The Lena River basin has an extremely cold continental climate, with winter air temperatures frequently below -30 °C from December to February. Positive mean air temperatures are generally limited to a thaw period of about five months (May–September). Winter precipitation is low and mostly stored as snow, whereas most annual precipitation occurs in summer, coinciding with peak biological activity and river discharge. Permafrost is predominantly continuous, with discontinuous and sporadic zones in the southern basin (Brown et al., 2002). Along the main stem, mean annual air temperature (MAAT) decreases northward from about -9 °C in the central basin to -13.3 °C in the north, while tributaries span a broader MAAT range from -4.7 to -15.9 °C. Annual precipitation is low, particularly in the northern basin (about 200–250 mm yr⁻¹), but its seasonal maximum occurs in summer. Winter precipitation is smaller and is stored mainly as snow until spring melt (Chevychelov and Bosikov, 2010).

The lithology of the Lena basin upstream of Yakutsk is highly diverse and includes Archean and Proterozoic crystalline and metamorphic rocks, Upper Proterozoic to Ordovician carbonate formations, Permo–Triassic volcanic rocks, and widespread Phanerozoic terrigenous silicate sedimentary units. This geological heterogeneity exerts a major control on riverine major ions, trace elements, and inorganic carbon. More detailed descriptions of basin landscapes, vegetation, and geology are available elsewhere (Rachold et al., 1996; Huh et al., 1998a, b; Huh and Edmond, 1999; Pipko et al., 2010; Semiletov et al., 2011; Kutscher et al., 2017; Ogneva et al., 2023; Juhls et al., 2020, 2025). Together, these climatic, permafrost, and lithological gradients provide the framework for interpreting spatial variability in solute export, greenhouse gas dynamics, and weathering regimes along the Lena River continuum and its tributaries.

In the middle and lower Lena (Yakutsk to Kyusyur), peak annual discharge typically occurs in June and is mainly driven by rapid snowmelt across the permafrost-dominated basin. Sampling was conducted from 5 to 22 July 2022, shortly after peak flow, during the early post-freshet recession. At this stage, river levels remain high but hydrological conditions are more stable than during peak flood, allowing integration of basin-wide snowmelt inputs while reducing short-term variability linked to flood-wave propagation and bank erosion. July conditions also correspond to active-layer deepening and enhanced subsurface connectivity, which promote mobilization of solutes, dissolved inorganic carbon (DIC), dissolved organic carbon (DOC), and greenhouse gases from soils and permafrost-affected terrains.

During the campaign, we navigated ∼ 1500 km aboard the research vessel Merzlotoved, from Yakutsk (62.15° N) to Samoilov Island (72.38° N) in the Lena Delta (Fig. 1A). A total of 33 sites were sampled along the main stem, together with two major tributaries (Aldan and Vilyui) and four smaller tributaries (Seen-Yurekh, Undyulyung, Bysyttakh, and Molodo). Regular stops at ∼ 50 km intervals along the main channel enabled systematic sampling of major hydrochemical parameters and greenhouse gases (GHG) along the longitudinal river gradient. This transect design aimed to resolve changes in solute composition, dissolved carbon parameters, and CO₂ and CH₄ concentrations in relation to northward permafrost gradients, lithological transitions, and contrasting tributary inputs. At selected confluences, sampling was performed 200–500 m upstream of the mixing zone to obtain representative tributary water and avoid backwater and incomplete mixing effects. Longitudinal transect sampling is widely used to assess river water chemistry in large and climatically extreme basins (Huh and Edmond, 1999; Spence and Telmer, 2005). Sampling during the high-flow season also provides the closest agreement with annual flux estimates derived from time series, because this period accounts for a large share of annual water and solute export (Qin et al., 2006).

