Unfiltered water samples for the determination of stable isotopes were collected in 10 mL HDPE (high-density polyethylene) vials free of any chemicals, sealed tightly, and stored in the dark at 4 ∘C. Measurements were conducted at the laboratory facility for stable isotopes at the Alfred Wegener Institute (AWI) Potsdam (Germany) using a Finnigan MAT Delta-S mass spectrometer equipped with equilibration units for the online determination of hydrogen and oxygen isotopic composition. Data were given as δD and δ18O values, which are the per mille difference to the Vienna Standard Mean Ocean Water (VSMOW). The deuterium excess (d-excess) was calculated as follows: d-excess = δD − 8⋅δ18. The measurement accuracy for hydrogen and oxygen isotopes was better than ±0.8 % and ±0.1 %, respectively .

Values of δ18O in combination with salinity can be used to distinguish water masses (e.g., river from ocean and meteoric from sea ice meltwater). Furthermore, within freshwater, O can reveal sources of the water that is transported by rivers. Figure 6c shows the ranges of observed δ18O values, indicating the spectrum of water masses (freshwater with low δ18O and marine water with higher δ18O) that was encountered during the four expeditions in the Mackenzie Delta region and adjacent coastal waters.

Water samples for the determination of absorption coefficients for colored dissolved organic matter (aCDOM) were filtered within 12 h of water collection using 0.2 µm GHP (hydrophilic polypropylene) filters (Acrodisc syringe filters, Pall Laboratory) prerinsed with 200 mL of Milli-Q water. Filtered samples were then pumped into the sample cell of an UltraPath liquid waveguide system that included both 200 and 10 cm cells (World Precision Instruments) using a peristaltic pump, and the absorbance was measured over the wavelengths ranging from 200 to 722 nm (see , and further modifications in , for the complete analytical procedures). For most samples, a 10 cm cell was used. For a limited number of cases, when low signal in the spectral domain was observed (mostly waters further offshore), a 200 cm cell was used instead.

Due to a complete absorption of light in the UV range, even when using the 10 cm pathlength, measurements from 200 to 280 nm were masked. To ensure the quality of the rest of the data, individual spectra were carefully checked for potential saturation, following recommendations by . A threshold of 1.2 absorbance units (AUs) was used, and only the data lower than the AU was included in this study. Measurements and sample processing were conducted following the protocol of Ocean Optics and Biogeochemistry from the International Ocean Color Coordinating Group . Uncertainty for most of the measurements was less than 5 % percent (determined using replicate scans). CDOM measurements were fitted using the following equation: 1aCDOM(λ)=aCDOM(λ0)×e-S(λ-λ0), where S is the spectral slope of aCDOM(λ), and λ denotes wavelength in nanometers between 350 and 500 nm . Coefficients of aCDOM at 443 nm ranged from 0.179 to 2.265 m-1 (Table 2) over the study region and broad seasonal extent, with the lowest median value observed during Leg 1 (Fig. 6d). While the highest values are likely associated with Mackenzie River water inputs during Leg 2, the lowest values were observed offshore beneath the ice cover in April–May (Leg 1).

In addition to CDOM measurements, the fluorescence of dissolved organic matter was measured at the National Institute of Aquatic Resources, Technical University of Denmark, Copenhagen, Denmark. The sampling and processing of samples was identical to those for CDOM measurements. Fluorescence excitation–emission matrices (EEMs) were collected using an Aqualog^® fluorescence spectrometer (HORIBA Jobin Yvon, Germany). Freshly produced Milli-Q water was used as reference. Fluorescence intensity was measured across emission wavelengths 300–600 nm (resolution 4.63 nm) at excitation wavelengths from 250 to 450 nm, with 5 nm increments and an integration time of 1 s. EEMs were corrected for inner-filter effects and for Raman and Rayleigh scattering . The underlying fluorescent components of DOM (dissolved organic matter) in the EEMs were isolated by applying PARAFAC (parallel factor analysis) modeling and validated with split-half analysis using the drEEM Toolbox . The fluorescent components derived from PARAFAC modeling were compared with PARAFAC components from other studies using the OpenFluor database .

To determine concentrations of dissolved organic carbon (DOC), 20–25 mL of water was collected into a sterile 30 mL syringe void of a rubber piston. The water was then filtered through a precleaned (acid-washed and Milli-Q-water-rinsed) filter holder containing a 25 mm Whatman GF/F (0.7 µm) and acidified with 20 µL Suprapur HCl (10 M) on the same day of sampling. DOC samples were stored and kept at 4 ∘C in the dark during transport until further analysis. A concentration of DOC was measured using high-temperature catalytic oxidation (TOC-VCPH, Shimadzu) at the AWI Potsdam, Germany. Blanks (Milli-Q water) and certified reference standards (Battle-02, Mauri-09, or Super-05 from the National Laboratory for Environmental Testing, Canada) were measured for quality control. The uncertainty in the DOC was derived from the deviation of the DOC standards that were used during the analysis. DOC standards with concentrations of 1.24, 4.64, 7.31, 25.0, and 103.4 mg L-1 were used. The uncertainty is the average percentage deviation from the measurement and the certified value. The results of the standards provided an accuracy better than ±5 %. Seasonal trends in DOC reveal a range of concentrations from a minimum of 1.7 mg L-1 to a maximum of 8.8 mg L-1 (Fig. 6e and Table 2) with the lowest median concentrations of DOC observed during summer (Leg 3).

