Empirical ocean color algorithms for estimating particulate organic carbon in the Southern Ocean

We have examined empirical algorithms for estimating surface concentration of particulate organic carbon (POC) from remotely sensed ocean color in the Southern Ocean using field data of POC, spectral remote-sensing reflectance, R-rs(lambda), and the inherent optical properties (IOPs) of seawater collected during a number of cruises. Several algorithm formulations have been considered, including direct relationships between POC and the blue-to-green band ratios of reflectance and a single-wavelength two-step algorithm that consists of relationships linking reflectance to the backscattering coefficient and POC to the particulate backscattering coefficient at 555 nm. The best error statistics among the algorithms tested were obtained for the power function fit POC (in mg m(-3)) = 189.29 [R-rs(443)/R-rs(555)](-0.87). This band ratio algorithm is based on 85 pairs of field data and shows a small mean bias of about 3%, the normalized root mean square error of 27%, and the determination coefficient of 0.93. These error statistics as well as the analysis of matchup comparisons of satellite-derived POC and in situ POC determinations support the prospect for reasonably good performance of this algorithm in the Southern Ocean. The two-step empirical algorithm operating at 555 nm shows inferior error statistics of the regression fits and matchup comparisons compared with the band ratio algorithm. The analysis of our data set also indicates that a general trend of variation in the blue-to-green reflectance band ratio over the examined range of POC values is driven primarily by the green-to-blue ratio of particulate absorption coefficient.