Progress in understanding the carbon cycle and its variability require us to take full advantage of both imperfect numerical models and limited data. Timeseries observations are crucial because they reveal carbon cycle variability in time, but do not directly allow study of spatial structures. Coupled physical-biogeochemical models estimate both spatial and temporal variability, but are imperfect representations of the real world. In this paper, we report on a new physical-biogeochemical model that is able to capture much of the observed carbon cycle variability from 1983 to 2005 at Bermuda (Bates, 2007). Preliminary results suggest that sea surface temperature at Bermuda has been stable since 1984 due to horizontal advection balancing decadal timescale warming and cooling trends in the local atmospheric forcing. The result is that temperature-driven pCO2 trends have been negligible. There is an increasing trend in surface ocean pCO2 driven by increasing atmospheric pCO2, and DIC anomalies due to changing vertical mixing also strongly influence surface ocean pCO2. Following detailed model-data comparisons, we use the model to consider the predictive power of timeseries observations at Bermuda for the rest of the subtropical gyre.