SeaBed NSW: Seabed Landforms Classification Toolset
An ArcGIS toolset for the classification of shelf seabed features
Here we present the Seabed Landforms Classification Toolset workflow. This toolset is available for download on the NSW Government SEED portal and GitHub . This toolset uses a semi-automated approach to extract seabed features from high-resolution bathymetry data.
Mapping the occurrence and spatial extent of submerged features provides critical information needed to effectively manage our coastal and marine environment.
To guide you through the classification approach, we will use this example dataset of marine lidar bathymetry data captured offshore of Moruya, NSW in 2018. Moruya is located approximately 300 km south of Sydney, Australia.
The Coastal and Marine Team, within the New South Wales (NSW) Department of Climate Change, Energy, the Environment and Water (DCCEEW, formerly Planning and Environment), have developed ArcGIS tools which are designed to classify seabed features into landforms – including reefs/banks, scarps (high slope areas), peaks, plains, depressions and channels.
To achieve this classification, a number of key layers are derived from the bathymetry digital elevation model (DEM), including: Ruggedness, slope, finescale Bathymetric Position Index (BPI) and broadscale BPI
Ruggedness represents surface roughness. It is the main layer used to define the outline of the reef or bank outcrop.
Ruggedness is calculated using the Benthic Terrain Modeler toolbox.
Slope measures the gradient of the surface. It is used to capture higher slope areas which are classed as “scarps”.
Slope is calculated using the Spatial Analyst toolbox in ArcGIS.
Finescale and broadscale BPI measure relative height across the seascape.
Broadscale BPI looks within a broad, larger distance to map whether features are high, flat, or low in elevation relative to the surrounding terrain.
Finescale and broadscale BPI are calculated using the Geomorphometry and Gradients Metrics Toolbox .
Finescale BPI looks within a smaller, localized distance to map the high, flat and low areas within the terrain.
Together, finescale and broadscale BPI can map the uppermost features within the seascape – defined as ‘peaks’ in the landforms classification, as well as the flat areas, which may become ‘plains’, and low areas, which may become ‘depressions and channels’.
Using all these variables – the user can explore different thresholds which can be used to break apart the surface
Example thresholds are shown here – where the user may set a specific ruggedness value – for example 0.00008 – to classify what they determine as rugose outcrops.
These threshold values are input into the ArcGIS toolset to run the surface elements classification
This shows the resulting surface element classification, based on the threshold values shown previously for each of the variables.
This breaks apart the surface into categories such as 'broadscale high smooth' or 'finescale low rugose'
Using the slider here – you can explore the input DEM for Moruya and the resulting surface elements classification
Comparison of input DEM and classified surface elements.
Once you have the surface elements classification, our tools guide the user through a series of steps which translate the surface elements classification into the landforms classification. These steps including identifying potential noise within the dataset, and aggregating features based on spatial and attribute queries.
Here, you can use the slider to view how the surface elements classification can ultimately become the landform classification.
Comparison of classified surface elements and final edited landforms classification.
The toolset also includes functionality to classify the plain areas in greater detail.
This breaks up the plain surface into localized high and low areas, broadscale low and high areas, as well as flats.
These tools, developed by the Coastal and Marine Unit within NSW Department of Climate Change, Energy, the Environment and Water (DCCEEW, formerly Planning and Environment) and are freely available for users to explore with their own datasets.
Please cite these tools as: Linklater, M, Morris, B.D. and Hanslow, D.J. (2023) Classification of seabed landforms on continental and island shelves. Frontiers of Marine Science, 10, https://doi.org/10.3389/fmars.2023.1258556 .
These tools have been designed for capturing prominent features within a continental or island shelf setting.
Funding for developing these tools was provided by the NSW Climate Change Fund through the Coastal Management Funding Package and the Marine Estate Management Authority.
These classification tools were used to complete a statewide classification of seabed landforms from the NSW marine lidar data. The marine lidar bathymetry data and classified seabed landforms can be viewed on the NSW SEED data portal .
To explore the toolset functionality in greater detail, you may scroll on to view a worked example and explanation of the key classification stages of the toolset.
The Seabed Landforms Classification Toolset is an ArcGIS Toolbox. They utilise two other freely available ArcGIS toolboxes: the Benthic Terrain Modeler and the Geomorphometry and Gradients Metrics Toolbox .
