Satellite-based prediction of pCO2 in coastal waters of the eastern North Pacific

Continental margin carbon cycling is complex, highly variable over a range of space and time scales, and forced by multiple physical and biogeochemical drivers. Predictions of globally significant air–sea CO₂ fluxes in these regions have been extrapolated based on very sparse data sets. We present here a method for predicting coastal surface-water pCO₂ from remote-sensing data, based on self organizing maps (SOMs) and a nonlinear semi-empirical model of surface water carbonate chemistry. The model used simple empirical relationships between carbonate chemistry (total dissolved carbon dioxide (_TCO2) and alkalinity (T_Alk)) and satellite data (sea surface temperature (SST) and chlorophyll (Chl)). Surface-water CO₂ partial pressure (pCO₂) was calculated from the empirically-predicted _TCO₂ and T_Alk. This directly incorporated the inherent nonlinearities of the carbonate system, in a completely mechanistic manner. The model’s empirical coefficients were determined for a target study area of the central North American Pacific continental margin (22–50°N, within 370 km of the coastline), by optimally reproducing a set of historical observations paired with satellite data. The model-predicted pCO₂ agreed with the highly variable observations with a root mean squared (RMS) deviation of <20 μatm, and with a correlation coefficient of >0.8 (r = 0.81; r² = 0.66). This level of accuracy is a significant improvement relative to that of simpler models that did not resolve the biogeochemical sub-regions or that relied on linear dependences on input parameters. Air–sea fluxes based on these pCO₂ predictions and satellite-based wind speed measurements suggest that the region is a ∼14 Tg C yr⁻¹ sink for atmospheric CO₂ over the 1997–2005 period, with an approximately equivalent uncertainty, compared with a ∼0.5 Tg C yr⁻¹ source predicted by a recent bin-averaging and interpolation-based estimate for the same area.