Deeper waters are changing less consistently than surface waters in a global analysis of 102 lakes

The study investigates how lake vertical thermal structure, particularly deepwater temperatures and stratification strength, has changed globally and what factors drive these changes. While global analyses have documented rapid and consistent warming of lake surface waters, much less is known about deeper waters, which are critical habitats for temperature-sensitive organisms and sites of key biogeochemical processes. Existing global work on deepwater temperatures is limited and shows highly variable trends across lakes, contrasting with the more uniform surface warming. The authors aim to close this knowledge gap by analyzing long-term summertime thermal profiles from 102 lakes across five continents. They pose two questions: (1) How have deepwater temperature and thermal stratification changed globally? (2) Do lake thermal region, geography, morphometry, or water quality explain observed trends? They hypothesize that high-latitude and high-elevation lakes would warm fastest, small lakes would show prominent deepwater cooling with increased stratification, and clearer lakes would exhibit more pronounced thermal changes, especially where transparency has declined.

Prior studies show consistent rapid warming of lake surface waters globally, linked to meteorological drivers such as increasing air temperatures, changes in solar radiation, reduced wind speeds, and declines in water clarity. Evidence indicates deepwater temperature responses can differ from surface responses; a global analysis of 26 large lakes reported modest average deepwater warming (+0.04 °C decade⁻¹) but high inter-lake variability (−0.22 to +0.25 °C decade⁻¹). Morphometry influences thermal responses: shallower lakes may warm faster at depth; lakes larger than ~5 km² may show faster and more consistent deepwater warming than smaller lakes. Water transparency (Secchi depth), DOC, and chlorophyll-a affect vertical heat and light distribution and thus thermal structure; clearer lakes can be more sensitive to changes. Browning (increasing DOC) and eutrophication can reduce light penetration, strengthening surface warming while cooling deep waters via thermal shielding. Classification by lake thermal region provides a globally relevant framework for comparing lakes based on seasonal surface temperature dynamics and may help predict thermal trends.

Study design and data: The dataset includes long-term (1970–2009) summertime vertical temperature profiles from 102 lakes across 18 countries and five continents, spanning wide ranges in latitude (68.9° N to 38.8° S), elevation (−210 to 1,882 m asl), surface area (0.005–32,500 km²), maximum depth (2.5–1,642 m), and water quality (Secchi 0.5–31 m; chlorophyll-a 0.1–60 µg L⁻¹; DOC 0.1–18.4 mg L⁻¹). Seven of nine global lake thermal regions were represented. Sampling and profile selection: Temperature profiles were taken in pelagic, deepest zones using manual probes or automated profilers (median 1.0 m vertical resolution; frequency ranging from annual to sub-daily; median 9 profiles per year). For each lake-year, one profile representing strong, stable summer stratification was selected based on the relative thermal resistance to mixing (RTR) metric, using the median day of maximum RTR per lake and selecting within ±21 days. Profiles were quality-checked and interpolated/binned to 0.5 m where needed. Thermal metrics: Five metrics were computed for the selected summer profile each year: (1) Surface water temperature at 2 m; (2) Deepwater temperature at the deepest consistently sampled depth; (3) Mean water column temperature (depth-weighted average from 0 m to deepwater depth); (4) Density difference between deep and surface layers; (5) Thermocline depth (depth of maximum density gradient > 0.1 kg m⁻³ between adjacent 0.5 m layers) using rLakeAnalyzer. Time periods and inclusion criteria: Two analysis windows were defined: 1970–2009 (n=30 lakes meeting criteria) and 1990–2009 (n=99 lakes). Inclusion required at least one sample within five years of the start and end years and a minimum of 20 (1970–2009) or 15 (1990–2009) summer profiles. Sample sizes varied across metrics if full-depth data were unavailable. Trend analysis: For each lake and metric, Sen’s slope estimated the median linear rate of change over time for each period. One-sample Wilcoxon rank-sum tests (α=0.05) assessed whether the distribution of slopes differed from zero across lakes. Random forest analysis: To identify drivers of trends, the 1990–2009 Sen’s slopes (transformed as x_T = sign(x)·log(|x|+1)) were modeled using random forests (1500 trees per metric). Predictors (10 total) included: thermal region (categorical), absolute latitude, elevation, log(surface area), log(max depth), Secchi depth, chlorophyll-a, DOC, browning region (yes/no), and mixing type (polymictic vs. other). Collinearity among numeric predictors was low (|r|<0.7). Variable importance was quantified via relative increase in MSE upon permutation; pseudo-R² indicated explanatory power. Partial dependence plots were produced for key predictors (relative importance >0.8). All analyses were performed in R 3.5.0 using packages randomForest, wq, and ggplot2.

