Global lake anoxia is projected to intensify under climate change

Climate change, the paramount challenge of the 21st century, is affecting virtually every ecosystem across the globe. Lakes, essential aquatic ecosystems, offer critical ecosystem services, including the provision of water for human use and habitats favorable to diverse flora and fauna^1,2. Lakes underpin the United Nations’ Sustainable Development Goals (SDGs) by connecting diverse elements of their shared vision for a sustainable, equitable, and prosperous future worldwide^3,4. Moreover, lakes are particularly sensitive to climate change⁵, with climate change-mediated water quality declines, including hypolimnion deoxygenation and potential eutrophication of lakes^6,7, underscored at length in literature at varying geographical scales^8,9. Oxygen (O₂), a vital water quality variable essential in the biological and biogeochemical functioning of lake systems^10,11,12, plays a crucial role in lake management efforts^13,14, and has been severely negatively affected by climate change^8,15. Deoxygenation is especially pronounced in the hypolimnion (deep layers) of stratified lakes, adversely affecting aquatic biodiversity^16,17, promoting increased greenhouse gas emissions^18,19, accelerating internal nutrient loading^11,20, and amplifying the burden and expense of providing safe drinking water^21,22. Thus, with the advent of climate-driven changes, there is a need for a concise understanding of future O₂ dynamics in lakes to ensure water security globally. However, there is a general lack of global modeling studies examining the impact of future climate-driven changes on lake water quality, particularly regarding their O₂ state.

Global studies on climate-driven changes to lakes are dominated by characterizations of lake physics. These include the documented loss of ice cover in lakes²³, the predicted increase in lakes’ evaporation rates²⁴, the anticipated intensification of lake warming driven by heatwaves^25,26, and the expected alterations in lake mixing patterns²⁷. The prolongation of stratification has been identified as one of the main causes of lake ecological degradation, as it adversely affects O₂ concentrations in lake hypolimnia by prolonging depletion of O₂^9,27. These detrimental impacts are further amplified by increasing thermal stability of lakes²⁸, further limiting vertical gas fluxes, influenced by the varying degree of warming in the surface and bottom of lakes^29,30, and the decreased gas solubility emanating from the increase in temperature⁸.

Furthermore, climate change-mediated increases of surface temperatures simultaneously affect lake habitat conditions, conclusively altering and disrupting their ecological communities³¹ and metabolism^32,33,34. This includes established links between lake metabolism and intensified internal nutrient loading³⁵, where climate warming and eutrophication act synergistically to increase productivity^36,37 despite nutrient management efforts^33,38,39. Warmer waters lengthen phytoplankton growth periods and accelerate metabolism^32,37, leading to increasing phytoplankton productivity and subsequent biomass sinking and decomposition, which depletes hypolimnion O₂ during stratification^8,40,41. Increased water column stability that promotes anoxia is linked to reduced fish yields in lakes⁴². Despite advances in understanding the combined effects of climate-mediated changes (increased temperatures and stratification durations) and their interaction with lake O₂ dynamics^15,43,44, existing research focuses on temperate zone case studies⁴⁵ while an assessment of future trends in O₂ concentrations in lake hypolimnia at a global scale is lacking.

In this study, we leverage recent advancements in lake hydrodynamics modeling to project hypolimnetic O₂ concentrations of 73 widely-distributed lakes (Table 1). To accomplish this, we employ an ensemble of three process-based lake models, each driven by outputs from five general climate models (GCMs), to simulate future lake thermal regimes and stratification patterns. We subsequently integrate these predictions into a lake O₂ depletion model⁴³ that incorporates stratification duration and temperature-driven metabolic processes through a generalized and validated framework. Stratification duration is derived from water temperature profiles and corresponding density gradients for each given year. We incorporate the temperature-dependence of metabolic dynamics by applying temperature scaling based on hypolimnetic temperatures, which we also use to calculate initial O₂ concentrations. Empirical data on relationships between O₂ consumption rates and observed trophic states are used to estimate volumetric hypolimnetic O₂ depletion (VHOD) rates. This validated modeling framework enables projections of hypolimnetic O₂ dynamics in each lake until the end of the 21st century.

