Episodic flooding causes sudden deoxygenation shocks in human-dominated rivers

Abstract

Dissolved oxygen (DO) sustains river ecosystems, but the effects of hydrological extremes remain poorly understood. While high river discharge (Q) enhances aeration, floods also deliver oxygen-consuming pollutants, making net impacts uncertain. Here, we analyze daily DO and its percent saturation (DO_%sat), and Q in 1156 Chinese rivers over three years. We show that DO and DO_%sat decrease with rising Q in 69.1% and 55.7% of rivers, respectively. Floods (Q > 95^th percentile) cause abrupt declines in both DO (19.7%) and DO_%sat (16.2%) in 80.1% and 69.4% of the rivers, respectively, with the sharpest declines in agricultural and urban areas. These abrupt deoxygenation events link to increased ammonium and land-use intensity, causing more frequent hypoxia in developed regions. Contrary to initial expectations, floods often reduce oxygen levels, with faster recovery in urbanized regions. As climate change intensifies flooding, such sudden deoxygenation shocks may degrade aquatic ecosystems particularly in human-altered landscapes.

Introduction

Oxygen is fundamental to sustaining aquatic life on Earth, as it supports aerobic biological (respiration) and chemical (redox) reactions, and is a key regulator of biodiversity, biogeochemistry, greenhouse gas emissions, and drinking water quality^1,2,3. However, the slow dissolution of atmospheric oxygen into water often limits freshwater ventilation rates and makes aquatic ecosystems vulnerable to hypoxic events (≤2 mg L^–1), which can suffocate aerobic aquatic organisms and cause mass fish kills. Decades of sustained deoxygenation have been widely reported for lakes, rivers, coastal and offshore marine environments^3,4,5. Although high rates of photosynthesis by phytoplankton in slow-flowing rivers can supersaturate some surface waters with DO⁶, prolonged eutrophication events also deoxygenate deeper and light-limited waters in river systems following the decomposition of phytoplankton biomass⁵. These systems are also often associated with the collapse of algal blooms, leading to the formation of dead zones, with anoxic conditions (≤0.5 mg L^–1) and subsequently favoring the release of phosphorus and metals from bottom sediments^7,8. In particular, high water temperature combined with stagnant or stratified waters reduces dissolved oxygen (DO) concentration and percent saturation (DO_%sat), especially when accompanied by increased chemical and biological oxygen demand⁹.

Future climate variability is expected to have substantial effects on freshwater oxygen content^3,4,5. To date, most studies have attributed long-term declines in river DO and DO_%sat to the effects of increasing water temperature, which reduces oxygen solubility and stimulates microbial respiration^4,10,11. In addition, DO in rivers should also be regulated by changes in discharge (Q), which affects the degree of turbulent mixing and the rate of atmospheric oxygen invasion^7,12. However, floods can also introduce nutrients and oxygen-consuming organic materials from land into rivers^12,13, affecting the rate of DO consumption, the intensity of biological processes, and riverine redox conditions¹⁴. Given the potentially opposing effects of greatly elevated Q, it is difficult to predict the role of hydroclimatic extremes, such as floods, on lotic oxygen conditions ^15,16. Occasional site-specific studies suggest that floods can increase biochemical (BOD) and chemical (COD) oxygen demand, creating localized declines in DO availability and abrupt hypoxic and anoxic events¹⁴. In contrast, little is known about the effects of Q, particularly flooding, on the short-term variability of DO and DO_%sat at the continental scale. When combined with chronic oxygen stress, sudden deoxygenation shocks can alter food webs, cause fish kills^5,12, and increase the likelihood of exceedance of critical planetary boundaries, particularly if geographically widespread⁵.

Here, we combined measurements of mean DO concentration and DO_%sat at 4-h intervals in surface waters (0.5 m) with coeval estimates of daily Q in 1156 rivers over 3 years to identify how flow, especially during floods, can affect oxygen content, induce abrupt oxygen shocks, and shape the oxygen environment in river ecosystems of the Asian subcontinent. While previous global syntheses¹² and single-river studies^17,18 have shown that hydrological extremes can increase oxygen consumption, our sub-continental scale analysis of 1156 rivers provides high spatial and temporal resolution to examine land-use effects, nutrient dynamics, and recovery patterns across diverse river systems. The study sites were categorized into four major basins; (i) Songliao + Hai, (ii) Yellow + Huai, (iii) Yangtze, and (iv) Pearl + Zhemin, and all rivers were further classified by size into small, medium and large rivers as defined by Strahler stream orders 1–3, 4–6, and 7–9, respectively (Fig. 1A–D). As floods may export large amounts of oxygen-consuming constituents (e.g., organic matter) from terrestrial environments^7,12,16, we hypothesized that flood events, especially in agricultural and urbanized catchments, can significantly reduce DO and DO_%sat levels and increase the occurrence of hypoxia and anoxia.

