Intermittent meromixis controls the trophic state of warming deep lakes

Vertical mixing modulates nutrient dynamics in lakes. However, surface warming reduces the range of vertical mixing and the probability of full circulation events. Important consequences of reduced vertical mixing include the sequestration of phosphorus (P) within a stagnant zone and the promotion of oligotrophication. Nevertheless, warming-induced shifts from full to partial mixing (meromixis) are not permanent and are partially reversible during exceptionally cold or windy winters. In this study, we investigated how intermittent meromixis affects lake P budgets. We examined the P cycle of a perialpine lake with variable mixing depths by pairing sedimentation and release flux measurements with sedimentary archives. We found that the amount of dissolved P surpassed that of the potentially mobile P in the sediments by a 13:1 ratio. At least 55% of the settled P was rapidly released to bottom waters isolated from flushing, illustrating the general biogeochemical mechanism that promotes deep-water P storage when lakes undergo warming. This storage process is abruptly inverted when meromixis suddenly retreats, deeper mixing introduces P pulses to the surface waters, thereby promoting phytoplankton proliferation. Our estimates showed that lakes containing up to 40% of the global freshwater volume could shift towards intermittent meromixis if the atmospheric warming trend continues. Thus, these lakes might accumulate 0–83% of their P load in irregularly circulating waters and are prone to large P pulses.

1. Basin bathymetry 2. Composition of sediments in trap bottles and cores 3. Mineralization rate estimate 4. Sediment characteristics across Lake Iseo 5. Deep mixing effect on EC25 6. Global meromixis prevalence and impact computation 7. Methods used for P pool and flux determination

Basin bathymetry
To account for spatial heterogeneity we collected samples at three locations (Station A-C). In general, we found water column parameters to be relatively similar across stations ( Figure 2). In contrast, sediment parameters differed more strongly (for example in dry mass and P content), likely because of different sediment compaction (the stations have different depths) and settling flux composition (the stations have different distance from the main inlet). Consequently, the basin was partitioned so that sub-basins can adequately reflect these differences. In brief, the results from Station A represent the deep sediments, because this basin (Central/North) includes all sediments below 175m, while Stations B and C represent the sediments that form within the shallow channel created by the island (East Channel), and the sediments that form in maximum distance from the main inlet (South), respectively. Therefore, we expect the most accurate lake-wide estimates when we treat the three stations not as replicates but as representatives of these specific sub-basins. The bathymetric curves of the sub-basins are shown in Figure S1 and they accumulate to 100% of the area of Lake Iseo ( Figure S1a). Lake-scale estimates of sediment characteristics (e.g. fluxes or pools) were derived by multiplication of values obtained in each station with the sub-basin's respective fraction of the total monimolimnion surface area (i.e. area below 90m, Table 1), followed by addition of these three local values.

Figure S1
Hypsographic curves of Lake Iseo shown for the whole lake (a) and for all sub-basins (b-d). Area below the dotted line (91 m) is considered as the monimolimnion area (2017), in each basin represented by one of the measurement stations.

Composition of sediments in trap bottles and cores
Element-specific content of collected material from traps and the sediment that formed below (from sediment cores, all at station B) are a convenient system to study sediment diagenesis and element recycling (see discussion section). Porewater concentration gradients and the according fluxes were determined as described in the methods section. Table S2 shows the individual fluxes of the reduced mineralization end products. Mineralization of carbon in organic matter (C org ) may proceed according to , assuming an average carbon oxidation state of 0. To infer the amount of organic matter mineralized in the absence of oxygen, fluxes of anaerobic metabolites may be normalized according to their oxidation stochiometry (Table S3).

Sediment characteristics across Lake Iseo
Analysis of sediment composition along a North-to-South transect in Lake Iseo. Main tributaries are in the North basin. 59% a regular station with detailed parameters for sediment speciation b additional stations with data for two sediment layers c sum of redox-sensitive P (BD), free and loosely adsorbed P (NH 4 Cl), and organic P and poly-P (NRP) as percentage of total P in all sediment fractions  Figure S3 shows sediment dry mass profiles. Cumulative dry mass may vary with sediment composition and depth. Cores from larger depths have a tendency to be more compacted with a higher dry mass per volume.

Figure S3
Cumulative dry mass profiles of sediment cores across all stations.

Deep mixing effects on EC25
We show in Figure 6 that deep-water SRP content drop when surface and deep layers mix. We infer the timepoint and magnitude of solute exchange from published conductivity and mixing depths time series. 2 Episodes, during which waters from the middle layer (50m) and the monimolimnion (e.g., 150m) mix show similar electrical conductivity, suggesting these waters are in a state of connection and solute exchange. Using the data from the same source, we expanded this analysis to other water layers between the depth of 100 and 245m. Figure S4 illustrates how conductivity (EC25) of waters are more likely to be similar (i.e., close to the 1:1 line) during and shortly after mixing events that reached down to the given depths.

Figure S4
Comparison of electrical conductivity (normalized to 25°C, EC25) in four different water depths (a-d) with EC25 closer to the surface (50m

Global meromixis prevalence and impact computation
Data for global lake bathymetry (maximum depth, medium depth, location, surface area) was obtained from the GLDBv2 dataset. We excluded lakes for which the mean depth was either not included or estimated by a simple approximation (volumetric factor of 1.500). The depth of the upper surface of the monimolimnetic water layer was derived based on an empirical model formulated by Bryhn 4 . He suggested that this may be used as a reference of the interface between monimolimnion and mixolimnion (i.e., the water layers that circulate in regular intervals) that can be expected to appear in meromictic lakes, irrespective of the reason for meromixis. Using this depth, for each lake we calculated the surface area of the monimolimnion from the bathymetric model of Håkanson 5 . To this end, the author uses the volumetric factor for each lake, a geometric relationship between the lake volume and surface area as follows 5 : with as the volumetric factor, the average depth, the maximum depth, and (0-1, no unit) as the areal fraction of monimolimnion surface area relative to the total lake surface. The absolute monimolimnion surface area can be derived by simple multiplication.
Accordingly, if sedimentation is assumed to be spatially homogeneous across the surface waters of a lake, equals the share of sediment material diverted towards non-mixing monimolimnetic waters (see discussion section).

Figure S5
Location of lakes included GLDBv2 and hydroLAKES datasets (n=243). Map was prepared by R software (Version 3.4.3) 76 using the package rworldmap.

Methods used for P pool and flux determination
Sampling strategies were carefully designed to produce estimates for phosphorus pools in sediment and various water layers of Lake Iseo, as well as the fluxes between these pools. Figure S6 illustrates the methodological portfolio with which the presented results were obtained.

Figure S6
Sampling strategies for P pools and fluxes in Lake Iseo.