Urbanization and seasonality strengthens the CO₂ capacity of the Red River Delta, Vietnam

River waters were sampled and analyzed monthly at 10 stations along the Red River and 11 along the Day River (figure 1; see supplementary site description). Physical and chemical parameters including temperature, pH, TA, turbidity, dissolved oxygen (DO), conductivity, total dissolved nitrogen (TDN), total dissolved phosphorus (TDP), nitrate (NO₃–N), nitrites (NO₂–N) and ammonium (NH₄–N) were collected and analyzed monthly at each station across the dry (November–April) and rainy (May–October) seasons of 2018–2019. Daily river discharge (m³ s⁻¹) was obtained from eight hydrological stations in the Red River corresponding to sampling points H1, H3, H4, H5, H7, H9, H10 and five stations in the Day River corresponding to sampling sites D3, D6, H6, D8, D10, provided by the Vietnam Center of Hydro-Meteorological Data [28].

Temperature, pH, DO, conductivity, and turbidity were measured in situ with a Hydrolab Sonde DS5 (USA). The pH electrode was calibrated between sampling campaigns using standard solutions (pH: 4.01 and 6.88, Merck) with a precision and accuracy of ±0.01. Surface water samples were collected and stored at 4 °C before analysis within 48 h. TDN, TDP, NO₃–N, NO₂–N, NH₄–N, and TA were determined according to APHA standard protocols [29] with filtration (Whatman Glass Fiber Filter(GF/F): 0.7 µm pore size) for soluble parameters. TA was measured on filtered samples and determined by the single-point titration using methyl orange and phenolphthalein indicators, respectively.

As a proxy of phytoplankton, chlorophyll and carotenoid pigments were extracted from filter papers collected monthly at each river sampling stations and stored at −4 °C for 12 h in acetone: methanol: water (80:15:5), followed by filtration through a polytetrafluoroethylene (PTFE) 0.22 μm filter, and evaporation under nitrogen gas. Dried extracts were then re-dissolved in a solution of (70:25:5) acetone, ion-pairing reagent (0.75 g of tetrabutylammonium acetate and 7.7 g of ammonium acetate in 100 ml water) and methanol. Samples were separated and quantified using an Agilent 1200 series high performance liquid chromatography (HPLC) separation module, quaternary pump, with an octadecylsilyl (ODS) Hypersil column (205 × 4.6 mm; 5 μm particle size) and photo-diode array detector, using modification conditions of Chien et al [30]. Pigments were quantified using commercial standard calibrations (DHI LAB PRODUCTS, Denmark) and identified using absorbance spectra and retention times. Analyzed pigments include those from siliceous algae (fucoxanthin, diadinoxanthin, diatoxanthin), chlorophytes (lutein), cyanobacteria (zeaxanthin, canthaxanthin) and from all primary producers (Chla and its degradation products the allomer Chla' and the epimer Chla").

Water samples collected monthly at each river sampling station were filtered through 0.7 µm GF/F Whatman filters. Filtrates were scraped from the filter papers and ground using an agate pestle and mortar. The percentage total organic carbon (% TOC) and percentage total nitrogen (% TN) were analyzed by sample combustion in a Costech Elemental Analyser at the National Environmental Isotope Facility, British Geological Survey. C:N ratios are reported as atomic mass ratio.

Monthly concentrations of pCO₂(μatm) at each sampling station were calculated from TA, pH, and temperature using the freshwater option in the CO2SYS program (version 2.0) [31]. pCO₂ values were then corrected for pH and alkalinity bias following Liu et al [32]. Bias in pH was estimated using a relationship between pH measurement error and river water ionic strength (conductivity). As low pH data is prone to higher errors owing to low ionic strength, we discarded any sampling data with pH < 6.4 (35 samples out of 367) before corrections [32, 33]. Organic alkalinity was estimated by applying a ratio to river water dissolved organic carbon (DOC) concentration. DOC was indirectly estimated by assuming an exponential relationship with TA [32].

