Introduction

During the past 50 years, large turnovers in diatom phytoplankton have been observed in some alpine lakes in the western United States (Wolfe et al. 2003; Saros et al. 2005, 2011; Hobbs et al. 2010) and in remote settings throughout the Northern Hemisphere (Rühland et al. 2010, 2013; Enache et al. 2011; Hadley et al. 2013). In many lakes, this shift has been marked by increases in araphid pennate species such as Asterionella formosa, with decreasing cyclotelloid or small fragilaroid diatoms. Causes have been variously attributed to either climate warming (Rühland et al. 2015), increased rates of atmospheric nitrogen deposition (Wolfe et al. 2003; Saros et al. 2005, 2011), and SO2 emissions that affect acid neutralizing capacity (Sickman et al. 2013; Heard et al. 2014), or a combination of these factors (Hobbs et al. 2010; Heard et al. 2014). Lakes of the western United States do not show a uniform response, and in the case of nitrogen deposition, may reflect either low sensitivity, or spatio-temporal variations in exceeding critical loads of nitrogen. Whereas southern and central Sierra Nevada lakes show responses that began in the early half of the 1900s (Heard et al. 2014), these effects are just beginning to manifest in places to the north. In the Lake Tahoe basin, monitoring showed that A. formosa was present in great abundance in several lakes within the watershed (Noble et al. 2013), but as of 2011 sediment cores from these lakes had yet to record increases (Johnson et al. 2017). Annual nitrogen deposition was comparatively low in the Tahoe basin and had just started to exceed critical loads in northern Sierra aquatic systems (< 3 kg N ha−1 yr−1; Fenn et al. 2010). Still farther north, responses of A. formosa and Fragilaria tenera in Olympic National Park, Washington, were interpreted to represent an instance where one lake had exceeded its critical load of atmospheric nitrogen deposition, beginning ~ 1969–1975, but may have been incipient in another (Sheibley et al. 2014). Additional data are needed to evaluate heterogeneity of recent turnovers in the context of individual lake histories, identify the associated drivers, and evaluate available courses of remediation.

This study used diatoms and organic geochemistry (C:N concentrations δ13C, δ15N) from a core taken at Castle Lake, CA, to identify past changes in lake conditions in relation to natural and anthropogenic drivers, and corroborated parts of the anthropogenic signal captured in monitoring data. Castle Lake (Fig. 1) is a dimictic glacially derived cirque lake in the Trinity-Siskiyou Mountains of northern California. It lies at an elevation of 1657 m, has a maximum depth of 35 m, a surface area of 0.20 km2 and a watershed area of 0.81 km2 (Axler and Reuter 1996). The watershed is characterized by dry summers, with 155 cm of annual precipitation coming largely from winter storms, and is forested with patches of conifers, shrubs, and alders in riparian areas (Mejica 2014). The lake is nitrogen-limited and oligo-mesotrophic, with a deep chlorophyll maximum between 15 and 20 m that persists until fall mixing (Priscu and Goldman 1983; Goldman and DeAmezaga 1984). Although naturally fishless, it was stocked to encourage tourism beginning in the early 1900s, and now supports populations of rainbow and brook trout, as well as golden shiners (Elser et al. 1995). Its location in a region of northern California from which there are few data of this nature makes Castle Lake a compelling study site for better understanding past responses to pollution and climate shifts, especially since Community Multiscale Air Quality modelling (CMAQ) simulations of nitrogen deposition show the Trinity-Siskiyou area is a hotspot (9–11 kg N ha−1 yr−1) relative to the surrounding parts of northern California (< 3 kg N ha−1 yr−1), with expectations that deposition will increase (Fenn et al. 2010). The core record provided an opportunity to compare and contextualize any modern shifts to past variations, as this record extends ~ 450 years back, into the Little Ice Age (LIA), a multi-centennial period of cooling and neoglacial advance that occurred in the Northern Hemisphere following the Middle Ages, until glacial recession ca. AD 1400–1850 (Mann 2002).

Fig. 1
figure 1

Bathymetric map of Castle Lake and the location of gravity core 17-CL-2

Full size image

Castle Lake has also been an experimental field station and long-term monitoring site going back to 1957 (Goldman and De Amezaga 1984), and although phytoplankton data collection has not been continuous, it does provide a limited basis for comparison to the core data. From these modern data, the experimental history, and sediment sampling prior to 2017, we hypothesize that increased atmospheric nitrogen deposition is a factor that has influenced the phytoplankton community in Castle Lake over the last 20–50 years. We further hypothesize that if the LIA affected this region, then it should be discernible in the diatom record of Castle Lake. The LIA is best known from climate records and historical accounts in Europe during the 16th to nineteenth centuries (Mann 2002), although original use of the term actually stems from work on neoglacial advances in the Sierra Nevada of California (Matthes 1939). Anomalies in pollen and tree ring records provide evidence of widespread cooling in western North America (Pitman and Smith 2012; Viau et al. 2012), although both the timing of the LIA and regional climate response is known to have varied spatially (Mann et al. 2009; Neukom et al. 2019) and requires additional scrutiny. A LIA diatom response was noted in Fallen Leaf Lake, northern Sierra Nevada mountains, characterized by an increase in Stephanodiscus alpinus and Aulacoseira subarctica (Johnson et al. 2017), and the age of neoglacial advance in the Sierra Nevada corresponds to the latter part of the LIA, ca. AD 1700–1780 (Bowerman and Clark 2011). Factors relating to neoglaciation that can affect diatom assemblages include ice cover, timing of ice melting, changes to lake stratification and mixing (Rühland et al. 2015), and reactive nitrogen additions from glacial runoff (Saros et al. 2011). All these factors should have influenced the diatom community if Castle Lake experienced neoglacial cooling during the LIA.