Impact of climate warming on lake temperature and oxygen dynamics

Climate scenario-dependent changes in hypolimnion temperature, initial O₂ level, volumetric hypolimnetic depletion (VHOD) rate, and stratification duration until 2099 are given in Fig. 1. Consistent warming of hypolimnetic waters was observed across all trophic states (Figs. 1a and S1), with the highest changes projected under pessimistic SSP5-8.5 (+0.14 °C decade⁻¹ in oligotrophic, +0.19 °C decade⁻¹ in mesotrophic, and +0.26 °C decade⁻¹ in eutrophic lakes). Due to their higher-latitude locations, oligotrophic lakes in our dataset experience colder average climate conditions than the lakes with other trophic states, contributing to differences in absolute hypolimnion temperatures and related variables across trophic states. As hypolimnion temperature increases, the initial O₂ available in hypolimnia will decrease due to the temperature-dependent gas solubility. Hence, we predicted a similar pattern of decline in initial O₂ concentrations in lakes of all three trophic states within each scenario (Fig. 1b). The decline over the simulation period was most pronounced in extreme climate scenarios, with SSP5-8.5 projections showing the strongest decreases (−2.49% in oligotrophic, −3.29% in mesotrophic, and −3.98% in eutrophic lakes), followed by SSP3-7.0 projections (−2.04% in oligotrophic, −2.14% in mesotrophic, and −2.67% in eutrophic lakes), and SSP1-2.6 projections (−0. 39% in oligotrophic, −0.57% in mesotrophic, and −0.55% in eutrophic lakes). The pre-industrial control (picontrol) scenario, representing a stable baseline of the natural climate system, highlights anthropogenic changes relative to natural variability and is included in subsequent figures.

Additionally, the projected hypolimnetic temperature increases accelerated O₂ depletion rates in lakes, with the more severe climate scenarios driving the strongest responses (Fig. 1c and see “Methods”). Higher surface-bottom temperature differences simultaneously extended the duration of stratification, thereby prolonging the period during which O₂ depletion occurred. Empirical VHOD measurements used to parameterize O₂ depletion rates (Fig. S2) showed a clear increase with trophic state, with eutrophic lakes exhibiting rates ~2.5 times higher than oligotrophic systems. In addition to productivity effects on VHOD, oligotrophic lakes experienced colder average temperatures (Figs. 1a and S1). Accordingly, we predict minimal increases of VHOD rates for oligotrophic lakes under all three scenarios (+0.14% year⁻¹ under SSP5-8.5), whereas in mesotrophic lakes the increases become steeper the more intense climate change becomes (+0.21% year⁻¹ under SSP5-8.5). This pattern is even stronger in eutrophic lakes, where a substantial and consistent increase is realized throughout the simulation period (+0.32% year⁻¹ under SSP5-8.5). Changes in stratification duration vary among trophic states (Fig. 1d; lake-specific changes in Fig. S3) but follow similar relative trends, increasing under SSP5-8.5 by 0.57 days year⁻¹ in oligotrophic lakes, 0.40 days year⁻¹ in mesotrophic lakes, and 0.32 days year⁻¹ in eutrophic lakes. Oligotrophic lakes exhibit longer stratification duration than eutrophic systems by virtue of the respective morphological features of the included selected lakes (e.g., greater average depth in the oligotrophic lakes), thus reflecting regional climatic features of each individual lake.

Oxygen depletion and rapid progression towards anoxia in lakes

Ensemble simulations revealed a consistent future decline of the time-to-anoxia, defined as the number of days for hypolimnetic O₂ to reach anoxic levels (O₂ ≤ 0.5 mg L⁻¹) after stratification onset, reflecting accelerated depletion and heightened anoxia risk throughout the 21st century (Fig. 2). Under SSP1-2.6, time-to-anoxia decreased by less than 10 days by 2099, indicating relatively stable O₂ conditions compared to other scenarios (Fig. 2a). In contrast, the time-to-anoxia progressively became shorter under both SSP3-7.0 and SSP5-8.5, with accelerated declines during the second half of this century. The most rapid declines occurred under SSP5-8.5, which consistently yielded the shortest time-to-anoxia across nearly all lakes at the end of the century. SSP5-8.5 accelerated anoxia onset by ~30 days, equivalent to a full month, relative to the other climate scenarios (Fig. 2a). Under the strongest warming scenario, stratification duration was projected to increase by ~50 days (Fig. 1d), substantially increasing the risk of anoxia by 2099 relative to present-day conditions. Lakes that currently remain marginally anoxic are particularly vulnerable, with many projected to cross the threshold into persistent anoxia by the end of the century. This trend was also reflected by a systematic increase in the anoxic ratio, defined as the fraction of the stratification period during which the hypolimnion is anoxic, with progressive climate warming (Fig. 2b). Under SSP5-8.5, even more oligotrophic systems exhibited rising anoxic ratios from 2080 onward. In eutrophic lakes, the anoxic ratio increased markedly from ~40 to ~60% by the end of the century. Although oligotrophic lakes reached a time-to-anoxia metric under picontrol and SSP1-2.6 (Fig. 2a), their high initial O₂ concentrations and ~180–210 day stratifications (Fig. 1a, d) prevented substantial anoxia development.

