On the variations of sea surface pCO₂ in the northern South China Sea: A remote sensing based neural network approach

Using a neural networking (NN) approach, we developed an algorithm primarily based upon sea surface temperature (SST) and chlorophyll (Chl a ) to estimate the partial pressure of carbon dioxide ( p CO 2 ) at the sea surface in the northern South China Sea (NSCS). Randomly selected in situ data collected from May 2001, February and July 2004 cruises were used to develop and test the predictive capabilities of the NN based algorithm with four inputs (SST, Chl a , longitudes and latitudes). The comparison revealed a high correlation coefficient of 0.98 with a root mean square error (RMSE) of 6.9 μ atm. We subsequently applied our NN algorithm to satellite SST and Chl a measurements, with associated longitudes and latitudes, to obtain surface water p CO 2 . The resulting monthly mean p CO 2 map derived from the satellite measurements agreed reasonably well with the in situ observations showing a generally homogeneous distribution in the offshore regions. The p CO 2 exerts a very dynamic feature in nearshore regions, especially in the coastal upwelling and estuarine plume regions. We identified three low p CO 2 zones (<330 μ atm), two of which are influenced by coastal upwelling: off Hainan island in the western part of the NSCS; and off Guangdong province in the eastern part of the NSCS. The path of the Pearl River plume on the shelf was another zone with low p CO 2 . For the monthly mean p CO 2 variations estimated based on the MODIS‐SST and ‐Chl a values, an RMSE of ∼6 μ atm may be attributable to the measurement errors associated with MODIS measurements. As a first order estimation, we used the same sampling periods of remote sensing and in situ measurements, and were able to estimate p CO 2 with an accuracy of 12.05 μ atm for onshore regions and 13.0 μ atm for offshore regions, but with combined uncertainties associated with the NN Testing algorithm and MODIS SST and Chl a measurements.