Hydroclimate Variability Affects Habitat‐Specific (Open Water and Littoral) Lake Metabolism

Lakes integrate the signals of changing hydroclimate in their surrounding basin, which substantially influence gross primary production (GPP), respiration (R), and net ecosystem production (NEP). Most research focuses only on the changes to the open water habitat despite the littoral habitat's importance to lakes and its sensitivity to hydroclimatic variability. We analyze how years with different ice‐out dates and snow water equivalent (SWE) triggered different metabolism responses in the open water and littoral habitats of a subalpine lake. The dry (early ice‐out date and low SWE) and wet (late ice‐out date and high SWE) years had lower GPP and R rates in both habitats compared to the average hydroclimatic year. Furthermore, consumer biomass decreased during dry and wet years. GPP and R reduced the most in the littoral habitat. Consequently, the relative contribution to total lake GPP and R of the littoral habitat decreased, and the contribution of the open water habitat increased during the dry and wet years. We demonstrate that hydroclimatic conditions affect productivity and consumer biomass and show that within‐lake habitats do not experience equivalent responses to climate forcing. Our study has implications for how ecosystem scientists and managers quantify the absolute and relative contributions of the littoral habitat to whole lake production in the context of climate variation.


Introduction
As sentinel ecosystems, lakes integrate the loading and environmental processes from their surrounding landscapes, reflecting the hydroclimatic contributions from precipitation and temperature (Adrian et al., 2009;Williamson et al., 2008Williamson et al., , 2009). In addition, more frequent extreme weather events (e.g., severe and prolonged droughts to intense precipitation in a short period of time) can influence the hydroclimatic exposure of lakes (Melillo, et al., 2014;Williamson et al., 2016). Lakes exhibit complex responses to these extreme weather events, which influence the supply of nutrients to the planktonic ecosystem or alter food web connections (Adrian et al., 2009;Shimoda et al., 2011). For example, warming conditions and changes in precipitation increases lake primary and secondary productivity by increasing lake heat content while at the same time reducing accessibility to productive lake habitats by fishes, which are sensitive to warming waters (Caldwell et al., 2020).
Lakes are spatially heterogeneous ecosystems with nearshore littoral and open water habitats providing different metabolic pathways for autochthonous energy sources (Devlin et al., 2016;Vadeboncoeur et al., 2003; Abstract Lakes integrate the signals of changing hydroclimate in their surrounding basin, which substantially influence gross primary production (GPP), respiration (R), and net ecosystem production (NEP). Most research focuses only on the changes to the open water habitat despite the littoral habitat's importance to lakes and its sensitivity to hydroclimatic variability. We analyze how years with different ice-out dates and snow water equivalent (SWE) triggered different metabolism responses in the open water and littoral habitats of a subalpine lake. The dry (early ice-out date and low SWE) and wet (late ice-out date and high SWE) years had lower GPP and R rates in both habitats compared to the average hydroclimatic year. Furthermore, consumer biomass decreased during dry and wet years. GPP and R reduced the most in the littoral habitat. Consequently, the relative contribution to total lake GPP and R of the littoral habitat decreased, and the contribution of the open water habitat increased during the dry and wet years. We demonstrate that hydroclimatic conditions affect productivity and consumer biomass and show that within-lake habitats do not experience equivalent responses to climate forcing. Our study has implications for how ecosystem scientists and managers quantify the absolute and relative contributions of the littoral habitat to whole lake production in the context of climate variation.
Plain Language Summary Lakes integrate climate variability at the catchment scale. These large-scale changes influence the food production and consumption (i.e., metabolism) of lake habitats both at the nearshore and open water. While the shallow nearshore habitat is recognized as an important site for food production, the offshore habitat remains the focus of many lake studies We analyzed how years with different climatic conditions (dry and wet) affected metabolism in the nearshore and offshore habitats of a mountain lake. During dry and wet years, the food production and consumption in both habitats was lower compared to a year with average climatic conditions. Furthermore, animal biomass decreased during dry and wet years and the nearshore bottom habitat was more sensitive to changes with a large reduction in food production. We demonstrate that climatic conditions lead to differential impacts within-lake habitats. Thus, lake management may be deficient if only one habitat is studied.

