Impact of mesoscale eddies on water mass and oxygen distribution in the eastern tropical South Pacific

. The influence of mesoscale eddies on the flow field and the water masses, especially the oxygen distribution of the eastern tropical South Pacific is investigated from a mooring, float and 10 satellite data set. Two anticyclonic (ACE1/2), one mode water (MWE) and one cyclonic eddy (CE) are identified and followed in detail with satellite data on their westward transition with velocities of 3.2 to 6.0 cm/s from their generation region, the shelf of the Peruvian and Chilean upwelling regime, across the Stratus Ocean Reference Station (ORS) (~20°S, 85°W) to their decaying region far west in the oligotrophic open ocean. The ORS is located in the transition zone between the oxygen minimum zone 15 and the well-oxygenated South Pacific subtropical gyre. Velocity, hydrographic, and oxygen measurements at the mooring show the impact of eddies on the weak flow region of the eastern tropical South Pacific. Strong anomalies are related to the passage of eddies and are not associated to a seasonal signal in the open ocean. The mass transport of the four observed eddies across 85°W is between 1.1 and 1.8 Sv. The eddy type dependent available heat, salt and oxygen anomalies are 7.6x10 18 J (ACE), 20 0.8x10 18 J (MWE), -9.4x10 18 J (CE) for heat, 23.9x10 10 kg (ACE2), -3.6x10 10 kg (MWE), -42.8x10 10 kg (CE) for salt and -3.6x10 16 µ mol (ACE2), -3.5x10 an imbalance between anticyclones and cyclones for heat and salt transports probably due to seasonal variability of water mass properties in the formation region of the eddies. Heat, salt and oxygen fluxes out of the coastal region across the ORS region in the oligotrophic South estimated on these eddy anomalies eddy 23 of satellite three 25 instruments recorded erroneous oxygen values, which could not be corrected after the recovery. The et al., 2015; Thomsen et al., 2016). It is suggested that anticyclones instabilities of the Peru Chile In of PCUC al., 2016). Observations as well as models show a weak seasonal variability of the PCUC off Peru which is stronger in austral summer and fall (Thomsen et al., 2016; Chaigneau et al., 2013; Penven et al., 2005) Available anomalies of heat, salt, and oxygen of cyclonic anticyclonic eddies relative eddy to fluxes of mass, heat, salt, and oxygen in the ETSP. By multiplying the amount of AHA, ASA, and AOA of the composite eddies with the number of eddies dissipating per year in a given area to divergence) mean heat (in W m -2 ), salt (in kg s -1 m -2 ) and oxygen release ( µ mol 25 s -1 m -2 ) were calculated. We define an area reaching in north-south direction from 10°-22°S. The transition area is bordered in the east by the Peruvian and Chilean coast and in the west by the longitude of the Stratus mooring (86°W) corresponding to a size of ~2 x 10 6 km 2 . Based on satellite

crossed the Stratus mooring position, from their formation areas near the coast to their decay eastwards of the Stratus mooring (Fig. 1). During their lifetime one of these eddies were also partly sampled by several profiling floats equipped with oxygen sensors which were deployed in March 2014 within the eddies (Fig. 1).
In general, the large-scale oxygen distribution in the ETSP is dominated by a strong OMZ at depths of 5 100-900 m (e.g., Karstensen et al., 2008;Paulmier and Ruiz-Pino, 2009 or Fig. 1b). In the ETSP the zonal tropical current bands supply oxygen (O 2 ) rich water to the OMZ (Stramma et al., 2010). In contrast, the mid-depth circulation in the eastern South Pacific Ocean is sluggish in the region of the OMZ. As the mean currents are weak, eddy variability strongly influences the flow and ultimately supplies oxygen-poor water to the OMZ (Czeschel et al., 2011). A rough estimate of the oxygen budget 10 of the eastern tropical Pacific ocean (Stramma et al., 2010) was used to estimate 22% by vertical mixing, 33% by advection and the largest component of 45% by eddy mixing (Brandt et al., 2015).
The mean upper ocean circulation of the ETSP is relatively complex exhibiting several surface and subsurface currents. It is described to be composed out of the South Pacific subtropical gyre with the north-eastern current band shown to be located south of 10 to 15°S called Humboldt Current, South 15 Equatorial Current, Oceanic Chile-Peru Current or Peru Oceanic Current (e.g. Kessler 2006; Ayon et al., 2008), a set of several zonal current bands between the subtropical gyre and the equator (.e.g. Kessler 2006;Czeschel et al., 2015) as well as poleward and equatorward current bands near the South American continent (e.g. Chaigneau et al., 2013). The shipboard zonal velocity component along about 86°W composed of three ADCP surveys showed larger regions with westward then eastward flow 20 between 13°S and 22°S (Brandt et al., 2015), however with influence by eddy features in ADCP measurements in November 2012 (Czeschel et al., 2015).
