Deepwater dissolved oxygen shows little ecological memory between lake phenological seasons

ABSTRACT

Depletion of deepwater dissolved oxygen (DO) in lakes has become increasingly prevalent and severe because of many external stressors, potentially threatening human-derived ecosystem services ranging from drinking water quality to fisheries. Using year-round, high-frequency DO data from 12 dimictic lakes, we compared 3 measures of deepwater DO depletion during winter and summer: DO depletion rate, DO minimum, and hypoxia duration. Hypoxia (DO < 3 mg L⁻¹) occurred in over half of the lakes and persisted an average of 83% longer in summer than in winter. While we found no difference in DO depletion rates between winter versus summer, these rates were significantly related to lake morphology in winter but trophic state in summer. In assessing cross-seasonal linkages, we found limited evidence for significant legacy effects in deepwater DO availability. Only fall mixing efficacy significantly responded to the previous summer’s minimum DO saturation, but it always reached moderate to high DO replenishment levels (>65%) regardless of the previous summer’s DO depletion severity. This lack of ecological memory in deepwater DO depletion across seasons suggests that deepwater DO largely resets during spring and fall mixing periods in most years in these dimictic lakes. Understanding the patterns and drivers in deepwater DO depletion in both winter and summer is a key step forward for predicting future chemical and biological consequences of seasonal DO depletion and managing lake ecosystem health, as well as the effects that climate change may have on these patterns.

KEYWORDS:

• climate change
• ecological memory; lake mixing
• legacy effects
• oxygen depletion
• oxygen minimum zones

Introduction

Dissolved oxygen (DO) has long been recognized as a critical component of lake ecosystem structure and function (Hutchinson 1957, Wetzel 2001) and has been used as an indicator of lake ecosystem health in management and monitoring programs across the world (Burns 1995, Sánchez et al. 2007, Marcé et al. 2016, Fukushima et al. 2019). DO is an important integrator of the physical, chemical, and biological processes in lake ecosystems across seasons (Adrian et al. 2009, Williamson et al. 2009) and, in turn, affects a range of human-derived ecosystem services such as freshwater quality and food provisioning (Marcé et al. 2016). Decreases in deepwater DO concentrations have been a globally pervasive pattern for many decades in lakes (Jenny et al. 2016, Jane et al. 2021) as well as in coastal oceans (Breitburg et al. 2018), with an increasing prevalence and expansion of low-oxygen “dead zones.” Increasing severity of low DO conditions extends not only to the dead zones in places such as Lake Erie and the Gulf of Mexico (Rao et al. 2014, Breitburg et al. 2018), but also to many pristine, well-protected lakes (Toro et al. 2007, Knoll et al. 2018), spurring widespread and growing concern for ecosystem health and drinking water quality. Low- or no-oxygen zones in lakes can lead to critical habitat loss (De Stasio et al. 1996, Craig et al. 2015), fish kills (Tonn et al. 1990, Rao et al. 2014, Kangur et al. 2016), and altered biogeochemical reactions, including dissolved phosphorus release from anoxic sediments (Mortimer 1942, Brothers et al. 2014), potentially stimulating harmful algal blooms and increasing greenhouse gas production (Marotta et al. 2014, Beaulieu et al. 2019). However, the connections between periods of DO depletion versus mixing and the potential ecological memory across seasons remains largely unknown.

In dimictic lakes, deepwater DO depletion can occur during both summer and winter when mixing to deep waters is limited. By contrast, spring and fall periods of mixing replenish deepwater DO and are considered “resetting” periods for DO. However, if DO depletion during winter or summer is high, if mixing periods are short, or if full vertical mixing is not complete, deepwater DO may not fully replenish before the subsequent summer or winter period (Couture et al. 2015, Dugan 2021). These cross-seasonal connections are highly understudied but could potentially lead to important ecosystem responses at subannual scales that can have long lasting influences on ecosystem function. Hence, understanding the severity and drivers of DO depletion as well as their connections across seasons in freshwaters is critical to understand DO dynamics. This information can be used to inform management decisions and predictive modeling of the ecosystem consequences throughout the year.