In 2009, an exploratory 20-cm core was collected from Castle Lake with the goal of determining whether the diatom record showed turnovers that might be related to anthropogenic effects. The top 5 cm of the core showed a sharp increase in the nitrogen-sensitive species A. formosa relative to cyclotelloids (Johnson et al. 2011). A 210Pb age model was attempted on this core, but was unsatisfactory because of coarse initial sectioning (2-cm-thick intervals) that did not extend to a depth where the unsupported/supported 210Pb boundary could be determined. Without a precise age model, it was difficult to correlate the observed diatom changes with the monitoring data set and determine exactly when the shift occurred. In fall 2017, a longer (39 cm) core was taken from the lake and enabled development of an age model using 210Pb, 137Cs, 241Am, and 14C dating techniques.

Materials and methods

Core 17-CL-2 was taken using a Pylonex gravity corer (Renberg and Hansson 2008). The core site was a deep location in the lake, far from the anchors of the Castle Lake Research Station dock, to avoid possible sediment disturbance caused by anchor movement (Fig. 1). The core was extruded in 1-cm increments in the field. In the lab, each section was analyzed for organic matter content (loss-on-ignition) in the University of Nevada Palynology Lab. Homogenized freeze-dried samples were analyzed for δ13C and δ15N at the Nevada Stable Isotope Lab at the University of Nevada, Reno. Samples were first placed into a HCl fumigation desiccator for 30 days to remove carbonate, following the method of Yamamuro and Kayanne (1995). A Eurovector EA 3000 elemental analyzer interfaced to a Micromass IsoPrime stable isotope ratio mass spectrometer was used to generate organic carbon and nitrogen concentrations, δ15N, and δ13C using an acetanilide standard, following the method of Kornexl et al. (1999). Stable isotope results are reported relative to the Vienna PeeDee Belemite (VPDB) standard for δ13C and atmospheric nitrogen for δ15N. Six duplicate analyses were run and the mean of the difference between each duplicate pair of analyses was calculated to establish uncertainty values. The mean of the difference between each duplicate pair was 0.05 ‰ and 0.11 ‰, and one standard deviation of the collective pairs was 0.17 ‰ and 0.21 ‰ for δ15N and δ13C, for both statistics, respectively.

Sediment chronology was derived from the naturally occurring fallout radionuclide 210Pb, with atmospheric nuclear weapons testing radionuclides (137Cs, 241Am) used to test the 210Pb age model. Activities of total 210Pb, 226Ra, 137Cs, and 241Am were measured simultaneously by gamma spectrometry (Baskaran and Naidu 1995; Fuller et al. 1999; Van Metre and Fuller 2009) on homogenized freeze-dried samples at the US Geological Survey Sediment Radioisotope Laboratory in Menlo Park, California. Sediment samples were sealed in 7-mL polyethylene scintillation vials and stored for at least 2 weeks before analysis to enable short-lived daughters of radon ingrowth to attain secular equilibrium with 226Ra. Total 210Pb was determined from the 46 keV gamma emission line, with the supported 210Pb activity, defined by the 226Ra activity, determined on each section from the 352 keV and 609 keV gamma emission lines of the short-lived daughters 214Pb and 214Bi, respectively. Self-absorption of the 210Pb 46 keV and the 241Am 60 keV gamma emission lines was accounted for using an attenuation factor calculated from an empirical relationship between self-absorption and bulk density (Cutshall et al. 1983). Self-absorption of the 214Pb, 214Bi, and the 661.5 keV 137Cs gamma emission lines was negligible. Detector efficiency was determined from National Institute of Standards and Technology (NIST) traceable standards. The reported uncertainty in the measured activity, calculated from the random counting error of samples and background spectra at the one standard deviation level, was typically within ± 10%. The measured activities of replicate analysis of material from the same sample agreed to within ± 15%. Measured activities of 137Cs were decay-corrected for the period between sample collection and analysis.