Specific ultraviolet absorbance at 350 nm (SUVA350; m2 g C-1) was calculated by dividing the decadal absorbance (absorbance/pathlength) at 350 nm by the DOC concentration . This index is commonly used as a proxy for assessing the chemical and the biological reactivity of the DOM pool (see references in ). Higher values of SUVA350 have previously been associated with a high proportion of aromatic compounds in the DOM pools , especially in estuaries and coastal areas, where the connectivity with the surrounding terrestrial landscape is elevated. In this study, the highest values of SUVA350 (Fig. 6f) coincided with high discharge rates of the Mackenzie River (Fig. 3a). In contrast, low values of SUVA350 were observed during Leg 1 at lower discharge. Lower SUVA350 during winter and summer could indicate organic matter sources from groundwater and/or lower soil horizons. High values during high discharge may indicate fresh and young organic matter from surface plant litter .

Water samples for the determination of suspended particulate matter (SPM), in addition to concentrations of total particulate carbon and nitrogen (TPC and TPN, respectively), were filtered through a glass filtration unit on pre-weighed, precombusted blank Whatman GF/F (0.7 µm) 47 mm filters. After filtering volumes of water ranging from 150 to 1000 mL, the filters were transferred to labeled petri dishes (previously acid-washed and Milli-Q water rinsed) and placed in the oven to dry overnight at 60 ∘C before being vacuum-sealed for storage and shipment. Analysis for SPM, TPC, and TPN was conducted at Laval University (Quebec city) at the end of 2019. Samples were weighed three times each with a Mettler Toledo microscale after a final overnight drying at 60 ∘C to remove any leftover moisture. Values for SPM were obtained by subtracting the initial weight of the blank filters from the final weight of the particulate-matter-ladened filter. Due to the presence of large amounts of matter, subsections of the filters were randomly taken with a specialized punching tool. Two punched replicates were placed into tin capsules and processed in a PerkinElmer elemental analyzer (PE 2400 Series-II CHNS/O Elemental Analyzer). Acetanilide was used as a calibration standard for the CHN analysis (carbon = 71.09 %, hydrogen = 6.71 %, and nitrogen = 10.36 %). Instrument blanks (empty tin capsules) are performed during calibration to stabilize and establish a baseline for the instrument. Considering that the average coefficient of variation for every pair of random punches was only 3 % for carbon measurements and 7 % for nitrogen measurements, concentrations of TPC and TPN were obtained by extrapolating data from the punched subsection diameter to the full filter diameter, assuming uniformity of particulate matter on the filter. As shown in Fig. 6g, Leg 1 was characterized by near-zero SPM concentrations, while Legs 2, 3, and 4 exhibited higher median values around 50 µg SPM mL-1.

Water samples for the determination of particulate organic (PO) carbon (POC) and nitrogen (PON) were processed according to methods described in the IOCCG Protocol Series . Samples were filtered through a glass filtration unit on precombusted Whatman GF/F (0.7 µm) 47 mm filters, with filtration volumes ranging from 250 to 1600 mL. The resulting filters were then placed into petri dishes (previously acid-washed and Milli-Q water rinsed) and transferred to the oven at 60 ∘C to dry overnight before being vacuum-sealed for storage and shipment to Laval University (Quebec city) for analysis. In order to solely preserve the organic fraction of the particulate matter, all samples were acidified with pure HCl (37 % w/w) placed into a container at the bottom of a dessicator. Following a 72 h exposure to HCl vapors, the filters were exposed to sodium hydroxide (NaOH) during 72 h to neutralize the samples. Using a punching tool, the filters were sectioned into two subsamples that were individually wrapped in tin capsules prior to analysis. The 47 mm filters had to be subsectioned, as they did not fit fully into the analysis capsules. These capsules were processed in a PerkinElmer elemental analyzer (PE 2400 Series-II CHNS/O Elemental Analyzer; accuracy ≤0.3 % and precision ≤0.2 %) to obtain the masses of carbon and nitrogen. A cross-product calculation was used to extrapolate POC and PON concentrations from the punched subsection diameter to the full filter diameter. As shown in Fig. 6h, Leg 1 was characterized by near-zero POC concentrations, while Legs 2, 3, and 4 exhibited slightly higher median values around 1 µgPOCmL-1. Basic information related to PON values can be found in Table 2.