Fig. 1: Nationwide multi-year mean of surface water dissolved oxygen (DO) and its percent saturation (DO_%sat) in Chinese rivers, along with their relationships with river discharge (Q).

A, B Mean DO concentrations and DO_%sat in 1156 rivers from the daily mean measurements in China from November 8, 2020, to December 31, 2023; C, D Pearson correlation coefficients (r) of linear fits between river Q and DO, and between Q and DO_%sat in the 1156 rivers in China. Results with p > 0.05 are shown as white circles. Stacked percentage histograms in (A–D) show different categories of DO and DO_%sat, and Pearson r of linear fits. E, F Boxplots showing different categories of Pearson r between Q and DO and DO_%sat in rivers with different Strahler orders, and G, H the relationships between the abovementioned Pearson r of linear fits and agricultural and urban land use (%). Each box shown in panels E, F represents the interquartile range, the horizontal line is the median, and the whiskers are 1.5× the interquartile range.

Results

Relationship between DO and DO_%sat, river discharge, and oxygen demand

We found the highest DO concentrations (mean ± standard deviation (SD) = 9.8 ± 1.1 mg L^–1; DO_%sat = 93.5 ± 13.2%) in the Songliao + Hai basin in northeastern China. The next highest DO was found in the Yellow + Huai basin (9.1 ± 1.3 mg L^–1; DO_%sat = 94.9 ± 14.1%), followed by the Yangtze River basin (8.6 ± 1.1 mg L^–1; DO_%sat = 95.6 ± 12.5%), and the Pearl + Zhemin basin (7.7 ± 1.0 mg L^–1; DO_%sat = 90.0 ± 15.0%) (Fig. 1E, F). The upstream watersheds of the Yellow + Huai and Yangtze basins, characterized by higher elevations and steeper slopes, generally had higher DO_%sat values (Fig. 1B, F). At 116 river sites, DO and DO_%sat exhibited spatial coherence, with upstream values predicting those obtained downstream, while the remaining 1040 river sites showed no spatial correlation (Supplementary Fig. 1).

Dissolved oxygen concentrations and DO_%sat levels were correlated significantly and negatively with Q at 69.1 and 55.7% of sites, respectively (Fig. 1A–D). On a river-specific basis, a linear fit better explained the variability in DO concentrations and DO_%sat than did a power function or logarithmic regression (Supplementary Fig. 2). The effects of Q on DO were particularly pronounced (Pearson r ~–0.4) for the Songliao + Hai, Yellow + Huai, and Yangtze basins, which are located in the East China Plain and have a high proportion of agricultural and urban land use (Fig. 1C–F and Supplementary Fig. 3). When comparing all rivers, the strength of the correlation (Pearson r) between Q and DO, and between Q and DO_%sat was itself correlated negatively and linearly with agricultural and urban land use within 30 km radius of each site (r² = 0.15, p < 0.001; r² = 0.23, p < 0.001, respectively) (Fig. 1G, H).

Strong inverse relationships emerged between oxygen metrics and oxygen-demanding constituents across the study sites. Both DO concentrations and DO_%sat showed significant negative correlations with ammonium (NH₄⁺-N) at 45.7 and 69.2% of monitoring locations, respectively (Fig. 2A, B). Similarly, oxygen metrics were correlated negatively with measured COD at 61.1% of sites for DO and 41.0% of locations for DO_%sat (Fig. 2C, D). These oxygen-depleting effects increased with the anthropogenic activity and population density (Pearson r ~−0.4), with the greatest values in the Songliao + Hai, Yellow + Huai, and Yangtze basins of the East China Plain (Fig. 2).

Fig. 2: Relationships between oxygen metrics—dissolved oxygen (DO) and its percent saturation (DO_%sat)—and water quality parameters.

