diff --git a/docs/_files/dea-coastalecosystems/CEM_flowchart.png b/docs/_files/dea-coastalecosystems/CEM_flowchart.png new file mode 100644 index 000000000..536f1aa9c Binary files /dev/null and b/docs/_files/dea-coastalecosystems/CEM_flowchart.png differ diff --git a/docs/_files/dea-coastalecosystems/Labelled_tile_regions.png b/docs/_files/dea-coastalecosystems/Labelled_tile_regions.png new file mode 100644 index 000000000..c504a73b1 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/Labelled_tile_regions.png differ diff --git a/docs/_files/dea-coastalecosystems/ML_Feature_Importance.png b/docs/_files/dea-coastalecosystems/ML_Feature_Importance.png new file mode 100644 index 000000000..d26e0434b Binary files /dev/null and b/docs/_files/dea-coastalecosystems/ML_Feature_Importance.png differ diff --git a/docs/_files/dea-coastalecosystems/Original_training_point_distribution.png b/docs/_files/dea-coastalecosystems/Original_training_point_distribution.png new file mode 100644 index 000000000..7a83ff95c Binary files /dev/null and b/docs/_files/dea-coastalecosystems/Original_training_point_distribution.png differ diff --git a/docs/_files/dea-coastalecosystems/Unlabelled_tile_regions.png b/docs/_files/dea-coastalecosystems/Unlabelled_tile_regions.png new file mode 100644 index 000000000..3c6b05f67 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/Unlabelled_tile_regions.png differ diff --git a/docs/_files/dea-coastalecosystems/classification_Shoalwater_2021.png b/docs/_files/dea-coastalecosystems/classification_Shoalwater_2021.png new file mode 100644 index 000000000..01af4a802 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/classification_Shoalwater_2021.png differ diff --git a/docs/_files/dea-coastalecosystems/classification_Shoalwater_2021_thumbnail.png b/docs/_files/dea-coastalecosystems/classification_Shoalwater_2021_thumbnail.png new file mode 100644 index 000000000..2b383da8f Binary files /dev/null and b/docs/_files/dea-coastalecosystems/classification_Shoalwater_2021_thumbnail.png differ diff --git a/docs/_files/dea-coastalecosystems/classification_bermagui_2021.png b/docs/_files/dea-coastalecosystems/classification_bermagui_2021.png new file mode 100644 index 000000000..09643c684 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/classification_bermagui_2021.png differ diff --git a/docs/_files/dea-coastalecosystems/classification_bermagui_2021_LTC.png b/docs/_files/dea-coastalecosystems/classification_bermagui_2021_LTC.png new file mode 100644 index 000000000..41c09a577 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/classification_bermagui_2021_LTC.png differ diff --git a/docs/_files/dea-coastalecosystems/connectivity_Mallacoota.png b/docs/_files/dea-coastalecosystems/connectivity_Mallacoota.png new file mode 100644 index 000000000..db20788d4 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/connectivity_Mallacoota.png differ diff --git a/docs/_files/dea-coastalecosystems/low_tide_composite_2021_bermagui.png b/docs/_files/dea-coastalecosystems/low_tide_composite_2021_bermagui.png new file mode 100644 index 000000000..dce828ecc Binary files /dev/null and b/docs/_files/dea-coastalecosystems/low_tide_composite_2021_bermagui.png differ diff --git a/docs/_files/dea-coastalecosystems/probabilities_Bermagui_2021.png b/docs/_files/dea-coastalecosystems/probabilities_Bermagui_2021.png new file mode 100644 index 000000000..9bb7e97b9 Binary files /dev/null and b/docs/_files/dea-coastalecosystems/probabilities_Bermagui_2021.png differ diff --git a/docs/_redirects/data-products.txt b/docs/_redirects/data-products.txt index ab952c955..02481070d 100644 --- a/docs/_redirects/data-products.txt +++ b/docs/_redirects/data-products.txt @@ -19,3 +19,8 @@ data/product/dea-intertidal-extents-landsat data/product/dea-intertidal # DEA Tidal Composites has superseded High and Low Tide Composites (HLTC) data/product/dea-high-and-low-tide-imagery-landsat data/product/dea-tidal-composites + +# A development URL for DEA Coastal Ecosystems + +data/product/dea-coastalecosystems data/product/dea-coastal-ecosystems + diff --git a/docs/_templates/product-v2.rst b/docs/_templates/product-v2.rst index 2b53281ed..da4b86fab 100644 --- a/docs/_templates/product-v2.rst +++ b/docs/_templates/product-v2.rst @@ -78,6 +78,7 @@ } %} {% set data_update_frequency_terms = { + "NONE": "None", "AS_NEEDED": "As needed", "DAILY": "Daily", "WEEKLY": "Weekly", @@ -89,6 +90,7 @@ {% set data_update_activity_terms = { "ONGOING": "Ongoing", + "NON_ONGOING": "Non-ongoing", "NO_UPDATES": "No further updates", "DEVELOPMENT": "Awaiting development", "PAUSED": "Currently paused", @@ -174,7 +176,9 @@ {% set coordinate_reference_system_term = coordinate_reference_system_terms.get(page.data.coordinate_reference_system, page.data.coordinate_reference_system) %} -{% set is_frequency_ongoing = data_update_activity == data_update_activity_terms.ONGOING %} +{% set is_activity_ongoing = data_update_activity == data_update_activity_terms.ONGOING %} + +{% set is_activity_non_ongoing = data_update_activity == data_update_activity_terms.NON_ONGOING or data_update_frequency_terms.NONE %} {% set is_cadence_yearly = data_update_frequency == data_update_frequency_terms.YEARLY %} @@ -268,10 +272,12 @@ {%- elif page.data.temporal_coverage_end %} :Coverage end: {{ page.data.temporal_coverage_end }} {%- endif %} - {%- if is_frequency_ongoing and is_frequency_multiple_words and page.data.is_latest_version %} + {%- if is_activity_ongoing and is_frequency_multiple_words and page.data.is_latest_version %} :Data updates: {{ data_update_frequency }}, {{ data_update_activity }} - {%- elif is_frequency_ongoing and page.data.is_latest_version %} + {%- elif is_activity_ongoing and page.data.is_latest_version %} :Data updates: {{ data_update_frequency }} frequency, {{ data_update_activity }} + {%- elif is_activity_non_ongoing and page.data.is_latest_version %} + :Data updates: {{ data_update_activity }} {%- elif is_frequency_multiple_words and not page.data.is_latest_version %} :Data updates: {{ data_update_activity_terms.NO_UPDATES }} (Previously: {{ data_update_frequency }}) {%- elif not page.data.is_latest_version %} @@ -735,7 +741,11 @@ - {{ coordinate_reference_system_term }} - The method of mapping spatial data to the Earth's surface. {%- endif %} - {%- if is_frequency_ongoing and page.data.is_latest_version %} + {%- if is_activity_ongoing and page.data.is_latest_version %} + * - **Update frequency** + - {{ data_update_frequency }} + - The expected frequency of data updates. Also called 'Temporal resolution'. + {%- elif is_activity_non_ongoing and page.data.is_latest_version %} * - **Update frequency** - {{ data_update_frequency }} - The expected frequency of data updates. Also called 'Temporal resolution'. diff --git a/docs/data/product/dea-coastal-ecosystems/_access.md b/docs/data/product/dea-coastal-ecosystems/_access.md new file mode 100644 index 000000000..a12cbaf5b --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_access.md @@ -0,0 +1,46 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/services/dea-knowledge-hub/operating-procedures/product-metadata-fields/#kh-product-metadata-fields + +:::{admonition} Streaming data from AWS is strongly recommended +:class: note + +DEA Coastal Ecosytems data is large with files up to 3 GB in size. We strongly recommend streaming data directly from the cloud rather than downloading the data. Please see the instructions below: **How to stream data from AWS** +::: + +:::{dropdown} How to explore DEA Maps (To be completed with StoryMap) + +Start exploring [DEA Coastal Ecosystems on DEA Maps](https://maps.dea.ga.gov.au/story/DEACoastalEcosystems). + +To add DEA Coastal Ecosystems to DEA Maps manually: + +1. Open [DEA Maps](https://maps.dea.ga.gov.au/). +1. Select **Explore map data** on the top-left. +1. Select **Sea, ocean and coast** > **DEA Coastal Ecosystems** > **DEA Coastal Ecosystems (Sentinel-2)**. +1. Click the **Add to the map** button on top-right. + +View the product layer required by selecting the layer from the **Styles** dropdown menu, and use the **Time** selector to view the annual time-series of each product layer. + +% TODO To query an absolute value for any of the product layers, click on a location to see a plain text summary of all the DEA Intertidal product suite data values at that pixel location. +::: + +:::{dropdown} How to stream data from AWS (Recommended) + +The easiest way to access DEA Coastal Ecosystems data is via our continental-scale cloud-optimised GeoTIFF mosaics (COGs). +The COG file format is a type of GeoTIFF raster file (`.tif`) that allows you to quickly and efficiently 'stream' data directly from the Amazon S3 cloud without having to download files to your computer. +This allows you to rapidly access data from the entire Australian continent without having to download large files. + +VRT (Virtual Raster) files are also provided alongside the mosaics. These files serve as lightweight wrappers around the main data and can be used to open data in GIS software with visual settings already applied. + +Learn more about [Continental Cloud-Optimised GeoTIFF Mosaics](/guides/continental-cogs-geotiff-mosaics/). + +::: + + +:::{dropdown} How to download Continental Mosaic data + + + +1. Open the [DEA Coastal Ecosystems](https://data.dea.ga.gov.au/?prefix=derivative/ga_s2_coastalecosystems_cyear_3_v1/AU/) directory in DEA's Amazon S3 bucket: +1. Enter a directory of a particular year, e.g. `2021--P1Y` +1. Click to download the product layer of interest, e.g. `ga_s2_coastalecosystems_cyear_3_v1-0-0_AU_2021-P1Y_final_classification.tif`. Learn more about file naming and product layers: [Technical Information](./?tab=description#core-product-layers). +::: + diff --git a/docs/data/product/dea-coastal-ecosystems/_credits.md