High-frequency monitoring reveals a CO₂ source-sink shift in a subtropical eutrophic urban lake

Abstract

Eutrophic urban lakes with CO₂-supersaturation represent potential carbon (C) sources; however, the drivers behind the reported C-source–sink shift remain poorly understood. This study provides a systematic assessment of daytime/seasonal pCO₂ and fCO₂ dynamics in a subtropical moderately eutrophic urban lake (Bailuwan, China), based on over a year of high-frequency monitoring, aiming to clarify the mechanisms regulating CO₂ exchange at the water–air interface in such ecosystems. Our work revealed consistent daytime declines in pCO₂ (and fCO₂) on 12 sampling days, though morning–afternoon differences were not significant (n = 24). Novel episodic undersaturation events were newly observed in October 2020 and March 2021, contrasting with the prevailing supersaturation. Annual mean values (n = 48) reached 1789 µatm (pCO₂) and 130 mmol m⁻² h⁻¹ (fCO₂). Critically, we identified a pronounced semi-annual divergence: pCO₂ from January to June significantly exceeded values from July to November. Both periods maintained a net source status (> 420 µatm), lacking the typical spring-sink/summer-source transition reported in previous studies. Key regulators, such as pH, chlorophyll a, and dissolved oxygen, influence C-sink-source dynamics, with eutrophication further modulating these shifts. These original findings highlight the need for targeted strategies to reduce pollutants and enhance carbon sequestration in urban lakes.

Introduction

Freshwater ecosystems, encompassing lakes, rivers, and wetlands, are pivotal in the biogeochemical dynamics of greenhouse gases (GHGs), with impacts extending from local to planetary scales^1,2,3,4,5,6. In particular, inland lakes, despite their limited spatial coverage (~ 3.7% of non-glacial terrestrial surfaces)⁷, are key players in carbon cycling through sequestration, transport, and transformation mechanisms^8,9,10. Empirical studies have consistently demonstrated that approximately 90% of global inland lakes exhibit carbon dioxide (CO₂) supersaturation relative to atmospheric equilibrium^11,12. Global estimates suggest that CO₂ emissions from lakes range from 0.07 to 0.15 Pg C year^-1, with upper estimates reaching as high as 0.57 Pg C year^-113,14,15. With a spatial extent representing 6.2% of global lake coverage (25°N–54°N), lakes in China release roughly 15.98 Tg C year^{− 1} in the form of atmospheric CO₂ emissions^16,17,18. Unexpectedly, shallow lakes with extensive surface areas but depths of less than 3 m have been identified as critical hot-spots for CO₂ emissions owing to their high biogeochemical activity and efficient gas exchange^19,20. The consistent pattern of CO₂ supersaturation in these aquatic environments, with CO₂ levels exceeding the atmospheric equilibrium range of 380−420 µatm, provides clear evidence of their role as net CO₂ sources to the atmosphere^21,22,23,24.

From a global spatiotemporal perspective, the CO₂ emitted into the atmosphere is rebalanced through biogeochemical mechanisms such as photosynthetic assimilation, sediment burial, and oceanic uptake, collectively contributing to the dynamic equilibrium of the global carbon cycle^25,26. From 2007 onward, the oceans have sequestered about 56% of human-induced carbon emissions, resulting in detectable ocean acidification (pH), with the residual 44% accumulating in the atmosphere^3,5,27,28. Growing empirical evidence documents the conversion of particular lake ecosystems from net carbon sources to net sinks, with some currently exhibiting transitional carbon dynamics^29,30,31. Systematic monitoring data presented by Xiao et al.³² showed a significant downward trend in CO₂ outgassing rates from Chinese lakes between the 1980s and 2000s, pointing to their gradual transformation from C-releasing to C-sequestering ecosystems. This transformation may be attributed to synergistic interactions among biogeochemical processes (e.g., enhanced C-sequestration and organic matter burial), shifting environmental conditions (e.g., nutrient loading and hydrological regimes), and targeted anthropogenic interventions (e.g., ecological restoration and eutrophication control)^{25,33,34,35,36}. Consequently, systematic documentation of C-source-sink dynamics and their regulatory controls in lacustrine ecosystems is imperative to refine predictive models of C-exchange and quantify their contributions to regional C-budgets under evolving climatic and anthropogenic pressures.

Watershed urbanization generates synergistic perturbations to aquatic ecosystem processes, wherein combined effects of anthropogenic nutrient loading and riparian habitat fragmentation lead to fundamental alterations in carbon transformation pathways and greenhouse gas exchange in urban water systems^37,38. Urban lakes, as highly sensitive freshwater ecosystems, are particularly susceptible to algal blooms, resulting in eutrophication^35,39,40,41. In this study, an urban lake is defined as a lentic water body situated entirely within a metropolitan area, whose hydrological processes, water quality, and ecological functions are primarily governed by anthropogenic activities^15,41,56,89. Key characteristics include: (i) heavily modified hydrology through water level control and engineered shorelines; (ii) significant nutrient inputs from urban runoff and wastewater; (iii) altered ecological communities due to habitat modification and recreational use; and (iv) serving dual roles in both receiving urban discharges and providing ecosystem services such as flood mitigation and recreation^25,26,42. Comparative studies reveal dramatic differences in carbon emissions between two Indian lakes: the hypereutrophic Belandur Lake exhibits exceptionally high CO₂ efflux rates (5711 ± 844 Tg C year^{− 1}), while Jakkur Lake, currently under ecological rehabilitation, demonstrates substantially lower emissions (24 ± 10 Tg C year^{− 1})⁴². While empirical evidence suggests that eutrophication could exert a modest positive influence on CO₂ sequestration in non-urban lakes^29,36, its role and mechanistic pathways in urban lakes remain inadequately elucidated. Consequently, to advance our comprehension of the underlying mechanisms regulating CO₂ exchange in urban lakes, particularly shallow lakes, it is imperative to conduct more extensive field measurements and systematic analyses.

Inland water CO₂ fluxes (fCO₂) are predominantly controlled by the interplay between aqueous CO₂ partial pressure (pCO₂) and the rate of gas transfer (kCO₂) across the water-air boundary in freshwater ecosystems^17,43,44. The pCO₂ parameter, recognized as a pivotal determinant in deciphering carbon cycle variability across urban water bodies at multiple spatial scales^11,45, demonstrates multifactorial regulation through: (i) environmental drivers (e.g., solar irradiance)^4,9,10, (ii) biogeochemical processes (particularly aquatic metabolism involving photosynthesis-respiration/P-R coupling)⁴⁶, (iii) hydrological dynamics (including thermal stratification and mixing regimes)^47,48,49, and (iv) allochthonous carbon inputs from catchment areas^50,51,52,53. Correspondingly, kCO₂ variability exhibits primary dependence on wind shear stress and thermal conditions⁵⁴. Distinct from the relative homogeneity of atmospheric pCO₂, lacustrine pCO₂ manifests pronounced spatiotemporal heterogeneity across daytime, monthly, and seasonal scales, with system-specific characteristics strongly influenced by morphometric parameters and trophic status^4,5,6,55,56. Empirical evidence from Lake Ulansuhai, a shallow urban waterbody, reveals eutrophication-induced functional shifts from CO₂ source to sink conditions⁵⁷. Moreover, rapid urban expansion around lacustrine environments introduces substantial complexity to carbon cycling processes, with notable impacts on CO₂ emission patterns^5,58,59. Despite these insights, critical knowledge gaps persist regarding high temporal resolution characterization of pCO₂ dynamics in urban shallow lakes, necessitating systematic investigations into urbanization-CO₂ emission synergies through integrated observational and modeling approaches.