To illustrate these patterns, we highlight nine proxy spotlight lakes representing different trophic conditions and climate contexts (Fig. 3; results for all 73 lakes’ time-to-anoxia and time-to-hypoxia (O₂ ≤ 2 mg L⁻¹) are shown in Figs. S4 and S5, respectively). The projected magnitude and trajectory of changes to time-to-anoxia were strongly modulated by trophic state, latitude, and the baseline thermal characteristics of a given lake. Despite interannual variability and uncertainty being large for several lakes (Figs. S4 and S5), a coherent directional response emerges across all lake systems, and its consistency with warming-induced stratification indicates an increasing risk of hypolimnetic O₂ loss rather than purely stochastic year-to-year fluctuations. Among the oligotrophic spotlight lakes, Kilpisjärvi (Fig. 3a) and Toolik (Fig. 3c) remained highly resilient, with projected time-to-anoxia exceeding 250 days, which far surpassed their future mean projected stratification periods (~78 and ~86 days, respectively), and hence remained oxic. In contrast, despite its oligotrophic state, Tarawera (Fig. 3b) was projected to exhibit anoxia patterns more typical of eutrophic systems, with a loss of over 40 oxic days by the end of the century under both SSP3-7.0 and SSP5-8.5. This stronger response was driven by higher hypolimnetic temperatures, reflecting local climatic conditions, which accelerated VHOD and thereby intensified O₂ declines. Hence, Tarawera was projected to display anoxic ratios ranging between ~40 and 48%, markedly higher than the other two oligotrophic systems, underscoring the potentially overriding influence of thermal regime over trophic state in shaping anoxia trajectories and persistence.

Conversely, the mesotrophic systems displayed intermediate and progressive declines in time-to-anoxia, with Zlutice (Fig. 3d) and Vendyurskoe (Fig. 3f) projected to lose more than ~20 days by the end of the century under SSP5-8.5, while Mt. Bold (Fig. 3e) exhibited a more pronounced decline in the latter half of the century, with time-to-anoxia falling below ~70 days. Among the spotlight mesotrophic lakes, Mt. Bold exhibited the highest anoxic ratios (~67%), ranking among the most anoxia-prone spotlight lake systems under SSP5-8.5. This exceeded the projected anoxic ratio of Zlutice (~31%) by more than twofold and was markedly higher than that of Vendyurskoe (~14%). Furthermore, under SSP5-8.5, Mt. Bold was projected to experience the longest mean stratification duration (~238 days), exceeding that of Zlutice (~191 days) and more than doubling that of Vendyurskoe (~87 days). Under SSP1-2.6 and SSP3-7.0, Mt. Bold and Zlutice were projected to display persistent (~65–67%) and episodic (~26–29%) anoxia, respectively. In contrast, Vendyurskoe exhibited comparatively low projected anoxic ratios (~6% under SSP1-2.6 and ~11% under SSP3-7.0), reinforcing the marked heterogeneity in anoxia responses across lakes of similar trophic states. Projected stratification durations under SSP1-2.6 and SSP3-7.0 were ~231 and ~236 days for Mt. Bold, ~185 and ~192 days in Zlutice, and ~75 and ~82 days at Vendyurskoe, respectively.

Eutrophic systems were predicted to show more pronounced declines in time-to-anoxia and sustained increases in hypolimnetic anoxia prevalence. By the end of the century, Mendota (Fig. 3i) and Hulun (Fig. 3h) were projected to experience a reduction in oxic duration of at least ~10 days under both SSP3-7.0 and SSP5-8.5 during summer stratification. Sau (Fig. 3g) exhibited the most pronounced decline in time-to-anoxia, decreasing from ~95 days at the start of the simulation period to under ~70 days by 2099 under both SSP3-7.0 and SSP5-8.5. Furthermore, Sau is projected to display the highest anoxic ratios among all the spotlight lakes, reaching ~71–74%, depending on the climate scenario. Despite these eutrophic systems being in differing climates with varying morphometries, they converged towards similarly short anoxic onsets and maintained persistent anoxic ratios, exacerbating the detrimental role of high productivity and elevated nutrient levels under warming climates.

