Authors / CoAuthors
Li, J. | Tran, M. | Siwabessy, J.
Abstract
Seabed hardness is an important character of seabed substrate as it may influence the nature of attachment of an organism to the seabed. Hence spatially continuous predictions of seabed hardness are important baseline environmental information for sustainable management of Australia's marine jurisdiction. Seabed hardness is usually inferred from multibeam backscatter data with unknown accuracy and can be inferred from underwater video footage or directly measured at limited locations. It can also be predicted based on two-class hardness data using environmental predictors, but no study has been undertaken for predicting multiple-class hardness data. In this study, we classified the seabed hardness into four classes based on underwater video images that were extracted from the underwater video footage. We developed an optimal predictive model to predict the spatial distribution of seabed hardness using random forest (RF) based on the point data of hardness classes and spatially continuous multibeam bathymetry, backscatter and other derived predictors. A novel model selection method that is the averaged variable importance (AVI) was used based on predictive accuracy that was acquired from averaging the results of 100 times replication of 10-fold cross validation. Finally, the spatial predictions generated using the most accurate model was visually examined and analyzed in comparison with previously published predictions based on two-class hardness data. This study confirmed that: 1) seabed hardness of four classes can be predicted into a spatially continuous layer with a high degree of accuracy; 2) model selection for RF is essential for identifying an optimal predictive model in environmental sciences and AVI select the most accurate predictive model(s) instead of the most parsimonious ones, and is recommended for future studies; 3) the typical approach used in pre-selecting predictors by excluding correlated variables (i.e. r 0.95 or the inflation factor 20) needs to be re-examined for identifying predictive models using machine learning methods, at least for the application of random forest in marine environmental sciences; 4) RF is an effective modelling method with high predictive accuracy for multi-level categorical data and can be applied to `small p and large n problems in the environmental sciences; and 5) the spatial predictions for four-class hardness data were similar with the predictions based on two hardness classes, with a high match rates. RF and AVI are recommended for generating spatially continuous predictions of categorical variables in future studies. In summary, AVI shows its effectiveness in searching for the most accurate predictive models and are recommended for future studies. This study further confirms the superior performance of RF in marine environmental sciences. RF is an effective modelling method with high predictive accuracy not only for presence/absence data and but also for multi-level categorical data. RF and AVI are recommended for generating spatially continuous predictions of categorical variables in future studies.
Product Type
nonGeographicDataset
eCat Id
83119
Contact for the resource
Custodian
Point of contact
Cnr Jerrabomberra Ave and Hindmarsh Dr GPO Box 378
Canberra
ACT
2601
Australia
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
Legal Constraints
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.