Building upon the aforementioned backgrounds, this work aims to address a central scientific question: how and to what extent do eutrophic urban lakes modulate the dynamics of CO₂ exchange across the water–air interface under high-frequency monitoring? To resolve the gap, we implemented a comprehensive study assessing daytime CO₂ patterns in a shallow, subtropical urban lake with elevated nutrient levels in southwest China. This work pursues three principal aims: (i) performing monthly diurnal monitoring of pCO₂/fCO₂ and associated physicochemical variables between October 2020 and November 2021; (ii) identifying key hydrological and environmental controls on aquatic carbon cycling; and (iii) evaluating the premise that subtropical eutrophic lakes demonstrate alternating C-source-sink behavior under high-frequency investigation. The findings of this work are expected to provide novel insights into the mechanisms governing CO₂ exchange in lacustrine systems and to refine the quantification of CO₂ emissions from urban lakes, thereby reducing uncertainties in regional/global C-budget assessments.

Results

Variations in pCO₂

Throughout our investigation period (07:00–18:00 CST), sustained/significant (p > .05) daytime pCO₂ decreases were observed during nine sampling days, with the exception of three dates (February 28, March 30, and November 22, 2021) that exhibited gradual increasing trends (n = 1; Fig. 1A–L). Considering individual field investigation days, the declining pattern remained predominant across most sampling occasions, except those illustrated in Fig. 1D, E, I, J, L. The composite analysis of four temporal sampling points revealed a statistically remarkedly (p < 0.05) reduction in mean pCO₂ levels (n = 12) between 11:00 and 18:00 CST (Fig. 2M; Table 1). Relative to the 07:00 CST reference (2043.91 ± 2033.26 µatm), pCO₂ exhibited pronounced (p < 0.05) daytime fluctuations as 42.63% at 11:00 CST, followed by successive reductions of 49.18% (14:00 CST) and 43.42% (18:00 CST). Interestingly, non-significant (p > 0.05) diurnal variation was detected between morning (07:00–11:00 CST; 2479.56 ± 3971.80 µatm) and afternoon periods (14:00–18:00 CST; 1097.66 ± 803.29 µatm) when analyzing the combined dataset (n = 24; Fig. 2A).

Fig. 1

Hourly variations of pCO₂ (solid line) and fCO₂ (dashed line) in the studied lake. Hollow circles denote individual measurements (n = 1; Fig. 2A–L) and mean values (n = 12; Fig. 2M–N). Field measurements originally scheduled for July 2021 were conducted on August 2 due to meteorological constraints.

Fig. 2

Diurnal comparison of mean pCO₂ (A), fCO₂ (B), T_water (C), pH (D), Chla (E) and DO (F) between morning (07:00–11:00 CST) and afternoon (14:00–18:00 CST) periods (n = 24). Distinct lowercase letters denote statistically significant differences (p < 0.05) between temporal periods. The vertical line on the bars represents the mean ± standard deviations (SDs).

Table 1 Temporal correlation analysis of daily pCO₂ and fCO₂ in Bailuwan lake (n = 4). x, time; y₁, pCO₂; y₂, fCO₂; r², regression coefficients.

Monthly analysis revealed unexpected temporal patterns, with peak pCO₂ concentrations (n = 4; 8148.02 ± 1882.12 µatm) occurring in October 2021, contrasting sharply with the minimum values recorded in October 2020 (198.14 ± 65.20 µatm). Throughout the annual cycle (January–November 2021), mean pCO₂ levels during the first half of the year (January-June; n = 28; 1239.34 ± 781.00 µatm) generally lowered (p > 0.05) those of the latter half (July-November; n = 16; 3143.15 ± 4673.69 µatm). However, January and March exhibited the lower mean concentrations (510.20 ± 101.94 µatm and 314.92 ± 145.48 µatm, respectively; Fig. 3A). The overall daytime mean pCO₂ across all measurements was 1788.61 ± 2919.44 µatm (n = 48).

Fig. 3

Monthly variations of pCO₂ (A) and fCO₂ (B) at the water–air interface in Bailuwan Lake. Upper panel displays median (black line) and mean (red line, n = 4), with whiskers spanning the range between [Q₁ − 1.5×IQR] (lower bound) and [Q₃ + 1.5×IQR] (upper bound). Q₁: first quartile; Q₃: third quartile; IQR: interquartile range (Q₃ − Q₁). Oct.0 and Dec.0 denote October and December 2020 sampling campaigns, respectively; remaining data were obtained in 2021. The vertical line on the bars represents the mean ± standard deviations (SDs).

Alterations in fCO₂

Similar to the pCO₂ patterns, the daytime mean fCO₂ concentrations (n = 12) exhibited a characteristic pattern of initial increase significantly (p < 0.05) followed by subsequent decrease. Specifically, compared to the baseline measurement at 07:00 CST (139.16 ± 175.86 mmol m^− 2 h^− 1), fCO₂ concentrations showed a 76.13% increase by 11:00 CST, followed by significant reductions of 55.93% and 45.78% at 14:00 and 18:00 CST, respectively. Analysis of individual sampling day (n = 1) revealed consistent and statistically significant (p < 0.05) daytime fCO₂ decreases across nine sampling dates, with the exception of February 28, March 30, and November 22, 2020 (Fig. 1A–L). Further, comparative analysis between morning (07:00–11:00 CST; 192.14 ± 418.09 mmol m^− 2 h^− 1) and afternoon periods (14:00–18:00 CST; 68.39 ± 76.65 mmol m^− 2 h^− 1) demonstrated non-significant (p > 0.05) diurnal variation in mean fCO₂ concentrations (n = 24; Fig. 2B).

Monthly analysis revealed distinct temporal patterns, with negative daytime mean fCO₂ values (n = 4) recorded in October 2020 and March 2021, contrasting sharply with the peak concentration observed in October 2021 (n = 4; 775.16 ± 873.01 mmol m^− 2 h^− 1; Fig. 3B). Remarkably, the calculated mean fCO₂ concentration was 130.26 ± 303.85 mmol m^− 2 h^− 1 (n = 48) across the entire study period.

Classification of trophic state

The Carlson’s trophic state index (TSI) was quantitatively assessed through the established methodology in Method S1, incorporating key limnological parameters including total phosphorus (TP), total nitrogen (TN), water transparency (TPC), chlorophyll a (Chla), and chemical oxygen demand (COD_Mn) as detailed in Table S1 and Fig. 4. The comprehensive TSI(∑) analysis yielded a mean value of 63.04 (Table 2), categorizing the lake within the moderately-eutrophic classification according to the standard limnological criteria. Temporal analysis of TSI(∑) dynamics revealed sustained values exceeding the eutrophication threshold of 60 across the majority of sampling intervals. However, notable deviations (p < 0.05) were observed during the spring sampling campaigns of March and April 2021, during which TSI(∑) values fell below this critical benchmark, as documented in Table S2.

Fig. 4

Temporal variations in water quality parameters of Bailuwan Lake: TPC (A,B), Chla (C,D), TN (E,F), TP (G,H), and COD_Mn (I,J). Additional details are provided in Fig. 3.

Table 2 Comprehensive TSI of Bailuwan lake (n = 48). The table presents NT, TP, COD_Mn, total dissolved nitrogen (TDN) and eutrophication evaluation criteria, as detailed in methods S2–S7.