b/docs/data/product/dea-coastal-ecosystems/_credits.md new file mode 100644 index 000000000..479c8ea71 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_credits.md @@ -0,0 +1,17 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +## Acknowledgments + +The authors would like to sincerely thank our collaborators on this project from James Cook University and the University of New South Wales, in particular Mitchell Lyons, Nicholas Murray, Alex Navarro and Robert Canto. + +We also like to acknowledge contributions and feedback from various academic institutions through the course of this product development, in particular from the teams at the Universities of Newcastle (Chris Owers and Rafael Carvalho) and Wollongong (Jeff Kelleway). + +Numerous Commonwealth, State and Local government departments and individuals have been generous with their time, data, and feedback through various workshops and engagement channels and their inputs have been crucial to the development of the DEA Coastal Ecosystems product suite. + + +## License and copyright + +© Commonwealth of Australia (Geoscience Australia). + +Released under [Creative Commons Attribution 4.0 International Licence](https://creativecommons.org/licenses/by/4.0/). + diff --git a/docs/data/product/dea-coastal-ecosystems/_data.yaml b/docs/data/product/dea-coastal-ecosystems/_data.yaml new file mode 100644 index 000000000..979c4a80e --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_data.yaml @@ -0,0 +1,123 @@ +# See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +# Specifications + +short_name: DEA Coastal Ecosystems +full_technical_name: Geoscience Australia Sentinel-2 Coastal Ecosystems Calendar Year Collection 3 +header_image: /_files/dea-coastalecosystems/classification_Shoalwater_2021_thumbnail.png + +version_number: 1.0.0 +is_latest_version: true +latest_version_link: /data/product/dea-coastal-ecosystems/ +is_provisional: false + +lineage_type: DERIVATIVE +spatial_data_type: RASTER + +spatial_coverage: Australia excluding external territories +temporal_coverage_start: 2021 +temporal_coverage_end: 2022 +temporal_coverage_custom: null + +data_update_frequency: NONE +data_update_activity: NON_ONGOING +is_currency_reported: false + +resolution: 10 m +coordinate_reference_system: EPSG:3577 + +product_ids: + - ga_s2_coastalecosystems_cyear_3_v1 + +parent_products: + - name: Sentinel-2A Surface Reflectance NBART + link: /data/product/dea-surface-reflectance-nbart-sentinel-2a-msi/ + - name: Sentinel-2A Observational Attributes + link: /data/product/dea-surface-reflectance-oa-sentinel-2a-msi/ + - name: Sentinel-2B Surface Reflectance NBART + link: /data/product/dea-surface-reflectance-nbart-sentinel-2b-msi/ + - name: Sentinel-2B Observational Attributes + link: /data/product/dea-surface-reflectance-oa-sentinel-2b-msi/ + +collections: + - name: Geoscience Australia Sentinel-2 Collection 3 + link: /search/?q=Geoscience+Australia+Sentinel-2+Collection+3 + +doi: 10.26186/150770 +ecat_id: "150770" + +licence_name: Creative Commons Attribution 4.0 International Licence +licence_link: https://creativecommons.org/licenses/by/4.0/ + +citation_data: "Phillips, C., Newey, V., Lyons, M., Sagar, S., Bishop-Taylor, R., DEA Coastal Ecosystems. Commonwealth of Australia (Geoscience Australia). https://doi.org/10.26186/150770" +# citation_paper: "Bishop-Taylor, R., Nanson, R., Sagar, S., Lymburner, L. (2021). Mapping Australia's dynamic coastline at mean sea level using three decades of Landsat imagery. Remote Sensing of Environment, 267, 112734. https://doi.org/10.1016/j.rse.2021.112734" + +# citations_custom: +# - name: Custom citation +# citation: "Bishop-Taylor, R., Nanson, R., Sagar, S., Lymburner, L. (2021). Mapping Australia's dynamic coastline at mean sea level using three decades of Landsat imagery. Remote Sensing of Environment, 267, 112734. https://doi.org/10.1016/j.rse.2021.112734" + +tags: + - geoscience_australia_sentinel_2_collection_3 + - coastal_ecosystems + - marine_and_coastal + - coast + - machine_learning + - intertidal + - coastal_change + +# Access + +access_links_maps: + - link: https://maps.dea.ga.gov.au/story/DEACoastalecosystems + name: null + +access_links_explorers: + - link: https://explorer.dea.ga.gov.au/products/ga_s2_coastalecosystems_cyear_3_v1 + name: null + +access_links_data: null + +access_links_code_examples: + - link: /notebooks/DEA_products/DEA_Coastalecosystems/ + name: View code examples (DEA Notebook) + +access_links_web_services: + - link: https://ows.dea.ga.gov.au/ + name: Web Map Service (WMS) + +access_links_custom: + - link: https://data.dea.ga.gov.au/?prefix=derivative/ga_s2_coastalecosystems_cyear_3_v1/AU + icon: database + label: Data sources + name: Access the data on AWS + description: "Learn **how to stream data from AWS below (the recommended method)** or `other ways to download from AWS `_." + class: null +# - link: https://github.com/GeoscienceAustralia/dea-coastalecosystems +# icon: code-branch +# label: Source code +# name: "Source code on GitHub" +# description: "The GitHub repository is GeoscienceAustralia/dea-coastalecosystems." +# class: null + +# History + +# previous_versions: +# - version_number: 1.0.0 +# title: Example product +# slug: example-product-1.0.0 + +# SEO + +# meta_description: This field can be used for SEO. + +# Settings + +enable_overview: true +enable_description: true +enable_quality: true +enable_specifications: true +enable_access: true +enable_history: true +enable_faqs: false +enable_credits: true + diff --git a/docs/data/product/dea-coastal-ecosystems/_description.md b/docs/data/product/dea-coastal-ecosystems/_description.md new file mode 100644 index 000000000..fc9738635 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_description.md @@ -0,0 +1,406 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +## Background + +Mapping coastal ecosystems is vital for understanding, managing, and protecting the diverse habitats that line our shorelines. These ecosystems — ranging from mangroves and saltmarshes to seagrass meadows and coral reefs — provide essential environmental services, including carbon storage, coastal protection, and habitat for marine biodiversity. From an economic standpoint, they sustain industries such as fisheries and tourism while reducing costs associated with coastal erosion and storm damage. Accurate and comprehensive mapping underpins management, restoration, and conservation efforts, allowing planners to identify priority areas, monitor ecological change, and assess the effectiveness of interventions. + +Achieving this at a continental scale, however, poses significant challenges. Traditional methods such as field surveys, aerial photography, and local-scale sampling are costly, labour-intensive, and constrained in spatial and temporal scope — limitations that are particularly relevant in a country as vast and remote as Australia. Earth Observation (EO) provides an efficient and consistent alternative, allowing large-scale, high-resolution, and repeatable mapping through the use of publicly available satellite imagery. + +The **Digital Earth Australia (DEA) Coastal Ecosystems** product suite was developed in response to this need for consistent, continent-wide monitoring. It provides the first national dataset to simultaneously map Australia’s mangrove, saltmarsh, intertidal, and intertidal seagrass ecosystems at 10 m resolution, using annual time-series of Sentinel-2 satellite imagery for consistent mapping at a continental scale. The mapping process employs a machine learning (ML) framework originally developed by the University of New South Wales (UNSW) and James Cook University (JCU) for global coastal ecosystem mapping on the Google Earth Engine platform (Murray et al., 2022). To tailor the approach for Australian conditions, JCU — commissioned by the Department of Climate Change, Energy, the Environment and Water (DCCEEW) — compiled a national training dataset of more than 40,000 annotated points across the continent (Canto et al., 2023; Becker et al., 2023). + +This collaboration ensures the resulting maps are ecologically meaningful and complementary to expert mapping undertaken by State Governments and academia. providing a strong foundation for environmental management and reporting, restoration planning, environmental-economic accounting, and long-term coastal conservation. + +## What this product offers + +The DEA Coastal Ecosystems product suite offers users a categorical classification of four coastal ecosystems, supported by probability layers for individual ecosystems and selected quality assurance layers to support product interpretation. +Key features of the product suite include: +- Sentinel-2 10 m resolution derived annual classification ecosystem maps, defining dominant ecosystem extents of mangroves, saltmarsh, intertidal, and intertidal seagrass. +- Probability layers for mapped ecosystems to enable user interpretation of mixed and transitional classes. +- QA/QC layers to provide confidence and assist in product interpretation. +- Publication of ancillary layers used for masking and constraints in the mapping workflow. +- Publication of the national training dataset, metadata, classification schema and acquisition protocol. + +% ## Access constraints + +% ## Use constraints + +## Applications + +- National Environmental Ecosystem Accounts +- National and Regional State of the Environment reporting +- A complementary mapping product to fill spatial and temporal gaps in higher resolution expert mapping (aerial and drone mapping, field surveys) +- Future integration with other terrestrial and ocean data sets (e.g. Land Cover and Marine Tenure) +- Coastal protection and Hazard modelling + +## Lineage + +DEA Coastal Ecosystems was initiated as a collaborative project to support