The magnitude of warming across climate scenarios was a significant determinant of shifting habitat conditions, with SSP5-8.5 inducing the strongest degradation of O₂ conditions. As a result, an increasing number of lakes at both ends of the trophic gradient (eutrophic or oligotrophic) were predicted to exhibit end-of-summer-stratification O₂ concentrations unsuitable for cold dwelling fish species (<5 mg L⁻¹)⁴⁶, with some lakes potentially becoming hypoxic (<2 mg L⁻¹), or even anoxic (<0.5 mg L⁻¹) (Fig. 4). About 95–100% of eutrophic lakes were predicted to fall below the 5 mg L⁻¹ threshold under all scenarios (Fig. 4a). Under SSP3-7.0 and SSP5-8.5, hypoxic eutrophic lakes were expected to rise to 80–96% (Fig. 4b), and anoxic eutrophic lakes to 75–90% (Fig. 4c). In contrast, oligotrophic lakes showed greater resilience, with 80–94% of them falling below the 5 mg L⁻¹ threshold across all three scenarios (Fig. 4a), with an increasing trend under SSP3-7.0 and SSP5-8.5. Under SSP5-8.5, the prevalence of hypoxia in oligotrophic lakes was projected to rise markedly, affecting 75% of these systems by the end-of-the-century, more than doubling from 32% at the beginning of the simulation period (Fig. 4b). Concurrently, end-of-summer-stratification anoxia increased from 13% of lakes to 57% of oligotrophic lakes over the same timeframe (Fig. 4c). Under SSP3-7.0, hypoxia in oligotrophic lakes increased from 38% at the start of the simulation to 63% by the end (Fig. 4b), while anoxia increased from 13% in 2015 to 45% in 2099 (Fig. 4c). Overall, these results reinforced that being oligotrophic does not confer full protection against hypoxia or anoxia: elevated hypolimnetic temperatures accelerated O₂ depletion even in oligotrophic lakes, although such lakes generally maintain higher O₂ levels than those with higher trophic states.

Translating changes in stratification duration and hypolimnion temperatures into O₂ dynamics allows us to provide lake-specific projections. A lake’s trophic state influenced the future development of its hypolimnion O₂ concentrations, where more productive lakes (mesotrophic and eutrophic) showed a stronger trajectory towards anoxia (Figs. 4 and 5). Compared with the first decade of our simulations (2015–2020), oligotrophic lakes experienced a milder progression in the direction of anoxia for the last decade of our simulations (2091–2099). In most of the oligotrophic lakes, the predicted changes were along the stratification axis, particularly in SSP1-2.6. Whereas under SSP3-7.0 and SSP5-8.5, an increasing number of the oligotrophic lakes were predicted to also shift along the hypolimnion temperature axis, signifying the impact of warming hypolimnia, and thus the trajectory towards anoxia in the last decade of the century (Fig. 5). A majority of the eutrophic lakes were already predicted to be anoxic in the first decade of the simulation, with their anoxic state predicted to worsen at the end-of-the-century. Eutrophic lakes, similar to mesotrophic lakes, showed sharper increases in anoxia due to prolonged stratification and rising hypolimnion temperatures. These effects were depth-dependent, with shallow lakes more impacted by the combination of temperature increases and extended stratification.

Our results quantify the widespread threat that a changing climate poses to lake hypolimnetic O₂ dynamics across trophic states, stratification patterns, and climatic regions. Hypolimnetic O₂ conditions are a fundamental indicator of lake health, with the absence of O₂ threatening biodiversity, aquatic habitats, and ecosystem services^6,9,47. Observational evidence of widespread hypolimnetic deoxygenation and the progression of hypoxia and anoxia, facilitated by prolonged stratification^9,48,49, aligns with our projected O₂ declines. For all 73 lakes, end-of-summer-stratification hypolimnetic O₂ concentrations are projected to decline, with earlier onset of anoxia and increased anoxic extent, intensifying under stronger warming scenarios. Among modeled oligotrophic lakes, at least 15% are projected to be anoxic in the last decade of this century under SSP1-2.6, increasing to at least 38% under SSP5-8.5 (Fig. 5). Comparable increases occur across the other trophic states, with anoxia in mesotrophic lakes increasing from at least 76% to 96%, and in eutrophic systems from at least 86% to 95% under SSP1-2.6 and SSP5-8.5, respectively. These projections are consistent with empirical evidence from approximately 8000 lakes, where anoxia occurrence has increased from ~39 to 61%⁹, and are consistent with documented positive feedbacks that intensify and prolong hypolimnetic deoxygenation⁵⁰, including the effect of stratification prolongation in increasing hypoxia by 0.9–1.7% per decade⁴⁸. Together, these lines of evidence indicate intensifying climate-driven hypolimnetic O₂ declines in both nutrient-rich and historically resilient systems.

In agreement with previous studies^27,30, we project lengthening stratification durations across all lakes through 2099, with severity dependent on the climate scenario (Fig. 1). Prolonged stratification extends the active period of hypolimnetic O₂ depletion⁴⁸, resulting in increased cumulative O₂ loss and expanded anoxic ratios (Fig. 2), and ultimately hypolimnetic O₂ concentrations falling below critical thresholds^8,11,51,52. As vertical mixing is suppressed during stratification, hypolimnetic O₂ cannot be replenished from surface waters.