Environmental parameters

Regarding diurnal variations, water quality parameters including aqueous temperature (T_water), TPC, dissolved oxygen (DO), chloride ion (Cl⁻), nitrate ion (NO₃⁻), and sulfate ion (SO₄²⁻) exhibited a progressive increase from 07:00 to 18:00 CST (Figs. 4, 5 and 6). Surprisingly, DO concentrations demonstrated a pronounced surge of 57.33% over this period, escalating from baseline levels at 07:00 CST to maximum values observed at 18:00 CST. This trend was corroborated by comparative analyses, which revealed significantly (p < 0.05) elevated afternoon DO levels relative to morning measurements (Fig. 2F). Conversely, non-statistically significant (p > 0.05) diurnal variations were detected in T_water, pH, or Chla concentrations between morning and afternoon sampling intervals (Figs. 2C–E; p > 0.05).

Fig. 5

Temporal variations in physicochemical parameters of Bailuwan Lake: T_water (A,B), pH (C,D), FNU (E,F), EC (G,H), DO (I,J), and TOC (K,L). Left panels (A,C,E,G,I,K): red lines indicate mean values per sampling time (n = 12); and right panels (B,D,F,H,J,L): black dots denote monthly means (n = 4) with whiskers representing mean ± SDs. Additional details are provided in Fig. 3.

Fig. 6

Temporal variations in anionic species of Bailuwan Lake: F⁻ (A,B), Cl⁻ (C,D), NO₃⁻ (E,F) and SO₄²⁻ (G,H). Additional details are provided in Fig. 3.

Monthly monitoring revealed a shared temporal trajectory among water quality indicators including TPC, turbidity (FNU), electrical conductivity (EC), anion concentrations (F⁻, Cl⁻, and SO₄²⁻), and metal levels (potassium/K, sodium/Na, chromium/Cr) all followed an initial ascending phase followed by measurable decreases. Conversely, Chla and TN demonstrated an initial decrease followed by an increase. The parameters pH and NO₃⁻ showed a gradual decline, while magnesium/Mg displayed a consistent upward trend (Figs. 4, 5, 6 and 7). Additionally, our analysis of monthly variations in CO₃²⁻ and HCO₃⁻ levels revealed distinct patterns. The CO₃²⁻ generally exhibited an initial decrease followed by an increase, with a notable surge observed in March 2021. In contrast, the HCO₃⁻ reached the lowest value in July 2021 (Fig. S1).

Fig. 7

Monthly changes in aquatic metals of Bailuwan Lake: K (A), Na (B), Mg (C), Cu (D), Zn (E), Fe (F), Mn (G), and Cr (H). Additional details are provided in Fig. 3.

Statistical analysis demonstrated strong inverse relationships of pCO₂ with pH and DO (p < 0.01, Table 3), contrasted by a direct positive association with Chla (p < 0.05, Table 4). These correlations facilitated the development of predictive linear models linking pCO₂ to the key parameters (pH, DO, Chla, solar radiation/SR, T_water, and COD_Mn), as visualized in Figure S2.

Table 3 Correlation analysis of pCO₂/fCO₂ with water quality parameters in the studied lakes. * and ** denote significant correlations at the 0.05 and 0.01 levels (two–tailed test), respectively. T_water, water temperature; EC, electrical conductivity; TPC, transparency; FNU, turbidity; DO, dissolved oxygen; SR, solar radiation.

Table 4 Correlation analysis of pCO₂/fCO₂ with nutrient status indices in the studied lake. Chla, chlorophyll a; TOC, total organic carbon; F⁻, fluoride; Cl⁻, chloride; NO₃⁻, nitrate; SO₄²⁻, sulfate; TN, total nitrogen; TP, total phosphorus; COD_Mn, permanganate index. Additional details are provided in Table 3.

Discussion

Notable shiftting patterns of C-sink-source in the investigated lake

According to recent data released by the United Nations, the global urbanization rate has increased from 30% in 1950 to 56% in 2020, and is projected to reach 68% by 2050^60,61. In China, more than 60% of the permanent population had achieved urbanization by the end of 2019^22,38. Previous studies have demonstrated that urban lacustrine systems receive substantial inputs of exogenous labile organic carbon derived from diverse anthropogenic activities, which significantly enhances heterotrophic metabolism in these urbanized aquatic ecosystems, consequently resulting in elevated CO₂ emissions. These mechanistic insights explain why metropolitan water bodies have been consistently documented^42,62,63,64 as focal points for carbon release within anthropogenic landscapes. Consequently, investigating the mechanisms by which urban lakes respond to urbanization is of significant importance for predicting urban greenhouse gas emissions, particularly CO₂.

Across all multi-year sampling days, the pCO₂ in the morning hours (07:00–11:00 CST; 2480 ± 3972 µatm; n = 22) was numerically higher than that in the afternoon periods (14:00–18:00 CST; 1098 ± 803 µatm; n = 22; Figs. 1 and 2), but the difference was not statistically significant (p > 0.05). Conversely, afternoon DO concentrations demonstrated a significant elevation compared to morning values (p < 0.05; Fig. 3F), whereas Chla levels remained stable without observable daytime fluctuations (p > 0.05; Figs. 3 and 4). These patterns align with prior work in local eutrophic lakes^4,5,6. Increased solar radiation after 07:00 CST spurs photosynthesis (P); later, declining light toward 18:00 CST shifts the balance toward respiration (R). The resultant drop in dissolved oxygen is documented in Table 1 and Fig. S3. Noteworthy, temperature-driven pCO₂ fluctuations demonstrated no direct significance⁶⁵ (Fig. 2C). Parallel observations by Potter and Xu in a subtropical North American lake revealed pronounced diurnal pCO₂ dynamics, with predawn peaks and evening troughs⁵⁸. Intriguingly, nocturnal CO₂ efflux rates nearly tripled daytime values, underscoring distinct day-night emission patterns. In our dataset, anomalous pCO₂ behavior was observed on February 28, March 30, and November 22, 2021, where morning pCO₂ (07:00–11:00 CST) slightly decreased compared to afternoon levels (Fig. 1). Measurements from February 28 and November 21 consistently demonstrated pCO₂ supersaturation, with all recorded values surpassing the characteristic atmospheric CO₂ range (380–420 µatm)^66,67. In contrast, significantly lower pCO₂ levels were recorded on March 30, 2021 (198 ± 65 µatm) and October 30, 2020 (315 ± 145 µatm; n = 4; Fig. 1), reflecting dynamic C-sink-source transitions. The underlying mechanism may be attributed to extreme precipitation events acting as a trigger. On one hand, rainfall directly reduced pCO₂ through the dilution effect. On the other hand, it introduced allochthonous nutrients and enhanced water column mixing, thereby stimulating intense algal blooms. The resulting high level of photosynthetic activity served as the key biological driver leading to significantly decreased pCO₂. Furthermore, as an urban wetland, its hydrology is likely influenced by anthropogenic regulation. The supplemental inflow of low-CO₂ external water (e.g., reclaimed water) during this period may have further reinforced and amplified the decline in pCO₂^6,15,68.

