Table 1: Site information and stable and clumped isotope results for freshwater carbonates used in this study. Mean annual air temperatures (MAAT) are averages of the long-term monthly means from each of our sites from 1981-2010, using the University of Delaware’s high resolution gridded air temperature dataset (Willmott & Matsuura, 2001) provided by NOAA (https://psl.noaa.gov/data/gridded/data.UDel_AirT_Precip.html). Samples from Eiler Lab at Caltech were processed in the CDES reference frame and projected into I-CDES following the methodology described in Bernasconi et al. (2021).
*Data from the Bernasconi Lab at ETH was recalculated using the methodology described in Bernasconi et al. (2021) and data from the Bergmann Lab at MIT was taken from Anderson et al. (2021).
ᵃ Recalculated from Wang et al. (2021)
ᵇ Recalculated from Huntington et al. (2015)
ᶜ Recalculated from Li et al. (2021)
ᵈ Recalculated from Huntington et al. (2010)
ᵉ Recalculated from Petryshyn et al. (2015)
ᶠ Recalculated from Santi et al. (2020)
ᵍ Recalculated from Bernasconi et al. (2018)
2.2 Sample Preparation
2.2.1 Biologic carbonates
Aquatic gastropod and bivalve shells were first separated by taxon. Organic material was removed from shells by scraping and sonicating in Milli-Q deionized water until clean. Samples were dried overnight at 50°C, and complete shells were powdered using a mortar and pestle, and reacted with 3% hydrogen peroxide for 60 minutes (Eagle, Eiler, et al., 2013) to remove any remaining organic material. Depending on carbonate content of the gastropod and bivalve shells and instrument sensitivity at the time of analysis, samples were weighed out for mass spectrometric analysis, typically in amounts varying between 5 and 10 mg for a single replicate.
2.2.2 Fine grained carbonates
Samples of unconsolidated calcareous particles, assumed to be micrite, were disaggregated in Milli-Q deionized water, after which the mixture was poured through a 212 μm steel mesh filter and left to settle in a beaker for 10 minutes. The residue was poured into a second beaker after filtration to remove any remaining suspended material, and this process was repeated until virtually no observable settling occurred. The final residue was treated with 3% hydrogen peroxide for 60 minutes to remove any remaining organic material (Eagle, Eiler, et al., 2013). Resulting micrite was collected on a 0.45 μm cellulose nitrate filter membrane and dried at 50°C. Depending on carbonate content and instrument sensitivity, the amount of sample utilized for mass spectrometry varied between 10 and 30 mg for a single replicate.
2.2.3 Biologically mediated carbonates
Tufas and microbialites were cut perpendicular to laminae, and polished slabs and thin sections were prepared in order to target specific zones for analysis. Samples were ground into a fine powder using a microdrill. To prevent potential bond reordering due to frictional heating, the drilling during this process was limited in duration and speed (rpm). Samples were reacted with 3% hydrogen peroxide for 60 minutes to ensure removal of any organics and dried overnight at 50°C (Eagle, Eiler, et al., 2013). Drilled samples were weighed out in 5 to 15 mg aliquots for mass spectrometry depending on the carbonate content of the sample and instrument sensitivity at time of analysis.
2.3 Stable Isotope Measurements
All samples were run at the University of California, Los Angeles on a Thermo 253 Gas Source isotope ratio mass spectrometer in the Eagle-Tripati Laboratory from 2012-2019, primarily between 2013 and 2015. Methods are described in detail in Upadhyay et al. (2021). Briefly, carbonate samples were first reacted with 105% phosphoric acid for 20 minutes on a 90°C online common phosphoric acid bath system, whereby solid carbonate reacts to produce CO2 gas for analysis. Acid temperature was monitored with a thermocouple throughout each analysis and checked daily for drift. Each sample gas was cryogenically purified using an automated purification system that was modeled on the previously described system at the California Institute of Technology (Passey et al., 2010). The liberated gas from each sample passed through two separate gas traps to ensure the removal of water and other compounds: the first (containing ethanol) is kept at -76°C by dry ice, and the second (containing liquid nitrogen) is kept at -196°C. After sample gas undergoes cryogenic purification, the sample gas is passed through silver wool to remove sulfur compounds  (e.g. halocarbons and hydrocarbons; Spencer & Kim, 2015) and remaining trace contaminants were separated by moving the resultant gas through a Thermo Trace GC Ultra gas chromatograph column, which is filled with a divinyl benzene polymer trap, Porapak Q at -20°C (L. M. Santi et al., 2020; Tripati et al., 2015). The purified sample gas was passed on to the mass spectrometer for analysis.