In general most of the eddies in the ETSP propagate westward originating from eddy generation hotspots near the coast following different eddy corridors (Fig. 1a). Coastal water properties are captured within the eddy-cores and transported on their way into the open ocean across several oxygen, 25 temperature, salinity and gradients ( Fig. 1 b, c, d). The coastal water mass properties differ, due to the upwelling, which is strongest in the austral winter months from a seasonal cycle. The upwelled water Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License. near the coast identified as Equatorial Subsurface Water (ESSW) (e.g. Thomsen et al., 2016) is colder, fresher and less oxygenated in austral winter than in austral summer. This paper describes the temperature, salinity and oxygen anomaly of the different eddy types in the ETSP and their efficiency to dissipate the existing gradients. Of special interest is the eddy type dependent isolation of the eddy cores during different eddy life stages. Knowledge about the initial 5 eddy-core conditions near the generation areas, measurements during the mid-age of the eddy due to Argo floats and measurements of the Stratus mooring at the end of the eddy lifetime allows us to investigate the lateral mixing from the eddy-core water masses with its surrounding waters.

Stratus mooring 10
Since October 2000 the Stratus mooring has been maintained at about 20°S, 85.5°W mainly to collect an accurate record of surface meteorology and air-sea fluxes of heat, freshwater, and momentum (Colbo and Weller, 2009). In addition velocity, pressure, temperature, conductivity sensors (for salinity computation) and 13 oxygen sensors were added to the mooring within the water column during the oxygen measurements were conducted. To avoid influence of a seasonal signal only the period 10 April to 9 April of the following year was computed and only instruments used which recorded the velocity for the entire period. These mean velocity profiles can be compared with the October 2000 to December 2004 mean velocity components (Colbo and Weller, 2007).
From the 13 oxygen sensors added to the 2014/2015 mooring period (Supplement Table S1), three 25 instruments recorded erroneous oxygen values, which could not be corrected after the recovery. The Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License. remaining ten oxygen sensors consist out of eight Aanderaa oxygen sensors in SeaGuard instruments, which were used with the manufacturers calibration (accuracy <8 µmol kg -1 or 5%) and two oxygenloggers, which received an additional lab-calibration. For the 15 MicroCats (pressure, temperature and salinity), a data calibration is done against shipboard CTD data during the service cruises (RV Ron Brown RB 14-01 and RV Cabo de Hornos) and later by comparison with the data overlap with the 5 previous mooring and by returning the instruments to SeaBird for laboratory calibration. The SeaGuard conductivity sensors in 107 m and 350 m depth have an offset of -0.13 psu and -0.18 psu, respectively.

Satellite data
Aviso (Archiving, Validation, and Interpretation of Satellite Oceanographic) satellite derived sea level anomalies (SLA) data were obtained and used to identify and track the different eddies passing the 10 Stratus mooring and to document the position of the floats within the eddies. The Copernicus Marine and Environment Monitoring Service (CMEMS, http://marine.copernicus.eu) has taken over the whole processing and distribution of the products formerly distributed by AVISO with no changes in the scientific content. The delayed-time "all-sat-merged" reference dataset of SLA is used which is mapped on an ¼° x ¼° Cartesian grid and has a temporal resolution of one day. The time period January 1993 to 15 December 2015 were chosen for the SLA and the geostrophic velocity anomalies also provided by AVISO.
For sea surface temperature (SST) the "Microwave Infrared Fusion Sea Surface Temperature" from Remote Sensing System (www.remss.com) is used. The data consist out of SST measurements of all operational microwave (MW) radiometer (TMI, AMSR-E, AMSR2, WindSat) and infrared (IR) SST 20 measurements (Terra MODIS, Aqua MODIS). Considered are here daily data with 9 km resolution from January 2013 to December 2015.
For sea surface chlorophyll (Chl) the MODIS/Aqua Level 3 data product mapped on a 4 km grid available at http://oceancolor.gsfc.nasa.gov provided by the NASA is used. The time period January 2013 to December 2015 with a daily resolution is chosen. Note, that the Chl data is cloud dependent as 25 it is measured via IR.