A, B Pearson correlation coefficients (r) of linear fits between river DO and ammonium (NH₄⁺-N), and between DO_%sat and NH₄⁺-N in the 1156 rivers in China. White circles show non-significant results (p > 0.05). Inserted are examples of site-specific relationships (colored dots map to locations). C, D Corresponding relationships between river DO and chemical oxygen demand (COD), and between DO_%sat and COD in the 1156 rivers in China. Relationships between DO and COD, and between DO_%sat and COD, with example sites shown.

Variations in DO and DO_%sat during floods

Contrary to the conventional assumption that floods aerate rivers, our analysis revealed that flood events (Q > 95th percentile of daily discharge) triggered acute oxygen depletion in >80% of the monitored Chinese rivers, with agricultural watersheds experiencing the most severe deoxygenation shocks. To assess the impact of flood events on river oxygen levels, we quantified the oxygen depletion during floods by comparing the ratio of DO and DO_%sat between flood and non-flood periods (i.e., DO_flood:DO_non-flood and DO_{%sat flood}:DO_{%sat non-flood}). We found that 80.1% of the rivers, excluding those in the Pearl + Zhemin basins, had a DO ratio <1, while 69.4% of the rivers had a DO_{%sat flood}:DO_{%sat non-flood} <1 (Fig. 3A–D). On average, the DO and DO_%sat ratios for the 1156 rivers, were 0.86 and 0.91, respectively, indicating that flood events reduced DO to 85.9% and DO_%sat to 91.0% of their low flow values (Fig. 3A–D). We further categorized rivers according to whether DO and DO_%sat increased or decreased during floods. For sites with decreasing DO concentration, DO was reduced by 19.7% and DO_%sat by 16.2% from baseline values. In general, the magnitude of decline in DO and DO_%sat during floods was highest in the Songliao + Hai, Yellow + Huai, and the middle and downstream reaches of the Yangtze River basins in the East China Plain (Fig. 3A–D). Thus, floods caused a sudden decrease in DO concentration and DO_%sat in most Chinese rivers, especially in the landscapes with the most agricultural and urban development.

Fig. 3: Dissolved oxygen (DO) and its percent saturation (DO_%sat) variability between the flood and non-flood periods and the occurrence of hypoxia and anoxia.

A, B Ratios of DO and DO_%sat between the flood period (discharge ranked above the 95th percentile threshold) and the non-flood period, i.e., DO_flood:DO_non-flood and DO_{%sat flood}:DO_{%sat non-flood} calculated from nationwide 1156 rivers monitored daily from November 8, 2020, to December 31, 2023. C, D Boxplots showing different categories of DO_flood:DO_non-flood and DO_{%sat flood}:DO_{%sat non-flood} in rivers with different Strahler orders. E–G Locations of sites with at least one occurrence of hypoxia (DO ≤2 mg L⁻¹) or anoxia (DO ≤0.5 mg L⁻¹) during flood events, and these sites were categorized into different rivers with different Strahler orders. Each box shown in panels C, D and F, G represents the interquartile range, the horizontal line is the median and the whiskers are 1.5× the interquartile range. H–K Relationships between DO_flood:DO_non-flood and NH₄⁺-N_flood: NH₄⁺-N_non-flood, and between DO_flood:DO_non-flood and agricultural and urban land use (within 30 km radius of each site); Relationships between DO_{%sat flood}:DO_{%sat non-flood} and NH₄⁺-N_flood: NH₄⁺-N_non-flood, and between DO_{%sat flood}:DO_{%sat non-flood} and agricultural and urban land use. These relationships were categorized for rivers with different Strahler orders.