The classification method is undertaken in several key stages:
- Preparation
- Surface elements classification
- Landforms classification
- Plains classification (optional)
Using the example dataset of Moruya, we will walk through each of the step tool
Starting with the input DEM, using the Moruya marine lidar example.
Default settings for the tool are based on a DEM with 5 m cell size
Data preparation tools include:
PREPARATION TOOLS
'PREP Create DEM from XYZ'
If you have XYZ inputs this will grid the DEM
PREPARATION TOOLS
'PREP Clip elevations'
It is recommended to clip to bathymetry data only (< 0 m elevation), and to exclude data beyond the shelf break
PREPARATION TOOLS
'PREP smooth DEM'
This smooths the data using a median filter with a 3 x 3 cell window. This is recommended for data which may have speckled noise such as marine lidar or depth derived from satellite imagery.
This is an important step to ensure the ruggedness variable is not too impacted by noise. You can see in this example, reclassified ruggedness at the same value with and without smoothing.
LANDFORM CLASSIFICATION
'LANDFORMS Step 01 - Terrain Variables'
This first step of the landforms classification calculates the four terrain variables you will need: Ruggedness (vrm), slope, finescale BPI and broadscale BPI.
Users will need to input the DEM, and the window size they desire for broadscale and finescale BPI. Users may re-run this tool to explore a suitable BPI window size.
Once you've run the 'terrain variables' tool, you will have the four layers which form the surface elements classification.
Users can interrogate these layers to determine suitable values to reclassify each variable.
For example - looking at ruggedness:
To determine what ruggedness threshold may be suitable for the surface elements classification, it is recommended users 'reclassify' the raster on-the-fly using the layers Symbology settings in ArcMap (right click - Properties).
In this example, the Symbology is set to 'Classified', using '2' classes. A threshold value of 0.00008 is set, and the user can see whether this captures the rugose features of interest (such as outcropping reef).
LANDFORM CLASSIFICATION
'LANDFORMS Step 02 - Surface elements'
One suitable thresholds are determined for each layer, the user can run the surface elements classification. This classification combines reclassified layers of ruggedness (vrm), slope, finescale and broadscale BPI.
The tool requires the user to enter threshold values which will be used to reclassify each layer. Default values are provided (based on a 5 m DEM with 3 x iterations of smoothing ) but the user must determine the most suitable values based on their individual dataset.
The reclassified ruggedness forms the foundation of the polygon extent of the “reefs/banks” polygons. The value selected should capture the full extent of the outcrops the user seeks to capture.
This example value of 0.00008 is based on a 5 m DEM with 3 smoothing iterations
Using a value of 0.00008, all areas > 0.00008 rugg are classed as “rugose” areas, which will eventually be labelled as “reefs/banks” in the final landform classification.
Later steps in the classification will identify polygons that occur within the reef/bank outcrop, and include them as part of the outcrop structure.
Therefore, the user should look for a value which captures the outer rim of the outcrop, so that polygons within the structure can be selected. This must be balanced with capturing excess noise.
For the slope reclassification value, the default value of 10 is based on a 5 m DEM with 3 smoothing iterations.
Using a value of 10, all areas > 10 degrees slopes are classed as “slope” areas, which will eventually be labelled as “scarps” in the final landform classification.
For broadscale BPI, the user must specify both a window size to generate the BPI grid as well as the values for reclassification.
The window size is the radius that the tool “looks” around each cell to determine whether surrounding features are high, low or flat in elevation relative to the central cell. A larger window size value will look across a greater distance of the seascape.
The default window size is 150 cells, with classification values of -100 and +100 are based on 5 m DEM with 3 smoothing iterations. Using these values, areas with positive values > +100 are the uppermost areas of the seascape, and will become “peaks” in the landform classification.
Areas with negative values < -100 are the lowest areas of the seascape, and will become depressions and channels or plains. Areas between -100 and +100 will be flat areas, that will become plains or flat areas of reef/bank outcrops
For finescale BPI, the user must similarly specify both a window size to generate the BPI grid as well as the values for reclassification.
Finescale BPI uses a smaller window size value, which will look across a shorter distance of the seascape.