July corresponds to the high-flow period of the Lena River, representing the recession phase of the spring freshet (Fig. 1B). Although peak discharge typically occurs in June, river levels remain elevated in July due to delayed snowmelt from northern and high-elevation sub-basins and continued active-layer development in permafrost-dominated terrains. Hydrographs for the studied large rivers were constructed using discharge data from the Automated Information System of the State Monitoring of Water Bodies (https://gmvo.skniivh.ru/, last access: 17 September 2026). These time series were used to position the sampling campaign within the seasonal hydrograph and to assess deviations from long-term mean discharge during the sampling period. To characterize hydrological and geomorphological controls on river chemistry, watershed-scale parameters were calculated for each sampling point (Fig. 1A), including total watershed area, river network density, and total watercourse length. These parameters were derived from HydroSHEDS shapefiles (https://www.hydrosheds.org/, last access: 17 September 2026) using standard GIS procedures.

Water surface area was quantified from Landsat 8–9 Level-2 surface reflectance products (Collection 2) acquired during 5–22 July 2022. Cloud-free scenes were atmospherically corrected, orthorectified, mosaicked, and processed in ArcGIS. Water bodies were delineated using the Normalized Difference Water Index (NDWI) and Modified NDWI (MNDWI), based on green, near-infrared, and shortwave infrared bands, with thresholds calibrated for Lena River conditions, including turbid waters. Classified rasters were converted to polygon shapefiles and manually corrected along shorelines, shallow areas, and zones affected by suspended sediments or aquatic vegetation. In total, 269 water polygons were identified. To improve area accuracy, polygons were subdivided using the Fishnet tool, and surface areas were calculated individually using planar geometry within appropriate WGS 84/UTM projection zones. Delineation accuracy was visually checked against high-resolution imagery (e.g., Google Earth) and river morphology.

Water discharge at each sampling point and date was estimated from gauging-station data by linear interpolation and hydraulic scaling along the main stem and tributaries, accounting for incremental drainage area and tributary inflows. Surface water discharge, watershed area, river network length, and latitude (used as an integrative proxy of climatic and permafrost gradients) were then compared with surface-water chemistry to assess controls on dissolved major elements, trace elements, and carbon species.

Dissolved CO₂ in surface waters was measured in situ using a portable infrared gas analyzer (IRGA; GMT222 CARBOCAP^® probe, Vaisala^®; accuracy ±1.5 %) operating in two ranges (0–2000 and 0–10 000 ppm). The underway probe was protected by a gas-permeable waterproof PTFE membrane/sleeve. The sensor was installed in the vessel's Kingston water intake system, where continuous flow (∼ 20 L min⁻¹) ensured constant renewal of river water during navigation. A copper mesh was added to reduce potential biofouling (Yoon et al., 2016), although such effects were expected to be minimal in the cold, low-impacted Lena waters.

Independent field checks of water pCO₂ were performed at 24 stations using a second, separately calibrated Vaisala GM70 meter equipped with a GMP222 probe enclosed in an independent gas-permeable PTFE membrane assembly, which allows rapid CO₂ diffusion while preventing water intrusion (Johnson et al., 2009). The control probe was immersed directly at ∼ 20 cm and allowed to stabilize before recording. Temperature and pressure corrections were applied using the same procedure as for the underway sensor. These stationary measurements were used as external checks of the underway record rather than as exact replicate observations, because they were not strictly synchronous or spatially co-located with the Kingston-intake measurements.

At each station, the sensor was equilibrated for ∼ 10 min before measurement. CO₂ concentration (ppm), water temperature (°C), and atmospheric pressure (mbar) were logged every minute using a Campbell data logger during 5 min measurement sequences repeated every 30 min, yielding 153 individual records. For the main stem, values were averaged over three consecutive 5 min sequences. At tributary confluences, measurements were performed 200–500 m upstream of the mixing zone. Sensors were calibrated before and after the expedition against certified gas standards (0, 800, 3000, and 8000 ppm), yielding linear relationships with R2 > 0.99. Instrumental drift ranged from 0.03 % d⁻¹ to 0.06 % d⁻¹, and post-processing corrections for temperature and pressure followed Johnson et al. (2009). Overall uncertainty of individual pCO₂ measurements was ≤ 10 %.