Argo floats
Seven profiling Argo floats with Aanderaa oxygen sensors were deployed in March 2014 at 19°36'S, 84°58'W; 19°27'S, 83°01'W; 19°15'S, 80°30'W and 18°58'S, 76°59'W. The deployment locations ( Fig. 1a) were chosen to be close to anticyclonic or cyclonic eddies determined from SLA figures. The floats were deployed in pairs with drifting depth at 400 and 1000 dbar and cycling intervals of 10 days, 5 except for 18°58'S, 76°59'W were only one float was deployed at 400 dbar drifting depth. From that seven Argo floats, four floats remained for a longer period within eddies which later crossed the Stratus mooring and are therefore used in more detail for our calculations in the paper (the four Argo floats are: 6900527, 6900529, 6900530 and 6900532). Typically a full calibration of the oxygen sensors on the Argo floats is not available. The different manufactures of Argo float oxygen sensors specify their 10 measurement error at least better than 8 µmol kg -1 or 5%. Additionally the Argo float profiles of temperature, salinity and oxygen are compared and calibrated against the measurements of the Stratus mooring and against each other giving a relative accuracy.

Methods
From the Stratus mooring time series of velocity, temperature, salinity and oxygen (from 8 March 2014 15 to 25 April 2015) eddies of each type are identified and followed back and forward in time with the help of satellite data. The focus is set on one mode water eddy (MWE), two anticyclonic eddies (ACE1, ACE2) and one cyclonic eddy (CE) as they are also sampled by Argo floats (including oxygen sensors) in before.

Heat, salt, and oxygen anomaly at the Stratus mooring
Available heat, salt and oxygen anomalies (AHA, ASA and AOA) were calculated as described in Chaigneau et al. (2011) and Stramma et al. (2014). At the Stratus mooring eddy core anomalies were estimated by the difference between the mean of temperature, salinity and oxygen within the eddy boundaries and the background field estimated from the annual mean for the period 10 April 2014 to 9 25 April 2015. Eddy boundaries are determined for every depth by the mean of the maximum absolute Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License.
values of the maximum 90 h low-pass filtered southward and northward velocity. The mean westward propagation of the eddies estimated from SLA measurements is used to convert the time axis to a space axis leading to a mean radius. The vertical extent is defined as the depth of the coherent structure of the eddy, which is the ratio between the swirl velocity U and the propagation velocity c of the eddy. If U/c > 1, the feature is nonlinear and maintains its coherent structure while propagating westward (Chelton et 5 al., 2011). The swirl velocity is derived from the mean of the absolute values of the maximum 90 h lowpass filtered southward and northward velocity.
At the time when the mooring was deployed, part of the MWE had already passed the mooring.
Therefore, the measurements of the eddy were mirrored to obtain the full coverage of the MWE.

Determining of properties of the MWE, ACE1/2 and CE conducted from satellite data 10
The eddy shape is identified by analysing streamlines of the SLA-derived geostrophic flow around an eddy centre (high/low SLA). An eddy boundary is defined as the streamline with the strongest swirl velocity (for more information on such an eddy detection algorithm see e.g. Nencioli et al., 2010). Note that the identified areas are irregularly circular therefore the circle-equivalent area is used to estimate the eddy radius. Due to the resolution of the SLA data the eddy radius must be at least 45 km to 15 unambiguously state that the identified area is a coherent mesoscale eddy and not an artificial signal.
Clearly identified individual eddies may have a smaller radius than 45 km to get tracked. Eddies are tracked forward and backward in time following the approach described by Schütte et al. (2016a). To estimate the percentage of eddy coverage in the ETSP eddies are identified and tracked between 1993 and 2015. In the following it was counted how often a grid point (0.5° x 0.5°) was covered by an eddy 20 structure. For the identification of eddy generation areas every newly detected eddy closer than 600 km off the coast is counted in 1° x 1° boxes. The sum of all these boxes is taken to compute the seasonal cycle of eddy generation. The Argo float profiles and the mooring time series are separated into data conducted within cyclones, anticyclones and the "surrounding area" which is not associated with eddylike structures also following the approach of Schütte et al. (2016a). In addition the relative position of 25 the mooring or Argo float profile in relation to the eddy center and eddy boundary could be computed.
Furthermore, the composites of the eddy surface signatures (SLA, SST and Chl) consist of 150 x 150 km snapshots around the identified eddy centres. To exclude large-scale variations, the used SST data is low-pass filtered (cut-off wavelength of 15° longitude and 5° latitude) and subtracted from the original data to preserve only the mesoscale variability (see Schütte et al., 2016a for more details).

