Graduation Year


Document Type




Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Marine Science

Major Professor

Chuanmin Hu, Ph.D.

Committee Member

Robert H. Byrne, Ph.D.

Committee Member

Lisa L. Robbins, Ph.D.

Committee Member

Mark E. Luther, Ph.D.

Committee Member

David F. Naar, Ph.D.


dominant controls, remote sensing, sea surface salinity, surface pCO2


Surface ocean partial pressure of CO2 (pCO2) is a critical parameter in the quantification of air-sea CO2 flux, which further plays an important role in quantifying the global carbon budget and understanding ocean acidification. The demand for a clearer understanding of how, and how fast, the ocean is changing due to atmospheric CO2 absorption, requires accurate and synoptic estimation of surface pCO2.

Surface ocean pCO2 is mainly controlled by four oceanic processes – thermodynamics, ocean mixing, biological activities, and air-sea CO2 exchange. Surface ocean pCO2 is therefore closely related to environmental variables that characterize each oceanic process. These variables include sea surface temperature (SST), sea surface salinity (SSS), chlorophyll-a concentration (Chl), diffuse attenuation of downwelling irradiance (Kd), and wind speed. Ocean color satellites provide a means by which the relationship between these environmental variables and surface pCO2 can be developed. Yet, remote estimation of surface pCO2 in coastal oceans has been difficult due to the dynamic and complex biogeochemical processes. To date, most of the published satellite-based pCO2 models are developed for single-process dominated regions, therefore having poor applicability in other oceanic regions. Particularly, there is no unified approach, let alone unified model, to remotely estimate surface pCO2 in oceanic regions that are dominated by different oceanic processes.

This work provides solutions to these challenging issues for the remote estimation of surface pCO2 in the Gulf of Mexico (GOM), with the following objectives: 1) Develop satellite-based surface pCO2 models and data products for single-process dominated subregions of the GOM, and quantify the sensitivities of the pCO2 algorithms to the input environmental variables; 2) Quantify the oceanic processes in controlling surface pCO2 in the GOM, analyze the relationships between environmental variables and surface pCO2, and understand the mechanisms of seasonal and interannual variations of surface pCO2 and its driving factors; 3) Develop an improved SSS model and data products for most GOM waters, and quantify the sensitivities of the SSS model to the input variables; 4) Develop a unified pCO2 model and data products for the GOM waters, and quantify the sensitivities of the pCO2 model to the input environmental variables and their relationships; 5) Quantify the temperature and non-temperature effects on surface pCO2 at different latitudes, analyze the dominant controls and the corresponding the driving factors of surface pCO2. The data used in this dissertation include those from extensive cruise surveys, buoy measurements, and long-term measurements by the Moderate Resolution Imaging Spectroradiometer (MODIS).

Specifically, for single-process dominated regions, two separate algorithms are developed and validated, respectively, from MODIS measurements. One is focused on the ocean current- dominated West Florida Shelf (WFS) (Appendix A), and the other is on the river-dominated northern GOM (Appendix B). The former utilizes a multi-variate nonlinear regression approach to establish the relationship between surface pCO2 and environmental variables of SST, Chl, and Kd. The latter relies on a mechanistic semi-analytical approach (MeSAA), modified from an existing algorithm published earlier. Both algorithms show satisfactory performance, yet the latter requires SSS as the model input, which is difficult to obtain from ocean color satellite measurements. Therefore, a multilayer perceptron neural network-based (MPNN) SSS model is developed and validated, which generates SSS maps at 1-km resolution for the GOM using MODIS measurements (Appendix C). Finally, with the availability of SSS from MODIS for the GOM, a unified pCO2 algorithm is developed and validated. The machine-learning algorithm is based on a random forest regression ensemble (RFRE), which is able to estimate surface pCO2 from MODIS measurements with a Root Mean Square Error (RMSE) of < 10 µatm and R2 of 0.95 for pCO2 ranging between 145 and 550 µatm (Appendix D). Using this approach, The RFRE algorithm is shown to be applicable to the Gulf of Maine (a contrasting oceanic region to GOM) after local model tuning. The results show significant improvement over other models, suggesting that the RFRE approach may serve as a template for other oceanic regions once sufficient field-measured pCO2 data are available for local model tuning.

To further improve the accuracy of satellite-derived surface pCO2 from coastal oceans and to increase its capability in capturing the interannual variations of surface pCO2 resulting from anthropogenic forcing, the dominant controls of surface pCO2 over seasonal and interannual time scales need to be better understood. As such, in situ pCO2 time series data along the coasts of the United States of America at different latitudes are analyzed (Appendix E). On a seasonal time scale, surface pCO2 tends to be dominated by the temperature effect (pCO2_T) through SST and wind speed (with some exceptions) in tropical and subtropical oceans, but appears to be dominated by the non-temperature effect (pCO2_nonT) in subpolar regions. In contrast, in tropical and subtropical waters on interannual time scales, surface pCO2 is primarily moderated by the non- temperature effect (through air-sea CO2 exchange via atmospheric pCO2), but conversely dominated by the temperature effect (i.e., SST increase) in subpolar regions. The effects of biological activities (i.e., algal blooms) need to be further investigated in the future.

Overall, this dissertation has developed several algorithms to estimate SSS and surface pCO2, among which the unified pCO2 algorithm for multi-processes dominated regions appears to be able to serve as a template for many other regions after local model tuning. The derived surface pCO2 data products for the GOM provide a fundamental basis to assess air-sea exchange of CO2 and understand the carbon chemistry under a changing climate.