Methane concentrations were determined from unfiltered water collected in 60 mL serum bottles without headspace, preserved with 0.2 mL saturated HgCl₂, and analyzed in the laboratory by headspace equilibration with ultra-high-purity N₂. Two 0.5 mL aliquots of the equilibrated headspace were measured on a Bruker GC-456 gas chromatograph equipped with flame ionization and thermal conductivity detectors. Calibration was performed every 10 samples using certified standards (Air Liquide; 145 ppmv), and duplicate injections showed reproducibility within ±5 %. Dissolved CH₄ concentrations were calculated from headspace measurements using temperature-dependent solubility coefficients (Yamamoto et al., 1976) and mass-balance equations. Combined uncertainty of CH₄ concentrations was estimated at 7 %–12 %. These analytical uncertainties are small relative to the observed spatial gradients in CO₂ and CH₄ and do not affect the interpretation of longitudinal and tributary-scale patterns.

Diffusive CO₂ fluxes (FCO2) were measured using two floating chambers (ca. 30 cm diameter, ∼ 10 L volume) equipped with non-dispersive infrared CO₂ sensors (SenseAir). Sensors were calibrated in the laboratory against certified N₂–CO₂ gas mixtures prior to field deployment. During measurements, chambers floated freely at the water surface, and headspace CO₂ was recorded over 5-min intervals. Fluxes were calculated from the initial linear accumulation phase using the first 30 min of deployment (Serikova et al., 2018; Vorobyev et al., 2024).

In addition to direct chamber measurements, diffusive CO₂ fluxes were independently estimated using the bulk gas exchange equation of Cai and Wang (1998): 1FCO2=KhkCO2Cwater-Cair, where Kh is the Henry's constant corrected for temperature and pressure (mol L⁻¹ atm⁻¹), kCO2is the gas exchange velocity at a given temperature, Cwater is the water CO₂ concentration, and Cair is the CO₂ concentration in the ambient air. Conversion between dissolved concentration and partial pressure followed Wanninkhof (1992) and Lauerwald et al. (2015). For the July 2022 flux calculations, atmospheric CO₂ was represented by the 2022 global annual mean of 417.9 ppm from the WMO monitoring network, consistent with the nearest Tiksi station (422.5 ppm; https://community.wmo.int/wmo-greenhouse-gas-bulletins, last access: 17 September 2026; https://www.meteorf.gov.ru/product/infomaterials/90/, last access: 17 September 2026). Temperature-dependent solubility constants followed Wanninkhof (1992).

A median gas transfer velocity of 4.46 m d⁻¹ was adopted from measurements in four major Western Siberian rivers (Karlsson et al., 2021), consistent with previous estimates for the Lena River (Vorobyev et al., 2021a), the Kolyma (Denfeld et al., 2013), the Yukon (Striegl et al., 2012), and global syntheses for large rivers (Alin et al., 2011; Raymond et al., 2013). Wind speeds during sampling were below 3.7 m s⁻¹ and water surfaces were visually smooth, supporting the use of this literature-constrained KT value (Jähne et al., 1987; Alin et al., 2011; Guérin et al., 2007).

Instantaneous diffusive CH₄ fluxes were calculated using an equation analogous to Eq. (1), with gas transfer velocities derived from chamber-measured CO₂ fluxes and in situ CO₂ concentrations, following approaches used for Western Siberian rivers (Serikova et al., 2018; Lim et al., 2022; Vorobyev et al., 2024). Dissolved CH₄ concentrations were combined with atmospheric equilibrium concentrations assuming a background atmospheric pCH₄ of 1.8 ppm and the 2022 mean atmospheric CH₄ concentration from Mauna Loa Observatory (https://gml.noaa.gov/webdata/ccgg/trends/ch4/ch4_annmean_gl.txt, last access: 17 September 2026), following Serikova et al. (2018, 2019). Flux uncertainty was dominated by variability in gas transfer velocity rather than by analytical uncertainty in dissolved gas concentrations. Concentration uncertainty contributed comparatively little (≤ 10 % for CO₂ and 7 %–12 % for CH₄), whereas kCO2 may vary by ±40 %–60 % depending on hydraulic conditions (Raymond et al., 2013). Sensitivity analysis showed that using a conservative kCO2 value of 3 m d⁻¹ instead of 4.46 m d⁻¹ would decrease calculated CO₂ emissions by about 30 %, while preserving the spatial patterns along the transect. Because the objective of this study is to resolve relative gradients between tributaries and main stem sections rather than derive absolute annual budgets, the adopted kCO2 parameterization provides a robust and internally consistent basis for comparative flux analysis.