General eddy generation and its seasonal cycle in the ETSP
In the ETSP 5244 eddies (49% cyclones; 51% anticyclones) are found between January 1993 to December 2015 (requirement: having a radius between 45 km and 150 km and visible for more than 7 days). Both types of eddies have an average radius of about 70 km and on average 15 % of the ETSP are covered everyday with eddies ( Fig. 1a). Most of the eddies are generated close to the Peruvian or 10 Chilean coast, where large horizontal/vertical shears exist in an otherwise quiescent region. In almost entirely consistence with Chaigneau et al. (2008), hotspot locations of eddy generation are near the coast around 10°S and between 16°S to 22°S (Fig. 2a, b). The four eddies (MWE, CE, ACE1, and ACE1) described in detail below originate from the latter region. After their generation near the coast the anticyclonic eddies tend to propagate north-westward whereas cyclonic vortices migrate south-15 westward (e.g. Chaigneau et al., 2008) into the open ocean. The seasonal cycle of eddy generation, based on all eddy new detections closer than 600 km off the coast, peaks in March and has its minimum in September (Fig. 2c), whereas cyclonic eddies exhibit a stronger amplitude. However, both anticyclonic as well as cyclonic eddies have their seasonal peak of formation in austral summer/fall (February/March) and the lowest number at the end of austral spring (September; Fig. 2d). 20 The full eddy generation mechanisms are complex, whereby boundary current separation due to a sharp topographic bend is one important aspect of the eddy formation (Molemaker et al., 2015;Thomsen et al., 2016). It is suggested that anticyclones are generated due to instabilities of the Peru Chile Undercurrent (PCUC), whereas cyclonic eddies are formed from instabilities of the equatorward surface currents (Chaigneau et al., 2013). In this context the strength of the PCUC is essential (Thomsen et al., 25 2016). Observations as well as models show a weak seasonal variability of the PCUC off Peru which is stronger in austral summer and fall (Thomsen et al., 2016;Chaigneau et al., 2013;Penven et al., 2005) Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License. and might explain the higher number of eddy generation during this season. Other model simulations have revealed a seasonal cycle in eddy flux that peaks in austral winter at the northern boundary of the OMZ, while it peaks a season later at the southern boundary (Vergara et al., 2016). The PCUC also experiences relatively strong fluctuations with periods of a few days to a few weeks (Huyer et al., 1991). 5 Intraseasonal and interannual variability is another factor modulating the strength of the PCUC off Chile and therefore the formation rate of anticyclonic eddies (Shaffer et al., 1999). El Niño (La Niña) events deepen (shoal) the thermocline and intensify (weaken) the PCUC (e.g. Montes et al., 2011;Combes et al., 2015). Although the strength of the PCUC increases during El Niño events the transport of mode-water eddies decreases, which is due to a weakened baroclinic instability related to the 10 deepening of the thermocline (Combes et al., 2015). were related to a cyclonic eddy (CE) with a radius of 71 km. Water properties of the CE may be associated with the Eastern South Pacific Intermediate Water which is transported by equatorward surface currents (Chaigneau et al., 2011).

Eddy observations from March 2014 to April 2015 at Stratus mooring
According to the SLA satellite maps the centre of the MWE passed north of the mooring. The centre of the westward propagating ACE1 and the ACE2 passed the Stratus mooring only 14 km respectively 17 25 km south of it, hence the Stratus measurements were close to the centre of these two eddies (Supplement Movie M1). However, satellite data show that the mooring captured only the northern segment of the ACE1 (Fig. 4b), therefore the radius of 28 km determined from measurements at the Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License.
Stratus mooring is small in comparison to a mean radius of 40 km from satellite maps (Fig. 5a). Oxygen anomalies from January to March 2015 are related to by two consecutive cyclonic eddies explaining the long lasting and strong anomaly. The first eddy (CE) passed the mooring 43 km north of it and then merged with a second cyclonic eddy and passed to the south of the mooring.
Therefore, eddy events lead to a strong signal in water mass properties up to 196 (40)

Net transport of heat, salt, and oxygen via eddies in the ETSP
The question on how much anomalous water properties an eddy is able to trap and transport into the open ocean depends on the relation between swirl velocity and propagation velocity. The MWE and the 15 ACE2 have a similar propagation velocity of 4.3 and 4.2 cm s -1 , respectively. The CE propagates fastest (6 cm s -1 ) and the ACE1 propagates slowest (3.2 cm s -1 ) (Fig. 6a), which fits well to the mean westward propagation speed of 3-6 cm s -1 and 4.3 cm s -1 estimated for eddies in the region off Peru (Chaigneau et al., 2008;2011).