Deepwater DO depletion during ice-covered winter periods plays an important role in regulating lake biogeochemistry, vertical nutrient distributions, and organism habitat but is greatly understudied compared to depletion during the open-water season when data are more readily available (Hampton et al. 2017). However, in contrast to the more accessible open-water season, sampling during winter ice cover is often more restricted, leading to few lakes with detailed measurements of deepwater DO concentrations during winter and thereby limiting our understanding of winter DO dynamics and drivers for management applications. Only one multi-lake study that included 41 lakes in Ontario, Canada, has directly compared deepwater DO availability in winter versus summer and reported longer periods of anoxia during summer compared to winter using estimates from weekly to monthly DO profiles (Nürnberg 1995). Currently autonomous high-frequency sensors can provide detailed estimates of deepwater DO availability throughout the year and are vital to informing lake managers about the critical, short-term processes influencing lake ecosystem health and services (Marcé et al. 2016, Rose et al. 2016). While some recent studies in individual lakes have used automated sensing data to better understand DO dynamics during winter (e.g., Couture et al. 2015, Obertegger et al. 2017, Granados et al. 2020), a broad-scale effort to compare deepwater DO availability in winter versus summer across lakes and assess the potential cross-seasonal effects has not been undertaken.

In addition to being understudied, DO depletion during winter has the potential for a legacy carry-over effect from winter to summer or vice versa, particularly in dimictic lakes (Dugan 2021). In dimictic lakes, periods of inverse stratification and ice cover during winter and of stable stratification during summer promote deepwater DO depletion due to minimal vertical mixing and respiration rates exceeding production (Hemmingsen 1959, Golosov et al. 2012, Kirillin et al. 2012). By contrast, the intermediate periods of spring and fall mixing promotes replenishment of DO to deeper waters, and the efficacy of this replenishment during mixing can be important benchmark for the severity of low DO conditions during the subsequent DO depletion period in deep lakes. For example, a longer period of spring mixing in 1 year in Lake Langtjern, Norway, resulted in enhanced DO replenishment and, in turn, less severe deepwater DO depletion in the subsequent summer (Couture et al. 2015). Longer spring mixing due to earlier ice breakup also resulted in greater spring DO replenishment in several lakes in Wisconsin, USA, but led to earlier onset of low DO conditions the subsequent summer (Dugan 2021). These potential cross-seasonal connections in deepwater DO can be useful tools for forecasting and managing DO depletion in the subsequent season or predicting DO depletion in other systems at risk of low DO conditions where data are not readily available for some or all of the year.

Here, we use a modified space-for-time approach to compare the prevalence of and cross-seasonal patterns in deepwater DO depletion between winter, spring, summer, and fall in 12 seasonally ice-covered lakes across North America and Europe. Using year-round high-frequency data from these lakes, we address the following questions: (1) how does deepwater DO availability compare between winter and summer; (2) what are the drivers of DO depletion rates across lakes in winter and summer; and (3) are there cross-seasonal relationships in DO availability between winter, spring, summer, and fall that can be broadly applied? We predicted that winter and summer DO depletion would differ as a function of water temperature and trophic status because of the differences in metabolic and biological activity associated with these drivers than influence DO consumption. We predicted that winter and summer DO depletion metrics would have significant relationships that could be used to predict DO availability in subsequent phenological periods, but only if deepwater DO was not fully replenished in spring or fall.

Materials and methods

Data processing

We used high-frequency DO and water temperature data from sensors in 12 seasonally ice-covered lakes in the Northern Hemisphere (Fig. 1, Table 1). Across all lakes, a deepwater sensor was used for bottom DO and water temperature, with depth varying because of differences in lake depth and deployment strategies (Supplemental Table S1). Bottom sensors were below the summer thermocline depth and within <5 m of the maximum lake basin depth in all but 2 lakes (Harp and George) and ranged from <0.5 to 27 m (median = 4.5 m) from the bottom, representing 48%–96% (median = 68%) of the maximum lake basin depth. While neither sensor location in the lake nor bottom sensor vertical position were coordinated a priori across all study lakes, these data are representative of general hypolimnetic conditions for this set of lakes based on the data collectors’ knowledge of their system(s). Sensor manufacturers included Aanderaa, Hydrolab, Onset Computer Corporation, Precision Measurement Engineering (PME), Thermocouples, Yellow Springs Instruments (YSI), and Zebra-Tech Ltd., and recording intervals ranged from 30 s to 60 min (Supplemental Table S2). The sensor data were quality checked for periods where sensors were temporarily removed from the lake or had otherwise unreliable readings; these data points were removed with no interpolation. For each lake, we calculated the daily average bottom water temperature and DO from 0000 to 2359 h local time. The total time span of these data ranges from 5 months to several years depending on the lake, and all lakes spanned at least 2 consecutive periods of winter ice cover, spring mixing, summer stratification, and fall mixing per year.