the Commonwealth Department of Climate Change, Energy, the Environment and Water (DCCEEW) to underpin the creation of Environmental Economic Accounts (EEA) for ecosystems in Australia’s coastal zone. This followed experimental Ocean Accounts, delivered in 2022 and based on selected individual ecosystem earth observation derived datasets. + +Leveraging a methodology that mapped [global shifts in tidal wetlands](https://www.science.org/doi/10.1126/science.abm9583), Geoscience Australia has collaborated with James Cook University and the University of New South Wales to deliver a similar approach for Australia in the DEA Coastal Ecosystems product suite. Initial Beta versions of the product were used to underpin the February 2025 release of the [National Ecosystem Accounts experimental estimates for the Coastal Realm](https://www.abs.gov.au/statistics/environment/environmental-accounts/national-ecosystem-accounts-experimental-estimates/latest-release#coastal-realm), with the first v1.0.0 release of the DEA Coastal Ecosystems product suite expected to contribute to the upcoming 2026 National Ecosystem Account released by the Australian Bureau of Statistics (ABS) + +Targeted stakeholder engagement and product feedback sessions have been conducted throughout the development of the product to ensure, where possible, a complementary alignment of the DEA Coastal Ecosystems product suite to State Government and Academic mapping efforts. These engagements aimed to ensure ease of uptake and integration of the product into a range of environmental reporting applications across jurisdictions (e.g State of the Environment). + +Stakeholder engagement at all levels will continue to drive the development of DEA Coastal Ecosystems product in a cycle of continuous improvement, validation, feedback integration and inclusion of additional ecosystem types. + +## Technical information + +### Product Features + +The DEA Coastal Ecosystems product suite is a multi-layered dataset delivered as annual summary products. See [Specifications](./?tab=specifications) for individual layer metadata. + +### Core Product Layers + +Five core layers are included in The DEA Coastal Ecosystems product suite: the classification layer and four probability layers. + +#### Classification layer (classification) + +The primary layer of DEA Coastal Ecosystems is a categorical classification of the extent of mangrove, saltmarsh, intertidal and intertidal seagrass coastal ecosystems (Figure 1). The layer is built from thresholds of the ecosystem probability layers (below), and constrained, modified and masked with a range of ancillary datasets and QA/QC processes (See [Contextual Editing](#4-contextual-editing)). + +Annual classification layers are published with the following categorical class values: + +- Intertidal (2) +- Mangrove (3) +- Saltmarsh (4) +- Intertidal Seagrass (5) + +:::{figure} /_files/dea-coastalecosystems/classification_bermagui_2021_LTC.png + +**Figure 1.** Bermagui, NSW. Top: The 2021 low-tide [DEA Tidal Composite](/data/product/dea-tidal-composites/). Bottom: The 2021 DEA Coastal Ecosystems classification layer. + +::: + +#### Probability layers (mangrove-prob, saltmarsh-prob, saltflat-prob, seagrass-prob) + +Individual ecosystem probability layers are provided as supporting datasets to aid the interpretation of extents for four ecosystem classes: Mangrove, Saltmarsh, Saltflat and Intertidal Seagrass (Figure 2). Probability values in these layers represent the percentage of random-forest trees that vote at each pixel for a given ecosystem in the relevant [Machine Learning model](#model-training); the Ecosystem Model (Mangroves, Saltmarsh, Saltflat) or the Intertidal Model (Intertidal Seagrass). + +The Intertidal class within the Intertidal model has been fixed to the extent of the [DEA Intertidal Extents product](/data/product/dea-intertidal/?tab=description#core-product-layers) (see [Ecosystem Prediction](#ecosystem-prediction)), so the probability layer for this ecosystem is not provided. As the Intertidal Seagrass is modelled within this fixed extent, the corresponding probability layer is also constrained to these extents. + +Probabilities for layers in the Ecosystem Model (Mangroves, Saltmarsh and Saltflat) are masked below 20% probability to remove very low confidence pixels and noise from the dataset to improve interpretability. + +The Saltflat ecosystem has been excluded from the classification product layer, though the probability layer generated for this ecosystem as part of the Ecosystem Model has been provided. Reasoning and further discussion on this decision can be found in the [Caveats and Limitations](./?tab=quality) section. + +Further information on how these Probability layers can be individually interpreted and used to aid in interpretation of the classified ecosystem map, including for mixed or transitional classes, can be found in the [Caveats and Limitations](./?tab=quality). + +:::{figure} /_files/dea-coastalecosystems/probabilities_Bermagui_2021.png + +**Figure 2.** Bermagui, NSW. DEA Coastal Ecosystems probability layers. + +::: + +### Quality Assurance Layers + +Two quality assurance layers are included in the DEA Coastal Ecosystems product suite. + +#### Clear count layer (qa-count-clear) + +A layer showing the distribution of clear and valid satellite observations at each [Sentinel-2 pixel](/data/category/dea-surface-reflectance/) during the analysis period. Artefacts that can invalidate pixels include cloud cover/shadow and low geometric accuracy. A pixel only proceeds into the DEA Coastal Ecosystems workflow if it has more than 10 valid observations in a modelling year. + +#### Coastal connectivity layer (qa-coastal-connectivity) + +An accumulated cost-distance connectivity layer used as a [covariate](#coastal-connectivity) within the modelling process, and as an additional [contextual editing](#4-contextual-editing) layer for the Saltmarsh class. Values represent the cumulative elevation above Highest Astronomical Tide that must be traversed along the shortest path from tidally influenced coastal waters and mangroves. Lower values indicate likely coastal pixels, reflecting both distance inland and topography. The values of the coastal connectivity layer are dimensionless with low values representing high coastal connectivity and high values indicating lower coastal connectivity. + +### Data Format + +#### Cloud-optimised GeoTIFF + +Layers of the DEA Coastal Ecosystems product suite are provided as single continental scale cloud-optimised GeoTIFFs (COGs). For data access and use in geospatial information system (GIS) environments, [streaming](/guides/continental-cogs-geotiff-mosaics/) of these datasets is strongly recommended rather than downloading. Web mapping services (WMS) are also available for all [DEA datasets](./?tab=access). + +#### Product naming convention + +To stream COG versions of DEA data, you will need to identify the data layers by their full product name. The DEA standard [file naming convention](/guides/reference/collection_3_naming/) has been applied as follows: + +```text +{Organisation}_{Platform}_{Product}_{Reporting period}_{Collection}_{Version}_{Tile reference}_{Data date}--{Data period}_{Product status}_{Band name}.{File extension} +``` + +For example, the 2021 DEA Coastal Ecosystems classification layer is named + +```text +ga_s2_coastalecosystems_cyear_3_v1-0-0_AU_2021-P1Y_final_classification.tif +``` + +## Processing steps + +### Workflow + +DEA Coastal Ecosystems is a machine learning workflow comprised of four primary components (Figure 3): Training data, Covariate data, Machine learning and Contextual editing. + +:::{figure} /_files/dea-coastalecosystems/CEM_flowchart.png + +**Figure 3.** Primary components and outputs of the DEA Coastal Ecosystems workflow. +::: + +These components are discussed in further detail below. The validation process is discussed in the [Quality](./?tab=quality) section. + +### 1. Training data + +::::{dropdown} Data curation +[Continental training data](https://data.dev.dea.ga.gov.au/derivative/ga_s2_coastalecosystems_cyear_3_v1/auxiliaries/training_data/MultEcosy_TData_v1_0.geojson), developed for [this work](https://data.dev.dea.ga.gov.au/derivative/ga_s2_coastalecosystems_cyear_3_v1/auxiliaries/publications/JCU_Coastal_Training_Data_Report_1_27012023_FR.pdf) (Canto et al., 2023), comprised a point-record training set of 40,934 Australian coastal ecosystem type occurrences (Figure 4). Development of this dataset was completed for use with multi-ecosystem classification models. + +The dataset integrated occurrence records from four data sources: + +- Australia coastal ecosystem mapping (Becker et. al, 2023) +- Australia seagrass mapping (Navarro et. al, 2022) +- Australia saltmarsh mapping (Navarro et. al, 2023) +- Global intertidal change (Murray et. al, 2019; Murray et al., 2022) + +:::{figure} /_files/dea-coastalecosystems/Original_training_point_distribution.png + +**Figure 4.** (After Canto et al., 2023; Becker et al., 2023.) 40,934 expert labelled point occurrences of coastal ecosystem types, distributed continentally around the Australian coastline. +::: +:::: + +:::{dropdown} Ecosystem definitions +The training data definitions for Australian coastal ecosystems (Table 1; Canto et al., 2023) were sourced from the Blue Carbon Method (CER, 2022) and the International Union for Conservation of Nature Global Ecosystem Typology (Keith et al., 2022). + +Training point class descriptors (level 2, L2) were aggregated into broader level 1 (L1) class categories to form the basis of the DEA Coastal Ecosystems classification schema. The mapping of L2 to L1 ecosystem types, and their definitions, are described in Table 1. + +
+
Table 1. Ecosystem definitions and class mappings.