In situ dissolved oxygen, specific conductivity, and water temperature were measured at ∼ 20 cm depth using a WTW 3320 multimeter equipped with a CellOx 325 oxygen probe (±5 %), a TetraCon 325 conductivity sensor (±1.5 %), and a temperature sensor (±0.2 °C). pH was measured with a portable Hanna instrument fitted with a combined Schott glass electrode, calibrated daily with NIST-traceable buffers (pH 4.01, 6.86, and 9.18 at 25 °C), yielding an uncertainty of ±0.01 pH units. Buffer temperatures were kept within ±5 °C of river water temperature.

Surface water was collected from 20–30 cm depth near the channel center into pre-cleaned polypropylene bottles and immediately filtered through sterile single-use Sartorius filters (0.45 µm). The first 50 mL of filtrate were discarded. Milli-Q blanks confirmed negligible DOC contamination from the filters. Filtered samples were split into aliquots: one was acidified to 2 % (v/v) with bidistilled HNO₃ in pre-cleaned polypropylene vials for major and trace element analyses; the second, non-acidified fraction was used for DOC, DIC, and UV absorbance. Samples were stored at 4 °C and analyzed within 30 d.

DOC and DIC were measured with a Shimadzu TOC-VSCN analyzer (detection limit 0.1 mg L⁻¹; precision ±3 %). Nutrients were analyzed on frozen filtered samples by ion chromatography (Dionex ICS-5000+) with ±2 % uncertainty; certified reference materials Ion-915 and Ion-964 were within 10 % of certified values. Total bacterial cell concentration (TBC) was measured by flow cytometry after fixation with glutaraldehyde, following Marie et al. (1999). Major and trace elements were determined in acidified samples by ICP-MS (Agilent 7500ce, Ar and He modes). Precision ranged from 5 %–10 % at 1–1000 µg L⁻¹ to 10 %–20 % at 0.001–0.1 µg L⁻¹. Accuracy and reproducibility were checked using SLRS-6 (River Water Certified Reference Material for Trace Metals and other Constituents, Canada) and three in-house multi-element standards every 20 samples (Yeghicheyan et al., 2019), with agreement within 10 %–15 % for 40 elements.

Twelve selected samples were additionally processed by dialysis (1 kDa cutoff) to distinguish low-molecular-weight (< 1 kDa) and colloidal (1 kDa–0.45 µm) fractions. River water was pre-screened through a 20 µm nylon mesh and placed in cleaned 5 L polyethylene containers; pre-cleaned 50 mL dialysis bags were immersed for 3–5 d following established procedures (Vasyukova et al., 2010; Pokrovsky et al., 2016b; Kolesnichenko et al., 2021). Containers were kept dark at near in situ temperature and gently agitated. Dissolved solute concentrations before and after dialysis differed by < 10 % (p > 0.05), indicating negligible alteration of dissolved and colloidal composition during incubation. Nevertheless, dialysis provides operational rather than structural size separation. During the 3–5 d equilibration, some redistribution among colloidal subfractions, aggregation/disaggregation, adsorption/desorption on container or membrane surfaces, and limited microbial alteration cannot be fully excluded. Therefore, the 1 kDa–0.45 µm fraction is interpreted here as an operational colloidal pool, and not as an indicator of the exact in situ colloidal size spectrum.

