Thumbnail Image

Remote sensing of dynamics and aboveground biomass of seagrass in Tauranga Harbor (Bay of Plenty), New Zealand

Seagrasses are angiosperm plants that are completely adapted to life in seawater. They are distributed widely across the climatic regions, ranging from the tropics to temperate regions, in both inter-tidal and sub-tidal zones. Multiple ecosystem services are recognized as being supported by seagrass meadows, which recently has included appreciation role in carbon sequestration. Seagrass meadows, however, have been degraded in both terms of area and habitat quality across the globe, leading to a significant loss of ecosystem services and human livelihood support. This ongoing degradation has resulted in an urgent need to develop tools for assessing the temporal changes of extant meadows and accurate estimation of seagrass biological parameters, which will contribute to a sustainable conservation strategy into the future. This thesis describes the use of a range of freely available Earth observation products, including multi-spectral imagery from Landsat and Sentinel-2, and synthetic aperture radar (SAR) products from Sentinel-1, coupled with a range of machine learning (ML) and meta-heuristic optimization algorithms, to develop novel and advanced techniques for remote sensing of seagrass. The work used field validation data from Tauranga Harbor, New Zealand, and specifically targeted mapping, change detection, and estimation of seagrass distribution and biomass. The relatively small and patchy meadows of Zostera muelleri in this harbor can be mapped using a three-category classification (dense, sparse and none) with up to 91% accuracy for dense and 75% for sparse meadows using the machine learning algorithm Rotation Forest with Sentinel-2 imagery. Despite a slightly lower accuracy (90%), the algorithm Canonical Correlation Forest also shows merit for categorical seagrass mapping. Historic Landsat multispectral satellite data used with ML models was able to map accurately the change in distribution of seagrass meadows over 29 years (1989 - 2019). For this binary mapping application (presence/absence) the CatBoost model obtained over 90% accuracy. Historic imagery indicates an approximately 50% of seagrass loss, from 2,424 ha in the year 1989 down to 1,184 ha in the year 2019 in Tauranga Harbor. Most of the early loss was from the northern and southern parts of the harbor and results were consistent with published estimates of change based on aerial photography. In addition, a mapping scheme of seagrass distribution was developed from SAR data and a fusion of the multi-spectral and SAR data was developed for seagrass aboveground biomass (AGB) estimation. Optimal results were obtained using a combination of ML methods and metaheuristic optimization. The seagrass distribution was mapped at an accuracy over 90% using the Extreme Gradient Boost (XGB) whilst the AGB map at 10 m spatial resolution was produced at 75% accuracy using the XGB model together with Sentinel-2 images and Particle Swarm Optimization (PSO). The last part of the thesis describes the development of a web-based application to allow the advances in this research to be shared with a broader community and strengthen international and domestic collaboration in seagrass protection and conservation. This study provides in-depth and advanced methods for seagrass resource inventory, maximizing the utilization of remotely sensed data, state-of-the-art ML and metaheuristic optimization algorithms to accurately map distribution and estimate desired biophysical parameters. The proposed methods are open-source and applicable across the globe, providing a complete toolset for both scientist and managers in aquatic resource management.
Type of thesis
Ha, N.-T. (2021). Remote sensing of dynamics and aboveground biomass of seagrass in Tauranga Harbor (Bay of Plenty), New Zealand (Thesis, Doctor of Philosophy (PhD)). The University of Waikato, Hamilton, New Zealand. Retrieved from https://hdl.handle.net/10289/14586
The University of Waikato
All items in Research Commons are provided for private study and research purposes and are protected by copyright with all rights reserved unless otherwise indicated.