Résumé
We provide a proof-of-concept demonstration using a novel method for estimating depth-integrated distributions of chlorophyll from archives of data from ships, buoys or gliders combined with remotely sensed data of sea surface temperature (SST) and surface chlorophyll a (chl a) from satellites. Our area of application has contrasting hydrographic regimes, which include the dynamic southern Benguela upwelling system and the stratified waters of the Agulhas Bank, South Africa. The method involves using self-organising maps (SOMs), a type of artificial neural network, to identify 'typical' chl a profiles regardless of their statistical form, provided several of a similar shape have been found in the training set. These are arranged in a linear array, ranging from uniform profiles low in chl a to profiles with high surface or subsurface peaks. We then use generalised modelling to relate these characteristic profiles to remotely sensed surface features, viz. surface chl a and SST, as well as area, season, and water depth (a proxy for distance offshore). The model accounts for 87% of the variability in chl a profile and is used to predict the type of profile likely for each pixel in monthly remote sensing composites of SST and surface chl a and then to estimate integrated chl a and primary production with the aid of a light model. Primary production peaks in mid-summer, reaching 5 mgC m(-2) d(-1) locally, with an average over the whole area and all seasons of 1.4 mg C m(-2) d(-1). Seasonal variation is greatest in the southern part of the west coast, and lowest in the stratified southeast. Annual primary production for the southern Benguela region including the Agulhas Bank is ca. 156 million tC yr(-1). This is the most robust estimate of primary production in the Benguela system to date because it combines the spatial and temporal coverage provided by remote sensing with realistic vertical chl a profiles.