Pearson correlations were calculated to assess linear relationships between variables, and statistical significance was evaluated at p < 0.05. To further identify the dominant drivers of dissolved carbon variability (CO₂, DIC, and DOC), multivariate statistical analysis was performed using Principal Component Analysis (PCA). This approach allowed us to assess the combined influence of hydrochemical parameters (major ions, nutrients, trace elements), hydrological variables (discharge, watershed area), and climatic factors (e.g., latitude as an integrative proxy) on spatial patterns of dissolved carbon in river waters. The analysis was performed using STATISTICA 7 (StatSoft Inc., https://www.statsoft.pl/en/software/statistica/, last access: 17 September 2026). Principal components were extracted from the raw data matrix using the covariance structure of standardized variables.

PCA was performed on the correlation matrix of standardized variables. Given the limited number of main stem samples (n=26), the large number of hydrochemical variables, and the modest cumulative variance represented by the first two axes, PCA was used only as an exploratory ordination. The number of components was evaluated using the scree plot. The first two Varimax-rotated components, explaining 23 % and 13 % of total variance, respectively (36 % cumulative), were retained to visualize the clearest covariance structure. Interpretations were accepted only when consistent with pairwise correlations and independent size-fractionation evidence; no causal or discrete cluster assignment was inferred from PCA alone. PCA was therefore used as an exploratory tool to identify major axes of covariation, rather than as a complete explanation of hydrochemical variability.

Hydrograph analysis (Fig. 1B) indicates that river discharge during the study period was primarily sustained by groundwater contributions and rainfall inputs following the spring snowmelt peak. By July, the Lena River was in the recession phase of the freshet, characterized by gradually declining but still elevated discharge likely sustained by delayed runoff, rainfall, and subsurface contributions. The estimated water travel time between the upstream gauging station at Tabaga and the downstream station at Kyusur was approximately 12 d under the observed hydrological conditions. This relatively short transit time implies efficient downstream propagation of water masses and associated solutes along the main stem during high-flow conditions.

The drainage network of the Lena basin is well developed, with a mean river network density of 0.22 km km⁻². According to Strahler's classification, stream order within the basin ranges from first-order headwater streams to ninth-order channels along the main stem, reflecting the hierarchical organization and large spatial integration of tributary inputs. Hydrologically, 2022 corresponded to a relatively high-flow year when compared to long-term records prior to the 1970s. Mean annual discharge reached 10 501 m³ s⁻¹ at Tabaga and 19 745 m³ s⁻¹ at Kyusur, placing 2022 among the upper range of historical values. However, in the context of the past two decades, these values fall within the range of average annual discharge reported by the Russian Federal Service for Hydrometeorology and Environmental Monitoring (Roshydromet), consistent with recent assessments (Juhls et al., 2025). Thus, while elevated relative to early historical records, hydrological conditions during the expedition were representative of contemporary flow regimes in the Lena River. The combination of high discharge, rapid downstream water transit (∼ 12 d), and strong tributary integration created conditions favorable for efficient longitudinal transport of dissolved carbon species and greenhouse gases, setting the framework for interpreting spatial gradients in CO₂, CH₄, DIC, DOC, and associated emission fluxes.

Under the elevated but hydrologically representative discharge conditions of July 2022, the Lena River system provides an opportunity to evaluate how basin-wide connectivity, tributary integration, and northward climatic gradients regulate dissolved carbon pools and greenhouse gas dynamics. Below, we describe the spatial distribution of pCO₂, CH₄, DIC, and DOC along the main stem and tributaries, and quantify associated diffusive carbon fluxes.

The principal hydrochemical carbon parameters of the Lena River main stem (pCO₂, CH₄, pH, O₂, E.C., DIC, and DOC) are summarized in Table 1. Along the main stem, DIC showed a weak, statistically non-significant northward increase, whereas DOC showed no systematic longitudinal trend (Fig. 2A, B). Mean concentrations were 8.4 ± 2.7 mg L⁻¹ for DIC and 7.5 ± 2.7 mg L⁻¹ for DOC. Discrete pCO₂ measurements indicated a weak northward decrease, with a mean value of 1020 ± 420 µatm (median 892 µatm; range 600–2600 µatm; Fig. 2C). Chamber-derived CO₂ fluxes ranged from 0.11 to 0.82 g C–CO₂ m⁻² d⁻¹ along the main stem (Fig. 2D), whereas calculated fluxes across the main stem and tributaries ranged from near-zero values to 1.3 g C m⁻² d⁻¹. Methane concentrations were low (mean 0.22 µmol L⁻¹; median 0.08 µmol L⁻¹) and showed no significant longitudinal trend (Fig. 2E). Diffusive CH₄ emissions contributed < 0.5 % of total carbon emissions (Fig. 2F).