In our study, the overall diurnal mean pCO₂ level in the investigated lake (n = 48; 1789 ± 2919 µatm) was significantly higher than typical equilibrium CO₂ thresholds (p < 0.05) but markedly lower than global lacustrine average pCO₂ of 3230 µatm⁶⁹. This finding indicates net CO₂ supersaturation within the lake during our study, where in-lake CO₂ production exceeded consumption, driving a net efflux of CO₂ to the atmosphere and thus classifying the lake as a C-source^31,44,70,71. These results align with prior investigations of eutrophic urban lakes, including Beihu Lake (~ 960 µatm)⁶ in the same metropolitan region, as well as Capitol Lake (~ 736 µatm)⁹ and University Lake (~ 630 µatm)⁴⁸ in Louisiana, USA. However, the observed pCO₂ levels were significantly elevated compared to our earlier findings in the same lake from January to September 2020 (~ 707 µatm)⁴. Importantly, pCO₂ in October 2021 reached an exceptionally high value of 8148 ± 1882 µatm (n = 4; Fig. 3A), a period not captured in our prior study in 2020⁴. However, a prior two-year comparative study of CO₂ fluxes across different habitats in Lake Võrtsjärv also revealed pronounced spatial, seasonal, and interannual variability⁷². This anomaly underscores the potential for pronounced seasonal variability in water-air interface pCO₂ dynamics, likely influenced by temporal environmental drivers^48,55,59.

Previous work by Wang et al., investigating 43 eutrophic lakes across China’s climatic zones⁵⁶, revealed pronounced seasonal variability in pCO₂ across all studied systems, with lower mean values in summer and autumn, a pattern consistent with most lacustrine studies^11,12. This seasonality likely stems from synergistic effects of increased phytoplankton and submerged macrophyte biomass, coupled with thermal stratification, pH dynamics, solar radiation, and anthropogenic activities^4,9,73,74. Specifically, pCO₂ fluctuations in aquatic systems are governed by the balance between biological production (P; positive correlation with Chla in Table 4) and respiration (R; negative correlation with DO in Table 3)^21,75. Thus, we reason that increased primary productivity during warm seasons could facilitate the transition of eutrophic lakes from net CO₂ sources to sinks, whereas microbial and/or photochemical mineralization of organic carbon (e.g., TOC in Fig. 5) during cooler seasons may surpass photosynthetic CO₂ uptake (also known as the biological C-pump-effect)⁷⁶, thereby driving seasonal source-sink shifts^6,77. Specifically, intense photosynthesis during warm seasons significantly consumes dissolved CO₂, lowering pCO₂ below atmospheric equilibrium and leading to CO₂ influx; the lake functions as net autotrophic (i.e., an atmospheric C-sink) when gross primary production (GPP) exceeds the carbon released through ecosystem respiration. Moreovoer, our year-round monitoring (January–November 2021; Fig. 3A) showed lower mean pCO₂ in the first half-year (1239 ± 781 µatm, January–June) compared to the latter period (3143 ± 4674 µatm, July–November), though both phases exceeded atmospheric equilibrium, confirming persistent CO₂ supersaturation. Intriguingly, episodic CO₂ undersaturation occurred in October 2020 (winter for 198 ± 65 µatm) and March 2021 (spring for 315 ± 145 µatm), temporarily converting the system to a net sink (Fig. 3). Collectively, these findings highlight diurnal/seasonal source-sink transitions in subtropical urban eutrophic lakes, modulated by dynamic biogeochemical drivers.

Drivers of CO₂ uptake and release in response to environmental conditions

Further correlation analyses revealed statistically significant negative relationships (p < 0.05) between pCO₂/fCO₂ and both pH (− 0.727^**/−0.681^**) and DO (− 0.311^*), alongside significant positive correlations with Chla (0.295^* and 0.287^*, respectively; Tables 3 and 4, and Figs. 5, S2). These findings underscore that CO₂ uptake/emission dynamics in urban lakes are governed by multifaceted controls from aquatic environmental factors. In this moderately eutrophic autotrophic lake system, we posit that biological drivers, particularly Chla (as P) and DO (as R), play pivotal roles in modulating CO₂ fluxes⁷⁸. The enrichment of nutrients intensifies the coupling between CO₂ fluxes and biogeochemical cycling, as demonstrated by strong correlations with both biological indicators (e.g., Chla) and chemical factors (e.g., TP, TN and pH; Table 2, and Fig. 4), rendering the carbon dynamics more responsive to environmental changes^15,18. Specifically, enhanced organic matter mineralization increases the bioavailability of TP, thereby stimulating CO₂ emissions^14,49,79. The mechanistic linkage aligns with the observed inverse correlation between DO and pCO₂ in our study.

Prior studies have demonstrated that pH regulates the physicochemical environment of lakes by mediating the dynamic equilibrium and spatial distribution of carbonate species (CO₂, CO₃²⁻, and HCO₃⁻), thereby influencing CO₂ fluxes (quantified as pCO₂ and fCO₂) at the water-air interface^4,80,81. This chemical control is particularly pronounced in lakes with elevated pH (> 8), where aqueous CO₂ concentrations exhibit marked sensitivity to alkaline conditions¹². Mechanistically, higher pH promotes the conversion of free aqueous CO₂ into carbonate ions, reducing pCO₂ and creating undersaturation that enhances atmospheric CO₂ absorption. Conversely, lower pH destabilizes dissolved inorganic carbon species, driving CO₂ efflux to the atmosphere⁸². In our study, pH values ranged from 6.5 to 9.0, displaying distinct seasonal variability: maximal fluctuations occurred in autumn-winter (August–March), while minimal variability was observed in spring-summer (April–July; Fig. 5D). This pattern may reflect temperature-mediated modulation of hydrogen ion activity⁸³, which is related to pH, even in the absence of statistical significance (R = − 0.095 for T_water/pH in Table 3), a finding consistent with our earlier observations⁴. Notably, CO₂ flux exhibits strong pH dependence across diverse lacustrine systems. For instance, global analyses of 196 saline lakes revealed that lakes with pH ≥ 9 typically function as weak CO₂ sinks⁸⁴, while a 14-year study of six hardwater lakes in Canada’s Northern Great Plains identified a critical pH threshold of 8.6 for source-to-sink transitions⁸⁵. In our dataset, seasonal shifts between C-source-sink suggest the existence of a comparable pH threshold governing flux reversals, though its precise value requires further investigation.

Chla, a principal photosynthetic pigment in algae and phytoplankton, serves as a critical proxy for freshwater lake productivity⁷⁹. Empirical studies have demonstrated that correlations between Chla concentrations and greenhouse gas emissions reflect the metabolic equilibrium of lacustrine ecosystems^15,86,87, particularly in urban lakes influenced by industrial and domestic wastewater discharges⁸⁸. In our investigation of a moderately eutrophic lake during non-bloom conditions, significant positive correlations were observed between Chla and both pCO₂ (R = 0.295^*) and fCO₂ (R = 0.287^*; Table 4), aligning with our previous findings from Beihu Lake in the same urban system⁶. However, earlier studies on the lake failed to detect statistically significant associations (pCO₂/Chla with R = 0.202; fCO₂/Chla with R = 0.213)⁶. Such inconsistent patterns have been documented in other lacustrine systems. Interestingly, Xu and Xu reported substantial spatiotemporal variability in Chla concentrations in University Lake⁸⁹, a small urban waterbody in the southern United States, yet subsequent CO₂ investigations at the same site revealed no significant Chla-CO₂ relationships⁴⁸. Recent analyses of 33 small artificial lakes by Wang et al.⁶⁴ further demonstrate this complexity: spring algal blooms in over half of these systems resulted in annual mean Chla concentrations exceeding 10 µg L⁻¹, coinciding with enhanced CO₂ sequestration. These paradoxical findings likely stem from multiple interacting mechanisms, including vertical and lateral chemical transport dynamics in open aquatic systems, phytoplankton community composition and density variations, and methodological challenges in modeling Chla-CO₂ interactions^6,86. Among those, solar radiation exerts a critical regulatory influence on Chla concentrations in urban lacustrine systems through its modulation of phytoplankton photosynthetic activity^89,90. Diurnally, photosynthetic efficiency follows a parabolic trajectory: morning radiation intensification stimulates photosynthetic activation, reaching maximum capacity at solar noon before subsequent photoinhibition reduces both photosynthetic performance and Chla levels through late afternoon (ref., Fig. 2E)^91,92. Seasonally, spring radiation intensification creates optimal photothermal conditions for phytoplankton proliferation, enhancing lacustrine carbon sequestration potential. Conversely, summer hyper-radiation events coupled with elevated water temperatures may induce thermal stress responses, potentially triggering CO₂ re-release mechanisms that could transition these aquatic systems from C-sink to -source^58,93.

