ALT Map of the Agulhas Bank, delineated by the 200 m isobath, summarizing the main driving mechanisms (in blue - Agulhas current and Southeasterly winds) resulting in high productivity events. Red shaded area demarcates the main chokka squid spawning grounds and upwelling sites.

ALT Summary diagram of the 3x4 SOM (Self Organizing Maps) architecture and functioning described in section 2.2. During the pre-processing, the input dataset which consist of 2-D daily satellite data (U(tj), V(tj), SST (tj), Chl-a(tj) with j=1: n days) is concatenated into a row vector X (which makes up the input layer). X is composed of n=3479 arrays (one for each day of the observational record) each of p=4416 elements (1104 pixels *4 variables). Each SOM neuron (or pattern) gets assigned a weight vector ((w1,1… w1,p)… (w12,1… w12,p)) that changes with the iterative training (second layer). When a BMU (wining neuron, dark blue circle) is found its weight vector and neighbouring neurons (light green shaded area) are updated. During the post-processing, the SOM outputs are reshaped into 2-D images.