When continuous pCO₂ measurements along the main stem (153 individual data points, with mean distance of 7.25 km) are examined along the south–north transect together with discrete observations from tributaries (Fig. 3A), major tributary confluences emerge as clear discontinuities in the longitudinal pattern. These zones are characterized by abrupt increases or decreases in CO₂ concentration, reflecting the input of waters with contrasting biogeochemical signatures. The independent control measurements reproduced the broad magnitude and spatial pattern of the underway pCO₂ record but showed larger pointwise differences at several stations. These divergences occurred mainly across steep pCO₂ gradients and close to tributary confluences. Underway values represent moving averages over ∼ 1.25 km river segments, whereas the controls are stationary snapshots; differences can therefore reflect short-distance or cross-channel heterogeneity, incomplete tributary mixing, small depth offsets, and the different equilibration response of the two membrane assemblies. Because the controls showed no consistent directional offset, the mismatches were not interpreted as evidence of systematic sensor bias.

A similar pattern is observed for calculated continuous CO₂ fluxes (Fig. 3B). The Vilyui River exhibited the highest emissions (up to 1.3 g C–CO₂ m⁻² d⁻¹), highlighting its role as a major localized source of CO₂ to the main stem. In contrast, the Aldan chamber measurement was close to zero, whereas pCO₂ indicated supersaturation and a small positive calculated flux during the sampling period, indicating substantial spatial variability in tributary influence. Despite this pronounced local heterogeneity, the main stem displayed relatively stable dissolved carbon characteristics. DOC concentrations remained spatially uniform, and DIC showed only a weak longitudinal trend, whereas pCO₂ and FCO2 exhibited dynamic variability controlled by tributary inputs and in-stream processes. As a result, the Lena River evaded CO₂ to the atmosphere during July 2022. In contrast, CH₄ concentrations were low and varied non-systematically, contributing only marginally to total carbon emissions. This pattern further supports the decoupling between bulk dissolved carbon pools and CO₂ evasion, with tributaries acting as primary factors of short-scale variability along the river continuum, as was also evidenced in earlier study of upper and middle Lena reaches (Vorobyev et al., 2021a).

Previous estimates of areal CO₂ emissions from lotic waters of the Lena River basin were based on total river surface coverage for the entire watershed, estimated at 28 197 km² in 2016, including 5022 km² of seasonal water (Global SDG database; Vorobyev et al., 2021a). This value is broadly consistent with the Global River Widths from Landsat (GRWL, Allen and Pavelsky, 2018) Mask estimate of 22 479 km². In contrast, the present calculation was restricted to the Lower Lena sector between Yakutsk and Kyusyur, covering 1 551 445 km² of land area and 12 904 km² of water surface. This included the main stem (7529 km²) and the two largest tributaries, the Aldan and Vilyui Rivers, with water surface areas of 2377 and 2998 km², respectively.

The emission estimate corresponds to the hydrological conditions of the expedition, i.e. the transition from the end of the spring flood to the onset of summer baseflow. This sector was previously identified as a major source of uncertainty for Lena River emission estimates because of the lack of direct pCO₂ and flux measurements downstream of the Aldan confluence and across the Vilyui-influenced reach. During July 2022, the Aldan River showed near-zero chamber-measured CO₂ exchange (FCO2=-0.03 µmol m⁻² s⁻¹; pCO₂ = 1014 µatm), whereas the Vilyui River exhibited the highest chamber-derived emission measured during the campaign (1.3 g C m⁻² d⁻¹; pCO2=1340 µatm). Fluxes calculated from pCO₂ for these two tributaries were more similar, amounting to 0.31 and 0.485 g C m⁻² d⁻¹ for the Aldan and Vilyui, respectively.