DO, serving as a critical indicator of aquatic metabolic activity, is regulated by multiple environmental drivers including water temperature, organic matter, biological photosynthesis and water dynamics. Its dynamic equilibrium reflects compensatory atmospheric exchange and respiratory consumption processes^91,94,95. In moderately eutrophic lakes (Chla < 30 µg L⁻¹), algal growth exhibits relative moderation compared to hyper-eutrophic systems, where biogeochemical processes (particularly N–P interactions involving NO₃⁻ and NH₄⁺) dominate over Chla in regulating fCO₂ dynamics^15,96. This regulatory dominance is particularly pronounced in urban lacustrine systems along anthropogenic disturbance gradients, where significant linear correlations between nutrient concentrations and fCO₂ variations have been documented^12,97. Empirical evidence from Taihu Lake demonstrates that anthropogenic nutrient loading elevates CO₂ emissions through enhanced aquatic respiration⁴⁰. While theoretical models suggest nutrient enrichment could reduce CO₂ emissions via boosted primary production, most field observations indicate that N and P inputs predominantly amplify CO₂ release through stimulation of heterotrophic respiration in both water column and sediments^17,35,42,98. Our findings reveal a significant negative correlation between DO and pCO₂ (p < 0.05; Table 3), indicating that, in addition to biological factors (where elevated productivity reduces CO₂ efflux, leading to higher pH, increased oxygen, and lower pCO₂), anthropogenically mediated heterotrophic respiration may also serve as a driver of dissolved CO₂ supersaturation^68,99. Further, Zhang et al. propose that DO (or apparent oxygen utilization, AOU) may serve as a more direct predictor of CO₂ variability than Chla in moderately eutrophic lakes during non-bloom periods, demonstrating greater independent explanatory power for fCO₂ fluctuations at the water-air interface¹⁵.

Moreover, in shallow lakes where hydrodynamics are the dominant force, the movement and mixing of water constitute the core physical mechanism regulating the oxygen (O₂) budget⁶. The continuous inflow of riverine water imparts kinetic energy, and the resulting turbulence significantly enhances gas exchange efficiency at the water-air interface, thereby promoting O₂ input⁹. Concurrently, such hydraulic disturbance effectively disrupts thermal stratification, leading the lake toward a fully mixed state⁵⁰. This process transports O₂-rich surface waters to the lake bottom, preventing the formation of hypoxic conditions in the benthic zone. Further, hydrodynamics govern a relatively short hydraulic retention time, which not only exports partially decomposed organic matter^18,27, reducing the internal O₂ demand, but also continually replenishes nutrients to support moderate levels of photosynthetic O₂ production. Therefore, through these three synergistic mechanisms, enhancing reaeration, optimizing vertical distribution, and reducing net consumption, intense hydrodynamic processes positively maintain the high-O₂ equilibrium in shallow lakes.

Uncertainties in CO₂ evasion from subtropical lake systems and future work

Our synthesis demonstrates that CO₂ exchange rates in anthropogenically-impacted subtropical lakes vary considerably (− 15 to 130 mmol m⁻² h⁻¹; Table 5), reflecting strong geographic and temporal dependencies in emission patterns. The mean fCO₂ in our study markedly exceeded values reported for urban-dominated lakes in other regions, even surpassing previous findings by Yang et al.⁴ for the same lake system (29 ± 67 mmol m⁻² h⁻¹). However, our results align with the previous observations in Qinglonghu Lake (a moderately eutrophic urban lake in the same region; 108 ± 101 mmol m⁻² h⁻¹)¹⁰, likely attributable to intensified anthropogenic pollution inputs and aggravated eutrophication in our study lakes during the monitoring period. Notably, fCO₂ values from the highly urbanized tropical Rio Grande Reservoir in South America⁹⁵, i.e., 5.14 and 3.18 mmol m⁻² h⁻¹ for hypereutrophic and moderately eutrophic zones, respectively, were substantially lower than our measurements for the moderately eutrophic in this lake. These discrepancies may reflect not only differences in eutrophication status and nutrient levels but also hydrological seasonality, as exemplified by the previous work on subtropical University Lake in Louisiana, USA⁴⁸. While local anthropogenic forces are often the dominant driver of CO₂ dynamics in human-modified lakes, we hypothesize that a portion of the residual uncertainties in regional-scale CO₂ emission estimates for subtropical lakes is associated with patterns of regional climate change^15,31. Specifically, monsoon-driven climatic patterns characterized by high temperatures and/or heavy rainfall amplify interannual fluctuations in hydrological conditions (e.g., rainfall-evaporation balance), thereby modulating the dissolution-release equilibrium of CO₂.

Previously, Zhang et al.¹⁵ demonstrated that most lakes across diverse geographical regions including Taihu Lake, Lake Guadalcacín, Lake Bornos, Lake Alexandrina, and urban artificial ponds function as atmospheric CO₂ sources, with annual mean fCO₂ increasing alongside Chla concentrations (Table 5). These comparative findings further emphasize the variability of CO₂ dynamics across interannual, seasonal, and hour scales^6,58,100. For instance, seasonal analyses revealed a C-source-sink alternation pattern in subtropical urban lakes, where winter and spring (periods of low algal biomass) exhibit CO₂ production and release, while elevated summer primary productivity facilitates a transition to CO₂ sinks¹⁸. Contrary to this typical seasonal pattern, our study observed no strict as the summer-sink and winter/spring-source dynamic. Despite two months of pCO₂ levels significantly below atmospheric equilibrium (Fig. 3), the annual mean pCO₂ (1789 µatm) remained markedly supersaturated, indicating persistent CO₂ emissions. Further, discrepancies emerged when comparing our with the previous findings of Yang et al.⁴ for the same lake system (707 µatm), suggesting that such divergence in seasonal patterns contributes to uncertainties in urban lake CO₂ emission assessments. Similarly, Potter and Xu⁵⁸ highlighted that while summer predominantly manifests as a C-source and winter as a sink, spring sampling may prove valuable for assessing CO₂ evasion dynamics in shallow trophic lakes. As a hypothesis, early-spring likely represents a critical transitional window, our future studies accordingly will prioritize through systematic monitoring of this period.