Increasing hypolimnion temperatures further accelerate deoxygenation, rendering lakes that experience hypolimnetic warming particularly vulnerable. In particular, monomictic lakes exhibit climate-induced hypolimnetic warming that drives temperature-dependent alterations in O₂ dynamics. Temperature regulates internal biogeochemical processes in lakes by raising hypolimnion temperatures and reducing initial O₂ concentrations at stratification onset (Figs. 1 and S1)^53,54. Previous studies support these findings, reporting both observed and predicted increases in lake temperatures globally, suggesting deleterious implications on lake water quality^15,25,34. We project an increased likelihood of lakes falling below ecologically critical O₂ thresholds, with the strongest effects under SSP5-8.5 (Figs. 1 and 4). Hypolimnion temperature emerges as a crucial integrator of climate effects, given the non-linear increase in O₂ depletion with temperature. Warmer lakes, particularly those in tropical regions, are more likely in our projections to experience hypoxic and anoxic conditions (Figs. 3 and 5), even under an oligotrophic state, consistent with their heightened vulnerability to climate-driven warming²⁸.

Despite these consistent large-scale trends, we project substantial variability in hypolimnetic O₂ responses among lakes of the same trophic state. Differences in physical and climatic properties lead to pronounced variability in projected hypolimnetic warming and O₂ depletion, with lakes of the same trophic state exhibiting divergent future O₂ trajectories depending on their climatic context (Figs. 3, S4, and S5). Baseline thermal structure and stratification regimes, influenced by latitude and regional climate⁵⁵, further shape these outcomes, where warmer and lower-latitude systems are more prone to higher hypolimnetic temperatures and accelerated O₂ loss under extended stratification^27,48,50,55. In line with this, we predict wide within-group variability in time-to-anoxia in all trophic states (Figs. 3 and S4), identifying temperature as a dominant control on hypolimnion O₂ dynamics within our modeling framework. We also recognize that additional biological and biogeochemical processes not represented in the model further modulate lake-specific vulnerability, as discussed in the subsequent paragraphs.

Although our model captures the dominant physical and temperature-driven controls on hypolimnetic O₂ dynamics, a comprehensive quantitative assessment of all controlling factors was beyond the scope of our modeling framework, primarily because of data limitations across the full lake data set. Lake morphometry strongly modulates stratification responses^50,56 by controlling both hypolimnetic buffering capacity and the coupling between sediment oxygen demand (SOD) and the overlying hypolimnetic water column. While hypolimnetic volume is explicitly represented and benthic demand is included within the VHOD rates, the model does not resolve key lake-specific morphometric shape or hypsometry. Therefore, variations among lakes in the ratio of hypolimnetic volume to sediment surface area are not captured. Because this ratio regulates the balance between benthic O₂ consumption relative to the hypolimnetic O₂ pool, our model likely understates O₂ depletion in lakes with low hypolimnetic volume per unit sediment area, where a given O₂ demand produces more rapid drawdown⁵⁷. The same morphometric characteristics also modulate climate sensitivity, with small and shallow stratifying lakes shown to respond particularly strongly to prolonged stratification and warming^30,51, due to reduced hypolimnetic O₂ storage and greater sensitivity to hypolimnetic warming. Moreover, explicitly modeling SOD is dependent on other site-specific factors, including organic matter supply⁵⁸, deep-layer photosynthesis⁵⁹, and bioturbation activity⁶⁰, further contributing to uncertainty in hypolimnetic O₂ projections⁴¹.

In addition to these structural limitations, thermal structure and stratification responses are also influenced by optical and regional characteristics that are not explicitly represented in the model. Additional modifiers of thermal structure, such as water color and dissolved organic carbon, further influence lake thermal regimes by enhancing light attenuation and surface heat absorption^61,62, thereby potentially suppressing vertical mixing⁶³ and promoting hypolimnetic deoxygenation⁹. In the underlying physical models, each lake is represented using a lake-specific but static light extinction coefficient that distinguishes clear-water from brown or turbid systems, thereby neglecting both seasonal dynamics and long-term variability in optical conditions. In addition, heat fluxes associated with inflows and outflows are not represented, a limitation widely recognized in multi-lake temperature modeling efforts⁶⁴. Although morphometric effects on thermal structure are explicitly accounted for, these assumptions limit the ability to capture interannual and long-term variability in stratification and warming among lakes. Consequently, lakes exposed to similar climatic forcing may experience divergent thermal trajectories⁶⁵, partly driven by unrepresented changes in optical conditions and hydrological connectivity⁵⁵.

