Appendix: additional material for model-data comparisons

Fig.1A showsthe time-series comparisons of FC transport between model, cable data and HYCOM counterparts. FC transport is monitored by a cable between the U.S. east coast and the Bahamas (pink line in Fig.1), and the data is made available online at: with observational data suggest that the model was able to capture the variability of FC transport reasonably well. One caveat we note is that the simulated four-year mean FC transport was 28.4Sv (1Sv=106m3s-1), as compared tothe observed average of 31.3Sv. This bias is likely from the global HYCOMdata that wereused to drive the boundary inflows, which also underestimated FC transport, with a mean transport value of 26.5Sv.

We also compared model simulated and satellite (AVISO) observed mean Eddy Kinetic Energy (EKE) fields over the four-year study period (Fig.2A). The EKE wascalculated based on the geostrophic velocity anomaly and:

Both AVISO and the model presented stronger EKE in the Loop Current and Gulf Stream areas and weaker EKE along the coast and in the open ocean of the Sargasso Sea. We note that model shows weaker EKE than AVISOnear coastal GOM. This discrepancy can be attributed to the fact that AVISO has a spatial resolution of 1/3º and its ability to resolve the coastal dynamics is limited. Nonetheless, the model is able to resolve the kinetic structure of the circulation reasonablywell.

In addition, NOAA National Ocean Service (NOS) sea level data were also used to assess model performance in coastal areas. We compared 26 sea level stations along the IAS coast from the GOM coast of Florida to the SAB to compare model with observations. For demonstration purposes,only six stations are shown (Fig.3A). The model clearly captured major features of sea level variability.Correlation coefficients between model and observations were all statistically significant(>0.6, p=0.05), and the model was in good overall agreement with observations at both seasonal to interannual time scales.

Model hindcast solutions were further gauged against ship CTD casts in the GOM collected in 2010. A total of 1,643 temperature and salinity profiles collected during April 22, 2010 to October 18, 2010 were utilized (Fig.4A). The linear regression of temperature and salinity (not shown) showed that the model reproduced observed temperature and the water mass reasonably well. A more statistically-robust temperature and salinity comparison is given in a Taylor diagram (Fig.5A), on which the standard deviation (STD), correlation coefficient and root mean square errors (RMSD) were presented. The temperature comparison showsthat the model was able to reproduce most of the observations (1531 out of 1643 profiles) with RMSD much less than 1, STD close to 1, and correlation coefficients > 0.95. The salinity comparisons hada more scattered distribution in the Taylor diagram. Nevertheless, all modeled solutions were still within 2times the STD of observations and had an RMSD < 1. These statistical comparisons further confirmed that the model was in good agreement with observations.

List of Appendix Figures

Fig.1A. Comparsions between observed and modeled (by IAS ROMS and HYCOM, respectively) Florida Current transport. Upper panel show the absolute transport value; lower panel show the transport anomalies with means removed

Fig.2A. Four-year mean Eddy Kinetic Energy (EKE) comparison between IAS model hindcast and 1/3º AVISO Altimeter observations. EKE values are scaled in log10 for better visualizations (units: m2s-2).

Fig.3A. Sea level comparisons between model and tidal stations; x-axis is time (month/year).A 36-hour low-pass filter was applied to yield sub-tidal signals. In each panel, the correlation coefficient between model and observations is shown.

Fig.4A. Locations of ship CTD observations in the GOM during April 22-October 18, 2010.

Fig. 5A. Taylor diagram for temperature (left panel) and salinity (right panel) comparisons; the color coding responds to their locations in Fig.7. All points are normalized by their corresponding observation points denoted by an asterisk. The radial distances from the origin are proportional to the ratio of standard deviations; the azimuthal positions indicate the correlation coefficient; and the distance between the modeled and observed points indicates the centered root mean squared difference (RMSD).

Appendix Figures

Fig. 1A. Comparsions between observed and modeled (by IAS ROMS and HYCOM, respectively) Florida Current transport. Upper panel show the absolute transport value; lower panel show the transport anomalies with means removed.

Fig.2A. Four-year mean Eddy Kinetic Energy (EKE) comparison between IAS model hindcast and 1/3º AVISO Altimeter observations. EKE values are scaled in log10 for better visualizations (units: m2s-2).

Fig. 3A. Sea level comparisons between model and tidal stations; x-axis is time (month/year).A 36-hour low-pass filter was applied to yield sub-tidal signals. In each panel, the correlation coefficient between model and observations is shown.

Fig.4A. Locations of ship CTD observations in the GOM during April 22-October 18, 2010.

Fig.5A. Taylor diagram for temperature (left panel) and salinity (right panel) comparisons; the color coding responds to their locations in Fig.7. All points are normalized by their corresponding observation points denoted by an asterisk. The radial distances from the origin are proportional to the ratio of standard deviations; the azimuthal positions indicate the correlation coefficient; and the distance between the modeled and observed points indicates the centered root mean squared difference (RMSD).