The observed swirl velocity at Stratus mooring is in accordance with other values measured in the ETSP 20 (Chaigneau et al., 2011;Stramma et al., 2013; Below 250 m depth, the swirl velocity of ACE1 is significantly weaker than the swirl velocity of ACE2. Nonetheless, due to the much slower propagation velocity of ACE1 the fluid stays trapped within the eddy (U/c >1) leading to a deep vertical extent of both anticyclones of 504 m (ACE1) and 523 m (ACE2) as described for mean anticyclones in the ETSP (Chaigneau et al., 2011). 5 Within the eddy boundaries of the two anticyclones (ACE1 and ACE2), positive anomalies of temperature and salinity were observed between 100 and 600 m depth (Fig. 6b, c) Oxygen shows a mainly negative anomaly below 220 and 280 m depth respectively within the anticyclones MWE, AC1 and AC2. Both, the MWE and the ACE2 are having their largest negative 15 anomalies in 250 m depth and a second minimum in 450 m, which is just above and below the core of the OMZ indicating the transport of low oxygenated water masses from a region with a larger vertical expansion of the OMZ (Fig. 6d). In the core depth of the OMZ at 350 m depth only weak negative oxygen anomalies are possible as the oxygen content is already low, but still the passage of the stronger anticyclone ACE2 results in oxygen decrease by 14 µmol/kg. 20 Water mass anomalies within the MWE lead to an available heat, salt and oxygen anomaly (AHA, ASA, AOA) of 0.8 x 10 18 J, -3.6 x10 10 kg and -3.5 x10 16 µmol. These values are about five times smaller in comparison to a mode-water eddy that was also measured at the Stratus mooring in eddy observed by Stramma et al. (2014) is generated in April, when the PCUC, which transports oxygen-deficient ESSW (Hormazabal et al., 2013), has its poleward maximum (Shaffer et al., 1999;Penven et al., 2005;Chaigneau et al., 2013). Additionally, the mode-water eddy observed by Stramma et al. (2014) was generated in year 2011, which is considered as a La Niña-period, when generally lower oxygen and higher salinity values exist in the upper 100 m depth (Stramma et al., 2016) leading 5 to higher anomalies of salt and oxygen of the MWE observed by Stramma et al. (2014) in comparison to the actual MWE. In addition, the MWE passed to the north of the Stratus mooring during its deployment, hence the method to define the fully MWE parameter might lead to higher deviations to the real eddy parameters.
The volume of ACE2 (4.6 x10 12 m 3 ) is in agreement to the mean anticyclones (Chaigneau et al., 2011) 10 and the open-ocean anticyclone (Stramma et al., 2013) but three times larger than the weaker ACE1, which is partly due to the underestimated radius (Tab. 1). The AHA, ASA, and AOA of ACE2 are 7.6 x 10 18 J, 23.9 x10 10 kg and -3.6 x10 16 µmol and therefore far greater than the AHA, ASA and AOA of x10 10 kg. Oxygen shows a strong negative anomaly in the upper 320 m having a maximum anomaly of -69 µmol/kg in 180 m depth (Fig. 6d) due to the uplift of the main thermocline leading to a negative AOA of -6.5 x10 16 µmol for the 107 to 176 m depth layer. Although the volume of the CE is in good agreement with the mean values of Chaigneau et al. (2011), the estimated AHA and ASA are much higher (Tab. 1), which is likely due to strong seasonal variations during the generation of the CE. The 5 CE is formed in austral winter off Peru when coastal alongshore winds intensify leading to an enhanced upwelling of cold as well as nutrient-rich and oxygen-poor water due to high biological production.
Additionally, equatorward surface currents, which transport the relatively cold and fresh Eastern South Pacific Intermediate Water, are strongest during austral winter (Gunther, 1936).
Fluxes of mass, heat, salinity, and oxygen are estimated from the volume, AHA, ASA, and AOA (Tab. 10 1) for the period in which the MWE, ACE2, and CE cross the 85°W longitude at the Stratus mooring.
The ratio of the heat fluxes of the three different eddies mirror the differences between volume, AHA, ASA, and AOA of the respective eddies because of the similar duration of the passages of the eddies.