Coefficients for the absorption of light by particles were obtained within ca. 12 h of water collection using a spectrophotometer (Cary 100; Agilent Technologies, Inc.) equipped with a small (60 mm) integrating sphere. Samples were prepared by filtering volumes of water (from approximately 20 to 500 mL, depending on turbidity) onto 25 mm Whatman GF/F filters, with a minimum of five stations measured in duplicate or triplicate every day. To minimize biases due to backscattering by particles retained on the sample filter, the transmittance and reflectance were measured from 350 to 800 nm at 1 nm increments (the so-called T–R method; ). A recent study demonstrated that the T–R method is particularly useful for turbid waters , such as those found in the Mackenzie River and delta. The derived absorbance was then converted into absorption coefficients (ap(λ); m-1) by taking into account the volume of water filtered and the clearance area of the filter. The final ap(λ) was determined by applying a beta factor specific to our instrument setup , allowing the extrapolation of the absorption of particles concentrated on the filter to what would be in suspension. Throughout the Legs, the median uncertainty for the final ap(λ) determination was lower than 3 %. Figure 6i shows that median coefficients of ap(λ) at 443 nm were significantly higher during Legs 2, 3, and 4, following the spring freshet (after Leg 1). Trends in both ap(443) and POC concentrations were very similar, as was reflected in the significant linear relationship observed between these two variables (Fig. 7a). Both POC and total particulate carbon (TPC) showed strong linear relationships (R2=0.86; p<0.001) with particle absorption (Fig. 7a and b), suggesting that ap(443) may be a good proxy for particulate carbon. While Leg 1 values show a clear pattern of low POC, TPC, and ap(443), the overall fit of the data points agrees with the linear relationships observed in Fig. 7a and b.

In order to evaluate atmospheric correction algorithms and develop/evaluate algorithms for deriving in-water constituents from reflectance, vertical profiles of downwelling irradiance (Ed) and upwelling radiance (Lu) of the water were measured during boat sampling of Legs 2, 3, and 4, using a Compact Optical Profiling System (C-OPS; from Biospherical Instruments Inc.; see a complete description of the system in ). Measurements of C-OPS radiometric light levels could not be acquired in the waters of Shallow Bay and Mackenzie Bay (western sector) during Leg 2 due to the lack of space aboard the helicopter, in addition to safety reasons associated with free cables outside the cockpit of the hovering aircraft. The C-OPS system was composed of a set of highly sensitive radiometers, which acquire radiometric measurements in water and in air. In order to correct in-water Ed and Lu for changes in the incident light field during Lu profiling, above-surface downwelling incident irradiance (Es0+) was measured at about 2 m above sea level and above any boat structure during the time of profiling . From the Ed and Lu profiles, apparent optical properties (AOPs), such as the diffuse attenuation coefficient or the remote sensing reflectance, were computed . The C-OPS measurement accuracy has been fully characterized in , where the C-OPS underwater irradiance integrated error is found to be less than 2.5 % (less than 1.5 % for the in air irradiance), and where the C-OPS data have also been compared to other well established sensors data. The reported average unbiased percentage difference (UDP) ranges from -2.2 % to 1.8 % (depending on the radiometric parameter under consideration and on water type), which falls within calibration uncertainties. The UDP is defined as UDP=200×(Stest-Sref)/(Stest+Sref), where Sref are the reference sensor data and Stest are the tested sensor data. A more recent and complete instrument characterization (, and personal communication from Biospherical Instruments Inc.) gives very similar figures (see, for example, Sect. 8.6 and Table 16 in ). The uncertainty in the remote sensing reflectance, which is the ratio of the water-leaving radiance and the downwelling irradiance just above the sea surface, could be estimated, to a first approximation, with a quadrature combination. An average uncertainty of 2.5 % in both radiance and irradiance yields an uncertainty in the remote sensing reflectance of 3.54 %. This uncertainty, however, does not include the uncertainty related to the estimation of the water-leaving radiance from the upward-radiance profile.

The very shallow waters and the often strong currents made the usual free-fall deployment challenging and even risky for the instrument. A custom deployment method was therefore adopted, using a negatively buoyant frame that was manually lowered, with a horizontal telescoping mast and a block pulley. Deploying the sensors into the direction of the Sun in highly turbid waters (photon mean free path of less than 1 m) avoided any optical pollution from the small ship hull. The processing of the radiometric data was conducted according to recent protocols implemented in an open-source package in R (available at https://github.com/belasi01/Cops, last access: 5 April 2023; ). Due to the highly turbid waters, the so-called self-shadow correction was not estimated using the well-established method but was rather estimated using Monte Carlo simulations based on SimulO (; see Fig. D1 in Appendix D for further details).

