Seasonal overturn and stratification changes drive deep-water warming in one of Earth's largest lakes

Eighty-four percent of Earth's non-frozen surface freshwater is found in the 10 largest lakes, which act as sentinels of climate change due to their sensitivity and integration of watershed climate signals. Global lake surface water temperatures have increased on average by about 0.21 °C per decade, with some large lakes warming faster than oceans and regional air temperatures. However, subsurface observations are sparse, limiting understanding of how deep waters in large lakes respond to climate change. Subsurface waters integrate conditions across years, providing a climate memory and potentially signaling ecological and thermal regime shifts (e.g., dimictic to monomictic). Most prior knowledge relies on summer surface observations; translating these to deep subsurface dynamics is difficult due to stratification and mixing complexities. Here, using three decades of high-frequency subsurface observations in Lake Michigan, the study asks: (1) How do the deep waters of Earth's largest lakes respond to climate trends? (2) As surface temperatures rise and summer periods are extended, which winter subsurface characteristics are altered? (3) What can high-resolution, high-frequency observations reveal that is obscured in other long-term datasets? The analysis shows deep waters are warming in winter and elucidates how changes in overturn timing and winter duration propagate through the water column.

The paper situates its contribution within extensive work documenting surface warming of lakes via satellite and buoy observations, including accelerated summer warming in large lakes and spatial heterogeneity influenced by bathymetry and climate. While satellite-based analyses have improved characterization of year-round and seasonal surface trends and overturn timing in the Great Lakes, few studies have long-term, high-frequency hypolimnetic data needed for trend analysis at depth. Prior modeling studies have projected climate-driven changes in thermal structure and potential mixing regime shifts, but empirical subsurface records with adequate temporal and vertical resolution are rare. Observations from other large lakes (e.g., Tanganyika, Baikal, Superior) highlight ecological responses to warming (changes in productivity, community composition) and changes in ice cover and stratification duration. This study fills a gap by linking observed surface trends and atmospheric drivers to high-frequency, long-term subsurface temperature dynamics in a large dimictic lake.

Study site and in situ data: A thermistor array was deployed in 150 m water depth in southern Lake Michigan (42°40.493′N, 87°04.772′W). Subsurface temperatures were recorded nearly continuously from 1990–2019 at 3-hourly (1990–1994, Aanderaa TR7, ±0.03 °C) and hourly intervals (1994–2012, RBR TR-1000, ±0.05 °C; 2012–2019, Sea-Bird SBE56, ±0.002 °C). Sensor depths varied by year; effective transects were interpolated to fixed depths (30, 60, 75, 100, 110, 140 m). Data quality control removed inaccurate values; gaps due to sensor malfunctions were noted (e.g., 1993–1994). In 2013, a mid-line float failure sank part of the string; the top nine sensors remained over the bottom 40 m and were used. Surface temperatures: Daily lake surface temperature at the mooring location was obtained from GLSEA (AVHRR-based, 1.8 km resolution) for 1995–2019. Atmospheric drivers: Overlake air temperature, wind speed, and cloud cover were taken from GLCFS (hourly, 1988–present) supplemented by GLM-HMD (daily, 1948–1987), aggregated to annual values for trend analysis. Downward shortwave radiation was from the NOAA SURFRAD Bondville, IL station (monthly, 1995–2018). Analytical approach: Time series were averaged to daily and monthly values. Trends were estimated using (1) simple linear regression, (2) Theil–Sen robust estimators, and (3) seasonal-trend decomposition using LOESS (STL) with decomposition into long-term, seasonal (12 monthly components), and residuals. Linear fits to STL long-term components provided trend estimates and confidence intervals. Monthly trends by depth were assessed using linear regressions, normalized by maximum absolute monthly trend at each depth, and significance annotated by confidence thresholds. Deep-water dynamics metrics: At 110 m (verified with 100 m in sensitivity tests), fall overturn date (O) was the maximum temperature spike at depth; minimum temperature (M) was the subsequent minimum; cooling period duration was days from O to M; stratification date (S) was the post-minimum inflection where temperature stabilized for summer; stratified season duration was days from S to next O. Overturn dates were referenced relative to December 1. Relationships among O, cooling duration, minimum temperature, and stratification timing/temperature were evaluated with least-squares fits and R².

