Physical and biogeochemical oceanography data from underway measurements with an AquaLine Ferrybox during the Antarctic Circumnavigation Expedition (ACE) ***** Dataset abstract ***** This data set contains measurements from various sensors installed on the Aqualine Ferrybox system that was connected to the underway seawater supply in the Southern Ocean during the Antarctic Circumnavigation Expedition (ACE). Data was collected continuously except for periods when the pump of the underway system was switched off or the system was turned off. Data collection covers all three cruise legs in the period 24th December 2016 to 18th March 2017. Data collected with the CTG MiniPack CTD-F are temperature, salinity, pressure, and turbidity. Data collected by the Aanderaa oxygen optode include dissolved oxygen and oxygen saturation. An SBE 18 sensor measured pH. The CTG UniLux fluorometer measured chlorophyll-a concentration. All data has been quality controlled and post-cruise calibrated. Data is provided at 1-minute intervals along the cruise track. In addition, we provide satellite data (sea-surface temperature, sea-surface height, geostrophic velocity, sea-ice concentration) that was interpolated to the cruise-track and an estimate of frontal positions to supplement this underway data set where data was missing or for additional information. This circumpolar data set provides insights into the circumpolar surface ocean conditions and biogeochemistry of the Southern Ocean during one austral summer season. Note on version 1.0: The first version of this data set only contains temperature, salinity, pressure, and potential density in the post-processed file, since post-processing and quality control for turbidity, chlorophyll-a, dissolved oxygen, oxygen saturation, and pH have not been finalized. These variables will be added to the post-processed data file in a future release. ***** Original data collection ***** The AquaLine FerryBox system (Chelsea Technologies Ltd., CTG) on the Akademik Tryoshnikov continuously recorded seawater measurements from the underway line. It was located in the CTD wetlab. This system consists of four main sensor packages: CTG MiniPack CTD-F: SBE CTD sensor and Fluorimeter/Nephelometer that measures turbidity Variables: Temperature, Conductivity/Salinity, Pressure, Turbidity AANDERAA Oxygen Optode 3835 (https://www.aanderaa.com/media/pdfs/Oxygen-Optode-3835-4130-4175.pdf) Variables: Dissolved Oxygen, Oxygen Saturation, Temperature SBE 18 pH (https://www.seabird.com/ph-sensors/sbe-18-ph-sensor/family?productCategoryId=54627869929) Variables: pH CTG UniLux Chlorophyll fluorometer (https://www.chelsea.co.uk/unilux) Variables: Chlorophyll-a The Ferrybox sensors were sent to Chelsea Technologies Group LTD (manufacturer) for calibration prior to the cruise (September 2016). The sensors arrived at the ship the day before leaving port from Bremerhaven and were installed at sea. Chelsea sent technicians to the ship in Southampton to complete the installation. It was connected to the underway line, which pumped seawater into the ship at an intake at 4.5 m depth at the front of the ship. When the underway water was turned on (November 2016) it was discovered the pump was rusty and the sensors were quickly coated in rust and failed to perform to standards. Subsequently, all Ferrybox sensors were cleaned during leg 0 and re-installed. It was connected to the replacement pipes that consisted of PVC pipes insulated with a ca. 1 cm thick layer of foam. This system was installed at the beginning of leg 1 with a new pump (see Walton and Thomas (2018) for details). As the pump was not self-priming it went off several times in rough weather. During these times it was manually re-primed. The pump was also switched off in close proximity of land at some points due to too much organic material in the water or during the stops between legs. All these instances are reported in the pump log file (ace_pump_log.csv). The ferrybox was switched on December 24th. Until January 8th data was extracted manually from the ferrybox storage and archived automatically every day after that, with a CSV file written daily at 2am UTC. This data was automatically backed up to the ACE shipboard server. Note that the binary files from the Ferrybox have not been saved so the data did undergo some preliminary processing by the Ferrybox system itself before they are output in the ASCII files (ace_ferrybox_data_raw), but these data should be considered as unprocessed and of unknown quality. The data is stored separately for each output channel described in “ace_ferrybox_channel_description.csv”. It was found that the raw data most likely has a warm bias of approximately 1 degC due to the PVC pipe running through the ship and heating up the seawater before reaching the instrument. Thus, substantial post-processing of the data was needed as