Substantial increase of organic carbon storage in Chinese lakes

Lakes play central roles as recipients, regulators, reactors, and storages within the global carbon cycle, receiving large terrestrial carbon inputs and mediating burial, emissions, and export. Total organic carbon (OC) stored in lake waters—defined by OC concentration times water volume—influences light absorption and phytoplankton photosynthesis, oxygen dynamics, sediment carbon burial, and CO2 outgassing. Despite this importance, large-scale assessments of temporal changes in lake OC storage have been scarce; most prior work emphasized OC production/mineralization, spatiotemporal changes in OC concentration/composition, or carbon balances that assumed constant OC storage. However, OC stored in lakes varies with phytoplankton growth, hydrologic changes, and sewage inputs. Rapid environmental change has altered lake conditions worldwide, with increasing algal bloom risk and widespread lake volume increases since the early 2000s. These trends indicate that OC storage should vary, yet comprehensive spatiotemporal analyses across diverse lakes are lacking. China, with 24,366 lakes distributed among five limnetic zones that span major contrasts in climate and human influence and have seen both Tibetan Plateau lake expansion and widespread eutrophication in the Eastern Plain, provides an ideal natural laboratory to investigate spatiotemporal changes in lake OC storage in a changing world. This study uses extensive field data and multi-decadal satellite monitoring to retrieve dissolved (DOC) and particulate (POC) OC concentrations, combine them with water volumes, and quantify nationwide OC storage changes for 1984–2023.

Prior studies framed lakes and reservoirs as key carbon cycle components regulating burial and emissions, with global estimates of terrestrial carbon inputs to inland waters, OC burial rates, and CO2 emissions. Research has emphasized OC production and mineralization processes, changes in OC concentration/composition, and carbon balance estimation often assuming constant OC storage. Satellite-based monitoring at individual eutrophic lakes (e.g., Lake Taihu) revealed substantial DOC and POC fluctuations, suggesting storage variability. Recent decades have brought increased frequency of algal blooms globally and widespread increases in lake water volumes, notably on the Tibetan Plateau. However, there has been limited work on large-scale, long-term spatiotemporal variations in OC storage across diverse lakes, motivating the present nationwide analysis in China.

Data acquisition: In-situ measurements from 4,201 stations across 348 lakes (2004–2023) included DOC, DIC, POC, chlorophyll-a (Chl-a), total suspended matter (TSM), pH, temperature, and conductivity. Field protocols used a multi-parameter sonde and GF/F filtration; laboratory analyses applied high-temperature combustion for DOC/DIC/POC, gravimetry for TSM, and spectrophotometry for Chl-a. Public datasets included: daily Landsat-5/7/8/9 surface reflectance (1984–2023) via Google Earth Engine (GEE); daily water storage (1984–2023) from GloLakes; lake polygons/areas/depths and river networks/basins (HydroSHEDS/HydroLAKES, 2015); and monthly basin properties from ECMWF (population density, DEM, evaporation, LAI high/low vegetation, runoff, 2 m air temperature, precipitation, wind speed). Meta-analysis compiled OC accumulation rate (OCAR) data after the 1950s for 115 lakes. DOC retrieval: A multilayer back-propagation neural network (three hidden layers) used nine basin properties to simulate lake-level annual mean DOC. Two Random Forest models (freshwater vs saline, saline/freshwater identified via a weighted decision tree) then retrieved DOC from Landsat visible/NIR reflectances plus longitude/latitude and the simulated DOC. Hyperparameters were tuned via grid search and K-fold cross-validation. Model accuracy (MAPD) was 17.5% (freshwater) and 15.15% (saline) on 30% test data. Daily DOC was derived from cloud-masked Landsat, aggregated to annual means (1984–2023). DOC storage in 2015 used HydroLAKES water volumes; annual DOC storage for 1125 lakes used GloLakes volumes (1984–2023). POC retrieval: A two-step algorithm first produced rough POC estimates via empirical relations with DEM, longitude, red-band reflectance (Rred), and a normalized difference carbon index NDCI=(Rred+Rblue)/(Rred−Rblue). A Random Forest model then refined POC using those rough estimates, longitude/latitude, DEM, and Landsat reflectances. The model achieved MAPD of 16.7% (training) and 16.5% (testing). Daily and annual POC concentrations were derived for 24,366 lakes (1984–2023). POC storage in 2015 used HydroLAKES volumes; annual POC storage for 1125 lakes used GloLakes volumes. CO2 calculation: In-situ pH, temperature, and DIC were used to calculate CO2 concentration via carbonate equilibrium, with temperature-dependent dissociation constants. Statistical analyses: Random Forest analysis quantified relative contributions of anthropogenic and natural drivers to DOC/POC variability. Regression analyses assessed relationships among variables and temporal trends (p<0.05). Lake basins were delineated; saline vs freshwater identified via a weighted decision tree; Landsat reflectance matched to in-situ data. Data/code availability statements provide access to datasets and algorithms used.

