Estimation of primary production over large areas of the ocean requires information on the shape of phytoplankton profiles. In this study we develop a generic quantitative approach to describe the continuous variation of profile shape within a region. We illustrate this approach by application to the dynamic southern Benguela upwelling system. First, we describe profile shape by fitting a four-parameter shifted Gaussian model. We then use a model-building approach to relate each parameter to a suite of environmental variables that are either known for each point of the ocean in time and space (depth of the water column, season, and area) or are easily obtained from remote sensing (sea surface temperature and surface chlorophyll a). As these variables are highly correlated and non-linearly related to profile shape, we use generalised additive models to visualise the non-linear relationships between each parameter and all environmental variables simultaneously. These relationships are then parameterised using generalised linear models to obtain a predictive equation for each profile parameter. Relationships identified made intuitive sense in terms of the evolution of phytoplankton blooms in upwelling systems. We found strong predictive relationships for the depth of maximum chlorophyll (r 2 = 0.70) and the total chlorophyll in the peak (r2 = 0.74), the two most important parameters for estimating primary production. Predictive relationships were weaker for the width of the peak (r 2 = 0.21) and the background chlorophyll (r 2 = 0.15). The predictive equations identified can be applied on a pixel-by-pixel basis to concurrent sea surface temperature and ocean colour images to estimate profile shape, and can be imbedded within local algorithms to provide regional primary production estimates. This approach can easily be applied to other biogeochemical provinces.