described below. ***** Data processing ***** STEP 1: Manual correction of raw data file formatting and bad characters The raw data (ace_ferrybox_data_raw) was stored as an ascii (comma-delimited) table with 47 variables/columns in one row and each row representing one data record. The first issue that arises is that there are inconsistencies in the raw data record formatting. These include (1) rows with more or less than 47 variables per row, (2) string characters within floating point numbers, and (3) skipped or shifted data entries. Therefore, a Matlab script was written to identify these issues and manually correct them in the raw data file in the following steps: 1) Reading each row as comma separated string and checking the number of variables 2) Manually correcting the respective rows where there were not 47 variables by either replacing faulty “,” signs, beginning a new row where the change of line character was missing, or filling up missing variables with “NaN”. 3) Checking the string formatted data from column 6 (“Speed”) to column 46 “Spare Channel 20” for any characters other than “0-9”, “.”, “-“, or “NaN” 4) Manually fixing issues with bad characters by either replacing with the obvious correct character if appropriate or replacing the entire data point with “NaN” if in doubt. 5) Deleting entire rows with obviously wrong or missing date, time, longitude, latitude, information, except for numerous wrong latitude values that obviously have the longitude instead of latitude written in the latitude column and latitude missing throughout the data set since there were too many to delete manually from the file. This will be taken care of in step 2 (see below). 6) Deleting files that had all “0” data entries. 7) Saving the difference between “ace_ferrybox_data_raw” and step 1 processed files for documentation of the changes into file “ace_ferrybox_processing1_modified_data.txt”. All modified or deleted files are listed in “ace_ferrybox_processing1_modified_files.txt”. STEP 2: Filtering of bad or missing data In this second post-processing step we are filtering bad or missing data from the record. This procedure also includes periods when data was recorded despite the pump being off or if data is too noisy due to high waves. In addition, we filtered bad data of certain channels that arose from numerical issues that were induced for unknown reasons to the raw data files by the Ferrybox system and could not be recovered. This post-processing is done in the following steps: 1) Deleting rows with bad latitude values and rows with only “0” data values (reducing the number of records/rows from 353’912 to 345’414 during legs 1 to 3). 2) Deleting periods 5 minutes before until 5 minutes after the pump was switched off according to records (ace_pump_log.csv, see below) and visual post-inspection of especially the turbidity variable that shows unrealistic positive outliers during periods when the water flow in the pipe was interrupted or disturbed by bubbles (reducing the number of records/rows from 345’414 to 249’475 during legs 1 to 3). 3) Data from channels 26 to 47 are for unknown reasons influenced by numerical issues (see “ace_ferrybox_channel_description.csv” and Figure 1a). Investigating this issue revealed that for some reason the actual data is at certain points (regular intervals) mixed with data from other channels. It might be that records have been shifted after channel 26. We used the MiniPack supply voltage (channel 28, Figure 1b-c) to filter bad data from the pressure (channel 26) and turbidity (channel 27), whenever the voltage value was above 24V. This procedure removed all numerical issues but also reduced the amount of data from these two channels by 188’821 values. An alternative approach would be to filter everything with a MiniPack supply current (channel 29, Figure 1d-e) below 80A. However, it turned out that this approach would filter some additional records that seem accurate in the pressure and turbidity channels, suggesting that the numerical “shifts” might increase with increasing channel numbers. The effect of this filter is shown in Figure 1f, where all red records have been removed. 