+
+ +| L2 ecosystem type | L1 ecosystem type | L1 Definition | +|-------------------|-------------------|---------------| +| • Land
• Supratidal forest
• Freshwater wetland
• Sand
• Water | Terrestrial/Other | Includes natural or anthropogenic land cover types that occur in non-intertidal areas but within the coastal zone such as supratidal forests, frequently inundated croplands, freshwater wetlands and permanent sand dunes. | +| • Seagrass (subtidal) | Permanent water | Refers to a variety of permanent water bodies around the coast but not within the intertidal zone. These water bodies range from deep permanent ocean to freshwater rivers and lakes and include clear and turbid waters. Includes submerged vegetation and substrates such as subtidal seagrass if the bottom is visible. | +| • Mudflat
• Intertidal non-vegetated soft sediments | Intertidal | Ecosystem occurring in low-sloping areas of the intertidal zone. Includes muddy, sandy and rocky substrates. In soft sediments, can readily transition to seagrass, mangrove or saltmarsh ecosystems if conditions become favorable for the colonisation of vegetation through changes in sea level or freshwater inundation over time. | +| • Mangrove | Mangrove | Ecosystem comprised of trees and shrubs which a) occupy the intertidal zone or floodplains, including marine and estuarine areas; and b) grow in saline or brackish water. | +| • Saltmarsh
• Algal Mats | Saltmarsh | Ecosystem comprised of a) salt-tolerant herbaceous plants and occasionally woody shrubs; and b) occur on floodplains and in estuaries and can be flushed with water from a combinations of sources including rainfall, rivers, groundwater and seawater. This includes saltmarsh in sparsely vegetated saline or hypersaline ecosystems. | +| • Seagrass (intertidal) | Intertidal seagrass | Ecosystem comprised of grass-like plants that grow in shallow intertidal coastal waters. Seagrasses are aquatic flowering plants that form meadows in temperate and tropical regions of Australia. | +| • Saltflat | Saltflat | A subtype of saltmarsh ecosystems comprising bare saltmarsh. Saltflats include sparsely vegetated saline or hypersaline ecosystems | +::: + +::::{dropdown} Regional modelling +A bioregional approach was used to improve modelling performance by accounting for regional ecological variations while maintaining the ability to apply a consistent classification schema. + +Seven bioregions (Figure 5) were developed with expert consultation across Geoscience Australia, University of Queensland, James Cook University and University of New South Wales, guided by Natural Resource Management regions, tidal regimes, climate forcing and ecosystem dynamics. + +These regions enabled models to be developed with regional specific training data to account for differences across the common ecosystem classes (e.g. the varied ecological nature of saltmarsh across the country), as well as regional climate and seasonal impacts on the covariate data inputs. + +It is important that these regions be regarded as a pragmatic technical choice rather than an attempt at a comprehensive national bioregionalization. + +:::{figure} /_files/dea-coastalecosystems/Labelled_tile_regions.png + +**Figure 5.** Seven biogregions were used to model coastal ecosystems. +::: +:::: + +:::{dropdown} Training Point Expansion +Introducing modelling regions had implications for the size and balance of the training data sets once split across the seven bioregions, with many ecosystem training point classes falling below a reasonable representative sample number. + +The development of the original training data set on Landsat 30 m data, coupled with an acquisition protocol focused on spatially homogenous regions, presented an opportunity to expand the training data sets when considering the increased resolution of the Sentinel-2-based product. With Sentinel-2's 10 m resolution pixels falling within the tolerance established for the training data point validity (30–40 m), a semi-automated process was implemented to expand the existing training dataset. This semi-automated approach was particularly effective for densifying training data coverage in homogeneous ecosystem patches while maintaining data quality. + +Training data densification was achieved by resampling the original training data points to 10 m resolution. By sampling the surrounding 9 pixels of each original training point, the overall number of training data points was increased to approximately 368,000. + +This approach balanced the need for increased training data volume across all seven modelling regions while maintaining data quality and ecological validity (Table 2). The process also included more spatial and spectral variability in the modelling process, resulting in a small but consistent improvement in [accuracy](./?tab=quality#accuracy) when models used the expanded training data set. + +
+
Table 2. Expanded training data distribution across model regions. Modelling regions: North West (NW), North East (NE), Gulf of Carpentaria (Gulf), West, South West (SW), South East (SE) and Tasmania (Tas.). Note that no mangroves or saltflat are recorded in Tasmania.
+
+ +| Region | Water | Mudflat | Mangrove | Saltmarsh | Saltflat | Terrestrial/Other | Intertidal seagrass | Total | +|--------|-------|---------|----------|-----------|----------|-------------------|---------------------|-------| +| NW | 4959 | 3135 | 7893 | 4641 | 5583 | 28734 | 186 | 55131 | +| NE | 9354 | 6424 | 8617 | 4473 | 3169 | 20664 | 5072 | 57773 | +| Gulf | 2754 | 2473 | 3960 | 3267 | 2526 | 7818 | 3464 | 26262 | +| West | 1638 | 2385 | 2241 | 2481 | 2898 | 12471 | 1765 | 25879 | +| SW | 2718 | 816 | 441 | 1926 | 126 | 46512 | 1377 | 53916 | +| SE | 9234 | 3891 | 7514 | 15512 | 422 | 64705 | 9734 | 111012 | +| Tas. | 2697 | 1359 | 0 | 3120 | 0 | 5622 | 1319 | 14117 | + +::: + +### 2. Covariate Data + +:::{dropdown} Covariate Data Stack +Covariate data describes the input datasets used to train the ML models. Covariate data stacks were generated using a combination of Sentinel-2 Analysis Ready Data, remote-sensing indices and derived datasets (Table 3). These covariate stacks of 121 total variables were generated for each of the modelling years. + +
+
Table 3. Covariate datasets used to model coastal ecosystems.