For the main stem, the chamber-derived mean CO₂ flux was 0.33 ± 0.20 g C m⁻² d⁻¹ (n=12) along the south–north transect. Fluxes calculated from continuous and discrete pCO₂ measurements (mean 957 ± 230 µatm, n=153) using (KT=4.46) m d⁻¹ yielded nearly identical values of 0.32 ± 0.20 g C m⁻² d⁻¹ (Figs. 2C, D, 3). The experimentally derived piston velocity calculated from paired chamber fluxes and pCO₂ measurements averaged (KT=8.82 ± 6.76) m d⁻¹, indicating substantial spatial variability in gas exchange conditions. However, the close agreement between chamber-derived and pCO₂-derived mean fluxes supports the use of (KT=4.46) m d⁻¹ for reach-scale comparative upscaling. Emission estimates were then weighted according to the relative water surface areas of the main stem (58 % of the total Lower Lena water area), Vilyui (23 %), and Aldan (18 %). The resulting water-area-weighted CO₂ emission estimate for the Lower Lena during the transition from the end of spring flood to early baseflow was 0.36 ± 0.15 g C m⁻² d⁻¹.

All measured hydrochemical parameters for the Lower Lena River and its tributaries are reported in Table S1 of the Supplement. Based on longitudinal trends along the main stem, three principal groups of solutes were identified: (i) Low-solubility trivalent and tetravalent elements and strong hydrolysates, exhibiting a significant northward decrease in concentration (Fig. 4). This group includes Al, Sc, Ga, Y, rare earth elements (REEs), Ti, Th, U, Be, and Nb. These elements are typically associated with particulate or colloidal phases and exhibit systematic downstream decline along the river continuum. (ii) Labile major cations and anions (including host carbonate mineral-related solutes), showing increasing concentrations from south to north (Fig. 5). This group comprises DIC, SO42-, B, Mg, P, Ca, Sr, As, Se, and Sb, consistent with progressive downstream integration of groundwater inputs and deep bedrock/weathering inputs. Third group includes components without statistically significant longitudinal trends, defined by correlation coefficients |r| < 0.3 (p > 0.05) with latitude (Table S2). This category includes DOC, greenhouse gases (CO₂ and CH₄), major anions (Cl⁻, F⁻, NO3-), Si, K, V, Cr, Mn, Fe, Ni, Cu, Zn, Co, Cd, Rb, Mo, Cs, Ba, Hf, W, Tl, Pb, and Bi. These solutes exhibited spatial variability but no systematic south–north gradient along the main channel.

In addition to longitudinal patterns, distinct local anomalies were observed in certain tributaries. For example, Mn concentrations in the Undyulyung River (LIV 17) exceeded adjacent main stem values by a factor of 27. Similarly, Fe and Mn concentrations in the Bysyttakh and Molodo tributaries (LIV 28 and LIV 29) were elevated relative to the main channel by factors of approximately 2 and 5, respectively. However, such deviations were not widespread. For most elements, concentrations in tributaries were comparable to those measured in neighboring main stem sections, typically within ±30 % of adjacent values. This overall similarity indicates efficient mixing and strong integration of tributary inputs within the large-channel hydrodynamic framework of the Lower Lena.

Multivariate statistical treatment was restricted to the Lena main stem (n=26), as the number of sampled tributaries (n=6) was insufficient for robust statistical analysis. Pairwise Pearson correlation analysis revealed several significant relationships among dissolved constituents (p < 0.05; Table S2). pCO₂ correlated positively with F⁻, Cl⁻, and Na⁺. Diffusive CO₂ flux correlated positively with DOC (r=0.87, p < 0.01; Fig. 6A), nutrients (P, K), and selected trace metals (V, Ni). In contrast, CH₄ concentrations did not show significant correlations with the measured hydrochemical parameters. DIC correlated positively with Br⁻, Si, P, Rb, and As, as illustrated by the DIC–As relationship in Fig. 6B. Aluminum showed strong positive correlations with Ti, Ga, Y, and REEs (Fig. 6C, D), and weaker positive correlations with Th and U.