Conifer needles from three depths were pre-treated for 14C analysis in the University of Nevada, Reno Human Paleoecology and Archaeometry Lab, following the acid-base-acid (ABA) method. Samples were treated with aliquots of 1 N HCl for 30 min, followed by successive 30 min treatments with 1 N NaOH until the liquid remained clear, ending with another 30 min 1 N HCl wash. The samples were then dried, and vacuum sealed in quartz tubes with CuO and Ag. This was followed by combustion at 900 °C. The pre-treated CO2 samples were then submitted to the 14C lab at Penn State University for the final stage of processing and generation of 14C dates. An age-depth model was constructed using 210Pb and 14C data and the software package Plum (Aquino-López et al. 2018). This package was chosen because an age-depth model can be constructed using both 14C and 210Pb data, without pre-modelling the 210Pb dates. Plum takes a Bayesian approach to 210Pb dating, similar to the program BACON (Blaauw and Christian 2011), except that Plum uses the total 210Pb instead of exclusively using the unsupported 210Pb to determine the chronology. This allows the model to use the available data to decide on the supported levels of 210Pb, and also implies that the uncertainty related to the chronology is not as strongly affected by user input. The resulting chronology provides more realistic uncertainties, as shown by simulations in Aquino- López et al. (2018), especially in the lower part of the 210Pb data where age model uncertainties are generally highest.

Twenty-three core samples were processed for diatoms using a 30% hydrogen peroxide solution that was heated for 1 h, then acidified with a few drops of HCl, modified from Battarbee (1986). Samples were allowed to settle and decanted 6 times to remove chemicals, and strewn slides were made using the mountant Zrax. A total of 500 valves were counted per slide using an Olympus BX51 microscope with Differential Interference Contrast (DIC) and a 100× oil immersion lens. Diatoms were identified to the lowest taxonomic level possible using the Diatoms of North America online database and Cantonati et al. (2017). Enumeration data were analyzed through time series graphs and multivariate analyses to determine significant change in the flora throughout the core. To determine the positions of stratigraphic diatom zones, a stratigraphically constrained cluster analysis was applied, using the R package rioja (Juggins 2017), and a broken-stick model was used to determine the number of statistically significant cluster zones (Birks 2012).

A phytoplankton monitoring dataset collected by the UC Davis Limnological Research Station at Castle Lake from 1967–1970, 1972, and 1975–1984, was selected for comparison. Sample collection and enumeration methods are outlined in Jassby and Goldman (1974). Sampling time, frequency, and depth of samples varied within this dataset. During the first 3 years, weekly samples were enumerated from 8 discrete water depths between 0 and 30 m, in the center of the lake (Fig. 1). Subsequent years shortened the monitoring time following ice-out in the spring/early summer, until September and October, and the number of discrete depths also varied to some degree from year to year. To enable consistent intra-annual comparisons, only data collected from depths of 3, 10, 20, and 25 m were compared, as these depths were present for all years except 1978. Additionally, in the interval 1981–84, the depth of 10 m was not sampled, but instead 12.5 m was used, both of which represented the approximate base of the metalimnion (Bachmann and Goldman 1965). The relative abundances of diatom valve counts for each of these depths were tabulated to compare with the sediment core counts. The full 1967–1969 time series was examined to understand seasonal succession and depth distributions, and these data were tabulated from actual valve abundances per ml of water. These are the three most complete years, which contain weekly to near-weekly enumerations from 10 or 11 discrete depths from 0 to 30 m. Pearson correlation coefficients of ice-out dates and valve counts/ml water sample were calculated for the most dominant euplankton taxa to determine the strength and statistical significance of correlations. Pearson correlation coefficients were also calculated for ice-out and the relative abundance of both D. stelligera and the araphid euplankton to aid in interpreting the relative abundance counts from the core.

Results

Core chronology

Total 210Pb activity decreases exponentially with increasing depth to 33 cm, with the difference between supported and unsupported 210Pb less than the combined analytical uncertainty in 210Pb and 226Ra below this depth (Fig. 2). Sediment chronology was obtained using both 210Pb measurements (Electronic Supplementary Material [ESM] Table S1) and 14C dates (Table 1) simultaneously. 137Cs and 241Am profiles were used to validate the age model. In this core, subsurface maxima in 137Cs and 241Am are observed between 6–8 cm and 7–8 cm, respectively (Fig. 2). Measurable activities of 137Cs and 241Am extend to 31 and 14 cm, respectively. The broader maximum and deeper penetration of 137Cs, in comparison to 241Am, likely resulted from the greater post-depositional mobility of 137Cs, independent of sediment particles, compared to 241Am, which is more strongly bound to sediments (Appleby et al. 1991; Winberg and Garcia 1995; Wang et al. 2017). The unsupported 210Pb profile can be affected by changes in accumulation rate, sediment mixing, hydrologic regime, sediment focusing and acidification (Binford et al. 1993). For example, the deviation from the exponential decay at 5–8 cm depth may result from these processes. Smoothing of the curve by the Plum model, however, accounts for this deviation from ideal exponential decay. The age-depth model, derived from 210Pb and 14C dates using Plum is shown in Fig. 3, where the upper panels (Fig. 3a–e) show parameters and results that help evaluate model performance. The resulting posterior distribution for every parameter of the model is shown, including: accumulation rate (Fig. 3b), memory (Fig. 3c), and influx of 210Pb (Fig. 3e). See Aquino-López. et al. (2018) for a description of these parameters. The model was compared to the activity maxima of 137Cs and 241Am at 7–8 cm depth that is assumed to correlate with the maximum fallout from atmospheric nuclear testing in 1963–1964 (Fig. 2). The modelled age of the 7–8 cm section is between 1981 and 1968 (using the range of minimum and maximum age of the 95% uncertainty envelope for the top and bottom depths) with means at 1979 (7 cm) and 1970 (8 cm). The mean modelled age at 8 cm (1970) is about 7 years younger than the 1963 date inferred from the observed 137Cs and 241Am maxima in this core interval. This comparison does not account for possible delay in peak delivery of 137Cs and 241Am to the core site caused by transport from the watershed and from redistribution from shallow areas in the lake, by sediment focusing (Matisoff 2017), which cannot be constrained with the current data set. In contrast, because atmospheric fallout of unsupported 210Pb is continuous or steady state, 210Pb dating is not affected by these processes delaying input to the core site (Davis et al. 1984). For these reasons, given the uncertainty in all age models because of possible mobility of the fallout radioisotopes, changes in hydrologic regimes, measurement error, and sampling uncertainties, we assumed that the presented age model is the best available with the current data.

