Dataset: Argo SCVs

Based on theory and observations, we identify remotely-generated submesoscale coherent vortices (SCV) within Argo profiles using three primary criteria. First, a SCV must contain a vertically confined, Gaussian-shaped water mass anomaly. Second, the water mass anomaly must be weakly stratified, corresponding to doming/bowling of isopycnals above and below the water mass, respectively. Finally, horizontal velocities estimated from hydrographic properties must suggest a local maxima within the range of the SCV. We establish specific detection criteria corresponding to these conditions, based on: (1) spiciness, a measure of density-compensated thermohaline variability, (2) buoyancy frequency, a measure of the vertical stratification, and (3) dynamic height anomaly, a measure of the horizontal pressure gradient force.

We downloaded Argo float profiles from August 1997 to January 2020 from the U.S. Argo global data center (GDAC). Quality controlled temperature and salinity profiles were then linearly interpolated to a 10 dbar pressure grid from 0 to 2000 dbar, before deriving spiciness, absolute salinity, conservative temperature, and potential density profiles. To identify water mass anomalies associated with remotely-generated SCVs, we compared each profile with the nearest monthly climatological profile from the Roemmich-Gilson Argo Climatology (global gridded Argo climatology obtained from http://sio-argo.ucsd.edu/RG_Climatology.html). In order to better preserve water mass properties, we derived and interpolated climatological profiles of temperature, salinity, pressure, spiciness and buoyancy frequency to each cast potential density, before subtracting them to obtain anomaly profiles along isopycnals.

To quantify the water mass variability at each profile location, we gathered spiciness and buoyancy frequency anomalies from all profiles conducted within 220 km of the cast coordinates that were also conducted within +- 45 days of the cast date, regardless of year. We then calculated the interquartile range (IQR), or the difference between the 75th and 25th percentile, to quantify the statistical spread of anomalies at each density level. After organizing all values into percentiles, we defined upper and lower outlier thresholds at each density level as the 75% value (Q3) plus 1.5 times the IQR, and the 25% value (Q1) minus 1.5 times the IQR, respectively. The upper and lower IQR thresholds in spiciness anomaly were applied to identify isopycnal surfaces with anomalously high or low spiciness, which we labeled as spicy (anomalously warm and salty) or minty (anomalously cold and fresh) water mass intrusions, respectively. Likewise, lower IQR thresholds in buoyancy frequency anomaly were used to identify anomalously weak stratification within the core of each intrusion. Any cast which featured at least two consecutive isopycnal surfaces below the mid-pycnocline which exceeded either of the spiciness thresholds discussed above was kept for further analysis.

In order to detect the presence of anomalously weak stratification associated with spicy or minty water mass intrusions, we required an estimate of the vertical thickness of each intrusion. Since observations of SCVs suggest that their vertical structure is well represented by a Gaussian monopole, to estimate the vertical thickness or extent we fit a one-term Gaussian model to spiciness anomaly profiles (calculated along isopycnals) in pressure space. After successfully fitting Gaussian curves to spicy and minty intrusions, we sought anomalously weak stratification within the vertical extent of each water mass, indicative of bowling/doming of isopycnals around the intrusion. Since the weak stratification characteristic of SCVs should be most evident within waters just above and below the core of the feature, we defined an additional core vertical thickness or extent. Using the above estimate, we set two strict criteria to check for weak stratification. First, the buoyancy frequency anomaly profile must be less than the lower buoyancy frequency IQR threshold at a depth within the core limits, to represent the abnormally low stratification associated with SCVs. Second, the vertical mean of buoyancy frequency anomaly within this range must also be less than zero to avoid false detections due to occasionally noisy signals. Profiles that did not fit the above stratification criteria were rejected.

The dynamic height differences between two nearby casts can be used as an estimate for the geostrophic velocity streamfunction, assuming flow at depth is relatively weak compared to surface velocities. To test for a local extremum in the horizontal velocity, we calculated the dynamic height profile vs. pressure for both the detected SCV profile and the local Argo climatological profile. We started this calculation from the greatest cast pressure. We translated our velocity maximum criterion into finding a peak in the geostrophic streamfunction within the vertical extent estimate provided by the Gaussian model. Initial results indicated that slight misplacements of the pycnocline in pressure space between the SCV and the climatological profile can lead to large dynamic height anomalies along isobars, which dominated any SCV dynamic signature within the profile. This finding is not surprising given that, away from oceanic margins, the vertical structure of the ocean is almost entirely described by the barotropic and first baroclinic modes. Thus, heaving caused by internal wave motions is an additional source of error in the dynamic height anomaly calculation.

To isolate and remove the impact of first baroclinic mode structures, we decomposed the climatological profile into vertical modes. Briefly, we solved for the linear dynamical modes brought about by the standard eigenvalue problem of internal wave theory. We extracted the first baroclinic horizontal velocity mode produced by the routine and estimated the corresponding modal amplitude (alpha) by restricting the modal structure to be zero at the deepest level of the dynamic height profile before applying a non-linear least-squares optimization to solve for the best fit to the dynamic height anomaly profile. We then removed the fitted first baroclinic mode from the dynamic height anomaly profile and used the adjusted dynamic height anomaly to test for the presence of a local peak in dynamic height anomaly within the vertical extent of each intrusion, rejecting any profiles which did not exhibit this feature, while defining those that passed all the above criteria as spicy or minty-core SCVs.

Previous studies suggest that the SCV radius is comparable to their internal Rossby deformation radius. We used estimates of SCV scale height, core-isopycnal background buoyancy frequency, and the local Coriolis parameter for each detected SCV to infer the first baroclinic deformation radius and thus an idealized horizontal scale.

The submitted data reveal the date, locations, and identification information of Argo profiles that pierced spicy-core and minty-core SCVs. We also include statistics gathered for each SCV, such as core water mass characteristics and height/length scales.

Regarding data accuracy with Argo (https://argo.ucsd.edu/data/data-faq/#accurate):
The temperatures in the Argo profiles are accurate to ± 0.002°C and pressures are accurate to ± 2.4dbar. ;For salinity, there are two answers. The data delivered in real-time are sometimes affected by sensor drift. For many floats this drift is small, and the uncorrected salinities are accurate to ± .01 psu. At a later stage, salinities are corrected by expert examination, comparing older floats with newly deployed instruments and with ship-based data. Corrections are made both for identified sensor drift and for a thermal lag error, which can result when the float ascends through a region of strong temperature gradients.

Following this delayed-mode correction, salinity errors are usually reduced further and in most cases the data become good enough to detect subtle ocean change. The estimated accuracy of the delayed mode quality controlled salinity can be found in the PSAL_ADJUSTED_ERROR fields in the D profile files.