Principal Component Analysis (PCA) was applied to the standardized main-stem dataset. The first two Varimax-rotated components explained 23 % and 13 % of total variance, respectively (36 % cumulative; Fig. 7; Table S3). Along rotated component 1, positive loadings of Be, Al, Ti, Ga, Y, Nb, REEs, and Th contrasted with negative loadings of B, SO42-, DIC, Mg, and Sr. Rotated component 2 grouped DOC with P, K, V, Mo, Co, Ni, As, Sb, and Cs, whereas latitude showed only moderate loading. Thus, PCA summarized the main covariance structure of the dataset but did not account for most of the total variance. These correlation and PCA patterns provide the statistical basis for the operational size-fractionation analysis as presented in the next section.

Dialysis (1 kDa cutoff) enabled operational separation of nominal low-molecular-weight (LMW_< 1 kDa) and dissolved (< 0.45 µm) colloids with sizes > 1 kDa. The colloidal proportion of each element i was calculated as the percentage of its concentration in the 1 kDa–0.45 µm fraction relative to its concentration in the dissolved (< 0.45 µm) form following: [i]colloidal=100%×([i]0.45µm-[i]<1kDa)/[i]0.45µm where [i]_{< 0.45 µm} and [i]_< 1 kDa are concentrations of element in the 0.45 µm filtrate and 1 kDa dialysate, respectively.

Across the northern Lena basin, the relative contribution of colloidal forms was remarkably consistent between the main stem and tributaries (Fig. 8). The colloidal fraction ranged from negligible values (0 %–10 %) for highly soluble species such as DIC, Li, and Si, to dominant proportions (80 %–90 %) for trivalent and tetravalent hydrolysates. Based on colloidal abundance, three principal groups of solutes were distinguished: (1) Predominantly truly dissolved species (0 %–20 % colloidal fraction): Alkali and alkaline-earth metals (Li, Na, K, Rb, Mg, Ca, Sr), major inorganic carbon (DIC), Si, and selected trace oxyanions (Mo, As). These elements were largely present in low-molecular-weight forms and exhibited minimal association with colloidal phases. (2) Mixed LMW–colloidal species (20 %–60 % colloidal fraction): P, Al, and several trace metals (Cu, V, Sc, Ni, Cr, Co, U) showed intermediate colloidal contributions. Dissolved organic carbon (DOC) exhibited a mean colloidal proportion of 41 ± 14 %, indicating that a substantial fraction of riverine organic matter was associated with supramolecular or organo-mineral colloidal structures. Here, colloidal DOC refers operationally to the organic-carbon fraction retained between 1 kDa and 0.45 µm. This fraction should not be interpreted as a single physical class of particles, but as a continuum of humic supramolecular associations, high-molecular-weight organic macromolecules, microbial and plant-derived polymers, and organic coatings or complexes associated with Fe- and Al-rich mineral phases. The last, 3rd group of solutes, included predominantly colloidal elements (80 %–90 % colloidal fraction) such as certain divalent metals (Mn, Ba, Zn, Pb) and refractory lithogenic hydrolysates (Fe, Ga, Y, REEs, Ti, Zr, Hf, Th, Be, Nb). These were largely transported in colloidal form, consistent with strong hydrolysis tendencies and association with Fe-rich mineral or organo-mineral particles. No major difference in colloidal partitioning was observed between the main stem and tributaries. In particular, DOC exhibited a remarkably uniform colloidal proportion (41 ± 14 %) across the main stem and four tributaries, including the largest ones (Vilyui and Aldan). The only notable deviation occurred in a floodplain lake hydrologically disconnected from the main channel, where DOC, Al, and Co displayed 20 %–30 % higher colloidal fractions relative to fluvial waters. Overall, the Lena River system exhibited a highly structured and spatially homogeneous colloidal pattern, with refractory lithogenic elements transported predominantly in colloidal form, mobile weathering-derived solutes remaining largely truly dissolved, and DOC having an intermediate and remarkably stable colloidal proportion across the basin.