Fig. 2
figure 2

Down-core activity profiles for 210Pb and 226Ra (Left), and 137Cs and 241Am (Right). Horizontal error bars depict one standard deviation uncertainty in measured activities. Symbols are centered on mid-depth of each 1-cm-thick core interval

Full size image
Table 1 Carbon-14 samples and dates. Uncertainty is given as ± 1σ. 14C ages (BP) calibrated using CalPal_2007_HULU calibration curve
Full size table
Fig. 3
figure 3

Age-depth model obtained using Plum (Aquino- López et al. 2018). a log of objective function used in Bayesian methods to check the convergence of the method. b, c, and d comparison of prior (green lines) and posterior (gray areas) distribution for the overall accumulation rate, the memory of the model, and influx of 210Pb respectively. e supported levels of 210Pb at every depth, estimated by Plum. f resulting age-depth model with a red dashed line and its 95% confidence intervals (in black dashed lines). Blue plots show the calibrated 14C dates, green and red lines show the 210Pb and 226Ra profiles, respectively. Model parameters used are supplied in b-e (red text). (Color figure online)

Full size image

Uncertainty in the age model increases with depth in the core. Near the bottom of the core, at a depth of 36 cm, the minimum age of 1662 may be as old as 1567 and has a median age of 1615 ± 48 years, but at 32 cm depth, the median 1736 horizon ranges from 1769 to 1694 for an average uncertainty of ± 38 yrs. Near the top of the core, the 8-cm horizon ranges from 1968 to 1972 and has a median age of 1970 ± 2 (ESM Table S2).

Organic geochemistry

Down-core profiles of % weight organic nitrogen, carbon, and C:N are shown in Fig. 4. Nitrogen values range from 0.7–1.3% weight and organic carbon from 9.3–14.4%, with C:N ranging from 11 to 15. Uncertainty values from replicate analyses, given in parentheses as mean difference between values and mean % difference, are as follows: δ15N (0.2, 16%) δ13C (0.2, 0.7%), weight % nitrogen (0.1, 6%), weight % carbon (0.4, 3%). The mean values of each were calculated for 3 time intervals, pre-1900, 1900–1950, and post-1950, and are shown in Fig. 4, with the pre-1900 value used as a baseline. Visual inspection of down-core plots shows a change in values beyond the range of calculated uncertainties, with organic matter concentrations increasing and C:N decreasing beginning in the 1900s and accelerating after 1950 (Fig. 4). The visual change seen ~ 1900 is small and within the range of uncertainties calculated, yet is part of a trend towards accelerated changes post-1950, which do exceed the uncertainty values (Fig. 4). Similarly, plots of δ15N and δ13C show a change in values beginning in the 1900s and accelerating post-1950, with a trend towards less negative δ13C and less positive δ15N (Fig. 5).

Fig. 4
figure 4

Down-core plots of % weight organic nitrogen, carbon, and C:N for each of the 39 1-cm-thick sections analyzed. Mean values for pre-1900, 1900–1950, and post-1950, with calculated uncertainties, appear next to each profile separated by solid lines. Light gray shaded area near the top of the plot includes 3 values reversed from the overall post-1950 trend, interpreted to be the result of a whole-lake fertilization experiment conducted in the early 1980s. The time of neoglacial advance in the Sierra Nevada (dark-shaded area mid-plot) is from Bowerman and Clark (2011). (Color figure online)

Full size image
Fig. 5
figure 5

Down-core profiles of δ15N and δ13C for each of the 39 1-cm-thick sections analyzed. Mean values for pre-1900, 1900–1950, and post-1950 with calculated uncertainties appear next to each profile. See Fig. 4 For an explanation of shaded gray areas. (Color figure online)