Figure 8 shows the remote sensing reflectance determined following procedures by and . The highest values of reflectance are present between 550 and 700 nm (green to red), and the spectra show a strong influence of CDOM-absorbing light in the shorter spectral domain. This set of remote sensing reflectance measurements can be useful to test and develop algorithms for retrieving water constituents using optical remote sensing and to evaluate the performance of atmospheric correction algorithms. Note that the retrieval of Rrs from profiling downwelling irradiance (Ed) and upwelling radiance (Lu) measurements in optically complex waters is extremely challenging (see also ). The strong absorption by organic matter in low wavelengths can result in higher uncertainties in the Rrs. This might explain the unexpected higher Rrs(412) compared to Rrs(443) for a few spectra.

A so-called optical frame was also deployed during the boat sampling of Legs 2, 3, and 4. This frame was equipped with a BB3 from Sea-Bird Scientific, a FLBBCD from Sea-Bird Scientific, a HydroScat-2 from HOBI Labs, an AC-9 (with a 10 cm pathlength) from Sea-Bird Scientific, and a LISST-100X (sometimes fitted with a path reduction module, in very turbid waters) from Sequoia Scientific. These sensors allow the retrieval of profiles of the backscattering coefficients at six wavelengths, the concentration of Chl a and of CDOM, the absorption coefficients at nine wavelengths, and the beam attenuation at nine wavelengths, in addition to particle volume and size distribution between 1.25 and 250 µm. These data have not yet been fully processed and will be published at a later time.

The concentration of a suite of phytoplankton pigments, including Chlorophyll a (Chl a), a proxy for phytoplankton biomass, was determined by high-performance liquid chromatography (HPLC), following the method described in and further descriptions in . Volumes of water, ranging from ca. 20 to 500 mL, were filtered onto Whatman GF/F 25 mm filters that were immediately stored at −80 ∘C until analysis at the Ocean Ecology Laboratory of the NASA Goddard Space Flight Center in Greenbelt (USA). Samples were shipped in liquid-nitrogen-preconditioned dry shippers. Briefly, the HPLC used for pigment analysis was an Agilent RR1200 with a programmable autoinjector (900 µL syringe head), refrigerated autosampler compartment, thermostatted column compartment, quaternary pump with in-line vacuum degasser, and photodiode array detector, with deuterium and tungsten lamps, which collects the in-line visible absorbance spectra for each pigment. The HPLC was controlled by Agilent ChemStation software. Calibration was performed with individual pigment standards, whose concentrations have been spectrophotometrically determined using absorption coefficients in common with those used by most other laboratories and the commercial vendor, DHI (Hørsholm, Denmark). Thirty-six peaks were individually quantified by HPLC, from which 26 pigments were reported (some pigments contained individual components that were summed and reported as one pigment).

A full list of the accessory pigments analyzed in this study is available in Table 2. Median concentrations of Chl a over the study region were at their lowest during winter (Leg 1; Fig. 10), while the greatest interquartile range and highest median concentrations were observed throughout the summer stations (Leg 3; Fig. 10). Despite their seasonal variability, concentrations of Chl a were proportionally dominant over all other photosynthetic and non-photosynthetic pigments analyzed across the sampled stations (see an example of three selected stations in Fig. 11). At the time that the samples were analyzed, the analysis precision was 0.6 % (Chl a) and 2.3 % (all other pigments).

Samples for the determination of bacterial abundance were prepared immediately at the Aurora Research Institute (Inuvik) upon reception of the water. A volume of 1.5 mL of water was pipetted into a 2 mL Nunc^® cryotube containing 15 µL of 25 % glutaraldehyde. The cryotubes were vortexed for 5 s, allowed to sit at room temperature for 10 min, and then stored in cryoboxes at -80 ∘C before analysis by flow cytometry with SYBR™ Green I Thermo Fisher Scientific; at the Laboratoire d'Océanographie Microbienne, Banyuls-sur-Mer, France. The mean coefficient of variation for this quantification was below to 5 % (n=3). Median concentrations of bacterial cells (Fig. 10b) were lowest during Leg 1 and highest during the summer months (Legs 2 and 3).

Molecular samples for the determination of bacterial RNA–DNA were processed as follows. Volumes of 1.5 L of water were collected in clean, prerinsed Corning pyrex media storage bottles. Glass filtration systems or plastic magnetic filter funnels of 47 mm diameter were cleaned with ELIMINase^® and rinsed with Milli-Q water. A total of 10 mL of water was run through the filtration unit to condition it for each new sample; this procedure was repeated three times. Then, a 0.2 µm polyethersulfone membrane was placed on the filtration head using tweezers cleaned with ELIMINase^® and ethanol. Then, 500 mL of water was filtered through the glass system, and the filter was then immediately placed into a cryovial. A RNAlater buffer was added to the cryovial to submerge the filter completely. The filtration process was conducted in triplicate to obtain three separate filters for later analysis. Great care was taken during these procedures, including the use of lab coats, hair nets, and powder-free gloves cleaned frequently with ethanol. The cryovials were stored in -80 ∘C freezers, shipped in liquid-nitrogen-preconditioned dry shippers, and stored again in -80 ∘C freezers until their analysis at the Institut national de la recherche scientifique in Quebec city.