• Surface warming: Lake-averaged year-round GLSEA trends (1995–2019): 0.34 ± 0.21 °C/decade (linear), 0.31 ± 0.18 °C/decade (Theil–Sen), 0.40 ± 0.05 °C/decade (STL). At the mooring: 0.40 ± 0.26, 0.41 ± 0.20, and 0.49 ± 0.07 °C/decade, respectively, indicating the site reflects lake-wide behavior. Monthly surface trends show strongest warming in October; STL shows fall (Sep–Oct) increases, with relatively flat winter surface trends. - Subsurface long-term trends: Significant warming below the thermocline at 60–100 m: ~0.08–0.12 °C/decade depending on method. At 110 m: 0.05 ± 0.06 (linear), 0.04 ± 0.04 (Theil–Sen), 0.06 ± 0.01 (STL) °C/decade; linear-method confidence intervals include zero, suggesting nonlinearity and high interannual variability. At 140 m: no significant trend. - Seasonal subsurface patterns: Delayed fall overturn yields relative cooling trends in fall at 30 m (September) and at 60–75 m (November). Peak subsurface warming occurs in winter (January–April) at ≥60 m; linear warming at 75 and 100 m from January–August; at 110 m, warming primarily January and October. STL reveals nonlinear winter warming at 30–110 m and fall cooling linked to extended stratification. - Atmospheric drivers: Over 1990–2018, overlake air temperature increased 0.52 ± 0.22 °C/decade; wind speed increased 0.23 ± 0.04 m s⁻¹/decade; cloud cover decreased −4.55 ± 1.17%/decade; shortwave radiation increased 4.66 ± 1.79 W m⁻²/decade (1995–2018). Long-term means shifted after the 1997–1998 El Niño (air temperature 7.45→8.55 °C; wind 6.10→6.54 m s⁻¹; cloud cover 59.1→50.1%). - Overturn/stratification dynamics (110 m): Overturn dates ranged from mid-November to early January. Cooling period duration: 53–134 days; summer stratification: 162–263 days. A regime shift around 1997–1998 featured later overturn (from early to late December), shortened cooling (<100 days), and extended stratification (>200 days). - Deep-water relationships: Later overturn strongly correlates with shorter cooling periods; later overturn and shorter cooling correspond to higher minimum winter bottom temperatures, with notable cold outliers (1996, 2003, 2014, 2015). Minimum temperatures ranged 1.2–4.4 °C at 110 m. Stratification temperature increased and stratification onset occurred earlier with delays in fall overturn. - Implications: Observed cascade links surface warming and atmospheric changes to deep-water winter warming and timing shifts, indicating potential progression toward a dimictic-to-warm-monomictic regime under continued warming and reduced ice.

The high-frequency, three-decade subsurface record demonstrates that deep waters in a large dimictic lake warm primarily during winter due to a cascade initiated by delayed fall overturn and shortened winter cooling. These dynamics directly address the research questions by showing how surface warming and extended stratification translate to deeper layers: later overturn reduces the time available for hypolimnetic cooling, yielding higher winter minima and earlier, warmer onset of summer stratification. The analysis ties these subsurface responses to concurrent atmospheric trends (increased air temperature and shortwave radiation, decreased cloud cover, modest wind increases) and highlights a step-change around the 1997–1998 El Niño consistent with broader Great Lakes indices and modeled heat content jumps. The findings underscore that relying solely on surface trends obscures critical subsurface processes and nonlinearity (e.g., at 110 m), and that winter conditions exert strong control on deep-water temperatures. Ecologically, extended stratification and warmer deep waters may reduce oxygen replenishment and alter food webs, reinforcing concerns about approaching a mixing regime shift in large lakes.

Using an unprecedented 30-year, high-frequency subsurface temperature record from Lake Michigan, the study quantifies how climate-driven changes in overturn timing and stratification extend surface warming signals into the deep water. Key contributions include: (1) documenting significant winter warming at depth (notably 60–100 m) despite modest or nonlinear trends near 110 m and none at 140 m; (2) precisely characterizing overturn, minimum temperature, cooling duration, and stratification timing across years; (3) establishing robust relationships showing that delayed overturn shortens winter cooling, raises minimum deep-water temperatures, and leads to earlier, warmer stratification; and (4) linking these dynamics to observed atmospheric trends and a post-1997–1998 shift. The results indicate increasing likelihood of a dimictic-to-warm-monomictic transition with profound ecological implications. The authors emphasize the critical need for high-frequency, long-term subsurface monitoring to detect and interpret climate impacts on the majority of Earth's freshwater.

• Spatial limitation: Analyses are centered on a single long-term mooring site in southern Lake Michigan; while shown to be representative for surface trends and supported by sensitivity analyses across depths, spatial heterogeneity elsewhere is not directly observed. - Data gaps and instrument issues: Sensor malfunction created a large gap in 1993–1994; in 2013 a float failure altered the array, leaving only the upper sensors over the bottom 40 m. - Temporal coverage constraints: Surface GLSEA data begin in 1995; SURFRAD shortwave radiation begins in 1995, limiting long-term radiation comparisons. - Depth-specific trend uncertainty: At 110 m, linear trend confidence intervals include zero, indicating strong nonlinearity and high interannual variability; at 140 m, no significant trends detected. - Variable sensor depths and interpolation: Sensor depths varied year-to-year; temperatures at analysis transects were interpolated between sensors, potentially introducing small uncertainties. - Record length: Subsurface record begins in 1990; earlier conditions (pre-1990) are inferred from other sources and not directly observed.