Full size image

Diatom monitoring

Phytoplankton monitoring data (Table 2) indicate that the predominant diatom species throughout the monitoring period were D. stelligera, F. tenera grp. (reported as Synedra radians), and A. formosa. There was large inter-annual variability in F. tenera grp. (ESM Table S3), with wide swings from 0 to 59%. Other araphid taxa showed high abundances during a given year, but were otherwise a small component (< 1%). A more detailed view of the seasonal and vertical distribution from 1967–1969 weekly data show a succession of D. stelligera->F. tenera grp.-> A. formosa (Fig. 6). Peak abundances of D. stelligera occurred during the spring bloom, with the highest concentrations in the surface waters. Seasonal peaks of A. formosa and F. tenera grp. occurred during summer, with greatest concentrations within or below the deep chlorophyll maximum (Fig. 6; ESM Table S4). Pearson correlations for actual abundances based on number of valves/ml sample, showed a negative correlation between D. stelligera and ice-out (Pearson correlation coefficient =  − 0.62), such that earlier ice-out dates were associated with greater numbers of D. stelligera. A test of significance (p < 0.05) found that the correlation between ice-out and number of D. stelligera valves/ml was significant (p = 0.014). There were negative correlations for the actual abundance of araphids and ice-out (A. formosa =  − 0.48, F. tenera grp. =  − 0.29), but neither of these correlations was significant. In contrast to the actual abundance, a Pearson correlation of the relative abundance of D. stelligera and ice-out showed a significant positive correlation (r = 0.58, p = 0.023), indicating that its abundance relative to the other constituents in the assemblage decreases with earlier ice-out dates. Relative abundance counts are dependent on changes in the proportion of the other taxa present, unlike actual abundance counts. Thus, while increases in the actual numbers of D. stelligera valves significantly correlate to earlier ice-out, this signal is belied by increases in the other components of the sample, especially the araphids, so that the D. stelligera appeared to decrease with earlier ice-out when measured in relative abundance. The positive correlation between increased relative abundance of D. stelligera and later ice-out becomes useful in helping to explain relative abundance data in the core.

Table 2 Relative abundance of common diatoms (> 1%) from phytoplankton tows, Castle Lake deep station, reported as % abundance of total counts per sample
Full size table
Fig. 6
figure 6

Stacked histograms plotting valve count abundance per ml of water sampled from discrete water depths during 1967 through 1969. Seasonality and depth distribution are seen for the three most abundant euplanktonic species. For simplicity, only four depths are plotted: 3 m (epilimnion), 10 m (base of metalimnion), 20 m and 25 m (hypolimnion). See ESM Table S4 for data

Full size image

Diatoms in the core

Diatom assemblages over the past 450 years have been dominated by cyclotelloid taxa, with a substantial component of araphids (Fig. 7). The cyclotelloid D. stelligera dominated throughout the core, comprising between 33 and 78% of the assemblage. The centric Pantocsekiella ocellata grp. varied from 2% near the top of the core to 12% ~ 200 years ago. The primary araphid, Staurosirella pinnata, ranged from 3–29% with increases at the top and bottom of the core. The elongate araphid euplankton A. formosa and F. tenera grp. both increased during the last ~ 60 years in the core record, with spikes of F. tenera grp. (8%) occurring about 50 years ago and A. formosa (9%) in the last 10 years (ESM Table S4). A stratigraphically constrained cluster analysis indicates that there are three statistically significant diatom zones in the 450-year record (Fig. 7). The largest break, indicated by the cluster dendrogram, occurs at 8 cm, separating Zone 3 from the other zones.

Fig. 7
figure 7

Diatom stratigraphy of Castle Lake showing three diatom species zones (1-Little Ice Age, 2-Transitional Zone, 3-Anthropogenic Zone), with distributions of common diatom taxa. A stratigraphically constrained cluster dendrogram is shown to the right

Full size image

Discussion

Diatom stratigraphy

Zone 1: AD 1560–1736 (39–32 cm), neoglacial zone

Diatom Zone 1 corresponds temporally to the middle of the LIA, with the uncertainty envelope for the top of this interval (32 cm) at ~ 75 years (AD 1694–1769). Zone 1 is distinguished primarily by higher abundances of small araphid chain-forming fragilarioid taxa, especially S. pinnata (Fig. 7), which commonly grow in benthic habitats, attached or entangled with macrophytes in littoral zones, but may be tychoplanktonic, i.e. plankton entrained into the water column from benthic littoral habitats. These taxa can be abundant in environments with short growing seasons and low light attenuation, linked to prolonged ice cover (Rühland et al. 2015; Griffiths et al. 2017) and may represent an interval of neoglacial advance at Castle Lake. Within the uncertainties of the age model, this time interval is coincident with the neoglacial maximum in the southern Sierra Nevada, which is interpreted to have peaked about AD 1700–1780 (Bowerman and Clark 2011).

