Identification and classification of vertical chlorophyll patterns in the Benguela upwelling system and Angola-Benguela front using an artificial neural network

Abstract

Information on the vertical chlorophyll structure in the ocean is important for estimating integrated chlorophyll a and primary production from satellite. For this study, vertical chlorophyll profiles from the Benguela upwelling system and the Angola-Benguela front were collected in winter to identify characteristic profiles. A shifted Gaussian model was fitted to each profile to estimate four parameters that defined the shape of the curve: the background chlorophyll concentration (B 0 ), the height parameter of the peak (h), the width of the peak (σ) and the depth of the chlorophyll peak (z m ). A type of artificial neural network called a self-organizing map (SOM) was then used on these four parameters to identify characteristic profiles. The analysis identified a continuum of chlorophyll patterns, from those with large surface peaks (>10 mg m -3 ) to those with smaller near-surface peaks (<2 mg m -3 ). The frequency of occurrence of each chlorophyll pattern identified by the SOM showed that the most frequent pattern (~12%) had a near-surface peak and the least frequent pattern (~2%) had a large surface peak. These characteristic profile shapes were then related to pertinent environmental variables such as sea surface temperature, surface chlorophyll, mixed layer depth and euphotic depth. Partitioning the SOM output map into environmental categories showed large peaks of surface chlorophyll dominating in water with cool temperature, high surface chlorophyll concentration and shallow mixed layer and euphotic depth. By contrast, smaller peaks of subsurface chlorophyll were in water with warmer temperature, lower surface chlorophyll concentration, intermediate mixed layer and deep euphotic depth. These relationships can be used semi-quantitatively to predict profile shape under different environmental conditions. The SOM analysis highlighted the large variability in shape of vertical chlorophyll profiles in the Benguela. This suggests that an ideal typical chlorophyll profile, as used in the framework of biogeochemical provinces, may not be applicable to this dynamic upwelling system.

Generalised model of primary production in the southern Benguela upwelling system

“…Techniques such as self-organising maps (SOMs), a type of artificial neural network particularly adept at pattern identification (Kohonen 1997, Hewitson & Crane 2002 have been used to identify limited sets of characteristic chl a profiles from archives of vertical chl a traces (Silulwane et al 2001, Richardson et al 2002. However, these studies did not use raw chl a profiles, but first parameterized the profiles using the shifted Gaussian model and then based the SOM on these parameter values.…”

“…This not only constrains the patterns identified to be Gaussian in shape, but the more unusual profiles that do not fit the shifted Gaussian model also have to be removed prior to analysis (e.g. ~15% of the profiles in Silulwane et al [2001] and Richardson et al [2002]). Another approach has been to use generalised modelling to estimate the 4 parameters of the shifted Gaussian model from several environmental variables .…”

A Review of Self-Organizing Map Applications in Meteorology and Oceanography

“…They applied the SOM to a data set of 170 sediment samples for biological and geochemical conditions of a tidal flat in the southern North Sea, and demonstrated the efficiency of the SOM technique in analyzing multivariate data sets of complex natural system (Kropp & Klenke, 1997). Silulwane et al (2001) used the SOM to classify in situ vertical chlorophyll profiles from the Benguela upwelling system, and related the identified characteristic chlorophyll profiles to pertinent environmental variables, such as sea surface temperature, surface chlorophyll, mixed layer depth and euphotic depth. They pointed out that these relationships can be used semi-quantitatively to predict the subsurface chlorophyll field from known (water column depth) or easily measured variables from satellites, such as surface temperature or surface chlorophyll (Richardson et al, 2002).…”

“…Early SOM applications in oceanography community were mainly limited to satellite and in situ biological/geochemical data analyses by remote sensing scientiests or biological/chemical oceanographers (e.g., Kropp & Klenke, 1997;Ainsworth, 1999;Ainsworth & Jones, 1999;Yacoub et al, 2001;Silulwane et al, 2001). Since the introducion and demonstration of the use of the SOM to the oceanography community by Richardson et al (2003), SOM applications have been steadily increased in physical oceanography (e.g., Risien et al, 2004;Liu & Weisberg, 2005Leloup et al, 2007Leloup et al, , 2008Iskandar et al, 2008), and other disciplinary of oceanography as well (e.g., Chazottes et al, 2006Chazottes et al, , 2007Telszewski et al, 2009).…”

A Remote Sensing Approach to Estimate Vertical Profile Classes of Phytoplankton in a Eutrophic Lake

“…In ocean optics, many studies have focused on the vertical distribution of phytoplankton and chlorophyll-a [14,25,26] and its influence on the remote sensing reflectance [27][28][29]. Gaussian models [30][31][32] or shifted Gaussian models [14] have been used to represent the vertical profile of chlorophyll-a concentrations in marine environments.…”