The use of reflectance classification for chlorophyll algorithm application across multiple optical water types in South African coastal waters

Doctoral Thesis

2016

Permanent link to this Item
Authors
Journal Title
Link to Journal
Journal ISSN
Volume Title
Publisher
Publisher

University of Cape Town

License
Series
Abstract
Ocean colour remote sensing is a valuable tool for deriving information about key biogeochemical variables over inland, coastal and ocean waters at scales unachievable via in situ techniques. However, broader use of ocean colour data is still limited by the need for users to choose among a seemingly complicated range of available satellite products and to understand the limitations and constraints of these products across a wide range of water types. This issue could benefit from the capability to seamlessly apply and blend watertype appropriate algorithms into a single output product that provides optimal retrievals over a wide range of water types. The assessment of the fuzzy membership of satellite remote sensing reflectance (Rᵣₛ) to pre-defined regional optical water types (OWTs) provides a framework for application and blending of OWT-appropriate algorithms on a per-pixel basis. This study presents the first characterization of the OWTs in the coastal waters of South Africa. The OWTs are determined through stepwise fuzzy c-means clustering of a systematically expanding and modified database constructed from in situ, synthetic and regionally extracted Medium Resolution Imaging Spectrometer (MERIS) Rᵣₛ. A database division allows separate and more detailed clustering of phytoplankton-dominated Rᵣₛ and backscattering-dominated Rᵣₛ into six and five classes respectively. Chlorophyll α (Chl α) algorithms are assigned per OWT based on lowest error and uncertainty. The blended Chl α product consists of weighted retrievals from five different algorithms, including two 4th order polynomial exponential algorithms utilizing the blue-green spectral region, two red-NIR band ratio algorithms, and a neural network. The algorithm blending procedure retrieves satellite-derived Chl α concentration ([Chl α]) with lower RMS error and uncertainty compared to individual algorithms and provides improved capability to retrieve [Chl α] for different South African water types with a single product over a range spanning almost four orders of magnitude. The eleven OWTs are utilized in the classification and algorithm blending framework and applied to the full archive of MERIS Level 2 reflectance between the years 2002 and 2012 over South Africa's coastal waters. The persistence of the OWTs is presented and linked to the prominent environmental and physical drivers, whilst regions with low total class membership sums are discussed in terms of satellite data coverage and data quality. A time series of the blended [Chl α] product displays improved capability to capture the ranges of variability observed in the coastal, shelf and offshore environment compared to currently available regional and standard MERIS Level 2 products.
Description
Keywords

Reference:

Collections