Figure 1. Locations of lakes included in this analysis, spanning (a) north central to northeastern USA and southeastern Canada and (b) Europe. Symbols represent individual lakes and are consistent across all panels and figures.

Table 1. Geographic, morphometric, trophic state, and deepwater temperature summary data for each lake. Mean values across all year(s) are presented for winter and summer deepwater temperature with standard deviation in parentheses if more than 1 year of data was available, and “na” if no data were available. Trophic state is the average of Carlson’s trophic state index based on Secchi depth, TP, and chlorophyll a for available years (Carlson 1977). Additional lake characteristic data are in Supplemental Table S1.

Lake name	Elev. (m a.s.l.)	Surface area (ha)	Max. depth (m)	Mean depth (m)	Trophic state	Mean winter deepwater temperature (°C)	Mean summer deepwater temperature (°C)
Cimera	2105	4.7	9.4	4.6	Oligotrophic	4.0 (0.3)	na
Giles	428	48.1	23.0	10.0	Oligotrophic	3.6 (0.4)	5.3 (0.2)
Harp	327	71.4	37.5	13.3	Oligotrophic	3.6 (<0.1)	4.8 (0.7)
Kilpisjärvi	473	3730.0	57.0	20.0	Oligotrophic	3.1 (0.1)	7.0
Lacawac	439	21.4	13.0	5.0	Mesotrophic	4.1 (0.3)	6.9 (<0.1)
Lake George, Calves Pen	110	443.0	35.5	17.4	Oligotrophic	2.4	na
Lake George, Tea Island	110	132.0	56.7	19.5	Oligotrophic	3.5	na
Langtjern	516	23.0	12.0	2.0	Mesotrophic	4.0 (0.2)	4.7 (1.0)
Nohipalo Mustjärv	61	22.2	8.9	3.9	Eutrophic	3.7 (0.2)	7.6 (0.4)
Nohipalo Valgejärv	62	7.0	12.5	6.2	Mesotrophic	4.4 (0.1)	6.2
Lake Stechlin, South Bay	60	412.0	30.0	22.8	Mesotrophic	4.1 (0.5)	5.6 (0.2)
Waynewood	421	28.0	12.5	6.0	Eutrophic	3.4 (0.2)	6.3 (0.1)

Metrics of deepwater DO depletion

We calculated 3 response metrics of deepwater DO depletion during lake phenological winter and summer: deepwater DO depletion rate (change in % saturation per week), minimum DO saturation, and duration of hypoxia. For each lake and year, we used daily DO from the bottom sensor to calculate these metrics. We calculated daily average DO saturation from the bottom sensor based on its temperature and DO concentration readings, in conjunction with the elevation for each lake and assuming zero salinity per the Benson-Krause method (Benson and Krause 1984, USGS 2011), using the rMR package in R (Moulton 2018). The start of DO depletion in winter and summer was estimated from the data when DO saturation began to decrease, and the end of winter or summer was estimated when deepwater DO saturation substantially increased, indicating mixing. DO depletion rates were calculated using least squares linear regression, starting at the onset of DO depletion during the respective winter or summer period (usually shortly after ice freeze for winter or the onset of summer stratification for summer; Fig. 2). The calculation of DO depletion rates ended before DO increased during spring or fall mixing, or on the latest date before mixing when the DO reading was ≥2 mg L⁻¹, after which nonlinear decreases in DO are more likely to occur (Burns 1995, Zdorovennova et al. 2016, Yuan and Jones 2020). We transformed DO depletion rates by multiplying by −1 so that high positive values for this metric indicate rapid rates of deepwater DO depletion. Minimum DO saturation was defined as the lowest recorded DO saturation value during the winter or summer before mixing. Duration of hypoxia was defined as the number of days with deepwater DO concentration <3 mg L⁻¹ during winter or summer before mixing. This DO concentration is a key physiological threshold for a broad variety of fish species (Fang et al. 2004, Jacobson et al. 2010), and below which is a potentially important short-term refuge for zooplankton species (Wright and Shapiro 1990, Vanderploeg et al. 2009). Finally, we calculated the level of DO replenishment during the spring and fall mixing periods as the maximum deepwater DO saturation following winter DO depletion and before the onset of summer stratification (spring mixing efficacy) or following summer stratification before the onset of winter DO depletion (fall mixing efficacy).