4) Setting all data where pressure on the line was below 8dbar to “NaN” 5) Setting all data where pressure exceeded the mean by more than 4 standard deviations to “NaN” 6) Setting all data where temperature was either below -3°C or above 30°C to “NaN” 7) Setting exact “0” values to “NaN” Figure 2 shows the temperature (a), salinity (b), and turbidity (c) data after this filter step (blue) and the filtered data (red). STEP 3: Correcting biases 3.1 Temperature: We compare both the MiniPack and Optode temperature sensors to the CTD data (Henry et al., 2020) and the XBT data (Haumann et al., 2020b), in order to determine the time evolving temperature bias of the Ferrybox. For this comparison, we use an average temperature over the upper 15m of each profile (CTD up- and downcasts are treated separately). Profiles with a standard deviation larger than 0.1°C in the upper 15m have not been used. Note that the sensor biases are independent of any chosen depth threshold smaller or equal to 15m, i.e. similar if they would be referenced to a CTD/XBT level of 5m, but the noise in the data and the amount of reference points increases when averaging data over the upper 15m. The biases start increasing for values larger than 15m. This procedure results in a total number of 123 reference points (CTD: 91, XBT: 32) that can be used to evaluate and correct the Ferrybox temperature bias. For the comparison we averaged the Ferrybox data 10 minutes before and after the recorded start time for CTD downcasts, the recorded end time for CTD upcasts, or the release time for the XBTs. Figure 3 shows the results of the comparison. The MiniPack temperature sensor (blue) has a mean bias of +0.83 degC compared to the CTD sensor (blue dots), and +0.85 degC compared to the XBT data (blue crosses). The bias varies between +1.26 degC and +0.27 degC (blue dots and crosses in Figure 3). The Optode temperature sensor (gray) has a mean bias of +1.76 degC compared to the CTD sensor (gray dots), and +2.09 degC compared to the XBT data (gray crosses). The bias varies between +3.04 degC and +0.53 degC (gray dots and crosses in Figure 3). The bias might be caused by either a mean bias and drift in the sensor or by a difference between the temperature measured inside the ship and outside the ship. Given that the time evolution of the bias has a similar shape for the two different Ferrybox temperature sensors, an external source of the bias, i.e. an actual difference in the water temperature, seems to be the dominant cause of the bias. The latter is most likely the heating of the water as it flows through the underway line system in the interior of the ship, which would cause a time evolving bias that depends on the difference between the water temperature and the interior ship temperature. The first point of comparison occurs on Jan. 7th and the last on Mar. 17th, 2017. As one can see from Figure 3, the temperature bias of the first period during leg 1 (prior to Jan 7th, 2017) is difficult to assess due to the absence of CTD or XBT data. Therefore, a mean bias of the first CTD casts (first 5 points of comparison) on Jan. 7th 2017 is applied to this period. The bias of the period after Mar. 17th, 2017 is set to the last point of comparison. Between this period, we fit a 10th order polynomial to estimate the bias correction (thick dotted blue and gray lines in Figure 3). The estimated root-mean-square errors for the temperature corrections from the residuals towards the polynomial fit are ±0.14 degC for the MiniPack sensor and ±0.31 degC for the Optode sensor. The resulting mean bias of both corrected temperature sensors (blue in Figure 4) with respect to all CTD and XBT data are less than 0.005 degC after the bias correction has been applied. The corrected MiniPack temperature (blue in Figure 4a) offset towards the CTD and XBT sensors has a maximum of 0.37 degC and a minimum of -0.58 degC. The corrected Optode temperature (blue in Figure 4b) offset towards the CTD and XBT sensors has a maximum of 0.78 degC and a minimum of -1.01 degC. We estimate the final temperature error from residuals to be ±0.13 degC for the MiniPack sensor and ±0.28 degC for the Optode sensor (95% confidence interval). The corrected Optode temperature remains 0.11 degC warmer than the corrected MiniPack temperature (standard deviation of the difference is ±0.25 degC). These remaining differences are within the estimated uncertainty and mostly arise from periods with few or no reference points (CTD/XBT) are available to better constrain the bias, i.e. mostly leg 1. The Optode temperature sensor is expected to have a reduced accuracy and precision compared to the MiniPack sensor and clearly has a larger bias. Therefore, the temperature used for the final data set as well as all other corrections is the one of the MiniPack sensor. Temperature data prior to Jan. 7th, 2017 has a larger and unknown uncertainty in the bias correction. 