• Spatial patterns: DOC higher in northwest saline/arid lakes and lower in the southeast; POC lower in the northwest and higher in the southeast eutrophic regions.
• DOC statistics: Mean DOC in northwest IMXL and TPL was 25.71 ± 24.72 mg L^-1, 2.5× higher than southeast YGPL and EPL (10.29 ± 7.27 mg L^-1). DOC correlated positively with conductivity (r=0.56, N=1636, p<0.01) and with Chl-a in freshwater lakes (r=0.23, N=3624, p<0.0001). From 1984 to 2023, mean DOC increased from 14.79 ± 12.49 to 18.14 ± 21.75 mg L^-1; 65.4% of lake area showed increasing trends. In the southeast NPML/EPL, 72.2% of freshwater lakes increased in DOC; in the TPL, 48.7% of lakes decreased, likely due to increased precipitation.
• POC statistics: Mean POC was 0.83 ± 0.19 mg L^-1 in TPL, vs 4.85 ± 0.65 mg L^-1 (NPML) and 2.87 ± 1.08 mg L^-1 (EPL). TSM explained limited spatial variation in POC (R^2=0.14, N=2026, p<0.0001), while Chl-a explained 71.0% (N=2088). From 1984–2023, 16,007 lakes (80.6% of total area) increased in POC; 93.2% of lake area in IMXL/TPL increased, whereas 59.3% of NPML/EPL area decreased, likely due to pollution control since ~2000.
• 2015 storage: DOC storage totaled 39.43 Tg C (83.2% in TPL); DOC stock scaled with water volume (log–log R^2=0.81, p<0.01). POC storage totaled 2.14 Tg C, with 55.2% in NPML/EPL (only 29.3% of water). TPL held 55.3% of water but 25.6% of POC storage; POC stock scaled with water volume (log–log R^2=0.87).
• Temporal storage trends (1984–2023, 1125 lakes with volume time series): 63.7% increased in DOC storage; 92.0% of the total DOC increase occurred in TPL. Average DOC storage growth rate was 96,295 t C yr^-1 across these lakes. Extrapolated to all 24,366 lakes, DOC storage increased by 13.05 Tg C (+44.6%). For POC, 71.7% of lakes increased; 94.3% (TPL) and 88.9% (YGPL) showed upward trends; declines (28.1% of lakes) were weaker. Estimated total POC storage rose by 33.5% (average growth 4,033 t C yr^-1 across 1125 lakes).
• Drivers: Random Forest results indicate human activities explained 50.2% of spatial DOC variability, while climate factors explained 75.9% of spatial POC variability. Main annual drivers for OC changes were human activities, water input, and wind speed. Population density was the principal driver for POC in 72.3% of lakes (mean contribution 33.1%). Water input factors (precipitation/runoff/evaporation) contributed on average 37.3% to DOC changes in IMXL/TPL. In NPML/EPL/YGPL, wind speed was the main driver of DOC changes in 70.6% of lakes.
• Contribution of concentration vs volume to storage variability (1125 lakes): Changes in DOC concentration explained on average 68.2% of annual DOC storage variability; changes in POC concentration explained 38.4% of annual POC storage variability.
• Carbon burial: OCAR was higher in southeast lakes (22 ± 19.87 g C m^-2 yr^-1) than in northwest (7.78 ± 7.77 g C m^-2 yr^-1). OCAR after the 1950s related exponentially to climatological mean POC (OCAR = 3.06 × e^(0.49 × POC); N=115; R^2=0.35; p<0.01). Estimated total OC burial increased from 1.01 Tg C yr^-1 (1980s) to 1.13 Tg C yr^-1 (2020s), an 11.0% rise.
• CO2 implications: In-situ CO2 correlated with DOC/POC ratio (ln[CO2] = 0.35 × ln[DOC/POC] + 1.35; N=110; r=0.49; p<0.01). From 1984–2023, 67.2% of YGPL/EPL lakes likely had elevated CO2 due to increased DOC, whereas 71.1% of IMXL/TPL lakes likely had decreased CO2, driven by DOC dilution from lake expansion and enhanced algal uptake.