Transports of mass anomaly for the CE, ACE2 and MWE are again similar and range between 1.1 Sv for the CE to 1.8 Sv for ACE2 (Tab. 2). Due to the local conditions and different water masses during 15 its formation, the CE shows a strong negative transport of heat (-4.0 x 10 12 W) and salt (-18.1 x 10 4 kg s -1 ) across the mooring, whereas the ACE2 transports the highest positive amount of heat (3.0 x 10 12 W) and salt (9.5 x 10 4 kg s -1 ) per year. Whereas the transport of heat and salt of the MWE is relatively small in comparison to the CE and ACE2, the transport of low oxygen water of -1.8 x 10 10 µmol s -1 is in the same range as the CE and ACE2 due to a thick lens of low-oxygen water within the MWE. 20 Available anomalies of heat, salt, and oxygen of cyclonic and anticyclonic eddies gained from the Stratus mooring and from the literature (Table 1)

Properties of the observed eddies MWE, ACE1/2 and CE during their lifetime 15
With the help of satellite data the four eddies (MWE, ACE1/2 and CE) could be identified and followed As expected from the polarity depending meridional deflection of all eddies (anticyclonesequatorward, cyclones -poleward) also the individual pathways of the ACE1 and ACE2 show a northwestward direction whereas the CE migrates more south-westwards (Fig. 1a). Note, that the MWE shows no clear meridional deflection on the way to the west.
Anticyclonic eddies (MWE, ACE1/2) are associated with a positive SLA, wherein ACE2 shows the 5 strongest mean elevation of all anticyclonic eddies of 8 cm (Fig. 4c) and cyclonic eddies are identified by a negative SLA, wherein the CE shows a mean minimum SLA of -2 cm in the centre of the eddy (Fig. 4d). Nevertheless, the CE showed the largest SLA differences at the Stratus mooring (Fig. 3a). In general, mode-water eddies are difficult to detect by satellite altimetry due to a relatively weak velocity at the near-surface (Fig. 6a) which is generated by the typical distribution of the isopycnals. Therefore, 10 it is noteworthy that the MWE has a stronger SLA (7 cm) than the relatively weak ACE1 (6 cm). The SLA of the MWE indicates higher variability during its lifespan than the other eddies (Fig. 5e). The maximum SLA of the anticyclonics is obtained during their mid-age, whereas the SLA of the CE decreases after the very beginning.
Due to the uplift of the seasonal pycnocline in both eddies, MWE as well as CE, cold and nutrient rich 15 water is upwelled into the euphotic zone leading to enhanced biological production, which is reflected by negative SST anomalies of -0.04°C (MWE, Fig. 4e) and -0.08°C (CE, Fig. 4h) and high chlorophyll production of 0.19 mg m -3 (CE, Fig. 4l). Surprisingly, the mean SST (Chl) of the ACE1 and ACE2 are negative (high) and around zero, respectively, as one would expect positive (low) SST (Chl) anomalies due to the depression of the thermoclines. The development of the SST predominantly shows negative 20 anomalies with short periods of positive anomalies for the anticyclonics (Fig. 5f).
The development of further eddy properties (radius [km], rotating velocity [m s -1 ], nonlinearity parameter and Rossby Number) during the normalized lifespan are shown in Figure 5 indicating that the observed eddies pass the Stratus mooring during their mid-age (MWE and ACE2) and during the end of their lifetime (ACE1 and CE). The anticyclonics ACE1/2 have their maximum radius during the last 25 third of their lifetime, whereas the development of the radius of the MWE is symmetric and the radius of the CE increases during the first third of its lifetime (Fig. 5a).
Water mass anomalies can only be preserved within an eddy if the feature is nonlinear and maintains its coherent structure. During their full lifetime the nonlinear parameter U / c > 1 for all eddies confirming the coherent feature (Fig. 5c). MWE and CE show stronger fluctuations of the nonlinear parameter than the ACE1/2, which mirrors the higher variability of the swirl velocity of both eddies (Fig. 5b).
All eddies indicate a Rossby number R o <1 describing the typical scale for mesoscale eddies (Fig. 5d). 5 The mean life cycle of an eddy consists of a growth and decaying phase, both lasting about 20% of its lifetime and a stable phase in between (Liu et al., 2012;Frenger et al., 2015;Samelson et al., 2014), which is not consistent with our observations showing Rossby numbers of less than 0.1 for ACE1 and ACE2 reflecting a stable phase over 90 % of the lifetime of the eddies. In contrast to this, MWE and CE show a longer growing phase of 30 % and also a longer decaying phase of 40 %, where the stable phase 10 with a Rossby number of less than 0.1 remains short.  (Fig. 8a). The warm, salty and 20 oxygen-depleted water mass of the core of the ACE2 coincides with the water mass of the likely formation region of the ACE2 obtained from the MIMOC climatology (Fig. 9a, b) reflecting the characteristics of the oxygen-depleted ESSW. The ESSW is carried poleward by the secondary southern subsurface countercurrent (Montes et al., 2014), feeds the subsurface PCUC (Hormazabal et al., 2013) and is then transported along the Peruvian and Chilean coast where anticyclonic eddies are likely 25 generated (Chaigneau et al., 2011).