Figure 2. Time series plots of dissolved oxygen percent saturation for all lake-years included in this analysis. Blue trend line represents the linear fit of the winter depletion rate estimates, and red trend line represents the linear fit of the summer depletion rate estimates.

In total, we calculated oxygen depletion metrics during winter in all 12 lakes across 30 lake-years, during spring mixing in all 12 lakes across 28 lake-years, during summer in 10 lakes across 27 lake-years, and during fall mixing in 9 lakes across 22 lake-years (Fig. 2). Lake George (Calves Pen and Tea Island) was limited to only winter and spring data and had no adequate data to calculate summer oxygen depletion metrics or fall mixing replenishment, and Lake Kilpisjärvi had no data from fall mixing periods because of deployment timings.

Statistical analyses

We first compared metrics of deepwater DO depletion in winter versus summer across lakes using nonparametric paired Wilcoxon signed rank tests with α = 0.05. We assessed which lake characteristics might influence winter and summer DO depletion metrics using nonparametric Kendall correlation tests with variables of morphology (maximum depth, surface area, surface area:volume), trophic state index (average of Carlson’s trophic state index based on Secchi depth, total phosphorus, and chlorophyll a; Carlson 1977), and water temperature as covariates, with α = 0.01 (correcting α = 0.05 for 5 comparisons). These variables were selected based on data availability and previous literature suggesting their role in oxygen consumption and thereby deepwater DO depletion rates.

To test for cross-seasonal relationships between deepwater DO depletion metrics (winter–summer) and potential legacy effects via mixing periods (winter–spring, spring–summer, summer–fall, and fall–winter), we conducted a series of generalized linear models with α = 0.0045 (correcting α = 0.05 for 11 comparisons). Variables of interest included DO depletion rate, minimum DO saturation, hypoxia duration in winter and spring, and maximum DO saturation in spring and fall. We transformed response variables not suitable for normally distributed models because they were count data or bounded by zero as percentages. Response metrics of DO saturation and hypoxia duration were run using quasi-Poisson models because the variance was up to 2 orders of magnitude greater than the mean. Response metrics of DO depletion rates remained untransformed and were run in Gaussian models. All analyses were completed in R 4.0.2 (R core Team 2021), and figures were created using the gplot2 (Wickham 2016) and ggpubr (Kassambara 2020) packages in R.

Results

Comparison of deepwater DO depletion in winter and summer

Across all lake-years, oxygen depletion rates and minimum oxygen saturation values were not different between winter and summer (p = 0.643, n = 23 lake-years; p = 0.144, n = 23 lake-years, respectively; Fig. 3a and b). However, hypoxia duration was significantly longer in summer than in winter (p = 0.019, n = 23 lake-years; Fig. 3c), averaging 62.6 d (72.7 d standard deviation [SD] compared to 34.1 [47.3] d). Four lakes experienced hypoxia in both winter and summer (Lacawac, Langtjern, Nohipalo Valgejärv, and Waynewood), and high-elevation Cimera, which had the longest periods of winter ice cover, was the only lake to experience hypoxia in winter but not in summer. Three lakes never experienced hypoxia in winter or summer during the period of study (Harp, Kilpisjärvi, and Stechlin).

Figure 3. Comparison of deepwater DO depletion metrics during the winter versus summer stratified periods: (a, d) rate of DO depletion, (b, e) minimum DO saturation, (c, f) and duration of hypoxia. For cross-seasonal comparisons of winter and summer values within a lake-year (d, e, f), generalized linear models were run with DO depletion as Gaussian, and with DO saturation and hypoxia duration as quasi-Poisson. Symbols represent an individual lake and are consistent across all panels and figures.