Zone 2: AD 1736–1970 (32–8 cm), intermediate zone

The base is marked by a rapid increase in the abundance of small euplanktonic cyclotelloids D. stelligera and P. ocellata grp. and a corresponding decrease in the small benthic araphids (Fig. 7). D. stelligera begins to decline near the top of Zone 2 after its peak abundance ca. 1932, and benthic araphids also begin to increase in abundance at that time. The highest totals of Aulacoseira occur in this zone from the base to the middle of the zone. Small euplanktonic cyclotelloids like D. stelligera bloom in spring to early summer (Fig. 6) and are generally successful in lakes that exhibit an extended ice-free period during the summer months (Hobbs et al. 2010; Rühland et al. 2015), and recent increases in cyclotelloids and decreases in small benthic littoral araphids in Arctic lakes have been linked to warming and climate change (Rühland et al. 2015), as a result of earlier ice-out dates, increasing spring temperatures, and longer periods of spring turnover (Wiltse et al. 2016). These boreal lake responses provide a good analog for cyclotelloid increases at Castle Lake following the LIA.

Zone 3: AD 1970–2017 (8–0 cm), anthropogenic zone

This zone is distinguished by an increasing predominance of araphid phytoplankton. F. tenera grp. reaches its peak abundance of 8% (8–7 cm core depth), succeeded by peak abundance in A. formosa (2–1 cm), ~ 20–30 years afterward (Fig. 7). Prior to these increases, neither species reached an abundance > 0.25%, signifying more than a tenfold increase. Accompanying the araphid increase is a decline in abundance of D. stelligera and P. ocellata, which reach their minimum abundances at the top of the core (Fig. 7). The D. stelligera and P. ocellata decline in abundance and reach a minimum of 33% at the top of the core. This shift is interesting because both F. tenera grp. and A. formosa have been associated with multiple stressors, including climate warming and increased rates of atmospheric pollutants. The ice-out correlations for A. formosa ( − 0.44) and F. tenera grp. ( − 0.30) were negative, suggesting a preference for longer growing seasons, but these correlations, however, were not significant at the 95% confidence level (i.e. p > 0.05), suggesting multiple factors were at work.

Monitoring data covering this same period show a correlation between total numbers of D. stelligera and earlier ice-out. D. stelligera peaks earlier in the season than the araphids, increasing its abundance rapidly after ice-out, and is most dominant in the uppermost part of the epilimnion, whereas the two araphids are most dominant in the deep chlorophyll maximum, later in the season (Fig. 6). The decline in D. stelligera, coinciding with a successive rise in araphid euplankton, has been observed in lakes throughout the western US and can be associated with the combined effects of increased temperatures and nutrient availability caused by post-industrial-age pollution (Wolfe et al. 2003; Saros et al. 2005, 2011; Hobbs et al. 2010; Hundey et al. 2014). Observations from monitoring data support the proposition that warming temperatures and more prolonged open-water seasons in montane dimictic lakes favor increases in araphid euplankton.

Organic geochemistry and nutrient experiments

C:N ratios range from 11 to 14 and indicate that aquatic algae are a more dominant component of sediment organic matter than terrestrial organic matter from the surrounding watershed. The post-1950s trend in C:N indicates an increasing algal component commencing in the mid-1900s (Fig. 4). As a basis for comparison, C:N values for aquatic algae range from 6 to 9 (Meyers and Lallier-Vergès 1999) and C:N values of organic matter in soil range from 22 to 27 in northern California forests with comparable elevation and vegetation (Maxwell et al. 2018). The δ15N values also show a trend towards more negative values after 1950. Wolfe et al. (2001) noted the same directional decrease in alpine lakes in the Colorado Front Range and interpreted the cause to be increased atmospheric nitrogen deposition from strongly negative anthropogenic NH3 sources. Wet deposition values of δ15N derived from these sources, while highly variable because of post-emission fractionation processes, range from 0 to  − 20‰ (Stratton et al. 2019) and are a plausible explanation for the negative trend. Values of δ13C also show an increasing negative trend in the twentieth century and may be related to isotopic depletion of atmospheric CO2 from fossil fuel burning (Keeling 1979). Since 1880, there has been a ~ 2‰ decline in atmospheric CO2 δ13C values (Francey et al. 1999) that has been incorporated into terrestrial and aquatic organic matter (Long et al. 2005).

In addition to the twentieth century trends, there are two noteworthy geochemical excursions, one at 38 cm and another at 8–6 cm. The first occurs during the LIA (38 cm, AD ~ 1585–1603), as seen in the nitrogen concentration and C:N plots (Fig. 4), the δ13C, and to a lesser extent, the δ15N (Fig. 5). These trends may be related to watershed processes that controlled nutrient sources and inputs to the lake. There is a strong positive excursion of ~ 2‰ in δ13C at 37 cm, also during the LIA, but an explanation for this shift is presently unknown. The second noteworthy excursion occurs near the top of the core between 8 and 6 cm, with a peak in nitrogen and carbon concentrations centered in the 8–7 cm sample (mean age = AD 1970–1979), and a peak in δ15N and δ13C centered in the 7–6 cm sample above (mean age = AD 1979–1987). Adding in the age uncertainty at the 95% confidence level from the Plum model (ESM Table S2), the 8–7 cm sample represents a 9-year interval within AD 1968–1981 and the 7–6 cm sample represents a 9-year interval within AD 1977–1989. This second excursion is much smaller than the LIA excursion, and in the case of the nitrogen proxies, falls close to the level of noise within the calculated uncertainty. If one considers the 8–6 cm excursion beyond the level of background variability in organic matter deposition processes, then one possible explanation can be found in the literature regarding Castle Lake experimentation.

