Hydrographic data

We conducted full-depth conductivity-temperature-depth (CTD) and Lowered Acoustic Doppler Current Profiler (LADCP) casts at 29 stations along four meridional lines (L1, L2, L3 and L4 in Fig. 1a) from 29 January to 10 February 2020. This survey was conducted aboard the ice-breaking research vessel ARAON (Korea Polar Research Institute, KOPRI). The distance between each line (each station) was ~17 km (7 km). The first baroclinic Rossby radius of deformation, theoretically considered as the lower boundary of the ocean circulation radius³⁴, was estimated to be ~6 km in PIB. The spacing of the 2020 observations (~7 km) is comparable to the first baroclinic Rossby radius of deformation, indicating that our sampling space is sufficiently small to capture the ocean gyre circulations at horizontal scales of a few tens of kilometres in this area. The four meridional lines nearly cover the region where the cyclonic gyre was detected in 2009²⁰ (Fig. 1a). Full-depth CTD/LADCP casts were also conducted along the IF line (nine stations) aboard the ice-breaking research vessel Nathaniel B. Palmer (National Science Foundation, NSF; cruise NBP-2002), on 20 February 2020, after the ‘B49’ iceberg calving event. We only used full-depth CTD profiles along the IF line to estimate the meltwater flux from PIIS in 2020. All CTD profiles were measured using a SBE911 (Sea-Bird Electronic incorporated, US) with dual temperature and conductivity sensors. A down-looking Teledyne/RDI 300 kHz Workhorse-type ADCP attached to the CTD frame was used as a LADCP to measure the profiles of horizontal currents at a 5-m depth interval with an accuracy of typically ±0.5 cm/s (http://www.teledynemarine.com/).

The CTD data from both surveys were processed using the method recommended by Sea-Bird Electronics Incorporated³⁵. All CTD profiles were arranged at a 1 m depth interval. The dissolved oxygen (DO) sensor data were calibrated using the following equation:

Equation (¹) is based on the linear regression between the DO measured in 40 bottle samples using the Winker method and DO sensor data from four stations in PIB (not shown). We used two DO sensors with calibration dates of 4 April 2019 and 1 June 2019. The R² value for the correlation between the bottle samples and sensor values was 0.999.

The LADCP data were processed using the standard method³⁶, and de-tiding was not applied to the LADCP data because the observed current velocities were significantly stronger than the tidal velocities (<1 cm/s) in PIB²⁰. The horizontal currents in PIB were also observed by ship-based ADCP (SADCP) from 29 January to 25 February 2020. The NBP had two Teledyne RDI ADCPs³⁷, both Ocean Surveyor models (phased array) operating at 75 kHz in narrow and broadband modes and at 38 kHz in narrowband mode. The lower-frequency ADCP could reach 1000 m in good conditions. However, the range was typically less in the PIB environment. A thick ice-protection window impacted the high-frequency ADCP. Therefore, it only reached 100–150 m in broadband mode and 400–450 m in narrowband mode. The University of Hawaii Data Acquisition System (UHDAS) combined ADCP and navigational data streams and used Common Ocean Data Access System (CODAS) processing to incrementally build a dataset of averaged (15 min) edited ocean velocities for each ADCP and ping type specified. The SADCP data, which had percent-good (percentage of available pings in an ensemble) values lower than 90%, were removed from this analysis. We used the horizontal ocean currents obtained by averaging the currents observed at three frequencies. In addition, historical SADCP data observed in 2009²⁰ and 2014 were also used to identify horizontal ocean circulations in PIB (Supplementary Fig. ¹).