+
+ +| Data Source | Variable | Percentiles | Total Count | +|-----|-----|-----|-----| +|[Sentinel-2 ARD Reflectance](/data/category/dea-surface-reflectance/) | nbart_blue | 10,20,40,60,80 | 5 | +| | nbart_green | 10,20,40,60,80 | 5 | +| | nbart_red | 10,20,40,60,80 | 5 | +| | nbart_red_edge_1 | 10,20,40,60,80 | 5 | +| | nbart_red_edge_2 | 10,20,40,60,80 | 5 | +| | nbart_red_edge_3 | 10,20,40,60,80 | 5 | +| | nbart_nir_1 | 10,20,40,60,80 | 5 | +| | nbart_nir_2 | 10,20,40,60,80 | 5 | +| | nbart_swir_2 | 10,20,40,60,80 | 5 | +| | nbart_swir_3 | 10,20,40,60,80 | 5 | +| Sentinel-2-derived indices | | | | +|• Automated Water Extraction Index (no shadow) (Feyisa et al., 2014) | AWEI_ns | 10,20,40,60,80 | 5 | +|• Automated Water Extraction Index (shadow) (Feyisa et al., 2014) | AWEI_sh | 10,20,40,60,80 | 5 | +|• Enhanced Vegetation Index (Huete et al., 2002) | EVI | 10,20,40,60,80 | 5 | +|• Leaf Area Index (Boegh et al., 2002) | LAI | 10,20,40,60,80 | 5 | +|• Modified Normalized Difference Water Index (Xu, 1996) | MNDWI | 10,20,40,60,80 | 5 | +|• Normalized Difference Vegetation Index (Rouse et al., 1973) | NDVI | 10,20,40,60,80 | 5 | +|• Normalized Difference Water Index (McFeeters, 1996) | NDWI | 10,20,40,60,80 | 5 | +|• Tasseled Cap Brightness (Crist, 1985) | TCB | 10,20,40,60,80 | 5 | +|• Tasseled Cap Greenness (Crist, 1985) | TCG | 10,20,40,60,80 | 5 | +|• Tasseled Cap Wetness (Crist, 1985) | TCW | 10,20,40,60,80 | 5 | +|• Tasseled Cap Wetness (GSO variant) (Nedkov, 2017) | TCW_GSO | 10,20,40,60,80 | 5 | +|• Water Index (Fisher et al., 2016) | WI | 10,20,40,60,80 | 5 | +| [DEA Tidal Composites](/data/product/dea-tidal-composites/?tab=specifications) | low_blue | | 1 | +| | low_green | | 1 | +| | low_red | | 1 | +| | low_red_edge_1 | | 1 | +| | low_red_edge_2 | | 1 | +| | low_red_edge_3 | | 1 | +| | low_nir_1 | | 1 | +| | low_nir_2 | | 1 | +| | low_swir_2 | | 1 | +| | low_swir_3 | | 1 | +| Coastal Connectivity | Cost-distance | | 1 | +|Total Covariates | | | 121 | + +::: + +::::{dropdown} Coastal Connectivity +The Coastal Connectivity layer (also published as a [QA product layer](#quality-assurance-layers)) was designed as an accumulated cost-distance dataset to indicate likely coastal pixels. This was achieved by determining the cumulative elevation above Highest Astronomical Tide that must be traversed along the shortest path from tidally influenced coastal waters and mangroves at each pixel location (Figure 6), using the following datasets: + +- Shuttle Radar Topography Mission Digital Elevation Model Version 1 (Gallant et al., 2011). +- Highest Astronomical Tide ([Branson, 2023](https://data.csiro.au/collection/csiro%3A61319v1)). +- The extent of the [DEA Mangroves](/data/product/dea-mangroves/) product. + +Coastal connectivity is used here as a covariate layer within the modelling process, and as an additional contextual editing layer for the Saltmarsh class. + +:::{figure} /_files/dea-coastalecosystems/connectivity_Mallacoota.png + +**Figure 6.** a) Mallacoota, Victoria low-tide [DEA Tidal Composite](/data/product/dea-tidal-composites/) for 2021; b) Low elevation areas, highly connected to tidally-influenced coastlines, have low connectivity values (light grey). Elevated areas are penalised with high connectivity values (dark grey), even if located close to tidally influenced areas. +::: +:::: + +### 3. Machine Learning + +:::{dropdown} Training Point Attribution +For the v1.0.0 implementation of the model, the full expanded Training Point dataset was attributed with covariate data from both the 2021 and 2022 modelling year covariate stacks. Training points where covariates were derived from <10 clear observations in either modelling year were removed from the relevant modelling year dataset. + +Both of these attributed annual Training Point datasets were then combined to form the final datasets for Model training. +::: + +::::{dropdown} Model training +The DEA Coastal Ecosystems workflow trains two [random-forest classifiers](https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html) for each of the seven [modelled regions](#regional-modelling) from the covariate-attributed stack of training data: + +- a multi-class ecosystem model (mangrove, saltmarsh and salt flat) and; +- an intertidal model (intertidal and intertidal-seagrass). + +The multi-class ecosystem model utilized all data in the training data library while the Intertidal model was developed using only training data points located within the [DEA Intertidal Extents high confidence intertidal region](/data/product/dea-intertidal/#core-product-layers). Functionally, this meant that within the bounded extents of the Intertidal modelled region, any training points included in the model that were not defined as Intertidal Seagrass were considered Intertidal. For a discussion on the implications of this aspect of the model classification see the [Caveats and Limitations](./?tab=quality#constraint-of-the-intertidal-model-and-definition-of-the-intertidal-class) section. + +Multi-class and binary random-forest models were fit for each region using: + +- Number of trees (800): A large ensemble size ensures stable, reliable predictions by aggregating insights from many diverse trees, reducing the impact of any single tree's errors. +- Minimum samples per split (2): Set to the minimum allowed value to give the model maximum flexibility in finding predictive patterns. This is particularly useful when relationships in the data are complex or subtle. +- Minimum samples per leaf (1): Allows the model to capture fine-grained patterns and edge cases in the data without artificially constraining individual tree decisions. The large number of trees (800) provides natural protection against overfitting. +- Maximum depth (10): Limits tree complexity to prevent overfitting while still allowing enough depth to capture meaningful multi-level interactions between variables. Strikes a balance between model expressiveness and generalizability. + +A [k-fold cross-validation](https://scikit-learn.org/stable/modules/cross_validation.html#computing-cross-validated-metrics) approach was used to assess model behaviour and optimise the model parameters. Optimized model parameters were used to fit the final models using the full training data set for the multi-class model and the intertidal masked training data set for the seagrass model. A monte carlo resampling framework was used to validate the mapping products and calculate accuracy metrics including overall accuracy, class-based accuracy and relevant confidence intervals. Overall model accuracy statistics are reported in the [Quality](./?tab=quality#accuracy) tab. + +The combined 2021 and 2022 stacks of covariate-trained training data ensured the classifier models were suitable for prediction into both years. Common covariates that exhibited a top-10 feature importance for the ecosystem models included the NDVI and MNDWI percentiles as well as the coastal-connectivity. Common top-10 feature importance covariates for the seagrass models included the tasselled cap wetness percentiles and 'red_edge', 'nir' and 'swir' bands from the low-tide [DEA Tidal Composite](/data/product/dea-tidal-composites/) (Figure 7). + +:::{figure} /_files/dea-coastalecosystems/ML_Feature_Importance.png + +**Figure 7.** Commonality in the top 10 most important covariates for each of the seven regional ecosystem (top) and seagrass (bottom) models. The y-axis represents the number of regional models where the covariate features in the top 10. +::: +:::: + +:::{dropdown} Ecosystem prediction +Coastal ecosystems were predicted using Covariate data stacks created for each modelling year. + +Initial predictions were made using the multi-class ecosystem model, generating interim probability layers for mangrove, saltmarsh and salt flat ecosystems ([Figure 3](#workflow)). An interim classified ecosystem output layer was also generated to capture the highest probability ecosystem class predicted for each pixel and included mangrove, saltmarsh and saltflat classes. Water and Terrestrial classes in this interim layer were classified as `nodata`. + +The Intertidal modelling extent was then set using the high confidence intertidal extent from the [DEA Intertidal](/data/product/dea-intertidal/#core-product-layers) dataset. The Intertidal model was then used to predict Intertidal seagrass from the annual covariate stacks masked to this modelling extent. The output from this process is the interim Intertidal seagrass probability layer (Figure 3). +::: + +### 4. Contextual editing + +:::{dropdown} Rationale +The interim classified ecosystem layer defines classes based on the highest probability class for each pixel, which can result in a classification based on relatively low probability values were a pixel is modelled as a likely mixed ecosystem. Combined with the inherent uncertainty and resulting misclassification when applying a machine learning mapping methodology at a continental scale, this meant that a range of rulesets and post-processing operations were applied, collectively termed ‘contextual editing’, to deliver the final Classification layer. + +Probability layers were also masked in this contextual editing process, as described in the [probability layer description](#probability). + +The following section details the rules and processes that were applied, with justifications or links to further discussion points. +::: + +:::{dropdown} Classifier rulesets + +All classifier rulesets were applied to the interim ecosystem classification layer ([Figure 3](#workflow)) to produce the final Classification dataset. + +1. **Remove unused classes**: The ecosystem classification output layer was masked to remove all saltflat pixels (see [Caveats and Limitations](./?tab=quality#saltflat-mapping)). +1. **Apply probability thresholds**: A probability threshold of 50% was applied to the Saltmarsh and Mangrove pixels, with pixels below this threshold reassigned as `nodata`. A 50% probability threshold was selected based on the alignment of the outputs with external mapping products, and the