Prior studies has established that while long-term pCO₂ variations (days to months) significantly influence evasion rate estimates, gas transfer velocity (k₆₀₀) emerges as a critical regulator of diurnal CO₂ evasion at shorter timescales (minutes to hours)^100,101. Building on this understanding, both previous studies and our work observed distinct diurnal water-air CO₂ gradients from dawn to dusk, yet collectively underestimated pCO₂ owing to insufficient consideration of nocturnal dynamics. Through 24-h monitoring of subtropical urban lakes, previous study revealed pronounced diurnal fluctuations in pCO₂ and CO₂ degassing, particularly marked by nocturnal pCO₂ surges⁵⁸. Their findings identified 10:00 and 22:00 CST as periods of minimal deviation from daily mean pCO₂, recommending optimized sampling between 09:00–11:00 CST to balance accuracy and operational feasibility. Our study validated the efficacy of sensor-based continuous monitoring in lacustrine systems. Similarly, Wang et al.⁵⁵ demonstrated that daytime CO₂ fluxes in Tangxun Lake (ca. 8 mmol m⁻² day⁻¹) were remarkedly lower than nocturnal fluxes (ca. 10 mmol m⁻² day⁻¹), with 11:00–12:00 measurements best approximating daily means. It could therefore be inferred that during nighttime hours, when photosynthetic activity is minimal, enhanced CO₂ evasion is likely to occur. Nevertheless, only a limited number of studies have focused on nocturnal CO₂ release dynamics. For instance, analysis of long-term limnological data (1987–2006) from Lake Apopka, Florida, by Gu et al.¹⁰² revealed consistently higher average partial pressure of CO₂ (224 µatm) at nighttime compared to daytime levels. Similarly, Reis and Barbosa¹⁰³ reported significantly elevated mean nocturnal pCO₂ (565 µatm) relative to daytime values (436 µatm) in a tropical productive lake in southeastern Brazil. More recently, Reiman and Xu¹⁰⁴ further corroborated a consistent daytime pattern of pCO₂ in the Lower Mississippi River, characterized by a peak prior to sunset and a minimum during periods of maximum solar irradiance. These collective findings underscore the critical role of temporal resolution in evasion rate quantification.

Methodologically constrained by funding limitations, manpower shortages, and site accessibility challenges, our study lacked nocturnal monitoring and employed a limited sample size. While these findings provide preliminary insights into CO₂ dynamics at water-air interfaces, representing an initial exploratory step. For example, there are still the following urgent challenges that need to be addressed in our current work:

• (i) This study conducted high-frequency monitoring at a representative site located 2 m from the lake’s shoreline. This location was selected for its capacity to reflect the dominant air-water exchange processes in the lake’s open waters while avoiding known, strong point-source disturbances. Although spatial variability of CO₂ concentrations in the well-mixed central basin of this moderately sized urban lake is likely limited in the absence of major perturbations, we acknowledge that a single sampling point cannot fully capture the potential spatial heterogeneity of the entire lake. Further, its proximity to the shore may not adequately represent biogeochemical processes in the pelagic zone. For instance, nearshore areas susceptible to groundwater inflow or sediment resuspension may develop localized pCO₂ hotspots, which were not directly monitored in this study design. Additionally, the presence of submerged aquatic vegetation in the sediments at the sampling site can influence dissolved CO₂ concentrations and their diel fluctuations. The spatial heterogeneity of aquatic vegetation introduces uncertainty when extrapolating discrete point measurements to whole-ecosystem fCO₂. While the current sampling strategy adheres to standard protocols, it may not fully represent the metabolic diversity across different habitats. In essence, the net CO₂ flux in vegetated areas represents a dynamic balance between photosynthetic uptake and respiration, which varies nonlinearly with environmental conditions. Moreover, the inhibitory effect of vegetation canopies on the gas transfer velocity (k) is often overlooked in flux calculations, potentially leading to systematic overestimation in these areas. Consequently, the CO₂ values reported herein should be interpreted as the best available estimate of the dominant fluxes in the lake’s nearshore zone, rather than an absolute and precise whole-lake average. Future investigation should aim to better quantify this spatial variability and its impact on the integrated lake carbon flux budget by deploying more extensive sensor networks and integrating high-resolution pCO₂ mapping with habitat-specific parameterization of k.

• (ii) As noted previously, safety and technical constraints prevented our high-frequency monitoring from covering the nocturnal period. This may introduce uncertainty in our estimates of diel CO₂ fluxes, particularly in quantifying nighttime CO₂ emissions dominated by ecosystem respiration, potentially leading to an underestimation of total daily CO₂ emissions. To assess this uncertainty and provide reasonable flux estimates to the extent possible, we propose that the daily mean flux can be extrapolated based on a conservative estimate of nighttime flux, for instance, by assuming that nighttime fluxes are similar to the lower flux levels observed around sunset. An analysis of the potential systematic bias introduced by the absence of nighttime data indicates that, even in seasons when the lake consistently acts as a CO₂ source, and under a worst-case scenario where nighttime fluxes reach daytime peak levels, the core conclusion, that the lake may shift from a CO₂ source to a sink in certain seasons (e.g., during summer algal blooms), remains valid. This is because the strong daytime photosynthetic uptake is sufficient to offset the estimated upper bound of respiratory emissions at night. Therefore, we emphasize that the key finding of this study, the observed source-sink transition dynamics of the lake, is primarily driven by strong daytime biogeochemical processes (i.e., photosynthesis vs. respiration/chemical equilibria). Although the lack of direct nighttime observations introduces a degree of uncertainty, it does not undermine our understanding of the principal mechanisms driving these dynamics. Future investigations should place greater emphasis on investigating CO₂ dynamics during the nighttime.

• (iii) The pCO₂ in this study was calculated from pH, temperature, and alkalinity using thermodynamic equilibrium equations. This approach may introduce significant deviations under conditions of dissolved organic carbon (DOC) or extreme pH, potentially leading to overestimation of pCO₂. The combination of in-situ direct measurements and multi-method comparisons is necessary, mainly to reduce uncertainties and enhance data reliability. Moreover, the constraining the precise contribution of these allochthonous inputs to the lake’s CO₂ emissions entails considerable uncertainties. The pulsed nature of carbon delivery during rainfall events is poorly captured by our monthly sampling, likely leading to an underestimation of episodic inputs. Further, the heterogeneous composition and bioavailability of imported carbon (e.g., labile DOC from sewage vs. refractory DOC from soil erosion) make it difficult to predict its mineralization efficiency and thus its ultimate contribution to CO₂ evasion. Disentangling the effects of external carbon from in-lake processes remains a major challenge, as these drivers are often coupled (e.g., nutrient inputs stimulating productivity that consumes CO₂).

Overall, future work require standardized protocols incorporating temporal (diurnal/seasonal) and spatial (cross-regional urban lake selections) dimensions to reduce uncertainties in CO₂ flux estimates for urban lacustrine systems. Given the critical role of eutrophication in modulating C-source-sink transitions in urbanized lakes, we propose implementing dual-objective environmental management strategies, such as pollution mitigation (intercepting pollutant inputs through watershed management), and C-conscious restoration including optimizing hydrophyte communities through, C-sequestration species selection, biodiversity enhancement, and spatial configuration optimization. These measures should be prioritized in subtropical regions prone to algal blooms and climatic warming, aiming to simultaneously improve water quality and align lacustrine carbon budgets with global carbon neutrality targets.

Table 5 The global comparison of CO₂ fluxes heterogeneity in subtropical lakes.