To reduce variance in the data (tables S3 and S4), the median seasonal and annual values of the corrected monthly pCO₂, pigment, physicochemical parameter, C:N data, and river discharge data at each sampling station were calculated prior to statistical analysis. Conductivity and NH₄–N were highly collinear with the other parameters and excluded from all analyses. A Tukey Honest test (p < 0.05) was used to assess differences in physicochemical variables, water discharge, C:N data and pCO₂, at: (a) the annual between-river scale; (b) within-river seasonal scale; and (c) Hanoi City vs. less densely urban populated RRD provinces (Yen Bai, Phu Tho, Hoa Binh, Ninh Binh, Ha Nam and Nam Dinh). This test incorporates an adjustment for the unbalanced sample size to avoid inflation in the probability of declaring a significant difference between the groups.

Spatial gradients in the a/biotic parameters and pCO₂ variation were then assessed using Multiple factor analysis–MFA [34]. MFA clusters the different phytoplankton pigments (cyanobacteria, siliceous algae, chlorophytes and phytoplankton biomass), selected water quality parameters (i.e. excluding those used to calculate pCO₂), C:N, discharge, and corrected pCO₂ along with categorical variables (river name and season) into specific groups to simultaneously assess the amount of variation explained by each group. The quantitative group variables were scale-transformed in the MFA to reduce skewness and rescale the data into comparable units [34]. The MFA was performed in R using the package FactoMineR [35].

A partial redundancy analysis (pRDA; varpart function in 'vegan' R package) [36] was then conducted to partial out the unique contribution of the a/biotic data on pCO₂ variation. A parsimonious pRDA was achieved by selecting first, the most important a/biotic predictors of pCO₂ variation via forward selection analysis (forward.sel function in 'adespatial' R package) [37]. The variation explained by each component in the pRDA was corrected to adjusted R². The total variance of the pRDA was decomposed into four fractions: (a) abiotic, (b) biotic, (c) shared, and (d) unexplained variation. The shared fraction (c) represents variance that may be attributed to biological and/or abiotic descriptors together. The significance of each component was tested through 499 random Monte Carlo permutations under the reduced model.

Median values of the study a/biotic parameters and Tukey test results are presented in the tables S3–S5 respectively. Significant comparisons between these parameters (figure 2), indicated that annual median nutrient concentrations in the Day River were higher (TDN: 3.3 ± 1.6 mg l⁻¹; NO₂–N: 0.1 ± 0.1 mg l⁻¹; TDP: 0.3 ± 0.4 mg l⁻¹) than the Red River (TDN: 1.3 ± 0.2 mg l⁻¹; NO₂–N: 0.03 ± 0.03 mg l⁻¹; TDP: 0.1 ± 0.1 mg l⁻¹). Turbidity (150.7 ± 353.3 nephelometric turbidity unit [NTU]) and TA (1992 ± 243 µmol l⁻¹) were also higher in the Day vs. the Red River (turbidity: 29.5 ± 21 NTU; TA: 1705 ± 84.6 µmol l⁻¹ ). pH (6.8 ± 0.2) and DO (2.6 ± 1.0 mg l⁻¹) were in turn, lower in the Day vs. the Red River (pH: 7.1 ± 0.3; 3.9 ± 0.4 mg l⁻¹). C:N ratios of OM were higher in the Red (8.4 ± 1.4) than the Day (6.5 ± 0.6) River. Higher (∼10-fold) chlorophyll and carotenoid pigments from river phytoplankton were also found in the Day River than the Red River (figure 2).