- Surface waters warmed consistently: median +0.37 °C decade⁻¹ (1970–2009; p<0.001; n=30) and +0.33 °C decade⁻¹ (1990–2009; p<0.001; n=99). In 1970–2009, 90% of lakes increased; in 1990–2009, 69% increased. - Stratification strengthened: density difference increased by +0.08 kg m⁻³ decade⁻¹ (1970–2009; p<0.001) and +0.06 kg m⁻³ decade⁻¹ (1990–2009; p<0.001). Increases occurred in 87% (1970–2009) and 69% (1990–2009) of lakes. - Deepwater temperatures showed no significant overall trend: +0.06 °C decade⁻¹ (1970–2009; p=0.053; 63% increasing) and −0.05 °C decade⁻¹ (1990–2009; p=0.11; 38% increasing, 54% decreasing). Inter-lake variability was high, ranging from −0.68 to +0.65 °C decade⁻¹. - Mean water column temperature increased overall in 1970–2009 (+0.19 °C decade⁻¹; p<0.001) but not in 1990–2009 (+0.05 °C decade⁻¹; p=0.59). - Thermocline depth deepened slightly in 1970–2009 (+0.03 m decade⁻¹; p=0.004) but showed no significant change in 1990–2009. - No cross-lake relationship between deepwater trends and either surface temperature trends (r=0.09, p=0.12) or density difference trends (r=−0.08, p=0.17). - Random forest explanatory power was low for trend drivers: deepwater temperature trends (pseudo-R² ≈ 8.4%), mean water column (15.6%), surface water (3.5%), density difference (16.0%); thermocline depth (0%). - Key predictors and patterns: deepwater trends were most influenced by surface area, thermal region, elevation, and DOC. Small lakes (≈<1 km²) tended toward deepwater cooling; larger lakes showed slight deepwater warming. Northern Warm and Northern Hot regions tended to cool at depth; higher elevation lakes (>500 m asl) tended to warm at depth (edge effects due to small sample). Intermediate DOC (≈2.3–6.9 mg L⁻¹) associated with slight deepwater warming; very low or moderately high DOC with cooling. Maximum depth best predicted surface warming and density difference trends, with shallower lakes showing faster surface warming and larger increases in stratification strength.

Findings address the research questions by demonstrating that surface waters of lakes globally have warmed consistently and stratification has strengthened, while deeper waters exhibit heterogeneous and often weak trends with no clear global tendency. The absence of relationships between deepwater temperature trends and either surface warming or stratification increases indicates that deepwater responses are governed by drivers distinct from those affecting surface layers. Morphometry, particularly lake size and depth, influenced some patterns—small, shallow lakes showed the greatest increases in stratification driven by rapid surface warming and often deepwater cooling—yet overall explanatory power of static lake characteristics and thermal region was low. This suggests a prominent role for external temporal drivers such as regional climate variability (e.g., wind stilling, ice phenology changes), watershed inputs, and water transparency changes due to eutrophication or browning. Especially in Northern Warm and Northern Hot thermal regions, deepwater cooling and muted mean water column warming may reflect decreased transparency that enhances thermal shielding. Ecologically, stronger and longer stratification, particularly in small lakes, is likely to reduce deepwater oxygen, alter habitat availability for cold- and cool-water biota, impact nutrient cycling, and potentially increase greenhouse gas production from anoxic sediments. The heterogeneous deepwater temperature trajectories imply location-specific ecological outcomes and management needs.

This global analysis shows that, over recent decades, lake surface waters have warmed and water column stability has increased consistently across diverse lakes, whereas deepwater temperatures have changed minimally on average but vary widely among lakes. Deepwater trends are largely decoupled from surface warming and stratification metrics and are poorly explained by static lake characteristics or thermal region, pointing to the importance of temporal external drivers such as regional climate patterns and changes in water transparency from eutrophication or browning. Future work should expand geographic and thermal-region coverage (particularly Northern Frigid, Southern Temperate, and Southern Hot regions), integrate time series of climate and water quality (transparency-related) variables, and assess seasonal dynamics beyond peak stratification to improve mechanistic understanding and prediction. The magnitude and direction of changes in deepwater temperatures and stratification will determine impacts on thermal habitats, nutrient cycling, harmful algal bloom risk, deepwater oxygen conditions, and greenhouse gas emissions.

- Geographic and thermal-region coverage imbalances: over-representation of Northern Temperate lakes; under-representation of Northern Frigid; few lakes from Asia, Africa, and South America; missing Southern Temperate and Southern Hot regions. This limits generalizability, particularly for high-latitude systems. - Lack of time series for key water transparency variables (Secchi, DOC, chlorophyll-a) constrained testing of mechanisms (eutrophication/browning-driven thermal shielding) behind deepwater trends. - Thermocline depth trends had 0% explanatory power with the tested predictors; the metric’s seasonal variability and reliance on single peak-stratification profiles plus 0.5 m vertical resolution may mask subtle long-term changes. - Satellite data cannot capture deepwater temperatures; reliance on in situ profiles of variable frequency and vertical resolution introduces heterogeneity. - Potential edge effects in some predictor-response relationships (e.g., limited number of high-elevation lakes) reduce confidence in threshold inferences. - Polymictic or infrequently sampled lakes may be less well characterized by the single-peak stratification profile approach.