conservative consideration that 50% provides a clear majority vote by the machine learning models. +1. **Apply environmental corrections**: Saltmarsh pixels were masked to the observed coastal-connectivity range identified by the saltmarsh training data points. To exclude outliers, the saltmarsh connectivity mask was thresholded to the 99.5th percentile connectivity value (254) for all region-combined saltmarsh points from the training data set. This choice was made to remove misclassifications of saltmarsh in regions considered unconnected to coastal regions (see [Coastal Connectivity](#coastal-connectivity)). See [Caveats and Limitations](./?tab=quality#connectivity-masking-and-saltmarsh) for further discussion on the implications of thresholding based on existing distribution of Saltmarsh training data. + +::: + +:::{dropdown} Classifier masking + +1. **Apply mangrove mask**: Mangrove pixels that fell outside of the [Global Mangrove Watch (GMW) Habitat Mask](https://doi.org/10.5281/zenodo.74784913) (Bunting et al. 2025) were removed. This habitat mask defines an ecologically suitable niche for Mangrove occupation, and was used here to align our Mangrove mapping with global approaches and to minimize false-positive Mangrove classification in inhospitable environments. +1. **Apply Intertidal seagrass probability threshold**: Within the defined Intertidal model extent, the Intertidal seagrass probability layer was used to define the Intertidal seagrass classification at a probability of greater than or equal to 70%. The 70% threshold was selected as a conservative value to account for inter-annual variability in seagrass meadows (see [Caveats and Limitations](./?tab=quality#interpretation-of-probability-layers-and-implications-for-change-applications)). +1. **Apply manual mask**: A manual masking process was applied to the Classification layer, based on expert-identified misclassifications (including roads, urban areas, terrain shadow, and data noise). For full traceability, this polygon masking layer is provided here: [manual mask](https://data.dev.dea.ga.gov.au/derivative/ga_s2_coastalecosystems_cyear_3_v1/auxiliaries/contextual_editing/cem_manual_edits_3577_v1.parquet). +1. **Apply land use mask**: Industrial, urban or road areas as identified in the Australian Catchment Scale Land Use and Management dataset (ABARES, 2021) were removed. +1. **Remove and replace isolated pixel groups**: To reduce pixel-based noise in the data, the Classification layer was sieved to remove and replace isolated pixels in groups of 9 or less with the dominant surrounding class type. + +::: + +:::{dropdown} Probability masking +1. **Apply probability threshold**: Pixels with values of less than 20% were removed from the Mangrove, Saltmarsh and Saltflat probability layers to remove very low confidence pixels and noise from the dataset to improve interpretability. + +::: + +## Software + +- The DEA Coastal Ecosystems workflow leverages functions from [DEA Tools](/notebooks/Tools/) that have dependencies on [Scikitlearn](https://scikit-learn.org/stable/). +- The [DEA Tools](/notebooks/Tools/) library contains essential functionality required to generate random-forest models using [DEA Analysis Ready Data](/data/category/dea-surface-reflectance/). +- The [Scikitlearn](https://scikit-learn.org/stable/) library provided the random-forest classification methodologies. + +## References + +ABARES 2021, Catchment Scale Land Use of Australia – Update December 2020, Australian Bureau of Agricultural and Resource Economics and Sciences, Canberra, February, CC BY 4.0, DOI: 10.25814/aqjw-rq15 + +Branson, Paul (2023): Coastal carbon - Australia's blue forest future - Water Levels. v1. CSIRO. Data Collection. https://doi.org/10.25919/6672-jx11 + +Becker, M.L., Navarro, A., Canto, R., & Murray, N.J. (2023) [An Australian coastal ecosystem training library for remote sensing classification models.](https://github.com/GeoscienceAustralia/dea-coastalecosystems/blob/main/docs/publications/JCU_Coastal_Training_Data_Report_1_27012023_FR%20.pdf) Unpublished report to the Department of Climate Change, Energy, the Environment and Water. James Cook University, Townsville. + +Boegh, E., Soegaard, H., Broge, N., Hasager, C. B., Jensen, N. O., Schelde, K., & Thomsen, A. (2002). Airborne multispectral data for quantifying leaf area index, nitrogen concentration, and photosynthetic efficiency in agriculture. Remote sensing of Environment, 81(2-3), 179-193. + +Bunting, P., Rosenqvist, A., Hilarides. (2025) Global Mangrove Watch: Mangrove Habitat Mask v27.0 Dataset. https://doi.org/10.5281/zenodo.7478491 + +Canto. R., Navarro., A.,Becker, M., Murray, N.J., (2023) [Australian coastal ecosystem training library](https://github.com/GeoscienceAustralia/dea-coastalecosystems/blob/main/data/training_data_input/MultEcosy_TData_v1_0.geojson). Unpublished dataset. + +Crist, E. P. (1985). A TM Tasseled Cap equivalent transformation for reflectance factor data. Remote Sensing of Environment, 17(3), 301–306. https://doi.org/10.1016/0034-4257(85)90102-6 + +Feyisa, G. L., Meilby, H., Fensholt, R., & Proud, S. R. (2014). Automated Water Extraction Index: A new technique for surface water mapping using Landsat imagery. Remote sensing of environment, 140, 23-35. + +Fisher, A., Flood, N., & Danaher, T. (2016). Comparing Landsat water index methods for automated water classification in eastern Australia. Remote Sensing of Environment, 175, 167-182. + +Gallant, J.C., Dowling, T.I., Read, A.M., Wilson, N., Tickle, P. and Inskeep, C., 2011. 1 second SRTM Derived Digital Elevation Models User Guide. Geoscience Australia www.ga.gov.au/topographic-mapping/digital-elevationdata.html + +Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote sensing of environment, 83(1-2), 195-213. + +Keith, D.A., Ferrer-Paris, J.R., Nicholson, E. et al. (2022) A function-based typology for Earth’s ecosystems. Nature 610, 513–518 [https://www.nature.com/articles/s41586-022-05318-4](https://doi.org/10.1038/s41586-022-05318-4). + +Lovelock, C.E., Sippo, J., Fernanda Adame, M., Dittmann, S., Hickey, S., Hutley, L., Jones, A., Kelleway, J., Lavery, P., Macreadie, P., Maher, D., Mosely, L. and Rogers, K (2021) [Blue carbon accounting model (BlueCAM) technical overview](https://cer.gov.au/document/blue-carbon-accounting-model-bluecam-technical-overview). Report to the Clean Energy Regulator. University of Queensland. + +Lyons, M.B., Keith, D.A., Phinn, S.R., Mason, T.J., & Elith, J. (2018). A comparison of resampling methods for remote sensing classification and accuracy assessment. Remote Sensing of Environment, 208, 145-153. https://doi.org/10.1016/j.rse.2018.02.026 + +McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International journal of remote sensing, 17(7), 1425-1432. + +Murray N. J., Canto, R., Worthington, T. A., Lyons, M.B. (2022) Supplementary Data S1 to Losses and gains of Earth's tidal wetlands. Figshare. https://doi.org/10.6084/m9.figshare.19121660. + +Murray, N. J., Phinn, S. R., DeWitt, M., Ferrari, R., Johnston, R., Lyons, M. B., Clinton, N., Thau, D. & Fuller, R. A. (2019). The global distribution and trajectory of tidal flats. Nature, 565, 222-225 + +Murray, N.J., et al., High-resolution mapping of losses and gains of Earth’s tidal wetlands. Science 376, 744-749(2022). DOI: [10.1126/science.abm9583](https://www.science.org/doi/10.1126/science.abm9583) + +Navarro, A., Canto, R., Toor, M., & Murray, N.J. (2022) An Australian intertidal seagrass training library for remote sensing classification models. Unpublished report to the Department of Agriculture, Water and the Environment. James Cook University, Townsville. + +Navarro, A., Canto, R., Toor, M., Becker, M.L., Lyons, M. & Murray, N.J. (2023) Australian Saltmarsh Map. Unpublished report to the Clean Energy Regulator. James Cook University, Townsville. + +Nedkov, R., (2017). “Orthogonal transformation of segmented images from the satellite Sentinel-2,” Comptes Rendus De lAcad. Bulgare Sci., vol. 70, no. 5, pp. 687–692. + +Rouse Jr, J. W., Haas, R. H., Schell, J. A., & Deering, D. W. (1973). Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation (No. NASA-CR-132982). + +Xu, H. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International journal of remote sensing, 27(14), 3025-3033. + diff --git a/docs/data/product/dea-coastal-ecosystems/_faqs.md b/docs/data/product/dea-coastal-ecosystems/_faqs.md new file mode 100644 index 000000000..0be768310 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_faqs.md @@ -0,0 +1,4 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +% ## Frequently asked questions + diff --git a/docs/data/product/dea-coastal-ecosystems/_history.md b/docs/data/product/dea-coastal-ecosystems/_history.md new file mode 100644 index 000000000..6b1566294 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_history.md @@ -0,0 +1,8 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +## Changelog + +### DEA Coastal Ecosystems 1.0.0 + +In December 2025, [version 1.0.0 was released](https://github.com/GeoscienceAustralia/dea-coastalecosystems/releases/tag/1.0.0) of the DEA Coastal Ecosystems product suite. This version release included annual products for the 2021 and 2022 calendar years, to support National Ecosystem Accounts requirements of the Commonwealth Government. + diff --git a/docs/data/product/dea-coastal-ecosystems/_overview_1.md b/docs/data/product/dea-coastal-ecosystems/_overview_1.md new file mode 100644 index 000000000..6cf904a20 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_overview_1.md @@ -0,0 +1,7 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +## About + +The DEA Coastal Ecosystems product suite maps the extent of coastal ecosystems within Australia’s coastal zone, with a focus on those situated landward of the low-tide boundary. + +Developed using Sentinel-2 data, the product suite provides an annual 10 m resolution extent classification layer for four ecosystem classes, enabling users to better monitor and understand the distribution of these ecosystems at a continental scale. The classification layer is supported