Remote estimation of surface pCO2 on the West Florida Shelf

• Satellite-based surface pCO2 algorithm developed with RMS uncertainties <12 µatm for pCO2 between 300 and 550 µatm.
• Algorithm sensitivity to input variables (SST, CHL, Kd_Lee) analyzed.
• High-resolution surface pCO2 maps established for the West Florida Shelf for the first time.
• Spatial and temporal patterns of the WFS surface pCO2 examined and explained.

Surface pCO2 data from the West Florida Shelf (WFS) have been collected during 25 cruise surveys between 2003 and 2012. The data were scaled up using remote sensing measurements of surface water properties in order to provide a more nearly synoptic map of pCO2 spatial distributions and describe their temporal variations. This investigation involved extensive tests of various model forms through parsimony and Principal Component Analysis, which led to the development of a multi-variable empirical surface pCO2 model based on concurrent MODIS (Moderate Resolution Imaging Spectroradiometer) estimates of surface chlorophyll a concentrations (CHL, mg m−3), diffuse light attenuation at 490 nm (Kd_Lee, m−1), and sea surface temperature (SST, °C). Validation using an independent dataset showed a pCO2 Root Mean Square Error (RMSE) of <12 µatm and a 0.88 coefficient of determination (R2) for measured and model-predicted pCO2 ranging from 300 to 550 µatm. The model was more sensitive to SST than to CHL and Kd_Lee, with a 1 °C change in SST leading to a ~16 µatm change in the predicted pCO2. Application of the model to the entire WFS MODIS time series between 2002 and 2014 showed clear seasonality, with maxima (~450 µatm) in summer and minima (~350 µatm) in winter. The seasonality was positively correlated to SST (high in summer and low in winter) and negatively correlated to CHL and Kd_Lee (high in winter and low in summer). Inter-annual variations of pCO2 were consistent with inter-annual variations of SST, CHL, and Kd_Lee. These results suggest that surface water pCO2 of the WFS can be estimated, with known uncertainties, from remote sensing. However, while the general approach of empirical regression may work for waters from other areas of the Gulf of Mexico, model coefficients need to be empirically determined in a similar fashion.