3.2 Salinity: We compare the MiniPack salinity sensors data to the salinity samples collected from the underway line (Haumann et al., 2020a), in order to determine the time evolving salinity bias of the Ferrybox. For this comparison, we removed a number of salinity samples that have been flagged previously to be of questionable or poor quality and averaged the Ferrybox data 10 minutes before and after the time that the sample was taken from the underway line. This procedure results in a total number of 260 data points that can be used to evaluate and correct the Ferrybox salinity bias. Figure 5 shows the results of the comparison. The uncorrected MiniPack salinity sensor (red) has a mean bias of -0.147 compared to the salinity sample data. The bias varies between -0.556 and +0.096 (red dots in Figure 5). As the salinity has been measured from the same source water in the samples and by the Ferrybox, the bias is most likely caused by an offset of the MiniPack temperature and conductivity (and pressure) sensors. Since the temperature bias (see above) is largely related to the heating of the water in the underway line, it is impossible to assess which fraction of the temperature bias is associated with a bias in the temperature sensor and which fraction with the change in the water property itself. The latter effect would not affect the salinity bias itself as the heating of the water would equally change the measured temperature and conductivity. Therefore, we can also not assess the conductivity bias itself but only the final salinity bias. The bias during the first leg is much smaller and more constant than during the other legs (Figures 5), suggesting that the sensors considerably degraded over time. A linear bias correction seems not adequate since the bias seems to follow the temporal gradient of the salinity. This temporal evolution of the bias might be caused by an increased response time of the sensor to the water property change throughout the cruise. The first point of comparison occurs on Dec. 25th, 2016 and the last on Mar. 18th, 2017. We fit a 10th order polynomial to estimate the bias correction (thick dotted gray line in Figure 5). The estimated root-mean-square error for the salinity correction from the residuals towards the polynomial fit is ±0.074. The resulting mean bias of the corrected salinity sensor (blue in Figure 5) with respect to all sample data is less than 0.001 after the bias correction has been applied. The corrected MiniPack salinity offset towards samples has a maximum of 0.34 and a minimum of -0.29. We estimate the final salinity error from residuals to be ±0.142 for the MiniPack sensor (95% confidence interval). STEP 4: Temporal averaging and formatting Since filtering the data led to data gaps, the data is at irregular time intervals, and Ferrybox data is provided at a temporal resolution that is much higher than can be accurately provided, we proceed to average the data to regular 1-minute intervals. These 1-minute data are then matched up with the ACE GPS track dataset (Thomas and Pina Estany, 2019) using the Ferrybox timestamp. In order to filter noise and fill the 10-minute gaps that have been introduced by the filtering, we also apply a 15-minute running mean filter. We calculate the potential density anomaly with respect to 0 dbar pressure (sigma0) using the Matlab GSW toolbox (TEOS-10; McDougall and Barker, 2011). Data are then written to ascii (.csv) files. Pump log: A log file was kept during the cruise to note any time when the pump of the continuous underway water supply was turned on and off or failed. Reasons for a disruption in the pump were severe weather, waves, sea ice, or large amounts of organic matter in the underway system. This log file has been compiled into the file “ace_pump_log.csv”. It contains suggested periods of "bad" data from the underway system during ACE. These periods of "bad" data were identified based on the pump log kept on board (source = p), or by inspection of the Ferrybox data (source = v). Given the uncertainty associated with noticing the pump was off and the time taken for it to operate correctly following any problems, there is no clear cut-off of when data is "bad" and "good". Additionally, it typically took some time to completely flush the underway line with new water and some sensors needed time to adjust. Due to this transient behavior of the system, we suggest filtering out data within +/- 5 minutes of the "p/v" flags. However, this time window might be parameter, sampling, or sensor dependent and can be freely chosen by the user. There are also flags with a zero-second duration. These flags are typically based on single spikes occurring in the Ferrybox recording and users are free to choose their own time window around this spike to account for the tails of the spike. We again suggest a +/- 5 minutes time window around these zero-second duration "p/v" flags. Satellite data: We extracted satellite-derived sea-surface temperature, sea-ice concentration, sea-surface height, and surface geostrophic velocity (‘’ace_satellite_data_1min.csv”) along the cruise track of the Antarctic Circumnavigation Expedition (ACE) and interpolated them to the 1-minute resolution GPS data (Thomas and Pina Estany, 2019). All underlying satellite data are daily fields from the period December 20th, 2016 to March 20th, 2017. Daily Optimum Interpolation Sea Surface Temperature fields (0.25° resolution) are from the Advanced Very High Resolution Radiometer (AVHRR) infrared sensor (version 2; AVHRR-Only; Reynolds