by ecosystem probability layers for each class, to allow a more nuanced interpretation of the modelled ecosystem extents and to provide users flexibilty for tailored mapping approaches. \ No newline at end of file diff --git a/docs/data/product/dea-coastal-ecosystems/_overview_2.md b/docs/data/product/dea-coastal-ecosystems/_overview_2.md new file mode 100644 index 000000000..b2021cf58 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_overview_2.md @@ -0,0 +1,4 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +% ## Publications + diff --git a/docs/data/product/dea-coastal-ecosystems/_quality.md b/docs/data/product/dea-coastal-ecosystems/_quality.md new file mode 100644 index 000000000..e1f5012ff --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_quality.md @@ -0,0 +1,177 @@ +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +## Caveats and Limitations + +### Accuracy Considerations + +- **Training data composition:** Regional training sets remain unbalanced for rare classes (e.g. [salt flat, seagrass in Region 1](./?tab=description#training-data)), which contributes to wider accuracy intervals and should be considered when interpreting those classes. +- **Probability masking:** Accuracy statistics only include samples with greater than or equal to 50% model probability. Areas with persistent cloud, haze, or spectral ambiguity may be masked operationally and will not have associated accuracy estimates. +- **Spatial extrapolation:** Spatial stratification reduces autocorrelation bias, but accuracy can still degrade in locations far from labelled samples or outside the environmental gradients represented in the combined parquet. +- **Model uncertainty:** Reported medians and 95% intervals reflect Monte Carlo sampling only; additional uncertainty from sensor noise, ancillary covariates, or downstream post-processing is not captured here. + +### Sensor Tidal Bias + +- **Sensor coverage and tidal phase:** Sentinel 2 acquisitions do not uniformly sample the [tidal cycle](/data/product/dea-intertidal/?tab=quality#caveats-and-limitations), so in some regions intertidal habitats may be under-represented or not fully characterised in either the training data or the predictions. For example, in a region where the highest portion of the tidal range/cycle is not observed by the sensor, training data and the associated covariates for classes such as saltmarsh may be biased towards a lower tide representation of that class. +- **Intertidal Seagrass Training Points:** Due to sensor tidal bias, some Intertidal Seagrass training points fall outside of the Intertidal Extent that is observed by the sensor, and which is used to constrain the extent of Intertidal/Seagrass model. These training data points are excluded from the model, as they fall below the Lowest Observed Tide of the sensor, meaning they will only be observed as sub-tidal seagrass points. + +### Saltflat Mapping + +Saltflat mapping accuracy results in the Classification layer were observed to be highly variable between the regional models. Reasons for this may include: + +- limited proportional representation of Training Points for this class for regions where Saltflat is known to exist (e.g. in the South-East), +- lack of similar terrestrial class training points for differentiation resulting in false positives across Urban and Sand/Beach areas, and +- issues with cloud masking of highly reflective bright surfaces. + +For these reasons, the Saltflat class has been removed from the core Classification layer, so these issues can be addressed in future versions of the product. The Saltflat probability layer has been retained, so users can make their own interpretation of the mapping extents based on local and regional knowledge, whilst considering the issues highlighted above. + +### Constraint of the Intertidal Model and Definition of the Intertidal Class + +To align with the extents of the DEA Intertidal product, the extent of the Intertidal/Seagrass model has been constrained to the high-confidence DEA Intertidal Extents product. All training data points that are located in this extent, and are not Intertidal Seagrass, are considered Intertidal for the purposes of the binary Intertidal/Seagrass model. + +By constraining the extent of this model, we are forcing the classification of pixels in this Intertidal class that likely represent L2 classes beyond Mudflats and Intertidal soft sediments that are not represented in the training data (e.g. rocky intertidal substratum and exposed coral reefs/platforms). This may have implications on the binary classification of the Intertidal Seagrass in the model and is an additional reason why we have conservatively mapped this class at a 70% probability in the classification model. + +Future work will involve densification of the training data to include a more fully representative range of L2 classes considered in this Intertidal class. + +### Connectivity Masking and Saltmarsh + +The Saltmarsh connectivity thresholding assumes that the training data (which are located in predominantly near-shore coastal locations) encompass all possible saltmarsh locations and connectivity values. The limitations of this approach are acknowledged, as demonstrated by hard boundaries visible in the classified saltmarsh dataset such as in Princess Charlotte Bay, Qld. Future work will broaden the spatial range of saltmarsh training data that will improve the connectivity masking threshold used in this part of the workflow. + +### Global Mangrove Watch Masking + +The [Global Mangrove Watch (GMW) Habitat Mask](https://doi.org/10.5281/zenodo.74784913) (Bunting et al. 2025) is widely used in the coastal mapping and scientific community as a layer to constrain the mapping of Mangroves to suitable ecological conditions and zones. It is however acknowledged as a living dataset, with updates contributed by the community of new identified mangroves and mapping extents. For example, in the early stages of this project, new regions not captured by the GMW were identified in Batemans Bay, NSW, and these regions were added by GMW to the most recent version of the Habitat mask. + +Although the Classification layer has been masked by the GMW Habitat Mask, the Mangroves probability layer has not, and it is possible that users may be able to identify small areas of high-probability Mangroves not mapped by the masked classification layer. Users are encouraged to provide DEA with this feedback, so that we can continue to help the GMW initiative to provide the most suitable masking and mapping layers. + +### Interpretation of Probability Layers and Implications for Change Applications + +**Mixed and Transitional ecosystems:** Many ecosystems exist as highly mixed types or transitional ecosystems. These types of ecosystems, particularly when they exist in relatively even-cover proportions, are not well represented by categorical mapping products. These zones are often not captured in the 50% probability thresholded Classification layer, and may for example consist of pixels that are modelled with 34% Mangrove, 40% Saltmarsh and 26% Saltflat Probabilities. + +**Probability Thresholding and Interpretation:** The 50% probability threshold applied to the Classification of Saltmarsh and Mangroves in the Classification layer is designed to provide a high-confidence model of the dominant ecosystem in each coastal pixel. As described above, in mixed or transitional pixels not mapped by the classification layer; or in instances where local knowledge of ecosystem characteristics or vegetation density/canopy cover requires a more nuanced approach, the Ecosystem probability layers are designed to support this interpretation. + +Intertidal Seagrass is thresholded in the Classified layer at a 70% probability, primarily to account for inter-annual variability in this ecosystem, aimed at mapping regions of persistent cover over time. It is expected that the Probability layers for this ecosystem are used by seagrass mapping practitioners to enable a more detailed analysis of specific regions or ecosystem characterisation. + +**Implications for Change Applications**: Change-based applications utilising the Classification layers are to be approached with caution: + +- For classified ecosystems that are mapped at a probability of greater than 50%, these classes are still often in transition and may shift across those thresholds from year to year. Using the classified layer could potentially note the loss (or gain) of that ecosystem erroneously. Users should make use of the model probability layers to identify these areas and make case-specific judgement on usage. +- In mixed or transitional ecosystems it is expected that change metrics will be highly sensitive and potentially erroneous in these cases. Again, users should make use of the model probability layers to identify these areas and make case-specific judgement on usage. + +It is important to note that annual models of ecosystem extents are influenced by: + +- the seasonal characteristics of the year being modelled, +- the currency of the training data being applied to the modelling year, and +- the quality and number of EO datasets acquired and available over the modelling year. + +Whilst the modelling process in this product has been designed to mitigate these influences and differences between annual products where possible, each of these factors impacts the modelling accuracy and characteristics of both the probability and classified layers of the product for each year. This should be considered when using any of the Product layers for change applications, and we recommend utilising the full suite of QA/QC layers and Accuracy results to assist with any change applications or metrics. + +## Accuracy + +The DEA Coastal Ecosystems mapping was validated using a Monte Carlo resampling workflow with spatial stratification (32 km tiles) to limit spatial autocorrelation (Lyons et al. 2018). Each iteration trains on ~80% of tiles, tests on the remainder, filters predictions by a greater-than-or-equal-to 50% class-probability threshold (to