Observations of Argo floats within the eddy-core of ACE2 during its mid-age
In September 2014, about four months later and more than 6° further west at 83.4°W, 19.4°S, a second float (#6900530) was trapped in the same eddy ACE2 (supplement: movie M1). The core was still Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License. located at isopycnal 26.4 kg m -3 (~230 m depth) showing minimum oxygen values of less than 4 µmol kg -1 (Fig. 7b) and maximum salinity of more than 34.8 psu (Fig. 8b) between 200 and 280 m depth.
However, the vertical extent of anomaly high salinity and anomaly low oxygen has decreased (Fig. 9a,   b) which is likely due to lateral mixing by turbulent diffusion at the boundary of the eddy. Mixing mostly takes place above the core of the eddy between the density layers σθ=25.7 and σθ=26.3 kg m -3 5 (supplement Fig. S2) with largest changes in oxygen (0.5 µmol kg -1 day -1 ), temperature (-0.007°C day -1 ), and salinity (-0.002 psu day -1 ) at about σθ=26.0 kg m -3 . This density level corresponds to a depth between 100 and 170 m, where high velocities exist within the ACE2 (Fig. 6a), which are essential to keep up the coherent structure and therefore should inhibit lateral mixing.
Shortly after, in October 2014, a third float (#6900529) stayed in the ACE2 at about 82.8°W, 20.4°S. 10 The strongest water property anomaly is now located at isopycnal 26.6 kg m -3 (~310 m depth) showing minimum oxygen of less than 8 µmol kg -1 (Fig. 7c). The salinity anomaly transported within the eddy has declined furthermore to about 34.65 psu (Fig. 8c). The development of the water mass properties within the eddy points towards mixing along density surfaces between σθ=26.0 kg m -3 (~190 m) and σθ=26.5 kg m -3 (280 m). The changes are strongest above the core of the eddy at about σθ=26.3 kg m -3 15 (~205 m to 240 m) showing the mixing of oxygen-rich (2.8 µmol kg -1 day -1 ), colder (-0.07°C day -1 ) as well as fresher water (-0.017 psu day -1 ) into the ACE2 (supplement Fig. S2).
Decreased anomalies might also be related to the fact that the float did not capture the eddy centre as it was located in the southeast rim of the eddy (Fig. 4c). Whereas the salinity measurements of the float differ from those of the Stratus measurements obtained during the passage of the ACE2 (Fig. 9b), the 20 oxygen anomaly transported in the core agrees well (Fig. 9a).
After the ACE2 has passed the Stratus mooring in November 2014, the last of the four floats (#6900527) was trapped at the southern rim of the eddy (Fig. 4c) from December 2014 to January 2015 at about 86.4°W, 20.4°S. The eddy core is still clearly visible, although water mass properties within the core of the ACE2 has further changed (oxygen > 10 µmol kg -1 , Fig. 7d) and the vertical extent of the 25 eddy has declined. Mixing of slightly oxygen-richer water can be observed in the whole eddy (supplement Fig. S2a), whereas warmer and more saline water is entrained in the upper part of the eddy Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License.
(σθ > 26.3 kg m -3 ≅ 240 m depth) and colder and fresher water below. These changes might also be due to the location of the float outside the eddy boundary.

Discussion and conclusion
The ETSP is known for its high eddy frequency (Chaigneau et al., 2008). There is still limited knowledge in this region about the dynamics of eddies especially on their effective transport and their 5 dissipation. In this study the activity of three different types of eddies (mode water, anticyclonic, and cyclonic eddy) during their westward propagation was investigated from the formation area in the upwelling area off Peru and Chile into the open ocean. The focus was on the development of the eddies, seasonal conditions during their formation, and the change of water mass properties transported within the isolated eddies using a broad range of observational data such as SLA, SST, and Chl from satellites 10 as well as hydrographic data and oxygen from the Stratus mooring and from Argo floats.
