The NAFC regional data center has historical observations collected by Canada, the United Kingdom, France, and other countries. In addition, to retrieve ships of opportunity, profiling floats and other data collected in the region, but not necessarily shared with the regional data center, the NAFC created a routine to mirror DFO’s centralized database at MEDS and further converted them into the same format as the data processed directly at the NAFC.

These data includes historical (since 1912) and more recent casts collected by non-DFO programs (e.g., the groundfish survey realized in the Northwest Atlantic Fisheries Organization (NAFO) division 3M by the European Union (EU)). Thus, a large part of the data collected in this second step overlaps with the data present in the Climate database discussed in Sect.  . A duplicate removal scheme is described below. In addition, a large part of the data collected after the creation of the AZMP in 1999 have already been archived on CIOOS (see next Section), and therefore most of the data referred to in this section are pre-1999, although not limited to.

The casts retrieved were all converted into the same ASCII (American Standard Code for Information Interchange; a character-based encoding format) house-format commonly referred to as .pfiles. All casts collected recently using Seabird probes are automatically processed within the Seabird Data Processing software. This includes a variable-dependent low-pass filter, derivation of measurements from raw sensor data, and bin averaging. The metadata provided includes latitude, longitude, time, instrument type, instrument identification, trip identification, and data type. If the data type for a specific cast was Global Telecommunications System (GTS) (prefix TE- or BA-), it was removed from the dataset (see Sect.  for more details).

Unlike casts obtained from other sources, the temperature and salinity cast files from NAFC had not all been quality controlled. Therefore, additional steps were required before the casts could be combined with all other sources. First, casts that contained abnormalities were flagged and not considered for further analysis. These include casts with missing or non-sensical latitudes (outside [-90,90]° N range), longitudes (outside [-180,180]° W range), and time records (impossible recordings or time in the future, for example). Casts would also be ignored if there was no pressure record or if the temperature and salinity variables were empty. If no irregularities were found, temperature and salinity measurements would be bin-averaged according to pressure from surface to 5000 dbar using 1 dbar vertical resolution and converted from .pfile into yearly NetCDF files. Note that for some files, only the depth (in m) was available, so we used the latter as an approximation for the pressure. Latitude, longitude, cast identification, comments, instrument type, instrument identification, sounder depth, and time were also recorded. During the 1960s and 1970s, negative temperature measurements taken from MBTs were incorrectly recorded as either 0 values or switched to positive (-1.5 °C becomes 1.5 °C for example). These casts were manually removed from the data product after an exhaustive visual inspection of the meta data (this problem only affected data that originated from one source).

If two or more casts in the combined dataset were taken within the same 10 min period, and were within 0.01° latitude/longitude from one another, they were flagged as duplicates. Then, all flagged casts were visually inspected. The integrity of the profile, the number of measurements, the inclusion of temperature and salinity data, and the type of data were all considered during the visual inspection. If two comparative casts using different data types were flagged, such as an XBT cast and a bottle cast, both were kept. In cases where two or more reasonable casts of the same data type were flagged, the cast with the most temperature and salinity measurements would be kept. If all flagged duplicate casts had the same number of measurements and were all equally reasonable, the first cast was kept, and all others were removed.