replicate the mapping methodology), and then reports accuracy for both the ecosystem model and the final output after seagrass modelling is applied to the tidal flat area. The validation metrics are thus indicative of the accuracy of the final categorical classification product, and are not strictly relevant to the model probability layers. Final accuracy metrics are summarised from the sampling distributions as the median with a 95% interval (which can be used as area multipliers if desired). Validation statistics are reported separately for the 2021 and 2022 mapping. + +Comprehensive methodology notes are documented in the code repository (see the [Validation README](https://data.dev.dea.ga.gov.au/derivative/ga_s2_coastalecosystems_cyear_3_v1/auxiliaries/validation/Validation%20readme.pdf)), and the complete per-class/per-year outputs are available in the [full validation results](https://data.dev.dea.ga.gov.au/derivative/ga_s2_coastalecosystems_cyear_3_v1/auxiliaries/validation/coastal_ecosystem_validation_results.pdf). + +### Regional summary per year + +#### Year 2021 + +| Region | Overall Accuracy (Median [95% interval]) | +|--------|----------------------------------------| +| 1 | 0.949 [0.909, 0.971] | +| 2 | 0.915 [0.880, 0.944] | +| 3 | 0.883 [0.793, 0.930] | +| 4 | 0.932 [0.899, 0.970] | +| 5 | 0.973 [0.939, 0.994] | +| 6 | 0.947 [0.915, 0.970] | +| 7 | 0.941 [0.878, 0.976] | + +#### Year 2022 + +| Region | Overall Accuracy (Median [95% interval]) | +|--------|----------------------------------------| +| 1 | 0.945 [0.901, 0.968] | +| 2 | 0.911 [0.871, 0.945] | +| 3 | 0.875 [0.785, 0.926] | +| 4 | 0.936 [0.904, 0.969] | +| 5 | 0.972 [0.939, 0.993] | +| 6 | 0.945 [0.914, 0.970] | +| 7 | 0.946 [0.888, 0.979] | + +### Regional summary per class (combined years) + +#### Region 1 (North West) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.874 [0.671, 0.979] | 0.947 [0.859, 0.986] | +| Mangrove | 0.989 [0.960, 0.999] | 0.982 [0.963, 0.995] | +| Saltmarsh | 0.909 [0.512, 0.989] | 0.634 [0.348, 0.821] | +| Seagrass | 0.000 [0.000, 0.933] | 0.000 [0.000, 0.033] | + +#### Region 2 (North East) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.780 [0.641, 0.909] | 0.928 [0.839, 0.980] | +| Mangrove | 0.971 [0.933, 0.995] | 0.978 [0.934, 0.993] | +| Saltmarsh | 0.808 [0.537, 0.951] | 0.573 [0.417, 0.777] | +| Seagrass | 0.883 [0.746, 0.969] | 0.668 [0.433, 0.827] | + +#### Region 3 (Gulf of Carpentaria) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.620 [0.393, 0.843] | 0.870 [0.711, 0.977] | +| Mangrove | 0.989 [0.961, 1.000] | 0.990 [0.973, 1.000] | +| Saltmarsh | 0.904 [0.455, 0.983] | 0.764 [0.605, 0.913] | +| Seagrass | 0.891 [0.664, 0.983] | 0.685 [0.296, 0.897] | + +#### Region 4 (West) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.845 [0.671, 0.970] | 0.982 [0.930, 1.000] | +| Mangrove | 0.993 [0.907, 1.000] | 0.990 [0.941, 1.000] | +| Saltmarsh | 0.864 [0.582, 0.980] | 0.805 [0.501, 0.951] | +| Seagrass | 0.929 [0.066, 1.000] | 0.768 [0.059, 1.000] | + +#### Region 5 (South West) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.858 [0.000, 1.000] | 0.873 [0.000, 1.000] | +| Mangrove | 0.992 [0.000, 1.000] | 0.897 [0.000, 1.000] | +| Saltmarsh | 0.858 [0.608, 0.975] | 0.632 [0.294, 0.904] | +| Seagrass | 0.867 [0.104, 1.000] | 0.896 [0.306, 1.000] | + +#### Region 6 (South East) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.865 [0.701, 0.956] | 0.923 [0.791, 0.987] | +| Mangrove | 0.949 [0.843, 0.993] | 0.951 [0.813, 0.982] | +| Saltmarsh | 0.919 [0.816, 0.968] | 0.786 [0.584, 0.908] | +| Seagrass | 0.934 [0.859, 0.984] | 0.941 [0.874, 0.977] | + +#### Region 7 (Tasmania) + +| Class | User's Accuracy | Producer's Accuracy | +|-------|-----------------|---------------------| +| Tidal flat | 0.947 [0.786, 1.000] | 0.964 [0.654, 1.000] | +| Saltmarsh | 0.926 [0.729, 0.996] | 0.894 [0.684, 0.975] | +| Seagrass | 0.969 [0.519, 1.000] | 0.927 [0.764, 1.000] | + +## Quality assurance + +Code used to generate the DEA Coastal Ecosystem Map is run against automated integration tests to ensure that product quality is maintained as updates and improvements are made. These tests verify that the entire product generation workflow is performing as expected and track changes in product accuracy over time. + +% ## Known issues + +% :::{dropdown} 1 Jan 2020: Example issue +% :open: +% +% This uses the `:open:` option because it is an open issue. +% ::: + +## References + +Lyons, M.B., Keith, D.A., Phinn, S.R., Mason, T.J., & Elith, J. (2018). A comparison of resampling methods for remote sensing classification and accuracy assessment. *Remote Sensing of Environment*, 208, 145-153. + diff --git a/docs/data/product/dea-coastal-ecosystems/_tables.yaml b/docs/data/product/dea-coastal-ecosystems/_tables.yaml new file mode 100644 index 000000000..2bf373f59 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/_tables.yaml @@ -0,0 +1,60 @@ +# See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +# Bands + +bands_footnote: "For more information on these bands, see the `Description tab <./?tab=description>`_." + +bands_table: + - name: classification + aliases: [] + resolution: 10 m + nodata: 255 + units: Class + type: uint8 + description: "A high confidence categorical classifier of coastal ecosystem types. Classes: 2 = intertidal; 3 = mangrove; 4 = saltmarsh; 5 = intertidal seagrass." + - name: mangrove-prob + aliases: [] + resolution: 10 m + nodata: 255 + units: Percent + type: uint8 + description: A probability distribution for mangroves. + - name: saltmarsh-prob + aliases: [] + resolution: 10 m + nodata: 255 + units: Percent + type: uint8 + description: A probability distribution for saltmarshes. + - name: saltflat-prob + aliases: [] + resolution: 10 m + nodata: 255 + units: Percent + type: uint8 + description: A probability distribution for saltflats. + - name: seagrass-prob + aliases: [] + resolution: 10 m + nodata: 255 + units: Percent + type: uint8 + description: A probability distribution of seagrass inside the DEA Intertidal high confidence extent. + - name: qa-count-clear + aliases: + - count-clear + resolution: 10 m + nodata: 255 + units: Count + type: uint8 + description: The count of clear and valid observations per pixel. + - name: qa-coastal-connectivity + aliases: + - coastal-connectivity + resolution: 10 m + nodata: 65535 + units: Cost distance + type: uint16 + description: Accumulated cost-distance coastal connectivity, used to identify likely coastal pixels. + + diff --git a/docs/data/product/dea-coastal-ecosystems/index.rst b/docs/data/product/dea-coastal-ecosystems/index.rst new file mode 100644 index 000000000..bdd749e41 --- /dev/null +++ b/docs/data/product/dea-coastal-ecosystems/index.rst @@ -0,0 +1,4 @@ +.. See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html + +.. datatemplate:nodata:: + :template: product-v2.rst diff --git a/docs/data/theme/sea-ocean-and-coast/_data.yaml b/docs/data/theme/sea-ocean-and-coast/_data.yaml index 7a2773a15..a1dc735d8 100644 --- a/docs/data/theme/sea-ocean-and-coast/_data.yaml +++ b/docs/data/theme/sea-ocean-and-coast/_data.yaml @@ -2,12 +2,16 @@ title: Sea, Ocean, and Coast description: Explore Australia's marine and coastal environments, including dynamic coastlines and the intertidal zone. cards: + - name: DEA Coastal Ecosystems + link: /data/product/dea-coastal-ecosystems/ + image: /_files/dea-coastalecosystems/classification_Shoalwater_2021_thumbnail.png - name: DEA Coastlines link: /data/product/dea-coastlines/ image: /_files/cmi/thumbnail_vertical.png - - name: DEA Tidal Composites - link: /data/product/dea-tidal-composites/ - image: /_files/dea-tidal-composites/DEA_Tidal_Composite_Low_True-colour_2022_Montgomery-Reef_WA_cropped.png - name: DEA Intertidal link: /data/product/dea-intertidal/ image: /_files/dea-intertidal/DEAIntertidal_prettypic_Elevation_ShoalPointQLD_XS_Size.png + - name: DEA Tidal Composites + link: /data/product/dea-tidal-composites/ + image: /_files/dea-tidal-composites/DEA_Tidal_Composite_Low_True-colour_2022_Montgomery-Reef_WA_cropped.png + diff --git a/docs/guides/continental-cogs-geotiff-mosaics.rst b/docs/guides/continental-cogs-geotiff-mosaics.rst index 3076bd20e..98b4bb0e2 100644 --- a/docs/guides/continental-cogs-geotiff-mosaics.rst +++ b/docs/guides/continental-cogs-geotiff-mosaics.rst @@ -12,6 +12,7 @@ Not all DEA products provide continental-scale COGs. Currently, this access meth - `DEA Land Cover `_ - `DEA Intertidal `_ - `DEA Tidal Composites `_ +- `DEA Coastal Ecosystem Map `_ **VRT (Virtual Raster) files** are provided alongside the ``.tif`` mosaics. These files serve as lightweight wrappers around the main data and can be used to open data in GIS software with visual settings already applied. We use VRTs to provide: @@ -27,6 +28,7 @@ To stream DEA COGs directly in **QGIS**, follow these steps: - `Land Cover COGs `__ - `Intertidal COGs `__ - `Tidal Composites COGs `__ + - `Coastal Ecosystem Map COGs `_ 2. Choose a specific year (e.g. ``2024--P1Y``). diff --git a/docs/table_of_contents.yaml b/docs/table_of_contents.yaml index d9ccd7244..11bc3a8eb 100644 --- a/docs/table_of_contents.yaml +++ b/docs/table_of_contents.yaml @@ -129,10 +129,12 @@ entries: - file: data/theme/sea-ocean-and-coast/index title: All + - file: data/product/dea-coastal-ecosystems/index - file: data/product/dea-coastlines/index + - file: data/product/dea-intertidal/index - file: data/product/dea-tidal-composites/index title: DEA Tidal Composites - - file: data/product/dea-intertidal/index + - caption: Hazards entries: diff --git a/starter-kits/product/_access.md b/starter-kits/product/_access.md index 339d68e57..db6894f70 100644 --- a/starter-kits/product/_access.md +++ b/starter-kits/product/_access.md @@ -1,4 +1,4 @@ -% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/public_services/dea_knowledge_hub/product_metadata_fields.html +% See the Product metadata fields documentation: https://docs.dev.dea.ga.gov.au/services/dea-knowledge-hub/operating-procedures/product-metadata-fields/#kh-product-metadata-fields % :::{dropdown} How to access the data %