Data was collected over nine acquisitions consisting of 10 cycles each to determine δ13C, δ18O, Δ47, Δ48, and Δ49. During each acquisition on the mass spectrometer, sample isotope values were measured relative to high purity Oztech brand CO2reference gas (δ18O = 25.03‰ VSMOW, δ13C = -3.60‰ VPDB). Equilibrated CO2gas standards and carbonate standards were typically run every 3-4 analyses and used for standardization. Averages for our standard values can be found in Supplementary Table 2.
2.4 Data handling
Table 1 reports isotopic data for samples used within this study. Data processing is detailed in Upadhyay et al., 2021. Data are reported on the I-CDES scale which projects values into a 90oC reference frame. Acid digestion fractionation factors used for calcite and aragonite δ18O are reported in Swart et al. (1991) and Kim et al. (2007), respectively; for calculations of water δ18O, we use the equations of Kim and O’Neil, (1997) for calcite and Kim et al. (2007), for aragonite. For samples measured in the Eagle-Tripati Lab, raw mass spectrometer data was processed usingEasotope (Daëron et al., 2016; John & Bowen, 2016). Data is included in the supplement. We excluded replicates if results were consistent with high organic content, as indicated by anomalous Δ48 or Δ49 for a given correction interval, with samples having values that are more than 3 sigma from the standards being flagged for possible exclusion (Upadhyay et al., 2021). We also exclude replicates with anomalous values of Δ47(I-CDES), δ13C (VPDB) and δ18O (VPDB), of more than 3σ from the remaining replicates, which can reflect incomplete digestion or contamination (Tripati et al., 2015). We performed at least three replicate analyses of each sample unless the amount of material available limited the number of analyses; if less than three replicates were run for a sample, we propagated both the internal reproducibility of the sample and the average external reproducibility of the samples in this study to determine the uncertainty of the reported value.
Data from published studies (Anderson et al., 2021; Bernasconi et al., 2018; Huntington et al., 2010, 2015; H. Li et al., 2021; Petryshyn et al., 2015; Wang et al., 2021), were reprocessed using current data handling procedures and projected into the I-CDES reference frame, following methods described in Upadhyay et al. (2021) and Bernasconi et al. (2021). Mean sample values are in Table 1, and the replicate level recalculated values can be found in the Supplementary Material. We included modern authigenic carbonate samples from H. Li et al. (2021) that were previously run in the Eagle-Tripati Lab, with identical data processing. For samples and locations described in Li et al. (2021), we propagated the standard error of the regression used to constrain water temperatures and the temperature error reported in the original publication in quadrature to estimate error for lake temperature values. Modern gastropod samples from Wang et al. (2021), which were also run in the Eagle-Tripati Lab, were reprocessed with the Brand parameter set for inclusion in this work. Data from Huntington et al. (2010), Huntington et al. (2015), and Petryshyn et al. (2015) measured in the Eiler Lab at Caltech was reprocessed using Easotope (John & Bowen, 2016), and standard values for these analyses are reported in Supplementary Table 2. Tufas from Kele et al. (2015) run in the Bernasconi Lab at ETH Zürich were previously recalculated with the Brand parameter set by Bernasconi et al. (2018), and additional replicates were recently measured for the calibration in Anderson et al. (2021), thus, we include both sets of measurements on identical samples from both studies. These data presented in Bernasconi et al. (2018) are projected into the I-CDES reference frame using the methodology and new ETH values presented in Bernasconi et al. (2021). Methodology for clumped isotope analysis for these datasets can be found in the publication that initially reported the data. Comparisons to previously published calibration equations are also shown, with data brought into the 90°C reference frame here using AFF values reported in Petersen et al. (2019).
