Novel Maximum Carbon Fixation Rate Algorithms for Remote Sensing of Oceanic Primary Productivity.

Ocean remote sensing is the only technique capable of quantitatively detecting long-term and global variability of marine primary production. However, because the spectral curves of light derived from satellite sensors have a limited capability to identify the physiological state of phytoplankton, the direct estimation of photoadaptive variables, such as daily maximum carbon fixation in a water column (P{opt^{b}}), from the spectra of satellite sensors is difficult. This difficulty results in most of the errors in ocean color-based primary productivity models. In this paper, algorithms were developed to estimate $P_{{{\rm opt}}}^{{\rm b}}$ using statistical models known as support vector machines (SVM). The algorithms were run using three different inputs: sea surface temperature (SST); both SST and sea surface chlorophyll a concentration; and SST, sea surface chlorophyll a concentration and photosynthetically active radiation (PAR) together. The results indicate that the algorithm using the three inputs (SST, sea surface chlorophyll a concentration, and PAR) had the best performance. The algorithms based on the SVM had better results than the general statistical regression method. Using the new $P_{{{\rm opt}}}^{{\rm b}}$ algorithm, the performances of the previous primary production models, including the vertically generalized production model and the statistical model developed in our earlier research, were improved. In contrast to the previous algorithm, the global primary productivity estimated using the new $P_{{{\rm opt}}}^{{\rm b}}$ algorithm showed regional changes, with higher primary productivity in most open ocean areas and lower primary productivity in coastal waters. [ABSTRACT FROM PUBLISHER]