After the removal of empty (3032 profiles removed) and duplicate casts (1225 profiles removed), outlier measurements and profiles were flagged and removed. The process of removing outliers is as follows. First, a monthly spatial “climatology” was created by binning temperature and salinity records from individual casts into 1° latitude by 1° longitude by 1 dbar level grid cells, between 35–80° N and 42–100° W. First, all measurements present in the month and the spatial area of interest, and within ±10 dbar of the depth of the sea floor were isolated. In cases where the depth window was within 10 dbar of the surface or bottom, the depth window would only extend to the surface or bottom. The depth averages of each of the measurements were found to ensure that each sample was weighted equally and did not depend on the vertical resolution of the measurements. If more than three measurements were present, isolated measurements were randomly selected 100 times with replacement (i.e., bootstrap). If three or fewer measurements were present, the climatology for this pixel would remain empty (NaN) and no outlier check would be completed. The mean of the 100 randomly selected values was then found and recorded. This process was repeated 100 times, creating a list of 100 means. The mean and standard deviation (bootstrap standard error) of 100 repetitions were then calculated. This was repeated for each depth, in each grid cell, for each month. Each NAFC Oceanography cast was compared with the mean and bootstrap standard error. If a temperature or salinity measurement was outside the mean plus or minus 40 times the bootstrap standard error, then the measurement was flagged as an outlier. If 25 % or more of the measurements in a cast were flagged as outliers, then the entire cast was classified as suspect and removed. If both the temperature and salinity casts exceeded the 25 % threshold, then the resulting empty cast was also removed (169 profiles removed). The remaining clean casts were kept.

NAFC-sourced data acquired in the NL region through multiple ship-based monitoring programs from DFO (e.g., AZMP, Multispecies survey, cod-tagging, capelin/crab surveys), academia (e.g. Memorial University of Newfoundland and Labrador, the Fisheries and Marine Institute of Memorial University) or industry (e.g. Northern Shrimp Research Foundation) since 1983 have been archived in collaboration with MEDS on CIOOS . These data are regulary updated and cover the period 1983–2024, except for Station 27 data that cover the period 1946–2024. They are readily available in NetCDF format.

These casts were collected with instruments comparable to those from DFO, such as standard Seabird CTD/Rosette or XBTs, and processed within the NAFC using a set of routines described in the . They have been thoroughly quality-controlled.

Temperature and salinity casts (2009–2022) were gathered from the Aquaculture Section of DFO Newfoundland and Labrador Region. The metadata includes latitude, longitude, and time. CTD casts were part of field operations carried out in the south coast (Coast of Bays), southeastern coast (Placentia Bay), and Notre Dame Bay (South Arm) areas of Newfoundland, mostly in May and November each year, for approximately 10–15 d . These CTD data were mainly coastal, typically within 30 nm from shoreline. Individual CTD profiles were processed and quality-controlled using Sea-Bird Electronic data analysis software and averaged to 1 m bins as described in .

The Ocean Science Hydrographic Climate Database (or simply Climate) was a comprehensive open access collection of temperature and salinity data for the NW Atlantic and the Eastern Arctic . Its domain ranges from [35–80° N] and [42–100° W]. The data forming Climate comes from a variety of sources including hydrographic Niskin/Nansen bottles with reversing thermometer, CTD casts, profiling floats, spatially and temporally averaged Batfish tows, XBTs, digital bathythermographs (BT), and MBTs. Near real-time observations of temperature and salinity from the GTS were also included for later years, but were removed as part of this study due to the lower vertical resolution of the profile. At the time of publication, the database consisted of approximately 850 000 profiles and 35 million individual observations between 1913 and 2010. The vertical resolution ranged from 1 m (for CTDs) to ∼ 10–100 m for traditional bottle casts.

Climate has been a great asset for various scientists, researchers, industry, and government clients who were interested in accessing hydrographic data in Atlantic Canada from its inception in 2004 to the mid 2010s. The problems associated with this database prior to it being taken off line were primarily two-fold: (1) there was no centralized archive responsible for gathering all regional DFO data on an annual basis and Climate rapidly became incomplete following 2010; (2) Some observation profiles were duplicated (e.g., low resolution GTS data, raw full resolution profile and quality-controlled vertically binned data of the same profile may all be present in the database) without simple ways to filter them. These problems eventually led to the decision to put Climate offline in the late 2010s.