Lake characteristics versus deepwater DO depletion rates

DO depletion rates in winter were related to morphological variables, whereas DO depletion rates in summer were related to trophic state (Table 2, Supplemental Fig. S1). DO depletion rates in winter were positively related to surface area:volume (τ = 0.41, p = 0.002) and negatively related to maximum depth (τ = −0.40, p = 0.003) and surface area (τ = −0.49, p < 0.001). By contrast, DO depletion rates in summer were positively related to trophic state index (τ = 0.41, p = 0.004) and only weakly related to summer deepwater temperature (τ = 0.32, p = 0.022).

Table 2. Nonparametric Kendall correlation coefficients (τ) lake morphologic, trophic, and water temperature characteristics versus deepwater DO depletion metrics in winter and summer. Bold values with asterisks indicate statistically significant correlations (p < 0.01). Scatterplots of these relationships are in Supplemental Fig. S1).

Lake characteristic	Winter DO depletion rate (% week^−1)	Summer DO depletion rate (% week^−1)
Maximum depth	−0.40*	−0.21
Surface area	−0.49*	−0.27
Surface area: volume	0.41*	0.21
Trophic state index	0.06	0.41*
Deepwater temperature	0.16	0.32

Cross-seasonal relationships in deepwater DO metrics

We found no significant relationships between summer metrics and winter oxygen depletion rates (p = 0.428; Fig. 3d), winter minimum DO saturation (p = 0.764; Fig. 3e), or winter hypoxia duration (p = 0.755; Fig. 3f). Results from relationships using summer as the independent variable and the subsequent winter as the dependent variable produced similar results and coefficients (Table 3), with a similar lack of significance across all 3 metrics.

Table 3. Slope coefficients from generalized linear regressions coefficients of cross-seasonal deepwater DO depletion metrics across winter–summer, winter–spring, spring–summer, summer–fall, and fall–winter. Regression coefficients indicate the expected change (or percent change) in the corresponding dependent variable with a one unit increase in the associated independent variable. For winter–summer comparisons, coefficients preceding the slash indicate coefficients when winter metrics were the independent variables and the subsequent summer’s metrics were the dependent variable; values following the slash are comparable coefficients when summer metrics are the independent variable and the subsequent winter’s metrics are the dependent variable. Bold value with asterisk indicates statistically significant slope coefficient (p < 0.0045).

	Independent variable	Dependent variable	Transformation family	Coefficient
Winter vs. summer	Winter DO depletion rate	Summer DO depletion rate	Gaussian	0.16 / 0.02
	Winter minimum DO saturation	Summer minimum DO saturation	Quasi-Poisson	0.35% / 0.30%
	Winter hypoxia duration	Summer hypoxia duration	Quasi-Poisson	0.14% / 0.09%
Winter vs. spring	Winter minimum DO saturation	Spring maximum DO saturation	Quasi-Poisson	0.53%
	Winter hypoxia duration	Spring maximum DO saturation	Quasi-Poisson	−0.35%
Spring vs. summer	Spring maximum DO saturation	Summer minimum DO saturation	Quasi-Poisson	4.66%
	Spring maximum DO saturation	Summer hypoxia duration	Quasi-Poisson	−0.12%
Summer vs. fall	Summer minimum DO saturation	Fall maximum DO saturation	Quasi-Poisson	0.43%*
	Summer hypoxia duration	Fall maximum DO saturation	Quasi-Poisson	−0.12%
Fall vs. winter	Fall maximum DO saturation	Winter minimum DO saturation	Quasi-Poisson	0.70%
	Fall maximum DO Saturation	Winter hypoxia duration	Quasi-Poisson	1.59%

Spring mixing efficacy, measured as the maximum DO saturation observed between winter and summer, was not significantly related to the previous winter’s minimum DO saturation (p = 0.039; Fig. 4a) or to its hypoxia duration (p = 0.070; Fig. 4b, Table 3). When spring mixing efficacy was used as the predictor for the subsequent summer DO depletion metrics, it was similarly not significantly related to summer minimum DO saturation (p = 0.009; Fig. 4c) or to summer hypoxia duration (p = 0.148; Fig. 4d, Table 3).

Figure 4. Scatterplots of cross-seasonal deepwater DO metrics between (a–b) winter–spring (c–d) and spring–summer. Generalized linear models were run with response metrics of DO depletion as Gaussian, and with DO saturation and hypoxia duration as quasi-Poisson. Symbols represent an individual lake and are consistent across all panels and figures.

