Selecting optimal random forest models for predicting the spatial distribution of seabed hardness using underwater video images and multibeam backscatter data  

Authors / CoAuthors

Li, J. | Tran, M. | Siwabessy, J.

Abstract

Spatially continuous predictions of seabed hardness are important baseline environmental information for sustainable management of Australia's marine jurisdiction. Seabed hardness is often inferred from multibeam backscatter data with unknown accuracy, can be inferred based on underwater video footage at limited locations. It can also be predicted to two classes. In this study, we classified the seabed into four classes based on two new seabed hardness classification schemes (i.e. hard90 and hard70) for seabed video footage by. We developed optimal predictive models to predict the spatial distribution of seabed hardness using random forest (RF) based on point data of hardness classes and spatially continuous multibeam backscatter data. Five feature selection (FS) methods that are variable importance (VI), averaged variable importance (AVI), the combined, Boruta, and RRF were tested. Effects of highly correlated, important and unimportant predictors on the accuracy of RF predictive models were also examined. Finally, the most accurate models were used to predict the spatial distribution of the hardness classes and the predictions were visually examined and compared with the predictions based on two-class hardness classification. This study confirms that: 1) hard90 and hard70 are effective seabed hardness classification schemes; 2) seabed hardness can be predicted into a spatially continuous layer with a high degree of accuracy; 3) the typical approach used to pre-select predictors by excluding highly correlated predictors needs to be re-examined when using machine learning methods, at least, for RF, in the environmental sciences; 4) the identification of the important and unimportant predictors provides useful guidelines for further improving the predictive models; 5) FS is essential for identifying an optimal RF predictive model and the RF methods select the most accurate predictive model(s) instead of the most parsimonious ones, and AVI and Boruta are recommended for future studies; and 6) RF is an effective modelling method with high predictive accuracy for multi-level categorical data, can be applied to `small p and large n problems in environmental sciences, and is recommended for future studies. In addition, automated computational programs for AVI need be developed to improve its computational efficiency and caution should be taken when applying filter FS method in selecting predictive models in future studies.

Product Type

nonGeographicDataset

eCat Id

82762

Contact for the resource

Keywords

• External Publication

( Theme )

• marine

( Theme )

• NERP

Australian and New Zealand Standard Research Classification (ANZSRC)

• Earth Sciences

• Published_Internal

Publication Date

2015-01-01T00:00:00

Creation Date

Security Constraints

Australian Government Security ClassificationSystem    

Classification - unclassified

Legal Constraints

Creative Commons Attribution 4.0 International Licence    

Access - license

Use - license

Status

Purpose

Maintenance Information

unknown

Topic Category

geoscientificInformation

Series Information

Lineage

Unknown

Parent Information

Extents

Reference System

Spatial Resolution

Service Information

Associations

Downloads and Links

Source Information

Source data not available.