Increased DO due to changes in river turbulence or water temperature did not compensate for the sudden deoxygenation associated with flood events. To quantify how flood events affect the physical environment of Chinese rivers, we calculated the magnitude of decreases in DO, DO_%sat, and water temperature, as well as increases in turbidity during floods. Specifically, we compared the ratios of DO, DO_%sat, water temperature (WT), and turbidity during floods to those in a pre-flood period 4–9 days before the flood event (i.e., DO_flood:DO_pre-flood, DO_{%sat flood}: DO_{%sat pre-flood}, WT_flood:WT_pre-flood, and Turbidity_flood:Turbidity_pre-flood) (Supplementary Figs. 4A, B, 5A, B). We found that 55.8% of the rivers exhibited DO_flood:DO_pre-flood < 1, and 60.0% had DO_{%sat flood}: DO_{%sat pre-flood} < 1 (indicating decreased DO and DO_%sat) (Supplementary Fig. 5A–B). Correspondingly, 64% of the rivers showed WT_flood:WT_pre-flood < 1 (decreased water temperature), while 83% exhibited Turbidity_flood:Turbidity_pre-flood > 1 (elevated turbidity) during flood events (Supplementary Fig. 4A, B). For sites with decreasing DO and DO_%sat, floods reduced DO by 14.8% and DO_%sat by 14.9% (Supplementary Fig. 5A, B). For sites with decreasing water temperature, floods decreased water temperature by 9%, whereas for sites with increasing turbidity, floods increased turbidity by 516%. We conclude that the combined effects of increased physical mixing and cooler water were insufficient to prevent marked decreases in DO and DO_%sat during intervals of high discharge.

Land use practices within a 30 km radius around each monitoring site had a marked effect on the occurrence of deoxygenation shocks during flood events (Figs. 3,  4). For example, episodic river hypoxia (river n = 493) and anoxia (river n = 309) events during floods were mainly found in agricultural and urbanized rivers of the East China Plain, and in small and medium-sized rivers with Strahler orders of 1–6 (Fig. 3E–G). Sites with hypoxic and anoxic events were 258 and 249% more abundant in regions with intensive land use within a 30 km radius (agricultural and urban land use >30% cover) than those in undisturbed (agricultural and urban land use <5%) landscapes (Fig. 4A, B), although the exact radius had little effect on the relationship (Supplementary Fig. 6). Locally, agricultural and urban land use accounted for up to 90% of local land use at sites that experienced the most frequent hypoxic and anoxic events (Fig. 4A, B). Rivers with ratios of the frequency of hypoxic or anoxic events during flood relative to non-flood periods (i.e., hypoxia_flood:hypoxia_non-flood and anoxia_flood:anoxia_non-flood) exceeded five were mostly found in agricultural and urbanized catchments of the East China Plain (Supplementary Fig. 7).

Fig. 4: Agricultural and urban land use (%) and mean ammonium concentration (NH₄⁺-N_flood) for the rivers that experienced hypoxic (≤2 mg L^–1) and anoxic (≤0.5 mg L^–1) events during floods.

A, B Agricultural and urban land use (in %, within 30 km radius of each site) for the rivers that experienced hypoxic (A) and anoxic (B) events during floods (discharge ranked above the 95th percentile threshold). C, D The mean concentration of NH₄⁺-N_flood (mg L^–1) for the rivers that experienced hypoxic and anoxic events during floods. Similar NH₄⁺-N_flood concentrations were observed between hypoxic and anoxic sites during flood events. The results have been categorized into rivers with different Strahler orders.

Nutrient runoff from terrestrial sources may contribute to sudden deoxygenation of rivers during floods (Figs. 3H–K,  4C, D). During floods, mean ammonium concentrations (NH₄⁺-N_flood) increased by up to >1 mg N L^–1 (Fig. 4C, D). Further, the magnitude of the decline in both DO and DO_%sat was correlated significantly and positively with the ratio of ammonium concentrations during floods relative to non-flood periods (NH₄⁺-N_flood: NH₄⁺-N_non-flood) (Fig. 3H, J). These relationships were also found in relatively large rivers (Strahler orders >7), where the correlations between DO and DO_%sat ratios and NH₄⁺-N_flood:NH₄⁺-N_non-flood ratios were R² = 0.15 and 0.21, respectively (Fig. 3H, J). Additionally, DO_flood:DO_non-flood ratios were correlated negatively with the extent of agricultural and urban land use within a 30 km radius of each site (R² = 0.39, p < 0.001), as were DO_{%sat flood}:DO_{%sat non-flood} ratios (R² = 0.33, p < 0.001) (Fig. 3I, K).

Land-use and nutrient runoff also appeared to affect the rate of recovery of oxygen regimes in rivers from the effects of flooding (Fig. 5). For example, we estimated the rate of the recovery of DO and DO_%sat by calculating the linear regression slopes of DO and DO_%sat at each monitoring site that recorded at 1, 3, 5, 7, and 9 days after the flood event. The rate of recovery was then estimated as the slope of ratios for each site. Analyses reveal that river recovery from oxygen shocks was most pronounced in anthropogenic modified rivers of the East China Plain, with recovery rates reaching up to 0.5 mg DO day^–1 and 8% DO_%sat day^–1 (Fig. 5A, B). The recovery rates of both DO and DO_%sat exhibited a positive correlation with the extent of agricultural and urban land use within a 30 km radius of the rivers (p < 0.001), particularly in medium-sized rivers (Fig. 5C, D).

