Australian Seabed Sand Content Predictions from Machine Learning Models

Geoscience Australia's Marine Samples Database provided data for a 2010 study comparing methods to predict seabed sand content across the Australian Exclusive Economic Zone (AEEZ). The research evaluated 18 spatial interpolation and machine learning methods, including RFIDS and RFOK, across three distinct regions. Model averaging was found to improve prediction accuracy, with the best methods reducing error by up to 7%.

Use Cases

• Compare prediction accuracy of RFIDS, RFOK, and 6RFIDS methods for sand content across three Australian marine regions.
• Apply model averaging techniques to improve spatial interpolation accuracy for seabed sediment variables.
• Analyze the impact of search window size and mtry parameters on the realism of sand content predictions.
• Validate geostatistical predictions against the 2010 sample data from the Marine Samples Database for the AEEZ.

Strengths

• Evaluation of 18 distinct spatial interpolation and machine learning methods.
• Analysis covers three separate geographic regions within the Australian Exclusive Economic Zone.
• Model averaging demonstrated a concrete improvement in prediction accuracy.

Limitations

• Underlying sample data is from a single collection date in August 2010, limiting temporal analysis.
• Specific sample count, row count, and geographic coordinates for the source data are not provided.
• Primary deliverables are PDF and HTML reports, not the raw feature-level dataset used for modeling.

Provenance

Source

Australian Ocean Data Network, utilizing Geoscience Australia's Marine Samples Database.

Collection Method

Samples extracted from a marine database for a comparative study of spatial interpolation methods.

Time Range

Sample data from August 2010.

Freshness

Source data collected August 2010; metadata record last updated April 2026.

Geography

Australian Exclusive Economic Zone (AEEZ), analyzed across three specific regions.