et al., 2007) distributed by NOAA (https://www.ncdc.noaa.gov/oisst/) and created on March 23rd, 2017. Daily passive microwave sea-ice concentration fields (25 km resolution) are the NOAA/NSIDC Climate Data Record (CDR; version 3, release 1; Peng et al., 2013; Meier et al., 2017), which is based on gridded brightness temperatures from the Defense Meteorological Satellite Program’s (F17) Special Sensor Microwave Imager/Sounder distributed by NSIDC (https://doi.org/10.7265/N59P2ZTG) and created on November 30th, 2017. Daily sea-surface height and geostrophic velocity (zonal and meridional component) fields (0.25° resolution) are the gridded and merged SSALTO/DUACS Delayed-Time Level-4 multi-satellite altimetry observations measurements (Altika Drifting Phase, Cryosat-2, Jason-3, OSTM/Jason-2 Interleaved, Sentinel-3A; version 5.9) distributed by CMEMS/Mercator Ocean (http://marine.copernicus.eu) and created on June 14th, 2018. We use the absolute dynamic topography, which is the sea surface height above geoid, and derived products. We use the 1-minute GPS date, latitude, and longitude record (Thomas and Pina Estany, 2019) to find the closest points in space and time in the gridded satellite products. We interpolate the four spatially closest satellite data points to the ship’s location using a distance weighted mean. This is done for the two closest fields in time, i.e. the daily average (centered at noon UTC) before and after the GPS record. These two points are then interpolated to the GPS time using a distance weighted mean. Note that while the resolution of the record is 1-minute along the cruise track, the actual temporal and spatial resolution is determined by the original product. That means that the data set for example does not capture any daily cycle as the satellite data consists of daily means. Sea-surface temperature and sea-surface height (incl. geostrophic velocity) have a degraded quality in the sea-ice covered region and sometimes contain missing values that are set to ‘NaN’. When using this satellite-derived data, please cite the original data source (satellite product). Merged product, ocean zones, and fronts: We created a merged product containing the Ferrybox, sample, and satellite-derived data (“ace_merged_uw_data_1min.csv”). It contains data every minute between 2016-12-21 07:00:00 UTC and 2017-03-19 09:46:00 UTC along the cruise track whenever GPS data (Thomas and Pina Estany, 2019) was available. The only variables that were merged into one single variable are the surface temperature and salinity data. They correspond to the Ferrybox temperature (flag = 1, red in Figure 6) and salinity (flag = 1, red in Figure 7) whenever the corresponding Ferrybox data were available. Missing data is filled with either salinity sample data (Haumann et al., 2020a; flag = 2) or the satellite-derived temperature (flag = 3, blue in Figures 6 and 7) whenever no Ferrybox temperature was available within ±0.25days. Any remaining gaps are interpolated (flag = 8, green in Figures 6 and 7) using the Matlab Modified Akima method. Matlab’s “Modified Akima cubic Hermite interpolation” is designed to avoid overshoots and is based on a piecewise function of third-order polynomials. This procedure is applied to all temperature gaps, and salinity gaps are only filled with interpolated values, whenever at least one salinity measurement was available prior and after each time step within a 6-hour window. All other variables contained in the merged product correspond to the original Ferrybox or satellite data files. The merged product also contains estimates of ocean zones (provided in “ace_merged_uw_data_1min.csv”) that are defined as the regions before and after crossing an oceanic front. We identified 22 frontal crossings along the cruise track. The estimated times and locations of the frontal-crossings are provided in “ace_fronts.csv” and are based on the merged product. For the frontal identification, we first estimate the distance since the beginning of the corresponding cruise leg (first GPS record in file) by calculating the cumulative some of the distances between each 1-minute latitude and longitude record. The distance is calculated using the M_Map MATLAB package (by Rich Pawlowicz) and the function “m_idist”, which calculates geodesics on an ellipsoidal earth (WGS84). Then, we use the values 25km before and after each point to estimate the along-track gradients in surface temperature and SSH. The identification of the fronts can be depicted in the temperature (Figure 6), salinity (Figure 7), and SSH (Figure 8) for each leg. We estimated the crossing of the Subtropical Front (STF) by finding the maximum surface temperature gradient between a surface temperature of 11.5 and 13 degC. This definition is motivated by Orsi et al. (1995) who argue that subtropical waters are warmer than 11.5 degC and that