Fig. 5: Post-flood recovery rates of dissolved oxygen (DO) and its percent saturation (DO_%sat).

A, B The post-flood recovery rates of DO and DO_%sat were calculated as the linear regression slopes of these parameters at each monitoring site, measured at 1, 3, 5, 7, and 9 days following flood events. C, D Relationships between agricultural and urban land use (within a 30 km radius of each site) and the post-flood recovery rates of DO and DO_%sat. These relationships were categorized for rivers with different Strahler orders.

We further compared the values of DO, DO_%sat, water temperature, and turbidity during floods to those observed in a 4–9 day period after the flood event. (Supplementary Figs. 4C, D, 5C, D). Among the 1156 rivers, 61.7% had DO_flood:DO_post-flood <1 (decreased DO), and 62.9% had DO_{%sat flood}: DO_{%sat post-flood} <1 (decreased DO_%sat) (Supplementary Fig. 5C, D). In comparison, 57.6% exhibited WT_flood:WT_post-flood <1 (decreased water temperature), while 85.4% showed Turbidity_flood:Turbidity_post-flood >1 (elevated turbidity) (Supplementary Fig. 4C, D).

Discussion

Climate change is expected to increase the occurrence of flood events in the future⁷, with an increased likelihood of severe floods in monsoonal regions such as China ¹³. While flood-induced deoxygenation has been previously documented^12,17,18, our study systematically identified the continental patterns and causes of sudden DO declines. The strong relationship between the extent of DO shock and the degree of land-use development (farms, cities) and associated nutrient release provides clear guidance to managers seeking to reduce the damaging effects of deoxygenation events on river ecosystems. Specifically, such sudden deoxygenation shocks that arise during floods are likely associated with increased influx of oxygen-consuming particulate and dissolved constituents from land, highlighting the importance of effective land management in maintaining an oxic environment of river networks connected to intensive agricultural regions. Consistent with the importance of land-use practices in regulating changes in lotic oxygen environments during floods, we noted that DO and DO%_sat recovery rates were also greatest in anthropogenically-modified rivers of the East China Plain.

Aquatic DO content is regulated by a complex mixture of physical (atmospheric exchange, mixing, temperature), chemical (redox reactions), and biological (photosynthesis, respiration) processes⁵. Typically, gas exchange rates are much higher in turbulent waters (rivers) than in lentic systems (lakes, wetlands, ponds, etc.) due to, among other factors, greater vertical mixing, reduced boundary layer diffusion lengths, and limited water column stratification^12,19. In our study, we show that across most of China, increases in river discharge, particularly during floods, are associated with rapidly decreasing DO concentrations and DO_%sat, especially in fluvial systems within the developed East China Plain (Figs. 1, 3). Floods may reduce oxygen levels by reducing photosynthesis through scouring of benthic biofilms and plants, and by increasing turbidity, which attenuates light (Supplementary Fig. 4)^16,20. Increased turbidity mainly arises from the influx of eroded terrestrial materials during overland flow and floods^7,12 or resuspension of scoured sediments. In addition, periods of elevated discharge may favor DO consumption by increasing the export of oxygen-consuming substances from the terrestrial landscape, inducing metabolic stress in biota, and exposing chemically reduced buried sediments to the water column^21,22. In particular, biogenic solutes, such as dissolved organic carbon and NH₄⁺-N, are typically exported from the terrestrial environment during high rainfall and flow events and increase with water discharge^13,23,24, although the precise magnitude of export is also governed by differences in land use, water chemistry in shallow soils²⁵, and local sources of oxygen-poor groundwater¹⁵.

