This is a monthly gridded oxygen product for a transect across the Labrador Sea from August 2020 to July 2022, which was designed to calculate oxygen transport across this line. This product is created using a random forest regression trained on data from 60 carefully-calibrated oxygen optodes deployed along this line during this period. This algorithm was validated using both ship CTD and BGC-Argo oxygen profiles. The transect follows the OSNAP West line and provides mean monthly oxygen values across the entire water column.

Table of Contents

• Coverage
• Dataset Description
  • Methods & Sampling
  • Data Processing Description
  • Problem Description
• Related Publications
• Related Datasets
• Parameters
• Deployments
• Project Information
• Funding

60 calibrated and drift-corrected optodes deployed on the OSNAP array (for details on calibration see Miller et al., 2025) were used in this study. These optodes sampled every 15 minutes for the length of the 2-year deployment. Mean daily oxygen values from all 60 optodes spanning the 2-year deployment period (summer 2020-summer 2022) were used in this study. The dataset can be accessed here: https://www.bco-dmo.org/dataset/986667. 

The oxygen optodes were deployed with co-located CTD sensors (SBE37 microcats) for the duration of the 2-year deployment as a part of OSNAP. These co-located CTD sensors are used to generate daily and monthly, gap-free, gridded temperature and salinity fields across the OSNAP West and East lines (https://doi.org/10.35090/gatech/70342). OSNAP data were collected and made freely available by the OSNAP (Overturning in the Subpolar North Atlantic Program) project and all the national programs that contribute to it (www.o-snap.org).

Instruments

Oxygen:
Sensor: Aanderaa 4330 optodes integrated with RBR loggers
Oxygen optode dataset: https://www.bco-dmo.org/dataset/986667

CTD (co-located with oxygen optodes): Sensor: SBE37 microcats

CTD (from shipboard profiles): Specifications depend on specific cruise and research vessel. Please see relevant cruise reports for specifics.

BGC-Argo oxygen: Floats are equipped with oxygen optodes. The specific model (i.e. Aanderaa vs. SBE) depends on the float type. Data calibration and processing also depends on the float owner. For additional information see: https://archimer.ifremer.fr/doc/00354/46542/. 

Winkler-calibrated CTD oxygen profiles from optode deployment and recovery cruises were used to validate the trained random forest. 

 

AR45

Platform: R/V Neil Armstrong

BCO-DMO: https://www.bco-dmo.org/deployment/933794

Data access: https://www.bco-dmo.org/dataset/933743

 

AR46

Platform: R/V Neil Armstrong

BCO-DMO: https://www.bco-dmo.org/deployment/904871

Data access: https://www.bco-dmo.org/dataset/904721

 

AR69-01

Platform: R/V Neil Armstrong

BCO-DMO: https://www.bco-dmo.org/deployment/904879

Data access: https://www.bco-dmo.org/dataset/904721

 

AR69-03

Platform: R/V Neil Armstrong

BCO-DMO: https://www.bco-dmo.org/deployment/933797

Data access: https://www.bco-dmo.org/dataset/933743

 

MSM94

Platform: R/V Maria S. Merian

BCO-DMO: https://www.bco-dmo.org/deployment/990149

Data access: https://doi.pangaea.de/10.1594/PANGAEA.963675

 

M184

Platform: R/V Meteor

BCO-DMO: https://www.bco-dmo.org/deployment/990154

Data access: https://doi.pangaea.de/10.1594/PANGAEA.986330

Mean, daily oxygen values from 60 calibrated optodes deployed on the OSNAP array (for details on calibration see Miller et al., 2025) were used to train a random forest (RF) regressor to estimate oxygen from temperature, salinity, depth, longitude, bathymetry, year and day of year. Daily temperature and salinity measurements came from co-located CTDs that were deployed at the same depth as every optode. Bathymetry data was obtained from GEBCO (doi:10.5285/37c52e96-24ea-67ce-e063-7086abc05f29). For all datasets, the bathymetric depth was assigned using nearest-neighbor to the closest GEBCO grid point. No quality control filters were applied to the training data, and data were not normalized, etc. prior to training the RF, as this is not required for this machine learning algorithm. 

A random 80% of the data set was used for training the random forest, and the remaining 20% was used for testing the random forest. The root-mean-squared error (RMSE) for training and testing was 2.5 µmol/kg and 2.9 µmol/kg, respectively. The trained RF can be accessed here: https://github.com/ellenrpark/gohsnap_rf. 