Zoom In Zoom Out Reset image size

Comparisons in a/biotic parameters between the seasons at each river (figure 2) indicated that during the dry season, pH in both rivers was significantly lower (Day River: 6.6 ± 0.2; Red River: 6.8 ± 0.1) than in the rainy season (Day River: 6.9 ± 0.1; Red River: 7.3 ± 0.2). TDN and TA concentrations were also higher in the Day River during the dry season (TDN: 3.9 ± 1.6 mg l⁻¹; TA: 2564 ± 9230 µmol l⁻¹) than the rainy season (TDN: 2.3 ± 0.8 mg l⁻¹; TA: 1992 ± 243 µmol l⁻¹). Among river phytoplankton, diadinoxanthin from siliceous algae mainly occurred during the wet season in the Day River. Lutein from chlorophytes and indicators of total phytoplankton biomass in the Day River were also higher in the rainy season (lutein: 1.6 ± 0.6 nmol l⁻¹; Chla: 8.7 ± 4.1 nmol l⁻¹; Chla': 2.5 ± 1.4 nmol l⁻¹; Chla": 1.1 ± 0.7 nmol l⁻¹) than in the dry season (lutein: 0.8 ± 0.6 nmol l⁻¹; Chla: 2.8 ± 1.6 nmol l⁻¹; Chla': 1.2 ± 0.7 nmol l⁻¹; Chla": 0.5 ± 0.3 nmol l⁻¹). Zeaxanthin was, in contrast, higher in the Day River during the dry season (0.4 ± 0.2 nmol l⁻¹) vs. the rainy season (0.02 ± 0.05 nmol l⁻¹).

Discharge in the Red River during both seasons, was significantly higher in Hanoi City than in the other provinces (figure 3). During the dry season in the Day River, DO (1.9 ± 1.2 mg l⁻¹) and NO₃–N (0.7 ± 0.5 mg l⁻¹) were lower around Hanoi City than in the other provinces (DO: 3.8 ± 1.5 mg l⁻¹; NO₃–N: 1.9 ± 0.9 mg l⁻¹). During the dry season, TA (3362 ± 1332 µmol l⁻¹) and TDP (0.5 ± 0.6 mg l⁻¹) were in turn, higher in the Day River, around Hanoi City compared to the other provinces (TA: 2150 ± 189 µmol l⁻¹; TDP: 0.1 ± 0.03 mg l⁻¹).

Zoom In Zoom Out Reset image size

Phytoplankton biomarkers from cyanobacteria (zeaxanthin: 0.6 ± 0.2 nmol l⁻¹) were higher around Hanoi City in the Day River during the dry season than in the other provinces (zeaxanthin: 0.3 ± 0.1 nmol l⁻¹), while siliceous algal biomarkers (fucoxanthin) were lower in Hanoi City (0.7 ± 0.3 nmol l⁻¹) than in the other provinces (1.9 ± 1.1 nmol l⁻¹) (figure 3). In the rainy season, biomarkers representing siliceous algae (diatoxanthin: 0.2 ± 0.4 nmol l⁻¹), cyanobacteria (canthaxanthin: 0.1 ± 0.1 nmol l⁻¹), and total phytoplankton biomass (Chla: 7.1 ± 3.2 nmol l⁻¹, Chla': 1.8 ± 0.5 nmol l⁻¹) were lower around Hanoi City in the Day River vs. the other provinces (diatoxanthin: 1.2 ± 1.3 nmol l⁻¹, canthaxanthin: 0.2 ± 0.2 nmol l⁻¹, Chla: 12.0 ± 4.2 nmol l⁻¹, Chla': 3.7 ± 1.8 nmol l⁻¹).

3.3.1. pCO₂

After correction, water pH increased on average by 0.12 units across the Day River and by 0.14 units across the Red River (table S6). Alkalinity only declined in both rivers by <3%. Consequently, correcting pCO₂ indicated an overestimation of around 36 ± 8% for the Day River data and of 42 ± 7% for the Red River data (figure 4(a); table S6). The annual median concentration of both rivers combined (i.e. RRD) was 3976 ± 3144 µatm and there were significantly higher median values in the Day River (5000 ± 3300 µatm) than in the Red River (2675 ± 2271 µatm) (figure 4(b)). Concentrations in both rivers were much higher in the dry season (Day River: 6689 ± 3821 µatm, Red River: 4404 ± 1883 µatm) than in the rainy season (Day River: 3908 ± 1259 µatm; Red River: 1402 ± 579 µatm). No differences in pCO₂ concentrations were observed between Hanoi City and the other less urban-populated provinces in both rivers.