Floods may also disturb deep river sediments, thereby inducing strong redox gradients such as observed in the Loire River, France, where winter floods exposed anoxic sediments and enhanced benthic fluxes of NH₄⁺-N, iron, and methane into the water column²⁶. Although our study focuses on summer flood conditions—which may differ in terms of temperature, organic matter availability, and microbial activity from other seasons—the findings from the Loire system highlight the broader influence of hydrological extremes on sediment-water exchanges. Surges in NH₄⁺-N during floods likely fuel nitrification, a process with high BOD or COD¹², while particulate organic matter may further deplete DO via heterotrophic respiration¹⁴. Our quantification of NH₄⁺-N_flood:NH₄⁺-N_non-flood ratios directly links agricultural runoff to hypoxia, a mechanistic pathway that has been underexplored in earlier studies.

Despite cooling water temperatures during floods (Supplementary Fig. 4), DO reductions were significant, suggesting that thermal controls of oxygen content were secondary to the effects of terrestrial oxygen-consuming subsidies such as DOC and NH₄⁺-N. Our empirical analysis reveals a net negative relationship between Q and DO, and between Q and DO_%sat across most sites (Fig. 1C–F), indicating that oxygen-depleting processes (NH₄⁺-N loading, sediment resuspension) often outweigh DO gains from aeration and cooling (Supplementary Fig. 4). The mean (±SD) contribution of flood-induced water temperature changes to the observed decline in DO (DO_before–DO_flood) was estimated at 12.5 ± 16.2% (Supplementary Fig. 8). However, the balance of these positive and negative fluxes varies systematically with river size and land use intensity (Fig. 3).

Sudden deoxygenation events and the occurrence of biologically stressful hypoxic and anoxic conditions were most evident in the agricultural and urbanized catchments of the East China Plain, where high recovery rates of both DO and DO_%sat were also observed (Fig. 5). Flooding mobilizes point sources of pollutants from wastewater outflows, as well as non-point sources of pollutants from croplands and solutes and particulates from the urbanized landscape^7,12,20,27. Increased transport of agricultural fertilizers and sanitary sewer effluents into rivers is likely to increase BOD or COD by favoring oxygen-consuming microbial processes such as nitrification and respiration^12,28. These elevated fluxes will decline markedly after flooding, potentially leading to enhanced recovery rates at sites with intensive land-use change. Further, turbulence-induced mixing in large rivers can expose deep and anoxic sediments to overlying waters richer in oxygen. As well, the latter stages of flooding can exhibit elevated deposition of fine particulate material that may be rich in oxygen-consuming organic matter¹². We infer that this combination of benthic and planktonic processes increases the likelihood of abrupt decreases in DO, as well as the occurrence of hypoxia and anoxia, during flood events (Fig. 3A–G), leading to the risk of sudden deoxygenation shocks or hypoxic blackwater events¹².

Similar interactions between flooding, land use, and lotic deoxygenation may be expected in regions with similar hydrological and land-use characteristics (e.g., North American and European rivers¹⁰); however, further research is required to evaluate how controls of sudden oxygen shocks may vary on a planetary scale. While we studied 1156 rivers, the spatial distribution of sites over-samples lowland basins with intensive agriculture and urbanization (e.g., East China Plain) and may underrepresent mechanisms controlling flood—DO dynamics in steep, forested, or arid rivers where sediment loads ²⁹, groundwater inputs¹⁴, or flow regimes²⁹ differ. For example, mountainous rivers (e.g., upper Yangtze tributaries) may experience higher degrees of re-aeration during or after floods due to turbulence, potentially offsetting DO consumption and minimizing sudden oxygen shocks¹⁵. Similarly, the NH₄⁺-N—DO and COD—DO couplings identified in China’s most heavily-fertilized watersheds may not generalize well to colder sub-humid regions where fertilizer application is more modest, hydrological extremes exhibit strong seasonality related to snowmelt, and variation in groundwater supply affects the sensitivity of rivers to flooding (e.g., central Canada³⁰). However, while future studies should expand monitoring to diverse geomorphic settings to assess the generality of our conclusions, our main finding that intensive land-use development predisposes rivers to flood-induced deoxygenation remains an important insight for landscape management of lotic ecosystems (e.g., China’s Sponge Cities Initiative³¹).

