Towards Automatic Monitoring and Mapping of Benthic Vegetation using Machine Learning

Research output: Book/ReportPh.D. thesis

124 Downloads (Pure)


Seagrass is aquatic vegetation that constitutes a key element of the marine environment. It supports marine life, releases oxygen, helps in sediment stabilization, and is the food stock of many aquatic animals. Any marine monitoring program involves seagrass, as they are a key indicator of marine life. Underwater videos are one of the popular ways to map seagrass. Domain expert does the traditional way of seagrass estimation with visual estimation. This way is not only very subjective but also very time consuming and prone to errors. There exists no framework through which patchiness of seagrass meadows can be estimated in a better way.
There is a need to automate the process of seagrass estimation and detection from underwater videos and this thesis address the problem by proposing two methods for seagrass detection (presence/absence) and two methods for seagrass estimation (0 – 100%). Not only do the predictions obtained from the above methods match well with the domain expert’s estimation, but also detect rare errors made by the domain expert. The coverage estimated from one of the methods is then extended to estimate depth colonization or depth limit, which matches well with historical data on the site of study. Sensitivity analysis showed the depth limit
estimates are robust to noisy fluctuations. This thesis also attempts to quantify the bias in a visual coverage estimation task.
The status obtained from the videos is used on satellite images to extend seagrass segmentation on a larger scale. Aquatic vegetation mapping with satellite images is a challenging task. Atmospheric interference, cloud coverage, the turbidity of
water are some of the key factors that determine the success of such a task. This thesis attempts to use a deep generative model to segment seagrass meadows in the latent space on a shallow but turbid and optically deep water body. Finally, the thesis concludes with an easy to use online tool for marine ecologists, which has a simple plug and play format for seagrass analysis from underwater videos.
Original languageEnglish
PublisherTechnical University of Denmark
Number of pages110
Publication statusPublished - 2021


Dive into the research topics of 'Towards Automatic Monitoring and Mapping of Benthic Vegetation using Machine Learning'. Together they form a unique fingerprint.

Cite this