Zoom In Zoom Out Reset image size

The MFA showed the data are clearly separated by river, seasons, and degree of urbanization (Hanoi City vs. other provinces) (figure 5). Red River data negatively correlated with dimension 1 (DIM1), and Day River data positively. The a/biotic data from monitoring during the rainy season, were negatively associated (for both rivers) with dimension 2 (DIM2), but positively related to DIM2 during the dry season. An upstream-downstream spatial pattern, particularly for the Day River, was also depicted with downstream more saline (marine-influenced) areas converging towards the center of the plot and Hanoi city stations towards the right-hand side of the plot.

Zoom In Zoom Out Reset image size

MFA DIM1 explained 43% of the total variation with degree of urbanization, river phytoplankton pigment groups, river type, water quality, and C:N ratio having the greatest contribution (figure S1). DIM2 explained 17% of the total variation with season, pigments from cyanobacteria, and pCO₂ contributing the most (figures 1 and S1). A strong negative association with DIM1 was observed for river discharge, DO, and C:N ratio, while pCO₂, nutrients and turbidity, were all positively correlated. Pigments from river phytoplankton were all positively related to DIM1 but separated along DIM2. pCO₂, NO₃–N, and cyanobacteria (zeaxanthin) positively related to DIM2, while all siliceous algae, cyanobacteria (canthaxanthin) and phytoplankton biomass negatively relate to DIM2.

Forward selection analysis identified cyanobacteria (zeaxanthin) and river discharge, as the most significant variables contributing to pCO₂ adjusted variation (table S7). After accounting for the shared effects (14%), river discharge alone related to a significant 12% of the pCO₂ variation and cyanobacteria uniquely to 21% (figure 5(c)). A remaining 53% of the pCO₂ variance was unexplained by discharge and cyanobacteria.

Comparisons of two tropical rivers (Red and Day) which differ in size, discharge, degree of urban population, and water pollution allowed us to estimate the extent and causes of CO₂ supersaturation in the RRD, one of the most densely populated river deltas in Southeast Asia. Both river networks exhibited pCO₂ above atmospheric equilibrium (>410 μatm), consistent with other river studies in Southeast Asia, showing that they function as potential sources of CO₂ to the atmosphere. However, the 1.8-fold higher pCO₂ across the more hydrologically-regulated Day River, along with higher nutrients and lower DO across Hanoi City, highlight significant negative urban influences on river water quality and carbon dynamics. Urbanization effects on pCO₂ were further supported by the observed pCO₂ in the Day River during the dry season, which was almost double that of the Red River and was 2.4-fold higher than the median of calculated and directly measured pCO₂ (2811 ±  3577 µatm) in >20 global sub/tropical studied rivers (figure 6).

Zoom In Zoom Out Reset image size

Compared to the Red River, the Day River has a more populous catchment with abundant OM and TA sources, and waters that are turbid, hypereutrophic and with oxygen concentrations that are marginal for supporting aerobic life. The observed large differences in pCO₂ between the two rivers reflect the more stagnant/lentic conditions of the Day River resulting from river channelization and intensive ground and surface water use of approximately 70 l person day⁻¹ within Hanoi City [39, 40]. Urban domestic sewage effluents increase OM inputs enriched in organic nitrogen and heterotrophic micro/organisms [12, 18] and can reduce riverine organic C:N ratios below those typical of terrestrial sources [41]. The higher lability of DOC derived from urban watersheds is also thought to contribute to higher CO₂ production and DIC [11–13]. Aquatic pCO₂ can be higher in such narrower, smaller river channels with lower flow velocity and greater concentrations of DIC and OM [11–13], and the modifications of the Day River appear to have markedly enhanced pCO₂ in the upstream urban areas around Hanoi City. Higher [NO₃–N] outside Hanoi City during the dry seasons in the Day River (figure 3) likely reflects the influence of nitrogen, phosphorus and potassium (NPK) chemical fertilizers and urea that are heavily applied during the dry season in rice paddy fields, and which eventually leach into rivers/canals networks [42]. Alternatively, as observed in the Red River, OM and pollutant inputs from urbanization can be diluted (though still above equilibrium) in large fast-flowing rivers, leading to lower pCO₂ [43].