DNA extraction was performed using a DNeasy PowerSoil Pro Kit (Qiagen), following the instructions provided by the manufacturer, and DNA extracts were quantified on a Qubit fluorometer (the mean coefficient of variation for such quantification was about 30 %). Extracts were sent to the Integrated Microbiome Resource (Dalhousie University, Halifax, Canada), where library preparation, multiplexing, and sequencing were performed for bacteria. The sequencing targeted the V4–V5 regions of the 16S rRNA gene using the 515FB (5′-GTGYCAGCMGCCGCGGTAA-3′) and 926R (5′-CCGYCAATTYMTTTRAGTTT-3′) primer sets . Sequencing was performed on an Illumina MiSeq platform, and the sequences were then analyzed using the DADA2 pipeline v1.16 in R v4.1.2 . Taxonomy assignment was performed up to the genus level, with the SILVA reference database v138 . Details on the amplicon sequence variants (ASVs) of bacterial communities are given in Table 2.

For bacterial isolation, samples (0.5 mL) were placed in cryotubes containing 500 µL of 70 % sterile glycerol and stored at −80 ∘C before isolation at the laboratory. After thawing, 100 µL of each sample was spread in triplicate on R2A Agar plates adjusted to the salinity of the samples, and the petri dishes were incubated in the dark at 10 ∘C to allow the development of aerobic heterotrophic psychrotolerant marine bacteria. At regular intervals (1, 2, and 3 weeks), the colonies were counted and categorized based on their morphological characteristics (morphotypes). Representatives of each morphotype were selected for isolation by repeated streaking. After isolation and purification, each strain was cultivated in R2A Broth medium (Neogen Corporation) at 10 ∘C under agitation and darkness before DNA extraction using the Wizard^® Genomic DNA Purification Kit (Promega Corporation) and partial sequencing of the 16S rRNA using a Sanger 16 capillary sequencer AB3130XL (Applied Biosystems) to identify the bacterial strains . A total of 85 bacterial strains were isolated and identified.

Following DNA extraction (see above), fungal abundance was evaluated by a real-time quantitative polymerase chain reaction (qPCR) using the primer set FungiQuant-F (5′-GGRAAACTCCACCAGGTCCAG-3′) and FungiQuant-R (5′-GSWCTATCCCCAKCACGA-3′; ) and following the protocol described in . The mean coefficient of variation for such a quantification was about 25 % (n=3). Amplicon sequence variants (ASVs) of fungal communities were determined by Illumina MiSeq sequencing and metabarcoding analysis. After DNA extraction (see above), an amplification of the fungal internal transcribed spacer 2 (ITS2) region, library preparation, multiplexing, and sequencing were performed by LGC Genomics GmbH (Berlin, Germany). The fungal ITS2 region was amplified using the primer pair fITS7 (5′-GTGARTCATCGAATCTTTG-3′) and ITS4 (5′-TCCTCCGCTTATTGATATGC-3′; ). Sequence processing was performed following the R package DADA2 pipeline version 1.16 with R software (version 4.0.5; ). Taxonomic assignment was performed using UNITE (version 8.3; ) databases. Metadata related to fungal abundance and diversity can be found in Table 2. DNA sequences were deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) under accession number PRJNA822885.

Microbial respiration rates (see the metadata in Table 2) were derived from continuous dissolved oxygen (O2) measurements using a SensorDish^® Reader (SDR; PreSens, Germany) optical sensing system equipped with 24 glass vials of 5 mL containing non-invasive O2 sensors (OxoDish^®). The sensor vials were top-filled with sample water, then sealed without headspace or bubbles, and inserted into the SDR plate. The entire setup was placed in an incubator at a constant temperature of 10 ∘C. Dissolved O2 concentrations were derived from 10 min averages of quenched fluorescence measurements taken every minute over a period of 10 h. Linear regression analysis was performed on the dissolved O2 concentration data from each vial to determine microbial respiration (µmolO2L-1h-1). The mean coefficient of variation for this measurement was about 20 % (n=3).