The default window size is 27 cells, with classification values of -100 and +100 are based on 5 m DEM with 3 smoothing iterations. Using these values, areas with positive values > +100 are the uppermost areas of the seascape, and will become “peaks” in the landform classification.
Areas with negative values < -100 are the lowest areas of the seascape, and will become depressions and channels or plains. Areas between -100 and +100 will be flat areas, that will become plains or flat areas of reef/bank outcrops
The reclassification values depend on the resolution and level of smoothing of the DEM. Some example settings are provided in the tool user guide to assist users in adjusting settings.
These reclassified layers are combined to form the surface elements classification.
The many classes of the surface elements classification are then aggregated from 11 classes to 7 classes to form a summarised surface elements classification.
This offers a simplified representation of the prominent shelf features for the user's interpretation.
LANDFORMS CLASSIFICATION
'LANDFORMS Step 03 - OPTIONAL Depth reclassification'
'LANDFORMS Step 04 - OPTIONAL Drainage'
In addition to the surface elements classification, the user may optionally generate other layers to assist in the feature interpretation. These include a depth reclassification and drainage layer.
LANDFORMS CLASSIFICATION
'LANDFORMS Step 06 - Preliminary landforms - to review'
To get from the surface elements classification to the landforms classification, our tools perform a series of steps which flag potential noise, identify polygons within rugose outcrops, and aggregate surface element classes.
This creates 'preliminary landforms; which are required to be reviewed and edited by the user before finalising.
This step of the landforms classification asks the user for a new ruggedness threshold to flag potential noise polygons.The user must determine the most suitable value based on their individual dataset. The default value is based on a 5 m cell size DEM with 3 iterations of smoothing.
The first ruggedness value used in the surface elements classification ('Step 02 - surface elements' = ruggedness value 0.00008 used in this example) reclassifies the surface to capture rugose outcrops.
A second ruggedness value ('Step 06 - Preliminary landforms' = ruggedness value 0.005 used in this example) is used in this step to identify which of these rugose polygons are potentially noise.
To do this, a higher ruggedness value is used to identify the more rugose areas of the surface, which would be more likely to be real reef/bank outcrops. Smaller, bumpier outcrops would be more likely to be noise artefacts.
Within the tools, these two values are overlain over each other. Polygons which do not overlap are flagged as potential noise for the user to review at a later step.
The user can explore the ruggedness layer to determine the most suitable value prior to using this step.
The resulting classification transforms the surface element classification into a preliminary landforms classification.
From here, the user must review and edit this classification in detail to ensure the classes best represent the interpretation of features within the individual dataset.
The output of this stage of the classification requires detailed manual review and editing.
A number of editing steps are required. Here we can look at an example of required edits performed for the Moruya dataset.
This example area shows the preliminary landform labels and some areas where user's need to perform a manual review and editing.
This shows the resulting final landforms classification for the example area, with required edits performed.
The user may optionally undertake additional edits. These may include removing inferred soft sediment areas ('banks') from the classification to focus on 'reef' only areas. Additional channel areas may also be identified by cutting polygons and relabelling as 'depressions and channels'.
This shows the resulting final landforms classification for the example area, with optional edits performed.
When the manual edits and review are complete, the user can run 'LANDFORMS Step 07 - Final landforms' to run the final stages of the tool and create the final landforms layer.
PLAIN CLASSIFICATION
'LANDFORMS Step 09 - OPTIONAL Plain landforms'
If desired, the user may perform a more detailed classification of plain areas.
This is an optional step, recommended to be run on inferred soft sediment areas if the user has manually separated inferred reef areas as outlined in the optional edits.
To perform this classification, the user must use standard ArcGIS tools to select and export the plain polygon, then extract the DEM over plain areas only ('Extract by Mask').
This tool uses only the finescale and broadscale BPI to characterise the plain surface.
Similar to the surface elements classification, users must determine and input window sizes and reclassification values.
Default values in the tools are based on 5 m cell size DEM with 3 x iterations of smoothing.
The resulting classification captures the detailed surface of the plain surface, which may be useful in bedform interpretation and classification
We hope these freely available tools we have developed will assist users in classifying prominent features from shelf seabed data.
Please feel free to download the Seabed Landforms Classification Toolset and share these tools to those who may be interested in classifying seabed data