Fall mixing efficacy, measured as the maximum DO saturation observed between summer and winter, was positively related to previous summer’s minimum DO saturation (p < 0.001; Fig. 5a), where a 1% increase in summer minimum DO resulted in a 0.43% increase fall maximum DO (Table 3). Fall mixing efficacy was not related to the previous summer’s hypoxia duration (p = 0.012; Fig. 5b). When fall mixing efficacy was used as the predictor for the subsequent winter’s DO depletion metrics, it was not related to winter minimum DO saturation (p = 0.530; Fig. 5c) or to winter minimum hypoxia duration (p = 0.477; Fig. 5d, Table 3).

Figure 5. Scatterplots of cross-seasonal deepwater DO metrics between (a–b) summer–fall and (c–d) fall–winter. Generalized linear models were run with response metrics of DO depletion as Gaussian, and with DO saturation and hypoxia duration as quasi-Poisson. Trend line in (a) corresponds to Gaussian generalized linear model. Symbols represent an individual lake and are consistent across all panels and figures.

Overall, cross-seasonal connections in deepwater DO depletion were rare, although high variability was observed among lakes and, for lakes with multiple years of data, high variability among years. While summer minimum DO saturation could also predict fall mixing efficacy, maximum DO saturation during fall always reached at least 65%, indicating moderate to complete DO replenishment regardless of the previous summer’s DO depletion severity.

Discussion

Our results demonstrate that hypoxic and anoxic conditions were prevalent in many study lakes in both the winter and summer seasons, but hypoxic events lasted an average of 83% longer in summer compared to winter. Winter DO depletion rates were related to morphology, while summer DO depletion rates were related to trophic state, suggesting a more active role of biological influences on DO consumption in summer compared to winter. Further, while fall mixing efficacy could be predicted by the previous summer’s minimum DO saturation, other cross-seasonal relationships were not significant, suggesting that deepwater DO has little ecological memory across seasons and that mixing periods in dimictic lakes in most years can be considered “resets” of deepwater DO. Because winter and summer DO depletion rates were not significantly different, managers could use the depletion rates from previous seasons to infer and estimate the likelihood of increased greenhouse gas production (Marotta et al. 2014, Beaulieu et al. 2019) or low oxygen conditions in the future, which may help manage fish kills (Tonn et al. 1990, Rao et al. 2014, Kangur et al. 2016).

DO depletion rates were differentially related to lake characteristics in winter versus summer. In winter, shallower, smaller lakes had higher rates of DO depletion, as did lakes with a higher surface area:volume ratio, which approximates the sediment surface area to water volume ratio. These relationships reflect known lake-specific drivers of DO depletion, including sediment surface area to water volume ratio (Mathias and Barica 1980, Meding and Jackson 2001, Leppi et al. 2016, Bengtsson and Ali-Maher 2020), indicating greater sediment oxygen demand relative to water column oxygen production. In summer, more productive (i.e., higher values of trophic state index) and warmer lakes had higher rates of DO depletion. This increase in DO depletion rates is common in productive lakes largely as a result of increased biological activity and DO consumption (Mathias and Barica 1980). For example, the highest DO depletion rates in summer, which were 1.4 times greater than any other lake or year, consistently came from one of the eutrophic lakes in this study, which has the second highest trophic state index across all 12 lakes. Increased DO depletion under high temperatures reflects the established metabolic responses of DO consumption to water temperature (Kleiber 1932, Gillooly et al. 2001, Kraemer et al. 2017). The lack of consistency in the lake characteristics associated with DO depletion rates between winter and summer may be due to limited light and more consistent cold temperatures during winter compared to summer. Both factors would limit the range under which DO producers and consumers could operate, making morphology and its associated ecosystem variables, as opposed to productivity, more strongly associated with DO depletion in winter. These differences in DO depletion rates, even in winter, can also be affected by patterns in nutrient concentrations, organic matter availability, and other factors not measured here that may carry over from previous seasons (i.e., summer and fall before ice cover). Additional studies in lakes with high-frequency water chemistry data paired with deepwater DO data throughout the year are needed.