Sediment cores were retrieved using an UWITEC gravity corer fitted with predrilled liners, with holes every 1 cm for pore water (PW) retrieval. The corer was deployed from a tripod through ice holes during Leg 1 (Fig. E1) and from a davit secured on small boats during Leg 4 . Overall, 5 and 10 stations were sampled during Leg 1 and Leg 4, respectively (Fig. 12). Sediment cores from the western region (Mackenzie and Shallow bays) were brought back for processing in the laboratories of the Aurora Research Institute (Inuvik), while processing of the cores sampled in the eastern region (Kugmallit and Kittigazuit bays) was conducted in converted workspaces at the Tuktoyaktuk Learning Center. Pore water was sampled using rhizons (Rhizosphere Research Products) with 0.2 µm polyethersulfone PES;. The PW was divided into four different subsamples in vials prepared for specific analysis. Acid-washed high-density polyethylene (HDPE) centrifuge tubes containing 200 µL of ultrapure HNO3 were used for cations. Amber glass vials (4 mL) with polytetrafluoroethylene (PTFE)-lined caps were used for the analysis of DOC, after being acid-washed in hydrochloric acid (HCl) 10 % and burned at 500 ∘C overnight. Rinsed polypropylene gas chromatography vials were used to preserve anions for later analysis. Last, nitrogen-filled amber vials amended with 100 µL of 10 % zinc acetate solution were used to preserve dissolved sulfide.

Cations were analyzed with Agilent 8800 ICP-QQQ-MS, while DOC analysis was conducted on a Shimadzu total organic carbon analyzer, TOC-VCPH . Sulfides were analyzed using a HORIBA Aqualog^®, and the analysis of anions will be conducted by Dionex Integrion HPIC high-performance ion chromatography;. To ensure instrumental accuracy, certified reference materials were used when available. The accuracy and precision of each method was evaluated with the results from these analyses. When no certified reference materials were available, then control samples were prepared by another scientist to ensure that no systematic errors were applied to samples and that the calibration curves were accurate. When control samples were used, measurements were taken 10 times, and a variation of 10 % was accepted.

The sediment cores were sliced every 1 cm. The subsamples were placed in Falcon cups and kept frozen until treatment. The samples were then freeze-dried and homogenized with an agate pestle and mortar. Ground, freeze-dried sediment was mineralized using ultrapure nitric and hydrochloric acids using a microwave (Mars5 EasyPrep Microwave vessels). Results were validated against MESS-4 certified reference material (National Research Council Canada). The mineralization protocol used is based on . Major (Fe, Ca, Na, Mg, Mn, and K) and rare Earth elements (REEs) in the sediment were analyzed using a Thermo Fisher Scientific iCAP 7400 ICP-OES and an Agilent 8800 ICP-QQQ-MS, respectively.

For cation pore water analysis, SLRS-6 from the National Research Council Canada (Ottawa, Canada) certified reference material was used to validate the method (n=3). The values were always within 12 % of certified values. The precision, expressed as the coefficient of variation in replicates analysis was under 10 % for most analytes.

For cation sediment analysis, MESS-4 from the National Research Council Canada (Ottawa, Canada) certified reference material was used to validate the method (n=9). The values were always within 8 % of certified values. The precision, expressed as the coefficient of variation in replicates analysis, was under 10 % for all analytes.

Two sites near Tuktoyaktuk were investigated between 20 July and 5 August 2019 (Tuktoyaktuk Island, 69.4557∘ N, 133.0039∘ W; Peninsula Point in the Pingo Canadian Landmark, 69.4552∘ N, 133.0069∘ W; see Fig. 13). Massive ice, groundwater, and meltwater samples were collected on permafrost slumps. Seawater samples were also collected in front of each study site at 0.5, 1, 1.5, and 2 km from the coastline.

Naturally occurring radon (222Rn; half-life of 3.8 d) and radium (223Ra, 224Ra, 226Ra, and 228Ra; half-lives of 11.4 d, 3.6 d, 1600 years, and 5.8 years, respectively) isotopes in the environment are efficient geochemical tracers to map and quantify submarine groundwater discharge in nearshore waters. Groundwater is defined as water of any salinity that has circulated through the coastal aquifer , and groundwater is here synonymous with pore water. Massive ice, groundwater, and meltwater samples were collected on permafrost slumps. Seawater samples were also collected in front of each study site at 0.5, 1, 1.5, and 2 km from the coastline. Massive ice samples were collected and immediately stored in airtight buckets to limit radon loss. They were then kept at room temperature in the laboratory until they had melted completely. Groundwater, meltwater, and seawater were collected using a peristaltic or a submersible pump. Water was continuously pumped at an approximate flow rate varying from 0.2 to 0.8 L min-1 through a Teflon tube into an inline flow cell where temperature, practical salinity, and oxygen saturation were monitored with a daily calibrated multiparametric probe (YSI 600QS). For radon, 2 L of water was sampled in plastic bottles that were tightly sealed. Within the next 12 h, analyses were carried out, as described in . Briefly, the water was bubbled to allow 222Rn degassing, and the equilibrated air flowed through Drierite desiccant to a radon-in-air detector (RAD7; DURRIDGE Company, Inc.). The air volume of the closed-loop and water volume of the bottle were known and constant over the measurement. The air recirculated through the water and continuously extracted the radon until a state of equilibrium developed. Radon-in-water activities were finally corrected by temperature and humidity and by radon decay that took place between sampling and analysis. Analytical uncertainties were less than 10 % (2σ).

