map (suitability index 0 to 1/ expert vetted unoccupied areas set to 0) � applied threshold relaxed by 0.1 ��taxon_currentGSL.asc�;�d) 250m resolution distribution map (rescaled suitability index 0 to 1/ expert vetted unoccupied areas NOT set to 0) ��taxon_currentGuncut.asc�;�e) 250m resolution distribution map (rescaled suitability index 0 to 1/ expert vetted unoccupied areas NOT set to 0) � applied threshold relaxed by 0.1��taxon_currentGuncutSL.asc�;�f) 250m resolution distribution map (rescaled suitability index 1 to 2/ expert vetted unoccupied areas 0 to 1) ��taxon_currentCG.asc�;g) 10km resolution distribution map (1=suitable and likely occupied/ 0=unsuitable or vetted unoccupied) � taxon_currentF10km.asc;h) 250m resolution distribution map (1=suitable and likely occupied/ 0=unsuitable or vetted unoccupied) � taxon_currentF.asc;�i) 250m resolution distribution map (1=suitable and likely occupied/ 0=unsuitable or vetted unoccupied) � applied threshold relaxed by 0.1 � �taxon_currentFSL.asc�;�j) 250m resolution distribution map (1=suitable - may or may not be occupied/ 0=unsuitable) ��taxon_currentFuncut.asc�;�k) 250m resolution distribution map (1=suitable - may or may not be occupied/ 0=unsuitable) � applied threshold relaxed by 0.1��taxon_currentFuncutSL.asc�;�l) 250m resolution distribution map (2=suitable and likely occupied/ 1=suitable but vetted unoccupied/ 0=unsuitable) ��taxon_currentCF.asc�;Model versions with relaxed thresholds are intended to allow for users to estimate uncertainty around recommended cut off points for distributions, and aim to describe areas where the taxon �may occur� but is not necessarily �likely to occur�. These are to be used with caution as they are not based on expert vetting. For high resolution models, gradual and 0/1 outputs are furthermore provided as a (I) final version cut to areas recommended by experts (b/h), (II) pre-cutting versions that include areas predicted to be suitable but unlikely to be occupied (d/j), and (III) a combined version showing cut out areas as low suitability scores as well as actually vetted-occupied areas as higher suitability scores. Versions including unoccupied but potentially suitable areas were provided in case they are needed for translocation projects and to capture any remaining uncertainty but should be used with caution. For publically available 10km models only the final, vetted-occupied, vetted-threshold versions are provided.�Versions with 10 km resolu ion are intended for public use and are intended to be made available through the Atlas of Living Australia and the Department of Environment and Energy (DoEE) Environmental Resources Information Network (ERIN) public web resources, while all 250m resolution versions include detailed information on the distributions of sensitive species, should only be used for conservation purposes and are only intended to be distributed on request by JCU, DoEE ERIN and relevant state/ territory government departments that have been granted distribution privileges by JCU or ERIN.We furthermore produced hotspot maps that show how many taxa of conservation concern co-occur across our study region. Such hotspot maps were created for several taxa, functional groups, or groups of taxa with similar conservation listing and aim to indicate which areas may be in particular need of certain sets of conservation actions, how adequate current protected areas are for the conservation of certain groups, and/or how the distribution of threatened taxa corresponds to general distributions of biodiversity. For comparison with general distribution patterns of biodiversity we recommend products provided by CliMAS (Climate change and biodiversity in Australia; http://climas.hpc.jcu.edu.au/). Hotspot maps were created by summing all unique species (i.e. subspecies listed as threatened were only included once by using the merged overall species model) within each pre-defined group predicted to be present in each grid cell (according to 250m resolution vetted-occupied vetted-threshold model versions). Outputs were clipped to our study region because only northern taxa were modelled and any patterns outside of the study region would be misleading as they don�t include the set of taxa of relevance there. High resolution hotspot models are intended for conservation purposes as well as public use because any species-specific sensitive information is disguised.� This collection is comprised of raster layers of present day distributions of terrestrial and freshwater taxa that are of conservation concern in Northern Australia, which includes the Australian Conservation Management Zones �North Australian Tropical Savanna� and �North Eastern Australia Tropical Rainforests� and drainage basins intersecting these regions boundaries.� Distributions were derived either from Maxent models or from buffered occurrence records and were vetted by experts and modified accordingly. The additional conservation hotspot maps included in this collection show the number of taxa within each relevant taxonomic, functional or conservation group of organisms present across the study region. The collection also includes supplementary spreadsheets with further information on each taxon, such as certain organisms� traits, model details or conservation listings. The collection was created for the purpose of aiding stakeholders with conservation decisions in Northern Australia and was funded by the Australian National Environmental Science Program (NESP) as part of the Northern Australia Environmental Resources Hub. Show more