This nationwide, multi-decadal assessment shows that lake OC storage in China has increased substantially due to concurrent rises in OC concentrations and water volumes. The strong spatial differentiation—DOC high in arid, saline northwest lakes and POC high in eutrophic southeast systems—combined with widespread temporal increases, demonstrates that OC storage is dynamic rather than constant. The analysis attributes these changes primarily to intensified human activities (notably population-driven eutrophication elevating POC), enhanced water inputs (glacier melt and precipitation diluting DOC in the northwest while expanding lakes), and strengthened wind mixing (mobilizing DOC in shallow, eutrophic southeastern lakes). Given that changes in DOC concentration accounted for most DOC storage variability, lake biogeochemistry and mixing regimes are pivotal, whereas POC storage co-varied more closely with water volume. These shifts have significant implications: increased POC elevated OC burial rates, and changes in DOC/POC altered CO2 concentrations and likely emissions, with many northwest lakes transitioning toward lower CO2 due to dilution and productivity effects. The observed patterns align with global trends of lake expansion and rising bloom frequency, implying that lakes may increasingly function as carbon sinks through enhanced water-column OC storage and sediment burial, while CO2 emissions vary with eutrophication and hydrologic changes.

Using extensive in-situ datasets and Landsat time series (1984–2023), the study retrieved DOC and POC for 24,366 Chinese lakes, quantified their storage with observed water volumes, and revealed large, spatially structured increases in OC storage: DOC by 44.6% and POC by 33.5%. Human activities, water inputs, and wind speed emerged as dominant drivers of OC concentration changes, with concentration changes especially important for DOC storage variability. Enhanced POC levels contributed to an 11.0% increase in national OC burial, while shifts in DOC/POC altered CO2 dynamics, decreasing CO2 in most northwest lakes. These findings challenge assumptions of constant lake OC storage and underscore lakes’ growing role in carbon sequestration. Future work should leverage global-revisit satellite missions to jointly monitor water-column OC storage, atmospheric CO2 exchange, and sediment carbon burial, improving carbon budget estimates amid ongoing lake expansion and eutrophication.

• Time-series storage trends relied on water volume data available for 1,125 lakes (GloLakes); storage for all 24,366 lakes in 2015 used HydroLAKES volumes, introducing dependence on static volumes for that year.
• Remote retrieval algorithms, while validated, have non-negligible uncertainties (e.g., DOC MAPD ~15–18%; POC MAPD ~16–17%), which propagate into storage estimates.
• CO2 concentrations were calculated from carbonate equilibrium using pH, temperature, and DIC rather than direct in-situ CO2 measurements, potentially adding uncertainty.
• Effects of OC concentration on storage varied widely among lakes, indicating heterogeneity that may limit generalization at fine scales.