Future climatic regimes may intensify the occurrence or intensity of sudden river deoxygenation on sub-continental scales. Climate projections based on coupled atmosphere-ocean general circulation models indicate that continued greenhouse gas emissions will increase the frequency and intensity of flood events⁷, as well as the monsoon or cyclone systems that markedly influence weather patterns in China¹³. These floods and associated sudden deoxygenation shocks have the potential to reduce biodiversity, constrain fisheries, and even reverse the substantial improvements in freshwater oxygen status made in China through recent decades of improved nutrient management^32,33,34. Effects on biota will depend on the tolerance of individual species to low oxygen tensions, with some fish (e.g., trout, salmon) and invertebrates (e.g., mayflies and stoneflies) native to cold, fast-flowing streams or well-oxygenated environments exhibiting high vulnerability to sudden oxygen depletion³⁵. In contrast, non-native species are typically tolerant of hypoxia and can restructure aquatic communities following invasion, particularly under low DO tensions³⁶. Effects on biodiversity may be particularly profound given that sudden deoxygenation shocks occur over large geographic scales that can reduce rates of recolonization from regional population sources.

Methods

Daily DO and river surface water discharge (Q) in 2020–2023

To investigate how flood events drive the variability of DO in Chinese rivers, we defined flood events as the Q above the 95th percentile threshold of daily observations from November 8, 2020, to December 31, 2023. We obtained DO concentrations every 4 h (Beijing time at 0:00, 4:00, 8:00, 12:00, 16:00, and 20:00 h) from the Ministry of Ecology and Environment of the Republic of China at (https://szzdjc.cnemc.cn:8070/GJZ/Business/Publish/Main.html) from November 8, 2020, to December 31, 2023. Daily DO was derived by averaging the data for the six daily time points. NH₄⁺-N concentrations were measured photometrically, water temperature and turbidity were determined in situ with a probe, and DO was monitored with a DO probe, with all probes calibrated daily. The DO saturation concentration is defined as the maximum amount of oxygen dissolved in the water at a given water temperature, air pressure, and salinity. The daily DO saturation concentration at each site was calculated from the measured water temperature and air pressure using the ‘calc_DO_sat’ function included in the “streamMetabolizer” package in R^10,37. The effect of salinity on DO is negligible in river water and was therefore set to zero as suggested¹⁰. The ratio of the measured DO to the corresponding DO saturation concentration is calculated as DO_%sat. Daily mean Q for Chinese rivers in the corresponding time period was obtained from the National Water and Rainfall Information website at (http://xxfb.mwr.cn/sq_djdh.html). A total of 1156 paired rivers with DO, DO_%sat, and corresponding daily Q were obtained.

Land use within a 5, 10, 20, and 30 km radius of each site

The impact of land use practices on oxygen dynamics was investigated by quantifying agricultural and urban land use (% cover) within a varying radius of each water quality monitoring site. Land-use data were presented as the relative area of summed agricultural and urban land use (% cover) within 5, 10, 20, and 30 km radius of each site. Land use in China, including farmland, forest, grassland, waterbody, city, and bare soil, in 2020 at 500 m resolution, was obtained from Google Earth Engine at https://developers.google.com/earth-engine/datasets/catalog/MODIS_006_MCD12Q1#bands. Analyses of agricultural and urban land use (within 5, 10, 20, and 30 km radius of each site) were performed using ArcGIS 10.2.

Spatial coherence tests for river sites

As a small proportion of river water quality monitoring sites are not located entirely within distinct river catchments, autocorrelation may occur between upstream and downstream sites. To assess the spatial coherence of dissolved oxygen (DO) and DO saturation (DO_%sat) across the sites studied, the percentage contribution of the variability in DO and DO_%sat at each upstream site to its corresponding downstream site was calculated using the following formula:

$${{{{\mathrm{Contribution}}}}}=\left(\frac{{A}_{S1}\times {{{{\rm{Pre}}}}{{{\rm{cip}}}}}_{S1}\times {{{{\rm{DO}}}}}_{S1}({{{\rm{or}}}}\,{{{{\rm{DO}}}}}_{\%{{{\rm{sat}}}}\; {{{\rm{s}}}}1})}{{A}_{S2}\times {{{{\rm{Pre}}}}{{{\rm{cip}}}}}_{S2}\times {{{{\rm{DO}}}}}_{S2}({{{\rm{or}}}}\,{{{{\rm{DO}}}}}_{\%{{{\rm{sat}}}}\; {{{\rm{s}}}}2})}\right)\times 100\%$$

(1)

where A_S1 and A_S2 are the upstream catchment areas for sites S1 and S2, respectively, while Precip_S1 and Precip_S2 represent the multi-year (2021–2023) mean annual precipitation in the upstream catchments. DO_S1 and DO_S2 are the multi-year mean DO concentrations, and DO_{%sat S1} and DO_{%sat S2} are the multi-year mean DO_%sat at these two sites. Two sites were considered unconnected if they belonged to different river systems or had different stream Strahler orders.