An offline version of Climate was accessed from archived files at the Bedford Institute of Oceanography (BIO). This file contained all the data between 1913 and 2010, a year before the problems mentioned above started to become prominent. The metadata included latitude, longitude, time, cruise identification, cruise date, cruise time, station identification, flag, and data type (e.g., GTS, CTD, bathythermograph). At this stage, no further quality control was performed other than those described in . These include the assumption that casts originating from institutions outside of BIO were validated independently and the application of a suite of 25 tests including metadata and climatological checks based on real time quality control manual. Low resolution data originating from the GTS, that is, casts whose data type prefix is TE- (temperature salinity and current GTS message) and BA- (GTS low resolution for XBTs and Bathythermograph (BTs)), were removed from further consideration. A known issue concerning the rounding of time values to midnight or noon was investigated, but no additional duplicate casts were found because of this practice. The entire database was vertically averaged into 1 dbar pressure bins and converted into NetCDF file format.

The CIOOS – Environmental Research Division's Data Access Program (CIOOS-ERDDAP) temperature and salinity casts for the Maritime region were gathered from 1996 to 2024. The metadata includes latitude, longitude, time, project name, platform name, platform identification, cruise name, cruise number, station identification, and event number. Three NetCDF files were pulled from CIOOS-Atlantic, all sourced from BIO . They include the AZMP, the Atlantic Zone Off-Shelf Monitoring Program (AZOMP), and the Maritime Region Ecosystem Survey. Before 2016, the CTD files were quality controlled by visually checking metadata and profiles. On top of what was done previously; starting in 2016, quality flag fields were added to the files and have also been run through a suite of Global Temperature and Salinity Profile Programme (GTSPP) automated checks (code initially developed at Maurice Lamontagne Institute (MLI)). The file header indicates what type of quality control was performed on each file. The metadata included in the file also indicates if any of the sensor data were calibrated by a post-cruise calibration process. The retrieved files were bin-averaged according to depth from surface to 2000 m using 1 m vertical resolution and converted into individual NetCDF files.

After the shelving of Climate, a collection of profiles from databases in MS Access format were created at BIO, once per year between 1999 and 2018 for internal reporting on the Gulf of Maine, the Scotian Shelf and part of the Gulf of St. Lawrence (Roger Pettipas, personal communication, 2018). This data product is referred to as here as the BIO Ocean Monitoring and Modelling (BIO-OMM). The temperature and salinity profiles were collected directly from multiple organizations, including BIO (Climate), the Maurice-Lamontagne Institute, the US National Oceanic and Atmospheric Administration National Marine Fisheries Service, the Northeastern Regional Association of Coastal Ocean Observing Systems, and Ocean Tracking Network. Metadata including cruise number, latitude, longitude, time, maximum pressure, sounding, and station identification were also taken from each of the sources listed above.

This dataset was manually curated over the years and contains very little metadata to track the original source. For example, the station identification is not unique between databases, and two unrelated unique casts may share the same station identification. In recent years, the station identification field was constructed (where possible) as a maximum 12 digit double precision integer, with year (4 digits), cruise number (3 digits) and station number (4 digits). If a cruise number were duplicated, an additional zero was added. Various manual quality controls were performed on the combined dataset, including duplication, checks of nonsensical latitudes and longitudes, and time checks but it is difficult to determine specific methodologies. The Microsoft Access databases were converted into two .RData files and provided to us for the period 2008–2018.

Although this dataset overlaps with many other products used in this study, preliminary analyses suggested that it fills an important gap following the shelving of Climate and before the more recent data management effort described above, which led to archival with CIOOS. However, additional quality control of these data was needed.

Temperature and salinity casts from ocean gliders or moorings were ignored. CTD casts which are duplicates of GTS data could not be filtered out using the data type field because it is not provided in this data source. Instead we opted for a filtering strategy based on the number of temperature measurements per cast. Casts obtained from GTS in the Climate database between 1913 and 2010 have on average 3 or fewer temperature measurements. Based on this, all casts with 3 or fewer measurements between 2008 and 2018 were removed from the dataset. The remaining casts have an average of 155±38 measurements per cast. Therefore, we cannot be certain if all GTS-sourced casts were removed, but all casts have a higher number of measurements per cast compared to the average for GTS-sourced casts from Climate. As explained further, a duplicate removal scheme is applied to all data sources and gives lower priority to this dataset.