2.5 Regression Methodology
Recent work has shown that models derived using ordinary least squares (OLS) perform better than their error-in-variables counterparts (e.g. York regression, Deming regression), with higher accuracy and precision for regression parameters, and perform similarly to Bayesian tools for both calibration of clumped isotopes and for temperature reconstruction (Román Palacios et al., 2021). In this study, we evaluate the relationship between Δ47 and growth temperature using Ordinary Least Square regression models fit in the lm R function in the stats package (R Core Team, 2022). To evaluate material specificity within our dataset and compare our derived regression parameters to other studies, we utilize an ANCOVA. Specifically, we evaluate pairwise differences in slopes and intercepts between groups of data. We compare our composite calibration along with our material-specific calibrations to four additional studies: a recently published calibration that includes natural and synthetic samples (Anderson et al., 2021), a ‘universal’ calibration created from a synthesis of clumped isotope calibration studies (Petersen et al., 2019), a calibration derived from authigenic lacustrine carbonates (H. Li et al., 2021), and a recalculated travertine calibration (Bernasconi et al., 2018). All data are either in I-CDES or projected to CDES90 using AFF values in Petersen et al. (2019).
3 Results and Discussion
3.1 Seasonality of freshwater carbonate formation
Carbonates precipitate in many lakes and can form in various freshwater environments with the extent and depth of carbonate deposition determined by seasonal changes in water chemistry, as well as water depth, slope gradient, and circulation within the basin (Gierlowski-Kordesch, 2010; Platt & Wright, 2009). Carbonates that form at the lake margin include ooids, beach rock, shelly accumulations of gastropods and bivalves, microbialites, and tufa, while deep-water deposits are comprised largely of carbonate muds and grains (such as ostracods) that accumulate below the storm wave base (Platt & Wright, 2009). In most cases, carbonate accumulation is controlled by seasonal changes in saturation (Anadón et al., 2009; Hren & Sheldon, 2012; Kelts & Hsü, 1978; Street‐Perrott & Harrison, 2013).
Therefore, we aim to use data for samples within our study to evaluate the accuracy of reconstructing the seasonality of formation for different carbonate types. We use the recently-derived calibration from Anderson et al. (2021) that uses a combination of both synthetic and field-collected samples to estimate formation temperature (in this case, lake water temperature for lacustrine samples) using the clumped isotope results presented in this study. These estimates were then compared to predicted values of lake water temperatures for each location using the seasonal lake surface water temperature to mean annual air temperature (MAAT) transfer functions from Hren and Sheldon (2012) and warmest month lake surface water-to-air transfer function from Mering (2015). Riverine and spring samples were excluded from this analysis. MAAT estimates used within these transfer functions are averages of the long-term monthly means from each of our sites from 1981-2010, using the University of Delaware’s high resolution gridded air temperature dataset (Willmott & Matsuura, 2001) provided by NOAA. The analysis described below suggests that application of the calibration published by Anderson et al. (2021) to modern freshwater carbonates can, in some cases, yield inaccurate temperatures, given what is known about the seasonality of growth for different carbonate types.
Figure 2a shows our derived estimates of seasonal lake temperature using the methodology described above for biogenic carbonates. Biogenic taxa precipitate the majority of shell material during a well-defined “growing” season, typically initiating shell calcification during the spring, as long as food availability and water temperature exceed a species-specific threshold value (Gierlowski-Kordesch, 2010; Hren & Sheldon, 2012; Platt & Wright, 2009; Versteegh et al., 2010; Wilbur & Watabe, 1963). Prior research has shown that in the Northern Hemisphere, the April-October interval has been shown to encompass a majority of shell growth for freshwater mollusks (Apolinarska et al., 2015; Versteegh et al., 2010). However, most individual species have a restricted range of water temperatures that they can tolerate and that allows shell formation, thus, it is likely that calcification occurs within a narrower interval sometime between spring and early summer, when water temperatures fall within the species-specific temperature range (Hren & Sheldon, 2012; Versteegh et al., 2010).
When applying the calibration of Anderson et al. (2021), our Δ47-derived temperatures suggest that 37% of our biogenic samples reflect only warmest month water temperatures, implying a relatively narrow window of precipitation for shell precipitation (Fig. 2A). Although water temperature is the primary factor that controls carbonate growth in biotic carbonates, shell formation can be influenced by other environmental factors, such as food availability and primary productivity (Dunca & Mutvei, 2001). Therefore, utilizing a calibration based on both synthetic and natural samples may not capture the appropriate seasonality for freshwater mollusks. Specifically, the Anderson et al. (2021) calibration, when applied to biogenic taxa such as gastropods, may overestimate the water temperature in which the organisms formed, particularly at higher latitudes.
