This dataset describes the predicted probability of seagrass presence across the Great Barrier Reef World Heritage Area
and adjacent estuaries, based on six Random Forest models. The models have been mosaicked together into one raster dataset
with 30m resolution.
Managing seagrass resources in the GBRWHA requires adequate information on the spatial extent of seagrass habitat. The
enormous size of the GBRWHA (1000s of kilometres) and the remoteness of many seagrass meadows from human populations means
that models are a useful tool to predict the probability of seagrass for areas where data is lacking.
James Cook University’s Centre for Tropical Water & Aquatic Ecosystem Research (TropWATER) has been collecting spatial data
on GBR seagrass since the early 1980s. This project used TropWATER’s synthesis of seagrass site data (NESP Project 3.1 and
5.4: https://eatlas.org.au/data/uuid/5011393e-0db7-46ce-a8ee-f331fcf83a88) to predict potential seagrass habitat.
In making this data publically available for management, the authors from the TropWATER Seagrass Group request being contacted
and involved in decision making processes that incorporate this data, to ensure its limitations are fully understood.
The sampling methods used to study, describe and monitors seagrass meadows were developed by the TropWATER Seagrass Group
and tailored to the location and habitat surveyed; descriptions and references are available in the metadata for the GBRWHA
data composite ( https://eatlas.org.au/data/uuid/5011393e-0db7-46ce-a8ee-f331fcf83a88 ).
Environmental predictors used in the models were: depth below mean sea level (Beaman 2017), relative tidal exposure (Bishop-Taylor
et al. 2019), water type (Marine Water Bodies definitions version 2_4, Data courtesy of the Great Barrier Reef Marine Park
Authority; Dyall et al. 2004), proportion mud in the sediment (coast and reef models, https://research.csiro.au/ereefs/models/model-outputs/access-to-raw-model-output/ ) (see also Baird et al. 2020; Margvelashvili et al. 2018), dominant sediment (estuary models only; https://eatlas.org.au/data/uuid/5011393e-0db7-46ce-a8ee-f331fcf83a88 ), benthic geomorphology (Heap and Harris 2008), benthic light http://dapds00.nci.org.au/thredds/catalog/fx3/gbr1_bgc_924/catalog.html (see also Baird et al. 2016; Baird et al. 2020), water temperature, mean current speed and salinity http://thredds.ereefs.aims.gov.au/thredds/s3catalogue/aims-ereefs-public-prod/derived/ncaggregate/ereefs/gbr1_2.0/all-one/catalog.html?dataset=EREEFS_AIMS-CSIRO_gbr1_2.0_hydro_all-one.nc (Steven et al. 2019), wind speed ( http://thredds.ereefs.aims.gov.au/thredds/s3catalogue/aims-ereefs-public-prod/derived/ncaggregate/ereefs/gbr1_2.0/all-one/catalog.html?dataset=EREEFS_AIMS-CSIRO_gbr1_2.0_hydro_all-one.nc ) and Australian Bureau of Meteorology’s ACCESS data products (Bureau of Meteorology 2020; Soldatenko et al. 2018; Steven
et al. 2019), and latitude. Different models had different combinations of predictors after removing collinear variables
and excluding variables that did not extend into an area. For example, estuary models only include depth, relative tidal
exposure, dominant sediment, and latitude.
We modelled seagrass probability in six areas: Estuary Intertidal, Estuary Subtidal, Coast Intertidal, Coast Subtidal, Reef
Intertidal and Reef Subtidal. For each area we used the machine learning technique random forest to broadly examine whether
there were habitats within the GBRWHA and adjacent estuaries where seagrass was never likely to grow, using the binary classification
within the site data of seagrass present (1) or absent (0) irrespective of species. Random Forest models were implemented
using the randomForest package (Liaw and Wiener 2002) in R version 4.0.2 (R Core Team 2020). We used ArcGIS 10.8 to mosaic
the six rasters and create a single seagrass probability raster for the GBRWHA.
Seagrass data north and south of the GBRWHA were not included in the analysis. The model extends across the continental
shelf but excludes waters deeper than ~100m east of the shelf that were not surveyed for seagrass. Data were included when
sites extended west of the GBRWHA boundary into coastal and estuarine water immediately adjacent.
The site data used in this model is available here: https://eatlas.org.au/data/uuid/5011393e-0db7-46ce-a8ee-f331fcf83a88
Further information can be found in the upcoming publications of the final report for the NESP TWQ Project 5.4.
Limitation of the data:
The site data used in these models extends back to the mid-1980s. Large parts of the coast have not been mapped for seagrass
presence since that time. The seagrass probability raster is at 30m grid resolution, however some environmental variables
such as those from eReefs (wind speed, current speed, benthic light, water temperature) are from spatial data at 1km grid
resolution, and are likely to vary at much smaller spatial scales that we could not include in these models.
This dataset consists of a raster dataset with a geographic coordinate system of WGS84. The raster has been saved as a layer
package with symbology representing seagrass probability in 0.2 increments with a range of 0-1 (GBRWHA_seagrass_probability.lpk)
This dataset is filed in the eAtlas enduring data repository at: data\nesp5\5.4_Seagrass-Burdekin-region\
Additional licensing information:
TropWATER gives no warranty in relation to the data (including accuracy, reliability, completeness, currency or suitability)
and accepts no liability (including without limitation, liability in negligence) for any loss, damage or costs (including
consequential damage) relating to any use of the data. TropWATER reserves the right to update, modify or correct the data
at any time. The limitations of some older data included need to be understood and recognised. The TropWATER Seagrass Group
would appreciate the opportunity to review documents providing research, management, legislative or compliance advice based
on this data.