Summer hypoxia duration was longer than that in winter, although cross-seasonal relationships using summer DO depletion to predict the subsequent winter’s DO depletion, and vice versa, were not significant. Longer winter hypoxia duration is common in only high latitude or high elevation systems (i.e., Cimera), where ice cover is also long, resulting in a shorter available period for summer stratification and associated summer hypoxia to occur. Future projections of shorter ice cover (Sharma et al. 2019) suggest relief from critically low winter DO concentrations sooner than in the past, potentially relieving winter hypoxia duration. In summer, trends toward extended summer stratification will likely result in extended periods of summer hypoxic conditions (Fang and Stefan 2009, Foley et al. 2012, Rösner et al. 2012, Jane et al. 2023). Although the lack of cross-seasonal relationships limits our ability to use a previous season’s deepwater DO data to predict future season’s DO dynamics, a single winter or summer season’s DO depletion rates can be used and applied to the subsequent depletion period. From this, researchers or managers can estimate the time from the onset of DO depletion it may take to reach critically low DO conditions. This small potential tool can be a first step to forecasting future DO availability and proactively managing for ecosystem consequences of hypoxic conditions.

Spring and fall mixing periods in this set of dimictic lakes can be considered true periods of DO “resetting” given the near lack of significant cross-seasonal relationships with deepwater DO depletion metrics. The only significant cross-seasonal relationship with a mixing period was between summer minimum DO saturation positively predicting the subsequent fall maximum DO saturation. However, fall maximum DO saturation was always >65%, even when the previous summer reached 0% saturation. Hence, while this relationship is valid, it does not suggest severe consequences or poor mixing in fall, even following severe DO depletion in the previous summer. However, occasionally spring mixing efficacy was low (4 lake-years with <30% maximum spring DO saturation), and all 4 of these instances occurred when the previous winter reached 0% DO saturation. No significant relationship was found between these variables, and these lakes already consistently experienced anoxic conditions every summer, making potential influences of incomplete mixing difficult to determine. However, this finding suggests that incomplete spring mixing may be more common than in the fall given advances in the onset of summer stratification in many lakes (Woolway et al. 2021). Furthermore, oligotrophic lakes that do not already experience low summer DO conditions may be most sensitive to potential changes in spring mixing efficacy. The intensity, frequency, and duration of mixing is likely to change in many lakes with continued climate warming. As such, more intense and frequent mixing events that replenish deepwater DO are expected in some lakes (Flaim et al. 2020), while others may experience less intense or frequent mixing that limits deepwater DO replenishment before the summer (Woolway and Merchant 2019). Hence, more data on lakes with a range of trophic states and with incomplete spring mixing are needed to better address these cross-seasonal DO dynamics and implications for summer DO depletion.

While we found important differences in DO depletion between summer and winter along with a general lack of evidence for ecological memory across seasons, we note some important limitations to this study. Our dataset was limited to only 12 lakes in the temperate region that had typical phenology for dimictic lakes. Although our number of lake-years was up to n = 30, we recognize that the lakes in this study do not represent a full suite of lake characteristics or mixing phenologies. This limitation leads to some uncertainty common with studies with small sample sizes, so broad application of our findings should be done with caution. Incorporating data from a broader set of lakes spanning wider ranges in geography, size, mixing phenology, and trophic state can better assess if this lack of cross-seasonal ecological memory of deepwater DO is common across lakes or specific to those with certain characteristics (i.e., dimictic systems). We also recognize that other drivers of deepwater DO were not explored here, largely because available data were lacking. For example, time series of nutrient and light availability throughout the year, especially under ice when such data are uncommon, would be excellent additions to understanding the seasonal drivers of DO depletion and how they could be used to the benefit of management and forecasting efforts. Specifically, episodic events such as large algal blooms influence deepwater DO availability, and the lag in time and extent of DO consumption following algal death could be finely assessed using high-frequency deepwater sensors. Developing a predictive model from such data could help inform managers of the timing of anoxic conditions following short-term algal blooms to allow preventative management such as deepwater aeration. Studies building on this research that can integrate more variables and lakes to address these limitations will be important to our understanding of drivers of and connections between seasonal DO depletion.

