| Popis: |
In this thesis a newly developed 2–step neural network approach is used to reconstruct basin–wide monthly maps of the sea surface partial pressure of CO₂ (pCO₂) at a resolution of 1°X1° for both the Atlantic Ocean from 1998 through 2007 and the global ocean from 1998 through 2011. From those, air–sea CO₂ flux maps are computed using a standard gas exchange parameterization and high–resolution wind speeds. Observations form the basis of the studies conducted in this thesis. The neural network estimates benefit from a continuous improvement of the observations, i.e., the Surface Ocean CO₂ Atlas (SOCAT) database. Additionally, bottle samples were collected along the UK–Caribbean line to investigate the variability of the sea surface pCO₂ and its drivers. The neural network derived pCO₂ estimates fit the observed pCO₂ data with a root mean square error (RMSE) of about 10 μatm in the Atlantic Ocean from 1998 through 2007 and about 12 μatm in the global ocean from 1998 through 2011, with almost no bias in both studies. A check against independent pCO₂ data reveals a larger RMSE, in particular in regions with strong pCO₂ variability and gradients. Temporal mean contemporary flux estimates for the Atlantic Ocean (-0.45±0.15 Pg C ·yr⁻¹) and the global ocean (-1.54±0.65 Pg C ·yr⁻¹) agree well with recent studies. Trends and variabilities within the considered time periods are strongly influenced by climate modes. The global results from 1998 through 2011 reveal the strongest variability of the air-sea CO₂ fluxes in the Equatorial Pacific (±0.12 Pg C · yr⁻¹,±1σ), mainly driven by the El Niño Southern Oscillation (ENSO) climate mode. Trends towards a strengthening of the Southern Ocean carbon sink (-0.36±0.07 Pg C ·yr⁻¹ · decade⁻¹) from 1998 through 2011 are potentially linked to the recent weakening of the Southern Annular Mode (SAM) index. |
