Seafloor habitat mapping using machine learning and underwater acoustic sonar
The need for detailed spatial map of marine habitats is increasingly important and demanding in managing and preserving marine biodiversity. This study integrates machine learning technique with the in-situ dataset and underwater acoustic mapping data to produce habitat classification maps. For the...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Book Section |
| Published: |
Springer Science and Business Media Deutschland GmbH
2022
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/100531/ http://dx.doi.org/10.1007/978-981-16-8484-5_26 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The need for detailed spatial map of marine habitats is increasingly important and demanding in managing and preserving marine biodiversity. This study integrates machine learning technique with the in-situ dataset and underwater acoustic mapping data to produce habitat classification maps. For the acoustic data, high-spatial resolution bathymetry and backscatter data were acquired using Kongsberg EM2040C multibeam sonar echosounder. A set of derivative layers were computed as follows (from bathymetry); slope, aspect, rugosity, Benthic Position Index (BPI) (broad and fine scale BPI), whilst from backscatter layers were; Hue, Saturation and Intensity (HSI), Grey Level Texture Co-Occurrence (GLCM) layers (homogeneity, entropy, correlation and mean). Layers from inversion of acoustic properties using Angular Range Analysis (ARA) were also produced such as characterization (sediment class), phi, fluid factor, gradient, index of impendence, intercept, mean far, mean near, mean outer, mean total, roughness and volume homogeneity. Habitat classification map was derived using Random Forests decision trees. Full coverage benthic habitat maps at 1 m spatial resolution were successfully constructed which explained the spatial distribution of coral, fine sand and coarse sand. Bathymetry (and its derivatives) and GLCM mean were identified as the important variables to predict these habitats. This study demonstrated the contribution of machine learning technique to be integrated with underwater sonar data for seafloor habitat mapping. |
|---|