the front is marked by strong temperature decline of 4-5 degC over 100 km towards that value. The Subantarctic Front (SAF) resembles the northern boundary of the ACC and can be identified by a substantial southward decline in surface salinity and temperature as the fresher and colder Antarctic surface waters converge with saltier and warmer subtropical waters (Orsi et al., 1995). It is also associated with a steep decline in sea-surface height that turns negative (Kim and Orsi, 2014). We identify the SAF by finding the maximum temperature and SSH gradients between -0.25 and 0.1 m dynamic topography and 5 and 10 degC. We then visually subselect the maximum with the most northern extent of the low salinity surface water. The Polar Front (PF) and Southern ACC Front (SACCF) mark a further decline in the sea-surface height. However, there is not as a marked signal in the surface salinity as for the SAF and the surface temperature only shows weaker gradients during the summer-time. We identified the PF by selecting suitable absolute maxima in the surface temperature and SSH gradients between -0.35 m and -1 m dynamic topography and a surface temperature below 6 degC. We identified the SACCF by selecting suitable absolute maxima in the surface temperature and SSH gradients below -0.9 m dynamic topography and a surface temperature below 5 degC. We note that the choice of these fronts is somewhat subjective and could deviate from other estimates because of this subjective choice. Figure 9 shows a map of the frontal locations (x) and ocean zones (color), as well as the corresponding climatological mean frontal locations estimated by Orsi et al. (1995; light gray). Ocean zones are from north to south the Subtropical Zone (5, red) north of the STF, the Subantarctic Zone (4, orange) between the STF and SAF, the Frontal Zone (3, yellow) between the SAF and PF, the Antarctic Zone (2, light blue) between the PF and SACCF, and the zone south of the SACCF (1, dark blue). Note that the sea-ice covered ocean has not been separately declared in this definition. ***** Quality checking ***** Instrument uncertainties: pressure: ±0.05 dbar sea_water_temperature: ±0.003 degC Estimated measurement uncertainty: pressure: ±0.05 dbar sea_water_temperature: ±0.130 degC sea_water_practical_salinity: ±0.142 An independent comparison to satellite-derived sea-surface temperature (see above) reveals that the final corrected temperature from the Ferrybox is on average 0.195 degC warmer than the satellite data (Figure 10), where an overlap exists with a standard deviation of ±0.587 degC. This difference mostly results from the higher-latitude and colder regions during legs 2 and 3 (Figure 10). ***** Standards ***** If possible, we follow CF conventions version 72 (http://cfconventions.org/Data/cf-standard-names/72/build/cf-standard-name-table.html) ***** Dataset contents ***** Raw Data - ace_ferrybox_data_raw.zip Processed Data - ace_ferrybox_data_1min.csv Auxiliary Data - ace_satellite_data_1min.csv - ace_merged_uw_data_1min.csv - ace_fronts.csv Metadata - figure1.pdf, metadata, portable document format - figure2.pdf, metadata, portable document format - figure3.pdf, metadata, portable document format - figure4.pdf, metadata, portable document format - figure5.pdf, metadata, portable document format - figure6.pdf, metadata, portable document format - figure7.pdf, metadata, portable document format - figure8.pdf, metadata, portable document format - figure9.pdf, metadata, portable document format - figure10.pdf, metadata, portable document format - ace_ferrybox_channel_description.csv - ace_ferrybox_processing1_modified_data.txt - ace_ferrybox_processing1_modified_files.txt - ace_pump_log.csv - data_file_header.txt, metadata, text format - README.txt, metadata, text format ***** Dataset contacts ***** Jenny Thomas, Swiss Polar Institute, Switzerland. ORCID: 0000-0002-5986-7026. Email: jenny.thomas@epfl.ch, jen@falciot.net F. Alexander Haumann, Princeton University, USA; British Antarctic Survey, UK; ETH Zurich, Switzerland. ORCID: 0000-0002-8218-977X. Email: alexander.haumann@gmail.com Charlotte M. Robinson, Curtin University, Australia. ORCID: 0000-0001-8519-5641. Email: charlotte.robinson@curtin.edu.au, charlotte.mary.robinson@gmail.com ***** Dataset license***** This physical and biogeochemical oceanography dataset is made available under the Creative Commons Attribution 4.0 License (CC BY 4.0) whose full text can be found at https://creativecommons.org/licenses/by/4.0/ ***** Dataset citation ***** Haumann, F. A., Robinson, C., Thomas, J., Hutchings, J., Pina Estany, C., Tarasenko, A., Gerber, F., and Leonard, K. (2020): Physical and biogeochemical oceanography data from underway measurements with an Aqualine Ferrybox during the Antarctic Circumnavigation Expedition (ACE). (Version 1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.3660852.