4.1.1. pCO₂ and river phytoplankton

As the river water moves from the upstream urban areas towards the coast, the physicochemical conditions of the waters may naturally change towards an increase in pH and bicarbonates [15, 41]. Such changes may simply reflect lower pCO₂ in the coastal areas compared to the upstream urban areas [15, 41], hence confounding our interpretations. However, the spatial variation in pH and TA across our sampling sites does not entirely reflect such a gradient in water chemistry (table S3), suggesting the impacts from other driving factors such as the influence of river urbanization on pCO₂.

Direct measurements of water pCO₂ are scarce in tropical rivers. However, as water properties including T, pH, and TA are regularly monitored by governments and research bodies, estimations of pCO₂ from these monitored parameters remain important for assessing trends of C dynamics in response to human activities [38, 48]. Correcting the pCO₂ values for pH and organic alkalinity bias reduces estimations by 30%–50%; as recently reported in China's main rivers [48]. Although organic alkalinity is recognized as a main source of CO₂ calculation error in tropical acidic waters [32, 33], the high TA (>1000 μmol l⁻¹) in the RRD waters resulted in minimal influence. pCO₂ estimations also concur with previous assessments in the Day (calculated values) and Red (direct measurements) Rivers, especially during the rainy season (figures 4 and 6). In our study there was a marked exceedance of pCO₂ during the dry season in the Red River compared to that recorded in 2014 [20]. Such exceedances are most likely to be due to distinct pH and rainfall patterns between 2014 and 2018–2019 [19] with dry season pH (8.1 ± 0.2 [20]) and rainfall (1660 mm) being higher in 2014 than in 2018–19 (pH 6.8 ± 0.1 [20]; rainfall 1311 mm). Similar low pH values (7.3 ± 0.1) were also reported in 2019 for the Red River [49].

Carbon emissions from fossil fuels and land use change combined are estimated at around 11 500 ± 0.9 Tg C yr⁻¹ globally [50]. The natural contribution of tropical riverine CO₂ fluxes to the global C budget is comparatively minimal (<600 Tg C yr⁻¹). However, our findings are consistent with recent research showing that water quality degradation through increasing lentic/stagnant waters from river channelization, sewage treatment infrastructure, and pollution load in densely populated cities, could significantly strengthen the potential for river CO₂ outgassing [11–13, 51]. Annual pCO₂ in the Day River was 1.8-fold higher than in the Red River and up to two-fold greater during the dry season than the median of pCO₂ in >20 sub/tropical rivers. Thus, the rapid urbanization of rivers could result in hotspots of aquatic CO₂ sources, water pollution and HABs [11–13], increasing water quality concerns downstream for communities who are dependent on the water and for river biota [52]. The increase in water demands to support urban development coupled with periods of low rainfall and river flows as projected in future climates [53, 54] could further exacerbate such risks. Currently, 29% of total wastewater in Hanoi is treated [55] with near-future plans (2021–2025) aiming to increase it to 50% [56]. Such endeavor will undoubtedly help removing OM, and organic nutrients load into the rivers. However, the wastewater collection system and the pumping of untreated wastewater into the Red River for flood remediation, are still a big issue [42]. This means that water quality in the RRD needs to be monitored carefully in the long-term, especially as the release of mineralized nutrients from the expanding water treatments plants will likely exacerbate the HABs-pCO₂ feedback loop.

Traditionally, urban tropical river drainage systems include levees, dams, artificial lakes, and detention facilities to cope with the prevalent floods [22]. However, we demonstrate here that beyond flood control, water quality degradation and the associated increases in pCO₂ and HABs must also be considered to achieve environmental sustainability. To help mitigate these harmful effects introduced by urban development, we suggest carrying out responsive water course management actions to increase river flows in problem areas during the dry season [22, 40], along with the implementation of green infrastructure that prioritize the natural hydrological functions of the river [54]. Improving sanitation infrastructure and management in urban centers [51] and strengthening the engagement with local government to exchange research findings and support policy making on water management, including sustainable agricultural development (e.g. use of fertilizers) [42], could further help reducing nutrient load and greenhouse gasses.