Available heat, salt, and oxygen anomalies could be computed for the investigated eddies. Generally, heat and salt anomalies transported within eddies are positive for anticyclones and negative for cyclones and might be compensated as they are of about the same amount (Chaigneau et al., 2011). In contrast, in this study negative anomalies of the water mass properties within the observed cyclonic eddy are too 15 high for heat and salt (-9.4 x10 18 J, -42.8 x10 10 kg) as they could be balanced by positive anomalies of heat and salt transported within the anticyclonic (7.6 x10 18 J, 23.9 x10 10 kg) and the mode water eddy (0.8 x10 18 J, -3.6 x10 10 kg). Therefore, seasonal variability such as fluctuation of alongshore upwellingfavourable winds off Peru and Chile as well as interannual variability such as El Niño/La Niña have an impact on the water mass properties trapped and transported within eddies from the coast of Peru and 20 Chile into the open ocean reflecting the high variability of AHA and AHA. The AOA is negative for all types of eddies (MWE: -3.5 x10 16 µmol; ACE2: -3.6 5 x10 16 µmol; CE: -6.5 5 x10 16 µmol), whereby the transport of oxygen-low water from the upwelling region into the open ocean is more surface intensified due to the shallow structure of the cyclonic eddies.
Satellite-based estimate of the surface-layer eddy heat flux divergence, while large in coastal regions, is 25 small when averaged over the southeast Pacific Ocean, suggesting that eddies do not substantially Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License. contribute to cooling the surface layer in this region (Holte et al., 2013). In this study, the release of fluxes of heat (cyclones: -2.4 x 10 13 W m -2 ; anticyclones: 1.6 x 10 13 W m -2 ) and salt (-8.9 x 10 5 kg s -1 m -2 ; 5.0 x 10 5 kg s -1 m -2 ) estimated from eddies dissipating in the ETSP confirms the discrepancy between different types of eddies leading to a net transport of colder and fresher water from the formation regions off Peru and Chile into the open ocean. In contrast, all three types of eddies show a negative 5 oxygen flux of -2.0 x 10 11 µmol kg -1 s -1 m -2 for cyclones and -2.1 x 10 11 µmol kg -1 s -1 m -2 for anticyclones and mode-water eddies pointing towards an active role of eddies in maintaining and shaping the OMZ.
For the Atlantic Ocean the low-oxygen eddy cores has been attributed to high productivity in the surface (Schütte et al., 2016b), enhanced respiration of sinking organic material at subsurface depth 10 (Fiedler et al., 2016) and a strong isolation of the eddy core (Karstensen et al., 2017). An anticyclonic mode water eddy observed at the Stratus mooring in February/March 2012 indicated high primary production just below the mixed layer (Stramma et al., 2014). According to a global investigation the eastern South Pacific off Peru and Chile seems to have the highest amount of MWEs, which are also deep reaching compared to other regions (Zhang et al., 2017). Nevertheless, in the mooring deployment 15 period 2014/2015 only one MWE crossed to the north of the mooring and the results have to be regarded with caution. Even though the AHA and ASA of the MWE are small in comparison to both the anticyclonic eddies and the cyclonic eddy the transport of low oxygen water is in the same range as the other eddies due to the typical thick lens of low oxygen water within mode water eddies.
From a combination of satellite data and Argo profiles long-lived eddies (lifetime longer than 30 days) 20 in the Peru-Chile upwelling system 55% of the sampled anticyclonic eddies had subsurface-intensified maximum temperature and salinity anomalies below the seasonal pycnocline, whereas 88% of the cyclonic eddies are surface intensified (Pegliasco et al., 2015). The 55% subsurface-intensified anticyclonic eddies represent mode water eddies while the 45% surface intensified anticyclones are 'regular' anticyclonic eddies. Eddy generation off the coast between 8° and 24°S peak in austral 25 summer/spring, which agrees with the strengthening of the PCUC and the possible mechanism of the generation of eddies due to instabilities. However, this is not in agreement with model simulations showing an eddy-induced offshore transport off Peru that peaks in austral spring (winter) at the southern Ocean Sci. Discuss., https://doi.org /10.5194/os-2018-5 Manuscript under review for journal Ocean Sci. Discussion started: 31 January 2018 c Author(s) 2018. CC BY 4.0 License.
(northern) boundary of the OMZ (Vergara et al., 2016). As the PCUC also shows strong fluctuations lasting a few days up to a few weeks (Huyer et al., 1991) it might be difficult to determine a seasonal dimension between the PCUC and the rate of eddy formation. El Niño (La Niña) events deepen (shoal) the thermocline and intensify (weaken) the PCUC (e.g. Montes et al., 2011;Combes et al., 2015) and therefore might play a role for the amount of eddy generation on interannual time scales. and oxygen (OX) differs. It is important to note that the radius of the ACE1 might be underestimated.