Conclusions

This work employed a high-frequency observational program to examine temporal and spatial patterns of pCO₂ and fCO₂ in relation to key environmental variables within a subtropical urban lake with moderate eutrophic status. Temporal analysis identified consistent afternoon reductions in aquatic CO₂ parameters during daytime (07:00–18:00 CST), contrasting with anomalous measurements recorded on three specific dates. However, no statistically significant differences in pCO₂/fCO₂ levels were detected between morning and afternoon (p > 0.05, n = 24). During the investigation, the average pCO₂ and fCO₂ levels were measured at 1788.61 µatm and 130.26 mmol m⁻² h⁻¹ (n = 48), respectively. Monthly analyses revealed substantial variability in both parameters, with pCO₂ ranging from 198.14 to 8148.02 µatm, and fCO₂ from − 16.14 to 775.16 mmol m⁻² h⁻¹ (n = 4, monthly). The annual cycle (January–November 2021) showed significantly lower mean pCO₂ during the first half-year (1239.34 µatm, January–June) compared to the latter period (3143.15 µatm, July–November), both exceeding atmospheric equilibrium to function as net CO₂ source. Crucially, episodic CO₂ undersaturation occurred in October 2020 (198.14 µatm) and March 2021 (314.92 µatm), temporarily converting the system to a carbon sink. Statistical analyses identified pH, Chla (P), and DO (R) as key environmental drivers of pCO₂ and CO₂ flux variability, while eutrophication status and anthropogenic disturbances critically modulated source-sink transitions. These findings highlight the urgent need for improved management strategies in urban lake systems, such as reducing pollutants and mitigating carbon emissions, supported by standardized protocols that account for temporal (especially nocturnal), seasonal, and regional variations. Such integrated approaches will enhance the accuracy of CO₂ flux estimates and contribute to global carbon neutrality goals.

Methods

Site description

The study was conducted at Bailuwan Lake (104° 7′ 40.5″ E, 30° 34′ 56.01″ N), an urban water body situated in the peri‑urban transition zone of Chengdu, Sichuan, China (Fig. S4). This artificially established aquatic system functions as an integrated urban eco‑wetland complex under the management of the Jinjiang District Government, combining tourism with ecological conservation. Designated in 2017 as Chengdu’s first National Urban Wetland Park, the lake represents a typical urban water body in a western Chinese megacity and offers a valuable case study for examining common features and challenges, such as ecological functions, environmental pressures, and management practices, of urban lakes globally.

The lake covers a total surface area of 200 ha, with open water bodies accounting for approximately 33.5% of the area and exhibiting depth gradients ranging from 0.5 to 6.5 m. According to our previous study based on 2021 data, vegetation represented the largest land cover type (56.2%) in the study area, followed by bare land, lake surface, and roads (6.8%)¹⁰⁵. Hydrologically, the main inflow is an engineered tributary of the Dongfeng Canal system, while water export occurs mainly through evaporation and controlled discharge via constructed drainage infrastructure.

The study area is situated within a humid subtropical monsoon climate zone, characterized by pronounced seasonal thermal variability. Meteorological records indicate a mean annual temperature of 16.5 °C, with a distinct seasonal pattern featuring the lowest monthly temperatures in January and peak values during July–August, aligning with the broader regional climate regime. Throughout the monitoring campaign, diurnal air temperature fluctuations in the lake vicinity were substantial, varying from 0 °C to 35 °C (Fig. S3A), reflecting strong day-night thermal dynamics. In addition, the region experiences considerable solar exposure, with an annual cumulative solar radiation measuring approximately 161 kJ cm^− 2 (Fig. S3B). This high level of irradiance plays a critical role in driving both hydrological and ecological processes, underscoring the distinctive energy budget setting of this subtropical lacustrine environment.

Field measurements

This investigation employed a monthly field monitoring protocol to collect essential hydrological parameters from October 2020 through November 2021, with the exception of November 2020 and September 2021 due to logistical constraints. Additionally, the scheduled July 2021 field campaign was administratively rescheduled to August 2, 2021. In other words, a total of 12 in-situ measurements were collected over 13 months of investigation. To ensure methodological rigor and data reliability, a triplicate sampling approach (n = 3) was systematically implemented for each field collection event, with samples subsequently subjected to both in-situ measurements and comprehensive laboratory analyses.

To maintain rigorous data quality standards and ensure cross-comparability, standardized sampling was performed at four fixed time points daily (07:00, 11:00, 14:00, and 18:00 CST) using a plastic grab-sampler at a depth of 30–50 cm below the water surface. All trips were made on sunny days to minimize rainfall and/or stormwater runoff effects on water conditions. Moreover, one objective of this study was to capture CO₂ dynamics in the near-shore shallow water area, a typical ecotone, whose distinct characteristics are often homogenized and overlooked in whole-lake scale studies. Therefore, the sampling site was selected 2 m (Fig. S4) from the shore primarily because this location represents a sensitive zone for CO₂ exchange at the water-air interface and is also significantly influenced by anthropogenic activities (e.g., surface runoff and input from riparian vegetation). Further, our investigation revealed the presence of submerged aquatic vegetation, such as Ceratophyllum demersum and Potamogeton distinctus, distributed on the lakebed directly below the sampling point.

Water transparency was assessed through Secchi disk measurements (TPC; cm), employing a standardized 20-cm diameter disk, with concurrent turbidity determinations (FNU; NTU) performed using a calibrated HACH-TSS turbidimeter (Danaher Corporation, Washington, DC, USA). Concurrently, a Hanna-HI9829/HI98186 multiparameter probe (Hanna Instruments, Italy) was deployed for synchronous in situ measurements of fundamental water characteristics: pH, DO (mg L⁻¹), EC (µS cm⁻¹), and T_water (°C). The quantification of pCO₂ required precise determination of carbonate (CO₃²⁻; CB; mol mL⁻¹) and bicarbonate (HCO₃⁻; BCB; mol mL⁻¹) concentrations via acid-base titration, utilizing phenolphthalein and methyl orange as dual-endpoint indicators in accordance with the standardized analytical procedure (Method S2).

Additionally, during the dynamic monitoring phase, continuous in-situ measurements of water quality parameters were conducted only after the readings had stabilized and remained consistent for an additional 5 min. Moreover, the stabilization process typically required 15–30 min. For water samples intended for TOC and anion analysis, filtration through a 0.45 μm micropore membrane filter was performed prior to storage in pre-cleaned polyethylene bottles. All samples were stored in high-density polyethylene bottles that had been pre-acid washed, tightly sealed, and carefully inspected to prevent gas exchange. During transportation, samples were placed in coolers with sufficient wet-ice to maintain preservation conditions.

Furthermore, comprehensive laboratory analyses were conducted on water samples to quantify multiple physicochemical parameters. The measured parameters encompassed: (i) Chla (mg m⁻³) quantified per the National Environmental Protection Agency (NEPA) standards¹⁰⁶, (ii) nitrate (NT; mg L⁻¹) analyzed following Method S3, and (iii) TP (mg L⁻¹) assessed via Method S4. Additionally, TOC was fractionated into total carbon (TC) and inorganic carbon (IC) components (mg L⁻¹) using validated methodologies^4,5. Additional parameters assessed were TDN (mg L^{− 1}; Method S5), COD_Mn (mg L^{− 1}; Method S6). Anionic species (F⁻, Cl⁻, SO₄²⁻, NO₃⁻; mg L^{− 1}) were quantified following previously validated methodologies^4,5,6. The trophic status of water bodies was evaluated through the Carlson’s TSI (trophic state index; Method S1).

Besides, given its location within an urban setting, the concentrations and distribution patterns of metals in Baihuwan lake could serve as sensitive tracers of anthropogenic disturbances, such as discharges from urban wastewater and surface runoff. These indicators assist in evaluating the potential influence of allochthonous carbon inputs on the overall carbon balance of the system⁸⁷. For instance, the speciation and solubility of metals such as Fe and Mn are strongly dependent on ambient redox conditions^45,87. Variations in their concentrations and ratios could be used to infer dominant pathways of organic matter degradation (e.g., Fe and Mn reduction processes), which are closely linked to the production and consumption of greenhouse gases including CO₂. Therefore, in this study, we also measured trace metal concentrations (Cu, Zn, Cr, Fe, Mn, K, Ca, Na, Mg; µg L^{− 1}; see Method S7).