Station 27 (47°32.8′ N, 52°35.2′ W, 176 m depth) is located in the Avalon Channel just outside St. John's Harbour (Fig. ). It is downstream of the Coastal Labrador Current flowing southward along the Labrador Shelf.

Its local ocean parameters are considered a good proxy for the inter-annual fluctuations of the NL Shelf climate and the climate of the NW Atlantic as a whole . With regular occupations (near monthly basis) since the late 1940s, it is one of the longest hydrographic time series in Canada. On average, we cover 10.1 months yr⁻¹ since 1940. Since 1998, Station 27 is monitored as part of DFO's AZMP . Measurements taken before CTD and XBT were taken at discrete depths (0, 10, 20, 30, 50, 75, 100, 125, 150, and 175 m) .

Depth-averaged temperature and salinity anomalies, along with stratification anomaly between 0 and 50 m depth were calculated using all available Station 27 data (Fig. , first column).

Station 27 exhibits the coldest annual mean depth-averaged temperature among the stations presented in this report, with a climatological mean of 0.2±0.4 °C. Depth-averaged temperature also exhibit a strong pattern of multidecadal variability, with positive periods occurring from 1960–1970 and 2000 to present and negative periods from 1985 to 1995. The year 2021 exhibited the highest anomaly on record for the entire time period covered with a mean annual temperature of more than 1.5 °C (Fig. , first column).

The depth-averaged salinity anomalies show a relatively consistent 10-year oscillation pattern from 1945 until 1990. High-salinity periods can be seen during the mid 1960s, early 1980s, and early 1990s, while low-salinity periods can be seen during the mid 1970s and mid 1980s. More importantly, a significant freshening of the water column has occurred since the early 1990s, with a further increase since the early 2010s: six of the last eight years are fresher than average by more than one standard deviation. The freshest year on record is 2018 (2024 being the second freshest), surpassing 1970 which occurred during the Great Salinity Anomaly in the North Atlantic . This pattern is consistent with other observations from the North Atlantic , and may be a consequence of increased freshwater export from the Arctic since Station 27 is located downstream of the Davis and Hudson Straits.

The combined effect of increasing temperature and decreasing salinity could have an effect on water column density stratification. Defined as the density difference between 0 and 50 m, the stratification at Station 27 exhibits slow fluctuations on multi-decadal time scales (Fig. , bottom left). Overall, a stepwise increase appears to have occurred since the mid-1980s, although weak stratification comparable to pre-1980 conditions was also observed in the late 2010s. Differences in sampling procedure (bottle vs. CTD casts) may result in extra uncertainty in stratification calculations. Increased stratification may have widespread consequences for the ecosystem, such as changes in the timing of the spring phytoplankton bloom or prevalence of surface marine heat waves .

Halifax 2 (44°15.58′ N, 63°18.58′ W, 150 m depth) is located southeast of Halifax on the Scotian Shelf (Fig. ). It is the second sampling station along the AZMP hydrographic section Halifax Line . Measurements were sparse prior to its inclusion to the AZMP in 1998, although a relatively high number of measurements are present between 1950 and 1970. After 1998, temperature and salinity measurements were taken consistently. On average, we cover 5.9 months yr⁻¹ since 1952.

The climatological annual mean water temperature at Halifax 2 is 4.6±0.8 °C. Since 1998, the Halifax 2 depth-averaged temperature anomaly has risen by approximately 2 °C, while also exhibiting interannual fluctuations with a local minima occurring around 2008 (Fig.  in the 5-year running mean).

A similar decadal signal can be seen in depth-averaged salinity in the same phase as depth-averaged temperature. Low-salinity periods occurred during the late 1990s, mid 2000s, and after 2020, while high-salinity periods occurred in the early 2000s and 2010s.