The trained RF was validated by applying the RF to multiple validation datasets and comparing the RF-predicted oxygen values to the true oxygen values from these datasets. The following datasets were used for validation: 1) CTD oxygen profiles from deployment and recovery cruises (AR-45 (n = 150 profiles), AR-46 (n = 18 profiles), AR69-01 (n = 14 profiles), AR69-03 (n = 113 profiles), MSM94 (n = 23 profiles), M184 (n = 23 profiles)) within 50km of the OSNAP West line and 2) BGC-Argo oxygen profiles within +/- 3 degrees latitude and +/- 1.5 degrees longitude of the OSNAP West line with data with quality-control flags of 1, 2, or 8 (n = 219 profiles). The RMSEs for each data set are as follows: AR-45 (7.8 µmol/kg), AR-46 (6.6 µmol/kg). AR69-01 (n = 3.4 µmol/kg), AR69-03 (6.8 µmol/kg), MSM94 (4.6 µmol/kg), M184 (4.3 µmol/kg), and BGC-Argo (5.3 µmol/kg). 

The trained RF was then applied to the daily OSNAP gridded temperature and salinity fields and averaged to generate monthly oxygen estimates for this line. These daily temperature and salinity fields are gap-free and generated CTDs deployed across the OSNAP array using objective analysis (see Lozier et al., 2019 and Fu et al., 2023) at ~0.1º longitude and 20m resolution from the surface to bottom. Monthly versions of temperature and salinity fields are available (https://doi.org/10.35090/gatech/78023). OSNAP data were collected and made freely available by the OSNAP (Overturning in the Subpolar North Atlantic Program) project and all the national programs that contribute to it (www.o-snap.org).

We identified a potential bias in the RF predicted oxygen in the Denmark Strait Overflow Water (DSOW) due to a limited number of observations in this water mass. As a result a bias term (defined for the western, central, and eastern Labrador Sea) was included. This bias term was defined using World Ocean Atlas and CTD oxygen profiles from deployment and recovery. Both are included in this dataset and should be added (ex: DOXY + DOXY_BIAS_WOA) to get the final oxygen product. This bias term only applies to data with a potential density greater than 1028.XX kg/m3.

Coverage: Subpolar North Atlantic

NSF Award Abstract:
Every winter, frigid winds blowing eastward from the North American continent cool the surface waters of the Labrador Sea, which is situated between Canada and Greenland. As the ocean cools, oxygen and carbon dioxide are mixed from the atmosphere into a thick layer of water that ultimately spreads southward to fill a large volume of the North Atlantic and beyond. The presence of this water mass prevents the North Atlantic anywhere from becoming completely devoid of oxygen. Vertical mixing in the Labrador Sea also redistributes carbon dioxide into the deep ocean, where it can remain for hundreds of years, preventing it from contributing to the greenhouse effect. Yet, the processes governing the uptake of gases by the ocean are not well understood or quantified. Given that, over the last century, the ocean has become steadily more depleted in oxygen while also absorbing a large fraction of anthropogenic carbon dioxide, observing gas exchange processes is essential for understanding and predicting the evolution of the ocean and climate system. The circulation of the Labrador Sea has been monitored since 2014 with an array of instrumented cables extending from the seafloor to nearly the ocean surface. This project adds gas sensors to this array to investigate the rates and processes governing gas exchange. Through this project, a student and postdoc will be trained in interdisciplinary oceanography with a rich network of international collaborators. Responding to the need to increase public ocean literacy, the project scientists will work with University of Rhode Island’s Inner Space Center to broadcast live, interactive science sessions to educators at partner high schools and will follow-up with in-person science cafés at three participating schools.

Given the unique role of the Labrador Sea in providing a pathway for oxygen (O2) and carbon dioxide (CO2) to enter the intermediate depths of the ocean, a quantification and mechanistic understanding of the gas uptake and transport in the basin is a leading scientific priority. Oxygenation of Labrador Sea water prevents large-scale hypoxia from developing anywhere in the Atlantic Ocean and anthropogenic CO2 storage in the basin is the highest in the global ocean. The assumption that, in the Atlantic Ocean, O2 and CO2 uptake and their variability are tied to the dynamics of heat loss and the overturning circulation pervades the literature but has never been evaluated on the basis of direct observations. Thus, GOHSNAP (Gases in the Overturning and Horizontal circulation of the Subpolar North Atlantic Program) addresses this gap and the urgent need to better understand interactions between gas uptake, transport, and the overturning circulation. Specifically, this program will provide a continuous 2-year record of the trans-basin, full water column transport of O2 across the southern boundary of the Labrador Sea, leveraging the mooring infrastructure of the US-lead, international Overturning in the Subpolar North Atlantic Program (OSNAP). The addition of O2 sensors at various depths on this array, supplemented by observations collected by autonomous platforms will allow for the quantification of O2 export from the Labrador Sea. The data will further be used to empirically model carbon concentrations and estimate carbon export. Proposed instruments will also measure the mixed layer O2 and pCO2 for two winters, from which air-sea gas exchange will be calculated and compared against analogous observations in the convective interior of the Labrador Sea.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.