Temperature reconstructions from endogenic carbonate precipitation are typically biased towards the warmest period of the year, when carbonate precipitation is enhanced due to evaporation increasing carbonate saturation and photosynthetic uptake loweringp CO2, thereby increasing water pH (Hren & Sheldon, 2012; Oviatt et al., 1994; Platt & Wright, 2009). In the subtropical and polar Northern Hemisphere this corresponds to June – August, while tropical lakes have less variability in lake temperature resulting from decreased seasonality. Although tropical lakes experience decreased seasonality, seasonal changes in rainfall and evaporation within tropical lakes can also play a role in influencing carbonate saturation state.
The Δ47-temperature calculated using Anderson et al. (2021) for lower latitude micrite samples imply that early Spring (April - June) is the dominant interval for carbonate precipitation for 35% of the samples, instead of intervals of the most elevated temperatures (Fig. 2b). Estimated temperatures for higher latitudes shows carbonate precipitation mainly occurs during the estimated warmest month, and in some occasions Δ47-temperatures exceed the mean warmest month water temperature value. This could imply that precipitation of endogenic carbonate may be elevated in these settings during short intervals of intense heat and evaporation or indicate that the calibration of Anderson et al. (2021) does not work perfectly for all freshwater carbonates. In total, these observations indicate that the calibration of Anderson et al. (2021) yields temperatures that are broadly consistent with known freshwater carbonate growth seasons and systematics within mid-to-high latitudes, but underestimates lake water temperature in some cases and may not reflect what is known about season of growth in approximately half of the lakes examined here.
Biologically-mediated carbonates, including tufas and microbialites, precipitate as a result of local changes in water conditions and biological productivity (Capezzuoli et al., 2014). These carbonates are formed by both abiogenic and biogenic processes, with algae and other aquatic plants influencing their precipitation on organic and inorganic substrates (Capezzuoli et al., 2014; Flügel, 2004). Microbial activity within carbonates can increase rates of photosynthesis, thereby loweringp CO2 and increasing carbonate saturation state making carbonate mineral precipitation more favorable (Pacton et al., 2015; Platt & Wright, 2009; Solari et al., 2010). Additionally, microbial biomass provides a negatively charged surface to which ions may adhere, which locally increases calcium concentration and promotes the supersaturation of carbonates (Fein, 2017). These conditions are enhanced during the warmest interval of the year, when evaporation also plays a role in increasing carbonate saturation, eventually inducing precipitation of shoreline carbonates. Recent field studies of modern tufas and microbialites have shown elevated growth rates during warmer water temperature conditions (Brady et al., 2014; Marić et al., 2020; Pedley, 1990).
Reconstructions of water temperature from biologically-mediated carbonates from this study derived using the calibration of Anderson et al. (2021) (Figure 2c) seem to exceed predictions, with four samples matching or exceeding warmest month water temperatures. Prior work analyzing Δ47 in modern tufas in Japan has shown that tufas are able to discern seasonal changes in water temperature (Kato et al., 2019). Given that these carbonates are continuously calcifying (although precipitation amount is dependent on season), it is unlikely that precipitation solely occurred during the warmest water temperature intervals and that Δ47-derived temperatures should be reflecting a maximum. Therefore, the application of the Anderson et al. (2021) calibration does not appear to be capturing temperature variability that is contained within biologically-mediated carbonates.
Our analysis, on first order, shows that the lake water temperatures derived using the Anderson et al. (2021) calibration often results in unrealistic temperatures for carbonate precipitation for different types of freshwater carbonates. Our results show that the calibration of Anderson et al. (2021), which is dominated by synthetic precipitates and marine carbonates rather than natural samples, may over or underestimate a reasonable temperature range of carbonate formation and may not be ideal for terrestrial paleotemperature reconstructions for all freshwater carbonates. We note that the transfer function-based approach we used (Hren & Sheldon, 2012) does not, for lacustrine carbonates, take into account the size and setting of the water body and potential influences from snowmelt, both of which would impact the difference in temperatures between lake water and overlying air. Thus, in the next section, we derive calibrations that take these factors into account and are specific to different types of terrestrial freshwater carbonates.