Beyond the physical drivers explicitly represented in our modeling framework, future O₂ dynamics in lakes will also be shaped by interacting biogeochemical and landscape processes. Lake temperature increases may enhance productivity, potentially disrupting ecosystem functioning^36,66,67,68, and worsen O₂ conditions synergistically through increased organic matter export to sediments and accelerated microbial remineralization^41,69,70. These processes are strongly modulated by watershed inputs and hydrologic change. The lengthening of growing seasons and increased metabolic rates, stemming from rising temperatures enhancing lake productivity, can potentially amplify organic matter production and export to hypolimnia, where accelerated microbial respiration intensifies O₂ depletion. Alterations in precipitation regimes⁷¹ are likely to further intensify the effects of increasing temperatures, including more frequent extreme rainfall events⁷² that can increase terrestrial nutrient and organic carbon loading^71,73, and potentially alter lake residence times⁷⁴. Land-use changes within lake catchments^71,75, such as urbanization, agricultural intensification, and wetland loss, can further amplify external nutrient and carbon inputs. Collectively, these factors may shift lakes toward higher trophic states (i.e., from mesotrophic to eutrophic conditions), accelerating O₂ depletion beyond the rates projected by our model. Although our approach is simple, many of these climatic and watershed influences are partly captured by key model inputs: hypolimnion temperature, stratification duration, and trophic state.

In addition to physical and landscape controls, ecosystem structure can substantially modify hypolimnetic O₂ demand. Hypolimnetic O₂ demand and organic matter export efficiency can differ between diatom-dominated systems and more mobile/buoyant algal communities^76,77,78. Diatoms are often dominant in initial spring blooms^79,80 and are characterized by quick sinking to lake hypolimnia and sediment, where they cause an uptick in hypolimnetic O₂ demand early in the stratified season⁸⁰, setting the scene for hypoxic and anoxic conditions⁷⁶. In contrast, buoyant cyanobacteria largely remain in the epilimnion where a significant proportion of their organic matter is decomposed⁸¹, with later-season contributions to hypolimnetic O₂ demand arising from the fraction that sinks⁷⁶. Furthermore, algal community composition and the quantity and quality of organic matter exported to the hypolimnia are shaped by trophic interactions, including zooplankton grazing and fish predation, thereby feeding back on hypolimnetic O₂ consumption⁸². The pronounced within-trophic state variability observed in our simulations is consistent with well-established mechanistic controls on lake O₂ dynamics, and should be interpreted in the context that additional lake-specific physical⁵⁷, optical⁴⁶, and ecological properties⁵⁰, beyond those represented in our model, are also important determinants of divergent responses to climate warming. Future syntheses may integrate this additional complexity in morphometry, optical properties, and ecosystem structure to better identify lakes at greatest risk. But, following the concept of parsimony, we emphasize the value of our comparatively simple model, which captures key controls on O₂ dynamics and enables efficient application at the global scale.

Overall, our results indicate that lakes will face intensifying challenges from climate change, particularly with respect to hypolimnion O₂ depletion. Rising temperatures will intensify the combined impacts of reduced initial O₂ concentrations, prolonged stratification, and enhanced metabolic demand, producing impacts on hypolimnetic O₂ dynamics that increasingly resemble those of eutrophication⁸³. Consequently, lake water quality is likely to deteriorate further under continued warming. Hence, actionable strategies to mitigate the impacts of climate warming are necessary to safeguard lake ecosystems and their associated habitats, with a particular focus on reducing the risks of hypolimnetic hypoxia or anoxia. Short-term strategies, including hypolimnetic aeration, may help sustain high-value systems, such as drinking-water reservoirs or conservation-priority lakes. However, such engineered solutions are costly, constraining widespread application.

Our findings emphasize that mitigating future warming impacts on hypolimnia O₂ conditions could be partially achieved by targeting lake trophic states⁴⁴. This necessitates reducing the human footprint on nutrient pollution, the main driver of cultural eutrophication^35,84. Lake restoration and nutrient load reductions will therefore become increasingly vital, supporting recycling, remediation, and circular management⁸⁵ while also contributing to progress towards relevant SDGs⁴. Under more severe warming outcomes, proportionally greater and more stringent nutrient reductions will be required. Proactive and integrated water-quality management^3,4,86 will thus be essential to safeguard lake ecosystems amid increasing climatic and anthropogenic pressures.

Study sites

Simulations from 73 stratifying lakes across the globe from the “local lakes” category of the Inter-Sectoral Impact Model Inter-comparison Project (ISIMIP) were used in this study. These 73 lakes have water temperature observation data and hypsographic curves, although at varying temporal and vertical resolutions. Using the Carlson’s Trophic State Index, we grouped the lakes by trophic state, based on historical (within 10 years prior to 2015) annual means of either phosphorus, chlorophyll-a, or Secchi depth (26 oligotrophic, 25 mesotrophic, and 22 eutrophic). These lakes also range in mean depths (i.e., 1.7–304.8 m), elevation (i.e., −210 m a.s.l.–4300 m a.s.l.), area (i.e., 0.01–2700 km²), and cover 5 climate gradients (continental, dry, polar, temperate, and tropical) (Table 1 and Fig. 3), with lake climates assigned using the Koppen-Geiger system⁸⁷. Shallow lakes are included in our dataset since they provide critical insight into how stratification duration and intensity influence hypolimnetic O₂ depletion, as they are often overlooked⁸⁸. Despite their brief stratification, these lakes, representative of the majority globally⁸⁹, experience strong warming and consequently heightened O₂ depletion⁹.