In a world of rapid global change, a combination of long-term and high-frequency monitoring of DO throughout the year will be vital to improve our understanding of the drivers, mechanisms, and ecological consequences of decreasing deepwater DO availability (Sharma et al. 2020). Although we found limited evidence for ecological memory of deepwater DO depletion across seasons, year-round sensor deployment highlighted the greater prevalence of critically low DO conditions in summer than winter in our study lakes, which may be worsening in areas with increasing eutrophication, warming deep waters, or lengthening summer stratified periods. For example, lakes transitioning to mesotrophic and especially eutrophic states are likely to experience greater volumes of low-oxygen water during summer, as are lakes with extended periods of summer stratification (Jane et al. 2023), likely leading to altered biogeochemical processes or trophic interactions during subsequent seasons. As collections of high-frequency data, particularly outside of the open-water season, become more common, building longer time series within and across lakes will improve our ability to assess and forecast variability in deepwater DO depletion and to predict and manage for the many ecological implications (Marcé et al. 2016, Sharma et al. 2020), such as loss of oxythermal habitat for fishes (Fang et al. 2004, Guzzo and Blanchfield 2017) and the largely unstudied under-ice microbial responses (Bertilsson et al. 2013).

Author contribution statement

RMP, CEW, and EPO conceived the manuscript. RMP wrote the manuscript with substantial contributions and feedback from CEW, EPO, and KCR. RMP, CEW, EPO, KCR, SAB, R-MC, HAD, IG, H-PFG, GBK, AL, JCN, JAR, MWS, MT, and HY contributed to the data acquisition, analysis, and drafting of the manuscript. All authors approved the final submitted manuscript.

Acknowledgements

We thank the Lacawac Sanctuary & Biological Field Station for access to Lake Lacawac and use of research facilities; the Waynewood Lake Association for access to Lake Waynewood; A. Penske for maintaining the measuring devices attached to the IGB-LakeLab in Lake Stechlin, G. Mohr for ice cover observations, and the Lake Stechlin technician team of IGB Department 3 for further data; the Servicio Territorial de Medio Ambiente de Ávila of Regional Government of Castilla y León that granted the permissions for research in Cimera Lake (Regional Park of Sierra de Gredos) and provided the invaluable help of a helicopter flight to transport the heaviest field equipment; the personnel of the Kilpisjärvi Biological Station, whose support made available the long-term lake monitoring in the high Arctic; the staff at the Darrin Fresh Water Institute for assistance in sensor deployment and retrieval; and C. McConnell, T. Field, and R. Ingram for field assistance for Harp Lake. This work was conceived at the global Lake Ecological Observatory Network (GLEON) and benefited from continued participation and travel support from GLEON.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.7916515 under Creative Commons Attribution 4.0 International (Pilla et al. 2023).

Additional information

Funding

RMP, CEW, and EPO were supported by US National Science Foundation grants DEB 1754265, DEB 1754276, and DEB 1950170 and Ohio Eminent Scholar of Ecosystem Ecology funds. KCR was funded by NSF grants 1638704, 1754265, and 1761805. SAB, HPG, and JCN were supported by the German Federal Ministry of Education and Research (BMBF) within the Collaborative Project “Bridging in Biodiversity Science - BIBS” (01LC1501G) and HPG by the Leibniz Foundation. R-MC was supported by the Sentinel North Research Chair in Aquatic Geochemistry (Sentinel North, a Canada First Research Excellence Fund Program). HW received support from the Norwegian Research Council (Lakes in Transition 244558; Climer 243644) and the Nordic Centre of Excellence Biowater (Nordforsk, 82263). The long-term monitoring program of Langtjern is supported by the Norwegian Environment Agency. IG and MT were funded by the Spanish Ministry of Economy and Competitiveness through the projects PaleoNAO (CGL2010-15767/BTE) and PaleoModes (CGL2016-75281-C2-1-R). Multiprobes in Cimera were provided by Centre for Hydrographic Studies (CEDEX). GK was supported by the German Research Foundation (DFG): Projects KI 853-11/1-2, KI 853-13/1; EU Program on International Network for Terrestrial Research and Monitoring in the Arctic (INTERACT): Projects “ConCur,” “LACUNA,” and “IceWave.” AL was supported by the Estonian Research Council Grants PSG32 and PRG709. JR and HY were supported by the Inter-American Institute for Global Change Research (CRN3038) and the US National Science Foundation Grants GEO-1128040 and EF-1137327. MS was partially funded by the Helen V. Froehlich Foundation.