Samples for radium were filtered through a 1 µm Hytrex cartridge (seawater) or a 0.45 µm Pall Corporation high-capacity capsule filter (groundwater and melted ice) to remove suspended sediment and then through an acrylic fiber coated in manganese oxide (MnO2), which quantitatively scavenges Ra . Fibers were thoroughly rinsed with Ra-free water to remove any additional particles before analysis. Sample volumes ranged from 1–20 L for groundwater and thawed ice and up to 120 L for seawater. The short-lived 224Ra and 223Ra isotopes were measured using a radium delayed coincidence counter (RaDeCC) system within 3 d of sample collection. Fibers were reanalyzed after 4 weeks and after 2 months to determine the activities of 224Ra and 223Ra supported by parent isotopes 228Th (half-life of 1.91 years) and 227Ac (half-life of 21.8 years), respectively. Activities of 224Ra and 223Ra reported in Table 2 are the activities in excess of the parent isotopes (unsupported activities). Following RaDeCC analyses, fibers were sealed in an airtight housing for 3 weeks to allow 222Rn to reach equilibrium with 226Ra. The activities of 226Ra were then determined via 222Rn emanation and scintillation counting, following the methods described in . The efficiency of RaDeCC and scintillation counting was determined using MnO2-coated fibers spiked with known activities of radium.

In addition to radon and radium isotopes in water, the mineral-bound 226Ra activity of sediments was determined on four (4) sediment samples (two coastal permafrost cliff and two surficial Holocene inshore) collected at each site. These measurements were done as a first attempt to monitor the activity of 226Ra-supported 222Rn in groundwater. A total of 15 g of sediment samples were dried, crushed, and sealed in vials fitted for a high-purity Germanium gamma-ray spectrometer (ORTEC^® GMX50) at the ISMER (l'Institut des sciences de la mer de Rimouski) laboratory (University of Quebec at Rimouski (UQAR), Rimouski, Canada). They were left in sealed vials for at least 23 d to ensure radioactive re-equilibration between the 226Ra and the short-lived daughters of the 238U series . Counting time was fixed at 4 to 5 d to provide adequate counts for the peaks of 214Pb (using the 295.2 and 352 keV) and 214Bi (609 keV) peaks. The counting error was <10 %. Disintegrations per minute, (dpm g-1), were converted to becquerel per cubic meter (Bq m-3) of wet sediment.

Unfiltered water samples for the determination of stable water isotopes were collected in 30 mL HDPE vials free of any chemicals, sealed tightly, and stored in the dark at 4 ∘C. Stable isotopes of water (δ18O, δ2H) were analyzed by elemental analysis–isotope ratio mass spectrometry (EA-IRMS) at the Geotop laboratory (Université du Québec à Montréal (UQAM), Montreal, Canada). The precision was ±0.05 ‰ and ±1 ‰ (at the 1σ level) for δ18O and δ2H, respectively. Isotopic analyses are reported as compared to the international Vienna Standard Mean Ocean Water (VSMOW). Reference materials were used throughout the isotopic water analyses to ensure high-quality data.

Duplicate dissolved organic carbon (DOC) samples were collected at the outlet of a Teflon tube using acid-cleaned 60 mL polypropylene syringes and were directly filtered through combusted 0.7 µm glass-fiber filters. The filtered samples were acidified using high-purity HCl (37 %) to pH <2 in combusted borosilicate tubes with PTFE caps after being acid-washed in hydrochloric acid (HCl) 1 %, burned at 500 ∘C overnight, and stored in the dark at 4 ∘C until analysis. DOC was analyzed a few weeks post-collection using a total organic carbon analyzer (TOC-VCPN, Shimadzu), based on the method of , and combined with a total nitrogen measuring unit (TNM-1, Shimadzu) at ISMER (UQAR, Rimouski, Canada). The analytical uncertainties were less than 2 %, and the detection limit was 0.05 mg L-1. Fresh acidified deionized water (blank) and a standard solution (1.1±0.03 mgCL-1) were frequently analyzed during measurements to ensure the stability of the instrument's performance.