Multi-model projections of temperature and stratification data

Lake projections examined in this study consisted of a 15-member ensemble of lake-climate (three lake models driven by five climate models) model simulations. The three 1D physical lake models were applied, using LakeEnsemblR⁹⁰, to simulate the impacts of climate change on lakes in regards to their vertical temperature and stratification patterns⁶⁴. Of these models, two are turbulence-based models; the general ocean turbulence model (GOTM – lake-branch version 5.4.0)⁹¹ and Simstrat (version 2.4.1)^92,93 and the third is an integral energy model, the general lake model (GLM) (version 3.1.0)⁹⁴. The lake models were driven by bias-corrected climate forcing data from the Coupled Model Intercomparison Project (CMIP6), incorporating five global climate models (GFDL-ESM4, IPSL-CM6A-LR, MPI-ESM1-2-HR, MRI-ESM2-0, UKESM1-0-LL) and four experiments⁹⁵. The experiments were designed to simulate the projected evolution of the climate over the period 2015–2100 under different greenhouse gas (GHG) emission scenarios. For our analysis, we used data spanning 2015–2099 for lakes in the Northern Hemisphere and 2015–2100 for lakes in the Southern Hemisphere, thereby aligning with the timing of their respective summer seasons (Fig. 3). These scenarios included the picontrol scenario, representing pre-industrial climate forcing based on 1850 GHG levels, and three shared socioeconomic pathway (SSP) scenarios: SSP126 (SSP1-RCP2.6) for low GHG emissions, SSP370 (SSP3-RCP7) for high GHG emissions, and SSP585 (SSP5-RCP8.5) for very high GHG emissions⁹⁶.

The meteorological forcing data from the GCMs, provided at a daily resolution and used to drive each lake model included near-surface air temperature (°C), near-surface wind speed (m s⁻¹), surface downwelling long-wave and short-wave radiation (W m⁻²), precipitation (kg m⁻² s⁻¹), surface air pressure (Pa), and near-surface relative humidity (%) available at a spatial resolution of 0.5°. Additional lake model inputs included lake-specific temperature observations, bathymetries, and lake characteristics (elevation, latitude, longitude, maximum depth and light extinction). All three lake models were calibrated based on wind scaling (0.25–1.5), short-wave radiation scaling (0.7–1.3), and light extinction coefficient scaling (0.7–1.3, ~30% of lake specific default value), as well as three model specific parameters for each individual lake model to locally-observed water temperature profiles, as described by Feldbauer et al.⁶⁴. In GLM, calibration focused on the mixing efficiencies of hypolimnetic turbulence (0.1–2), convective overturn (0.1–0.3), and unsteady turbulence effects (0.35–0.65). For Simstrat, the calibrated parameters included the fraction of wind energy transferred to Seiche energy (0.0008–0.003), the geothermal heat flux (0.0–0.5, W m⁻²), and the bottom drag coefficient (0.00075–0.00325). In GOTM, calibration targeted the minimum turbulent kinetic energy (1.5 ⋅ 10⁻⁷–1 ⋅ 10⁻⁵, m² s⁻²), physical bottom roughness (500–3000, m), and constant eddy diffusivity (2.5 ⋅ 10⁻⁴–7.5 ⋅ 10⁻⁴, m² s⁻¹).

End-of-summer-stratification O₂ calculations

We applied a hypolimnion O₂ depletion model, devised by Nkwalale et al.⁴³, to project future hypolimnion O₂ concentrations at the end-of-summer-stratification (({{{rm{O}}}}_{2}left({t}_{{end}}right))) for the 73 lakes. The model predicts O₂ depletion during lake stratification using three key components; the volume and time-averaged summer hypolimnion temperature (as one whole compartment) (({{boldsymbol{T}}})), summer stratification duration (({t}_{{stratification}})), and the lake’s trophic state via the volumetric hypolimnetic O₂ depletion rate, (VHOD). The trophic state dependent standard rates ((VHO{D}_{{trophic}.{state}})) at a reference temperature (({T}_{{REF}})) were independently parameterized from literature^{43,50,97,98,99} (Fig. S2) using Carlson’s trophic state index (based on Secchi disk depth (m), total phosphorus (µg L⁻¹), or chlorophyll-a (µg L⁻¹)). Furthermore, the rates were temperature corrected (({{boldsymbol{k}}}))¹⁰⁰, from the reference temperature to the study lake’s yearly summer stratified hypolimnion temperature:

The model bases O₂ depletion on the initial O₂ concentration following spring mixing at 100% saturation (({{{rm{O}}}}_{2}left({t}_{{0}}right))). Volumetrically averaged hypolimnion temperatures from the lake models, obtained using rLakeAnalyzer¹⁰¹, were used to calculate initial O₂ concentrations, as described in Nkwalale et al.⁴³, following standard approaches in lake O₂ modeling¹⁰². Stratification duration inputs (({t}_{{stratification}})), obtained through the lake simulations, were defined as the longest continuous summer period of stratification exceeding 5 days²⁸. Following ISIMIP protocols, stratification was defined as a water column density difference greater than 0.1 kg m⁻³ between the bottom and the surface of a lake²⁷. We re-arranged our model assuming zero-order kinetics:

to:

To enable comparisons among the differing lake systems, we applied two metrics derived from this model. First, we calculated the shifts in time under stratification for a lake to reach anoxic (({t}_{{anoxia}})) (time-to-anoxia) or hypoxic (({t}_{{hypoxia}})) (time-to-hypoxia) conditions by rearranging the model equation as follows:

The numbers in the numerator represent the values of the O₂ thresholds (0.5 mg L⁻¹ for anoxia and 2.0 mg L⁻¹ for hypoxia). The anoxic period is defined as starting at the first onset of anoxia and persisting until the subsequent winter, with no recovery within the same season, and the hypoxia is defined likewise. Additionally, we employed anoxic ratio as an integrative metric to measure lake resilience, which captures the proportion of the stratified period during which O₂ concentrations are below the anoxic threshold, calculated as follows:

The O₂ model performance was previously validated using 5 German lake systems (see Nkwalale et al.⁴³), proving to be good (R² = 0.6), especially for large-scale and multi-lake systems modeling exercises. The model prioritizes deciphering general trends, simplicity in application, and broad transferability, hence it employs a zero-order kinetics approach through its assumption of linear oxygen depletion over time. Thus, negative O₂ concentrations may occur in the model, especially in warm climates or high trophic states, which accelerate O₂ depletion. Although the negative O₂ concentrations in the results seem unrealistic, a zero-order kinetics model allows for better reflections of a lake’s chemical O₂ demand, i.e., the O₂ needed to oxidize reduced substances like NH₄⁺, H₂S, Fe²⁺, and Mn^2+41,103. In essence, this O₂ deficit must be met before stable oxic conditions can be established. Thus, a highly negative O₂ concentration indicates the need for intense reaeration before replenishment.

Observational O₂ data from 11 lakes (Black Oak, Delavan, Green, Kinneret, Mozhaysk, Sammamish, Sunapee, Tarawera, Two Sisters, Washington, and Wingra) was used to validate our modeling framework following that described in Nkwalale, et al.⁴³. Although a moderate R² of 0.41 is achieved for the framework (Fig. S6), this level of performance is reasonable given its application across highly diverse lake systems spanning the equatorial to polar regions. Furthermore, uncertainty introduced by the O₂ model’s VHOD parameterization, along with the broader inherent uncertainties from hindcasting historical thermal and stratification patterns in the 11 lakes using the same approach as described above for future projections (three lake models driven by five climate models), contributes to the overall model uncertainty.

Limitations of the study

Our approach relied on several assumptions, notably the complete isolation of hypolimnia conditions during stratification. Although these assumptions may influence individual lake projections, robust conclusions on climate impacts and anoxia risk can be drawn from the modeling framework, which performs relatively well (Fig. S6) considering the global expanse covered in lakes occurring between the equator and the poles. Uncertainties from the climate/lake physics models are explicitly included in our results (Figs. 1 and 2) and further examined in Feldbauer et al.⁶⁴. Furthermore, our modeling framework neglects explicitly resolving for the seasonal deepening of the thermocline and the associated erosion of the hypolimnion during the course of the summer stratified period. However, hypolimnetic temperatures used to derive the VHODs were calculated as daily, volume-weighted means of the hypolimnion and subsequently averaged over the stratified period, such that late-summer warming and the concurrent reductions in hypolimnetic volume are indirectly incorporated in the empirical O₂ demand rates. Accounting for the changing hypolimnetic volume and sediment-water ratios could affect estimates of the timing of anoxia, particularly in shallow lakes⁴⁸, but is unlikely to alter predicted stratification duration or results for deep systems. This approach is centered on gaining broad applicability, lower uncertainty, and direct linkages to observable, management-relevant outcomes. Thus, it is an efficient and scientifically robust choice in global lake modeling applications.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.