Seabed Mud Content Simulation for Australian Margin Interpolation

A simulation experiment from Geoscience Australia compares statistical and mathematical techniques for spatial interpolation of seabed mud content. The study analyzes factors including regions, sample densities, and secondary variables like bathymetry and distance-to-coast to assess prediction accuracy using cross-validation metrics. It identifies a novel combined method, random forest and ordinary kriging (RKrf), as the most robust, achieving up to 17% lower relative mean absolute error than a control method.

Use Cases

• Compare the prediction accuracy of spatial interpolation methods like inverse distance squared, random forest, and ordinary kriging across different Australian regions.
• Analyze the effect of sample density and search window size on mean absolute error and root mean square error for seabed mud content prediction.
• Use secondary variables bathymetry, distance-to-coast, and slope as covariates in a random forest model to improve spatial predictions of physical properties.
• Assess the robustness of the RKrf combined method by examining its relative mean absolute error (RMAE) performance across north, northeast, and southwest regions.

Strengths

• Study evaluates five key factors affecting interpolation accuracy: regions, methods, sample densities, searching neighbourhoods, and sample stratification.
• Prediction accuracy is validated using ten-fold cross-validation and metrics including mean absolute error and relative mean absolute error (RMAE).
• Identifies a novel combined method (RKrf) with a relative mean absolute error up to 17% less than a control method and up to 30% lower than previously published best methods in one region.

Limitations

• The underlying dataset size, row count, and specific sample locations are not provided, limiting reproducibility of the simulation.
• Primary data outputs like prediction maps or interpolated grids are not directly available as tabular data; results are described in PDF and HTML formats.
• The simulation is based on samples from the Geoscience Australia Marine Samples database, which may contain inherent data noise as noted in the study.

Provenance

Source

Geoscience Australia Data

Collection Method

Simulation experiment using samples from the Geoscience Australian Marine Samples database, with data quality control criteria applied and accuracy assessed via cross-validation.

Time Range

null

Freshness

null

Geography

Australian marine margin, specifically north, northeast, and southwest regions.