Water samples for total dissolved Fe and CDOM were pumped through a Teflon tube and directly filtered using a Millipore Opticap^® XL 4 capsule with a Durapore™ membrane (0.22 µm porosity) connected to the outlet of the tube. Total dissolved Fe (Fetot) samples were stored in 15 mL Falcon™ tubes, acidified to pH <2 with an ultratrace parts per billion (ppb)-grade HCl (37 %) solution, and stored at 4 ∘C prior to total dissolved Fe analysis. Fetot was measured using the Ferrozine method modified by , with a detection limit of 0.3 µM. Measurements were done using a GENESYS 20 spectrophotometer (Thermo Fisher Scientific), using a quartz cuvette with a pathlength of 1 cm. CDOM samples were stored in acid-cleaned 15 mL glass tubes in the dark at 4 ∘C prior to analysis. Absorbance and fluorescence of CDOM were measured simultaneously. Before optical analysis, samples were allowed to equilibrate at room temperature (20 ∘C). CDOM absorbance in the UV-visible spectrum was measured using a LAMBDA 850 UV-VIS spectrophotometer (PerkinElmer) fitted with two 1 cm pathlength quartz cuvettes, with one used for the reference and one for the sample. Measurements were taken from 220 to 800 nm at 1 nm intervals, with a scanning speed of 100 nm min-1. Note that the method used to measure CDOM and DOC for these samples differs from samples described in Sect. 4.1.2 and 4.1.3. Fresh Milli-Q water was used as blanks and references during the analysis. The reference water was refreshed every 30 min. Before each analysis, the quartz cuvette was flushed first with 5 % HCl, then with deionized water, and finally with the sample. Absorbance metrics were then extracted from the different scans. Concomitantly, CDOM fluorescence was measured using a Varian Cary Eclipse fluorometer and a 1 cm pathlength quartz cuvette. Emission wavelengths (λEm) ranged from 230 to 600 nm and excitation wavelengths (λEx) from 220 to 450 nm. When the absorbance was higher than 0.3 at 254 nm, the sample was diluted to avoid saturating the fluorometer . Fresh deionized water was used as a blank, and absorbance measurements of the samples were used to correct the fluorescence data for inner-filter effects. All data were corrected for the Raman and Rayleigh effects, using daily fresh deionized water signals, and for the inner-filter effect, using absorbance spectra. Moreover, because the low concentrations of Fetot result in low total Fe:DOC molar ratios, of the order of 10-2, this suggested that the Fe effect on the absorbance and fluorescence of CDOM was negligible . Therefore, no Fe effect correction was applied to the DOM concentration, absorbance, and fluorescence. The collected data were then described as excitation–emission matrices (EEMs), and different absorbance and fluorescence metrics were calculated (Fig. 13). Briefly, EEM and fluorescent metrics were produced using the eemR package proposed by for R software. Prior to extraction of the fluorescent DOM metrics, a Raman calibration was performed for each EEM to remove the dependence of the fluorescence intensities on the measurement equipment, as proposed by . This database was completed by 50 more samples collected in the same region in summer 2021 (see , for details). Fluorescence spectroscopy can be indicative of the origin of the DOM pool. In this dataset, fluorescence spectroscopy was successfully used to determine the origin of the DOM pool. Fluorescence occurring at low excitation–emission wavelengths was associated with protein-like material originating from autochthonous production, whereas fluorescence happening at higher wavelengths was associated with humic-like material of higher molecular weight derived from terrestrial sources . In Fig. 13, different DOM fluorescence signatures can be observed across stations distributed along a north–south transect near Tuktoyaktuk. Strong signals of protein-like and humic-like fluorescence can be observed at seawater stations 17 and 18, as well as at the beach groundwater station 7, which represent stations characterized by a mixture of FDOM. In contrast, there are no obvious signals of in-situ-derived FDOM at station 4, which was derived from the thawing of massive ice.

Water samples for dissolved inorganic carbon (DIC or ∑CO2=CO2+HCO3-+CO32-; also called total carbonate) and methane were pumped and stored in 120 mL borosilicate glass bottles hermetically closed with a Teflon rubber crimped with an aluminum ferrule. A total solution of 0.2 mL of HgCl2 (7 mg L-1) was added into each bottle. The DIC concentration of samples were determined using a Apollo SciTech DIC analyzer. After reaching thermal equilibration at 25 ∘C, duplicates of 30 mL of the sample were injected into the instrument's reactor, where they were acidified with 10 % H3PO4. The CO2 formed was then carried to a LI-COR infrared analyzer by a stream of pure nitrogen. A calibration curve was constructed using Na2CO3 solutions, and the accuracy of the measurements was verified using the Dickson batch 122 standard. Reproducibility was typically of the order of 0.2 %. Methane (CH4) concentrations were determined by a Peak Performer gas chromatography with flame ionization detection (GC-FID; 2 mL sample loop; Peak Laboratories, LLC, USA), based on the static headspace method reported by . The samples were transferred into a 50 mL glass syringe, and 5 mL of CH4-free N2 was introduced to obtain a 1:10 gas : water ratio. The syringe was vigorously shaken for 4 min, and the equilibrated headspace gas was injected into the GC-FID for CH4 quantification. The methane concentrations in the headspace were calculated using the respective volumes of water and headspace in the vial and the solubility coefficient of methane of as a function of temperature and salinity. The analyzer was calibrated with a methane standard of 4.94 ppm by volume (ppmv; Air Liquide), traceable to the National Institute of Standards and Technology, and the precision of the technique was ±4 % (at 5 nmol L-1).