In the 1980s, whole-lake and 15N uptake experiments were conducted at Castle Lake with the goal of better understanding nutrient limitation and nitrogen cycling in oligotrophic lakes (Axler et al. 1984; Axler and Reuter 1996). In the summers of 1980 and 1981, the epilimnion was enriched with a 227 kg treatment of ammonium nitrate applied over a 6-h period as a concentrated solution, resulting in an increase in N concentration of 36 μg/liter for both NH4 and NO3 in the upper 5 m of the water column (Axler and Reuter 1996). Depletion to background levels took between 4 and 5 weeks and was hypothesized to have been largely attributable to uptake by benthic taxa (specifically epipelic diatoms and cyanophytes) and denitrification (Axler and Reuter 1996). We propose that the negative δ15N excursion may have resulted from fractionation during this denitrification. In contrast to the benthic response, the phytoplankton response was estimated to have been much smaller (Axler and Reuter 1996).

Scrutiny of the diatom data from both the core and phytoplankton monitoring dataset show subtle responses that may be related to the nitrogen experiments at Castle lake. The monitoring data show that the nitrogen experiment was most strongly manifested by an increase in S. pinnata (during year 1) and T. fenestrata (year 2), benthic forms that may have increased in the littoral zones of the lake, and then became entrained in larger numbers in the phytoplankton tows. In the core record, the 8–6 cm interval corresponds to the first abundance of Tabellaria fenestrata exceeding 1%, and increases in some of the small fragilarioids, including S. pinnata (Fig. 7). This is the same interval in which the δ15 N and δ13C excursions are centered (Fig. 5). The T. fenestrata response is very subtle and would easily be overlooked in the core data, except for prior knowledge from monitoring data summaries, which showed a spike in the T. fenestrata in 1981, when it reached 7.5% of the diatom phytoplankton counts (Table 2). In the years 1967–1979, and again afterward, in 1984, T. fenestrata represented 0.1 to 0.4% of the counts. The short-lived T. fenestrata spike was more muted in the sediment core, by virtue of the ~ 9-year averaging effect in that 1-cm-thick sediment slice. Nonetheless, the increase in T. fenestrata was discerned in the core during the time frame of the epilimnial experiment and organic geochemical excursions. The highest % of T. fenestrata in the sediment core was observed later, in the 2–1 cm sample, representing 2009–2014. Unfortunately, there are no monitoring data from the same time frame to compare with and corroborate the increase. A. formosa, which in the core increased steadily over time, increased beginning in the mid-1970s in the monitoring data, but with high interannual variability thereafter (Table 2).

The relationship between the increases in A. formosa and T. fenestrata relative to nitrogen addition in Castle Lake largely hinges on synchrony of diatom response with the nitrogen experiment. One other piece of information that supports a correlation with rapid nitrogen addition, however, comes from an experimental study by Suttle et al. (1987), who showed both S. radians (= F. tenera grp.) and T. fenestrata responded to nutrient additions, but that T. fenestrata dominated cultures with less frequent additions under P-limiting conditions. This result would support increased dominance of T. fenestrata following single large epilimnetic treatments and also suggests that continued increase of T. fenestrata in Castle Lake following the 1980s experiments may be related to episodic nitrogen delivery, such as that achieved by nutrient pulses delivered through spring snowmelt (Noble et al. 2013).

Drivers of change

Castle Lake is a good example of an aquatic system that is affected by multiple drivers that elicit varied responses. Ice cover is considered to be an important factor in explaining interannual variability in primary productivity and seasonal succession at Castle Lake today (Jassby et al. 1990). During the cooler climate regime of the LIA, there was likely more prolonged ice cover than the last century, and this was accompanied by a lower ratio of euplankton to benthic taxa, especially small benthic fragilarioids. Prolonged ice cover and enhanced mixing would have suppressed the spring cyclotelloid bloom during the LIA, in favor of the small benthic fragilarioids. Cyclotelloids began to increase relative to benthic taxa in the mid-1700s as the climate warmed (Fig. 7). Similarly, a long-term monitoring set from boreal lakes in Canada links an increased abundance in D. stelligera to a warming climate and earlier ice-out (Wiltse et al. 2016). In the boreal Canadian lakes, as well as numerous other Northern Hemisphere lake records over the last 150 years, small cyclotelloids are seen replacing tychoplanktonic species that require mixing to entrain them in the water column (Rühland et al. 2015). In contrast, over the last 50 years Castle Lake has experienced an overall decline in relative abundance of cyclotelloids. Interannual variation in the monitoring data points to a significant correlation between a higher absolute abundance of D. stelligera and earlier ice-out dates, illustrated in the 1967–1969 data showing that the year with the earlier ice-out date had larger numbers of D. stelligera (Fig. 6). The numbers of D. stelligera were greater in 1968, which had ice-out earlier, compared to 1967 and 1969 (Fig. 6, Table 2). This relationship appears inverted when comparing ice-out to relative abundance data in the monitoring and core counts. The A. formosa absolute abundance was also very high in 1968, which depresses the relative abundance of D. stelligera (Fig. 6) and accounts for the positive Pearson correlation between ice-out and relative abundance counts of D. stelligera in the monitoring data. Similarly, the core data show a decline in relative abundance in D. stelligera in zone 3 (Fig. 7) coincident with an increase in araphid relative abundance. It appears that the increased D. stelligera signal generated by earlier ice out is overprinted by the increased A. formosa signal over the last 50 years when examining relative abundance data, such as core data.