Statistical analyses

The relationships between river Q and DO and DO_%sat, the ratio of DO_flood:DO_non-flood, DO_{%sat flood}:DO_{%sat non-flood}, and the occurrence of hypoxia and anoxia were mapped using ArcGIS 10.2. We log-transformed the surface water discharge to obtain a normal distribution of river Q (Supplementary Fig. 2). Linear fits best explained the variability in DO and DO_%sat compared to other functions, so the Pearson correlation coefficient was therefore used to assess the relationship between river Q and DO, and between Q and DO_%sat (Supplementary Fig. 2). Linear regressions between river Q and DO, and between Q and DO_%sat, were performed using the “regress” function built into MATLAB R2021b. DO_flood:DO_non-flood, DO_{%sat flood}:DO_{%sat non-flood}, and NH₄⁺-N_flood:NH₄⁺-N_non-flood values were calculated as the ratio of mean values during the flood periods (Q above the 95th percentile threshold) to those during the non-flood periods over daily observations from November 8, 2020 to December 31, 2023.

To investigate how flood events alter DO, DO_%sat, water temperature, and turbidity, the ratios of DO, DO_%sat, water temperature (WT), and turbidity variability between the flood and pre-flood periods (nine to four days before the flood threshold) periods—i.e., DO_flood:DO_pre-flood, DO_{%sat flood}:DO_{%sat pre-flood}, WT_flood:WT_pre-flood and Turbidity_flood:Turbidity_pre-flood—were calculated from 1156 rivers across the country monitored daily from November 8, 2020, to December 31, 2023 (Supplementary Figs. 4,  5). Nine to four days before the flood threshold was chosen as the pre-flood period to allow a reasonable comparison between the flood and pre-flood periods and to capture the sudden increase or decrease in DO, DO_%sat, water temperature, or turbidity. Results for periods closer to the flood (e.g., 3 days) were not included in the comparison because Q may have been in a rapid growth phase during this part of the event.

To analyze post-flood recovery dynamics, we calculated post-flood recovery rates for DO and DO_%sat as the linear regression slopes of these parameters at each monitoring site, measured at 1, 3, 5, 7, and 9 days post-flood (Fig. 5). These recovery rates were then correlated with agricultural and urban land use (within 30 km radius of each site) to assess land use influences on post-flood DO recovery (Fig. 5). We further calculated the ratios of DO, DO_%sat, water temperature, and turbidity between flood and post-flood periods (four to nine days after the flood threshold)—i.e., DO_flood:DO_post-flood, DO_{%sat flood}:DO_{%sat post-flood}, WT_flood:WT_post-flood, and Turbidity_flood:Turbidity_post-flood—from the same 1156 rivers monitored nationwide (Supplementary Figs. 4, 5). The post-flood period was selected as 4 to 9 days after the flood threshold to avoid lingering flood impacts while minimizing the influence of long-term environmental changes.

To investigate how flood events will change the variability of hypoxic and anoxic events, the ratio of the frequency of hypoxic and anoxic events during flood (discharge above the 95th percentile threshold) to non-flood periods was calculated, i.e., hypoxia_flood:hypoxia_non-flood and anoxia_flood:anoxia_non-flood. Sampling sites that did not experience hypoxic and anoxic events during flood and non-flood periods were not included in the calculation of hypoxia_flood:hypoxia_non-flood and anoxia_flood:anoxia_non-flood. We further quantified the contribution of flood-induced water temperature changes to the observed DO decline. Using both the measured temperature difference (WT_before – WT_flood) and the known water temperature—DO relationship, we computed the water temperature-mediated DO loss. The percentage contribution was then determined as: (temperature-induced DO decline/total observed DO decline [DO_before – DO_flood]) × 100% (Supplementary Fig. 8).

Spatial variability of DO-related variables was mapped using ArcGIS 10.2. Means and standard deviations (SD) were calculated using R x64 4.0.5 software. Pearson correlations, stacked histograms, boxplots, and linear fits were performed using MATLAB R2021b. p < 0.05 was reported as statistically significant for linear fits.