Revisiting empirical ocean-colour algorithms for remote estimation of chlorophyll-a content on a global scale

ABSTRACT Remote-sensing data can be useful for investigating the bio-optical properties of the ocean. Among these bio-optical properties, chlorophyll-a content is of great importance. The standard NASA empirical ocean-colour (OC) algorithms are used widely to estimate global chlorophyll-a content. Despite their simplicity and effectiveness, these regression-based models have two shortcomings that we investigate here: (1) the general form of the models is a fourth-order polynomial that results in multicollinearity, and (2) the models have the same parameters for all ocean regions (i.e. they use global approaches). To resolve the first issue, we use partial least squares (PLS), which allows for an orthogonal transformation such that the covariance between the transformed independent variables and the dependent variable is maximized. To investigate the second issue, we use geographically weighted regression (GWR) to reveal the spatial variation of estimated parameters, demonstrating how the global model underperforms in some locations. GWR results show that model coefficients vary substantially between eastern and western portions of the same ocean basin. By including sea-surface temperature (SST) as an additional independent variable in the PLS model, we also develop a new approach that provides additional explanatory power and makes the global estimation of chlorophyll-a content more valid.