Nutrient addition appears to be another factor contributing to recent change, including the lower ratio of euplankton to benthic taxa from the 1970s onward. The nitrogen experiments provide convincing evidence of the sensitivity of Castle Lake to nitrogen stimulation, with phytoplankton showing a more muted response relative to the growth of benthic taxa. The second year of nitrogen addition had a stronger phytoplankton response than the first year (Axler and Reuter 1996) and perhaps represents a lag in establishing the algal assemblage that could best take advantage of the altered trophic conditions. Isotope data point to changes in nitrogen source, consistent with increased atmospheric nitrogen deposition in the Trinity-Siskiyou Mountains, as modelled in the CMAQ simulations of the Trinity-Siskiyou area (9–11 kg N ha−1 year−1) relative to surrounding parts of northern California (< 3 kg N ha−1 year−1; Fenn et al. 2010). Both A. formosa and F. tenera grp. have been shown to be responsive to nitrogen stimulation, yet there is a sufficient body of literature that points to a complex of factors, including warming and acid neutralizing capacity, that can account for these recent increases. The negative effects of earlier ice out on D. stelligera mentioned previously may have had a synergistic effect with nitrogen stimulation in driving the euplankton signal towards increased araphids.

Another anthropogenic factor that remains unexplored, yet is worth mentioning, is the effect of fish stocking. Near the top of Zone 2, the decline in cyclotelloid abundance coincides with the introduction of fish to the lake by the U.S. Forest Service in the 1930s. From the late 1950s until the early 1980s the lake was stocked annually with ~ 10,000 rainbow trout fingerlings in late summer (Elser et al. 1995). Following the top-down hypothesis, as apex predators, fish often control the trophic state of a lake system (Carpenter and Kitchell 1993). Fish directly restructure the food web by severely reducing the large-bodied zooplankton in a lake system (Tiberti et al. 2014), as well as indirectly causing an increase in nutrient content through sediment resuspension. Diatoms have been shown to be sensitive to the trophic changes associated with the introduction of fish into a previously fishless system (Drake and Naiman 2000; Sienkiewicz and Gąsiorowski 2016). Further analysis of historical data on fish from Castle Lake may clarify their importance as a driver in the diatom community change.

Conclusions

Castle Lake provides a new and interesting record of environmental change over the last 450 years in response to both natural climatic and anthropogenic drivers, and can be summarized as follows:

  • Diatom records, especially ratios of euplankton to benthic taxa, support neoglacial cooling that ends in the mid-1700s, with the lattermost part of the Little Ice Age (late 1700s to mid-1800s) showing decreased periods of ice cover. This is supported by a significant positive correlation between earlier ice-out and increased absolute abundances of D. stelligera in monitoring data. The timing of neoglacial cooling in the Trinity-Siskiyou Mountains appears coeval with that in the southern Sierra Nevada (Bowerman and Clark 2011).

  • The first anthropogenic influences on Castle Lake began in the early 1900s, about the same time fish stocking began, as did increased inputs of air-borne industrial pollutants, both of which may have had small, but discernible impacts during the early course of anthropogenic change. By the time Castle Lake Research Station was established in the 1950s, the lake was already undergoing change from its former pristine state.

  • Changes in diatom and geochemical variables accelerated in the 1970s. The diatom and δ15N records in the core point strongly toward atmospheric nitrogen deposition as a significant driver for the twentieth–twenty-first century trend, which may have been accompanied by increased periods of summer stratification during years with decreased ice-cover. A. formosa and F. tenera grp. dominate in the response to these drivers, such that any increased abundances in D. stelligera relating to decreased ice cover are overprinted in the relative abundance data.

  • Castle Lake’s whole-lake nitrogen fertilization experiment in the early 1980s provides additional evidence that the lake responds to nitrogen stimulation through increased growth of benthic algae, with a more muted response in phytoplankton (Axler and Reuter 1996). The 1980–1981 experiment produced a sufficiently large signal to be detected in the core geochemistry and diatom record, but more strongly in the δ15N than the diatoms.