Due to the scarcity of the data, the stratification can only be calculated for a few years prior to 1998. In the mid- to late-2000s, stratification at Halifax 2 reached low values similar to those measured in the 1960s. The highest stratification was reached in the late 1990s, late 2010s, and early 2020s.

Prince 5 (44°55.48′ N, 66°50.60′ W, 100 m depth) is located in the Bay of Fundy, north of the Gulf of Maine (Fig. ). The Bay of Fundy experiences some of the highest tidal fluctuations in the world, up to 16 m . Generally, the water column at Prince 5 is less stratified than than at Station 27 or Halifax 2 stations, mainly due to the presence of strong tidal mixing .

Prince 5 has been routinely sampled since 1920 and is Canada's oldest oceanographic station . Measurements are taken at Prince 5 once per month as part of the AZMP. On average, we cover 11.0 months yr⁻¹ since 1920.

The climatological annual mean water temperature at Prince 5 is 6.9±0.6 °C. Similar to Station 27, Prince 5 depth-averaged temperature anomalies also contain a strong multidecadal signal with colder periods from 1930–1945 and 1960–1970 and warmer periods from 1945–1955 and 2010 onward (Fig. , second column). Since the early 2010s, temperature at Prince 5 reached new record highs, with the warmest anomaly occurring in 2021 at a mean annual temperature of 9 °C.

Depth-averaged salinity anomalies exhibit a less clear multidecadal signal and instead oscillate more frequently (e.g., decadal time period), possibly influenced by the presence of the nearby St. John river. Consistent low-salinity periods can be seen during the 1960s, the late 1990s and early 2000s, and the early 2010s. No clear trends are observed. Prince-5 and Halifax 2 running average salinity in the last 20 years have shown a somewhat strong correlation (0.50).

The water at Station Prince 5 is generally less stratified than at other stations presented here (10±3 g m⁻⁴ compared to 23±3 and 38±6 g m⁻⁴ for Station 27 and Halifax 2, respectively) and stratification exhibits oscillations at multi-decadal time scales (Fig. , bottom right). When considering the running mean, the weakest stratification periods were the 1950s–1970s and from the early 1990s to the mid-2010s. In the 1930s–1940s, the late 1970s/early 1980s and during the late 2010s/early 2020s, the stratification was high.

Rimouski Station (48°40′ N, 68°35′ W, 330 m depth) is located in the Lower St. Lawrence Estuary of the Gulf of St. Lawrence (Fig. ). Regular sampling began at this location in 1991 and by 1993 it was occupied weekly from June through September . The weekly sampling was extended to include May in 1997 and sparser sampling occurred past September from 1999 onwards, and in April from 2006 onwards. It was incorporated into the AZMP in 2013. Since 1991, the three water layers (surface, CIL, deep) have experienced warming (Fig. ). The surface (0–50 m May–October average) and CIL minimum temperature (May–October average) have increased by 0.46 and 0.39 °C per decade. Bottom temperature (290 m) has increased since 2010 by 1.1 °C per decade, from 4.8 to 6.3 °C in 2024, due to changes in water masses that make up the mixture advected into the Gulf from the continental slope . The inter-annual variability of seasonal (May–October) stratification (0–50 m) is correlated with seasonally averaged runoff (1991–2024; r2=0.65; ), but shows no trend since 1991. Exact values for temperature, salinity and density reported in this paper slightly differ from similar metrics reported in annual AZMP reports because CASTS does not consider oceanographic buoy data when calculating monthly averages. Those differences do not affect the interpretation of the results.

Shediac Valley Station (47°46.8′ N, 64°01.8′ W, 84 m depth) is located on the Magdalen Shallows in the Gulf of St. Lawrence (Fig. ). It receives outflow from the St. Lawrence Estuary from the Gaspé Current. It has been sampled irregularly since 1947, nearly every year since 1957, and more regularly during the summer months since 1999 when the AZMP began. The surface (0–50 m May–October average) and bottom (84 m) temperatures have increased by 0.28 and 0.33 °C per decade, respectively, between 1999 and 2024 (Fig. ).