To ensure sample integrity, water specimens were collected in amber glass bottles and maintained at 4 °C during transportation. All field sampling procedures were strategically conducted during periods of meteorological stability (clear weather conditions) to eliminate potential confounding effects from precipitation or surface runoff on analytical outcomes.

Laboratory analyses

Calculation of pCO₂

Extensive studies has demonstrated that the equilibrium distribution of aqueous carbonate species, encompassing bicarbonate, carbonate, carbonic acid, and dissolved CO₂, is principally governed by a suite of physicochemical parameters, specifically hydrogen ion activity (pH), aqueous temperature (T_water), and ionic strength (IS) of the solution^4,23,107,108. Building upon this empirical foundation, the present study employs a thermodynamic carbonate speciation model (Eqs. 1–4) to computationally derive the aqueous carbon dioxide partial pressure (pCO₂, expressed in µatm), with rigorous implementation protocols detailed in Method S8.

$$p{\text{CO}}_{{\text{2}}} = \left[ {{\text{H}}_{{\text{2}}} {\text{CO}}_{{\text{3}}} } \right]/K_{{{\text{CO2}}}} \, = \,a\left( {{\text{H}}^{ + } } \right) \cdot a({\text{HCO}}_{{\text{3}}} ^{ - } )/\left( {K_{{{\text{CO2}}}} \cdot K_{{\text{1}}} } \right)$$

(1)

$$\alpha \left( {{\text{H}}^{ + } } \right){\mkern 1mu} = {\mkern 1mu} 10^{{ - [{\text{pH}}]}}$$

(2)

$${{\alpha (HCO}}_{3}^{-} {\text{)}} = {\text{[HCO}}_{3}^{-} {\text{]}} \times \mathop {10}\nolimits^{{-0.5\sqrt {\text{I}} }}$$

(3)

$$\begin{gathered} I\, = \,0.{\text{5}}(\left[ {{\text{K}}^{ + } } \right]\, + \,{\text{4}}\left[ {{\text{Ca}}^{{{\text{2}} + }} } \right]{\text{ }} + {\text{ }}\left[ {{\text{Na}}^{ + } } \right]\, + \,{\text{4}}\left[ {{\text{Mg}}^{{{\text{2}} + }} } \right] \hfill \\ {\text{ }} + {\text{ }}\left[ {{\text{Cl}}^{ - } } \right]\, + \,{\text{4}}\left[ {{\text{SO}}_{{\text{4}}} ^{{{\text{2}} - }} } \right]{\text{ }} + {\text{ }}\left[ {{\text{NO}}_{{\text{3}}} ^{ - } } \right] + [{\text{HCO}}_{{\text{3}}} ^{ - } ])/{\text{1}}000000 \hfill \\ \end{gathered}$$

(4)

where the terms α(H⁺) and α(HCO₃⁻) represent the chemically active fractions of hydrogen [H⁺] and bicarbonate [HCO₃⁻] ions, accounting for non-ideal solution effects, whereas I quantifies the cumulative electrostatic environment through ionic strength.

Furthermore, it should be noted that the concentration of CO₂ in the lake water was calculated from bicarbonate alkalinity, using pH and temperature as the relevant thermodynamic parameters. However, this computational approach could lead to significant overestimation of CO₂ under high DOC conditions (> 200 µmol L^− 1). Therefore, prior to each measurement, pre-screening was conducted to ensure that the hydrochemical conditions of the lake, including pH, alkalinity, and DOC (or TOC in this study) concentration, remained within a “safe range” for reliable estimation.

Estimation of fCO₂

Empirical studies have systematically demonstrated that the interfacial CO₂ flux across the aquatic-atmospheric boundary layer is predominantly regulated by a complex interplay of environmental variables, including thermal conditions (temperature), solute concentration (salinity), atmospheric turbulence (wind speed), and the pCO₂^69,109. In accordance with these established principles, we implemented a mechanistic transfer model (Eq. 2) to quantify the net fCO₂ at the water-air interface, with flux density expressed in standardized units of mmol m^− 2 h^− 1, following the comprehensive methodological framework in Method S9.

$$f {\text{CO}}_{{\text{2}}} \, = \,K_{{\text{T}}} K_{{\text{H}}} \left[ {p{\text{CO}}_{{{\text{2}}({\text{water}})}} \, - \,p{\text{CO}}_{{{\text{2}}({\text{air}})}} } \right]$$

(5)

where fCO₂ quantifies the net exchange rate of CO₂ per unit area at the water-air boundary interface. The parameters K_H and K_T correspond to the temperature-dependent Henry’s law constant (quantifying CO₂ solubility) and the gas transfer velocity (characterizing the water-air exchange coefficient), respectively. The determination of K_H, following established thermodynamic principles, incorporates a multivariate dependence on physicochemical conditions, specifically thermal regime (temperature), ionic strength (salinity), and hydrostatic pressure, as comprehensively characterized in the seminal work of Weiss¹¹⁰.

$$\mathop K\nolimits_{H} = \mathop {\text{e}}\nolimits^{{[\mathop A\nolimits_{1} + \mathop A\nolimits_{2} (100/T) + \mathop A\nolimits_{3} (T/100)]}}$$

(6)

Moreover, the normalized dimensionless Schmidt number (K₆₀₀) was computationally transformed to the CO₂-specific gas transfer velocity (K_T) using the functional relationship expressed in Eq. (7). This transformation incorporates the well-established functional dependence between the Schmidt number (Sc) and gas exchange kinetics, thereby facilitating the precise estimation of K_T across a range of environmental conditions, as originally demonstrated in the foundational work of Jahne et al.¹¹¹

$$\:{\text{}\text{K}}_{\text{T}}\text{=}{\text{K}}_{\text{600}}\times(\frac{\text{600}}{{\text{Sc}}_{\text{CO}\text{2}}}\text{)}{\text{}}^{\text{n}}\text{}$$

(7)

where the exponent of the Schmidt number, denoted as n, exhibits variability contingent upon wind speed conditions. The exponent n adopts a value of 0.5 for wind speeds > 3.7 m s^{− 1}, decreasing to 0.75 for calmer conditions (< 3.7 m s^{− 1}), as established by Guérin et al.¹¹² For the Schmidt number parameterization, we applied the widely accepted value of 0.67 following Cole and Caraco’s experimental determinations under reference conditions¹⁰⁷. Furthermore, the computational algorithm for K₆₀₀, as expressed in Eq. (8), accompanied by its comprehensive methodological elucidation, is presented in Method S9. In this study, U₁₀ denotes wind speed values corrected to the conventional 10-m reference height over the water body at sampling time.

$$K_{{{\text{6}}00}} \, = \,{\text{2}}.0{\text{7}}\, + \,0.{\text{215}}U_{{{\text{1}}0}} ^{{{\text{1}}.{\text{7}}}}$$

(8)

Statistical analysis

In the present study, the statistical framework employed the IBM-SPSS Statistics 22 (IBM Corp., USA) for comprehensive data analysis, adopting a 95% confidence level (α = 0.05) to ensure analytical robustness. High-quality data visualization was achieved using SigmaPlot 14.0 (Systat Software Inc., USA), enabling precise graphical interpretation. This dual-platform approach complies with contemporary standards for quantitative research methodology, guaranteeing both statistical validity and visual clarity.