SCORDO ET AL.
Zanden & Vadeboncoeur, 2020). The littoral habitat gross primary production (GPP) and ecosystem respiration (R) can be seven times higher compared to the open water habitat (Van de Bogert et al., 2007 and represent up to 50% of the total lake production in large lakes (Brothers et al., 2016). However, the littoral habitats in lakes show lower resilience to changes in hydroclimatic conditions, as they are proximate to the land's edge and have shorter retention times of water, nutrients, and detritus than open water habitats (Jeppesen et al., 2014;Peters & Lodge, 2009). The nearshore littoral habitats are also important for fishes as the benthic macroinvertebrates production is generally higher and more efficiently passed to fish than zooplankton production (Vander Zanden et al., 2007. Despite the importance of lake littoral habitats and their sensitivity to hydroclimatic variability, most monitoring programs and research continues to focus on the open water habitat (Cantonati & Lowe, 2014;Vadeboncoeur et al., 2014Vadeboncoeur et al., , 2021. Hydroclimatic conditions modify the balance between GPP and R in lakes (Ask, et al., 2012;Sadro & Melack, 2012), which affects whether an ecosystem serves as a source or sink for atmospheric carbon dioxide (Dodds & Cole, 2007;Engel et al., 2018;McCullough et al., 2018;Tranvik et al., 2018). For example, in temperate oligo-mesotrophic lakes, increasing run-off due to intense precipitation shifts lakes to negative net ecosystem production (NEP; quantified as GPP-R) during the summer when GPP is expected to dominate (Ask et al., 2012;Sadro & Melack, 2012). Greater run-off increases allochthonous organic compounds entering the lake, which stimulates heterotrophic bacterioplankton activity (Ask et al., 2012;Karlsson et al., 2007;McCallister & Del Giorgio, 2008). At the same time, suspended particles reduce light penetration and decrease benthic periphyton GPP (Ask et al., 2009(Ask et al., , 2012Vadeboncoeur et al., 2014). Increasing water temperature enhances GPP at the beginning of the growing season while providing higher potential growth rates for the zooplankton and benthic macroinvertebrates communities, which in turn, reduces GPP by the end of the summer season (Caldwell et al., 2020;Park et al., 2004). Therefore, understanding the role of lakes in processing carbon and whether they serve as net carbon sinks or sources to the atmosphere relies on understanding the drivers of a lake's metabolic processes and how those drivers vary under different hydroclimatic conditions (Staehr et al., 2012;Tranvik et al., 2018).
Climate and ecoregion characterize the influence of hydroclimate on open water productivity. Ice out and snow water equivalent (SWE) influence important drivers of carbon processing (i.e., water temperature, nutrient and light availability, biomass, and community composition of consumers) in western mountain lakes of the United States (Melack et al., 2020;Sadro et al., 2018). Accumulated SWE at the start of ice break-up functions as a surrogate for lake inflow during the spring (Goldman et al., 1989). Ice-out date and SWE at the start of ice break-up have been shown to affect inter-annual open water productivity, including the amount and timing of peak GPP and R (Goldman et al., 1989;Jassby et al., 1990;Park et al., 2004;Strub et al., 1985). When the ice break-up period starts, water influx flushes nutrients and algae from the lake, altering the initial conditions for summer production (Goldman et al., 1989). Higher accumulated SWE results in a stronger flush (Goldman et al., 1989;Melack et al., 2020). Late ice-out date lowers heat content during the summer and shortens recovery time for primary producers, resulting in lower open water productivity (Goldman et al., 1989;Park et al., 2004;Strub et al., 1985). Early ice-out allows producers more recovery time and the lake has more time to incorporate heat, resulting in higher metabolic rates (Goldman et al., 1989;Park et al., 2004;Strub et al., 1985). However, a long ice-free season may result in nutrient depletion and a high consumer grazing rate (Carter & Schindler, 2012;Jassby et al., 1990;Park et al., 2004), which will affect the open water pelagic primary producers, causing metabolic rates to decrease. Comparatively, little is known about the effects of variable hydroclimatic regimes on the littoral habitats within lakes or how yearly variability in hydroclimate may affect the relative contribution of the littoral habitat compared to the open water from a standpoint of GPP, R, and NEP.
Lake metabolism provides a framework for understanding the balance of carbon fixation (GPP) and oxidation (R) in lake ecosystems. There is a long history of quantifying metabolism in lakes to understand the spatial and temporal variation within and among lakes, drivers of metabolism at different spatial scales (Alfonso et al., 2018;Hanson et al., 2008;Idrizaj et al., 2016;Staehr et al., 2012;Stefanidis & Elias, 2019). However, quantifying how years with different hydroclimatic conditions influence metabolism in littoral and open water lake habitats has not been addressed to date despite the relatively larger contribution of littoral habitats to whole lake function. With changes to hydroclimate due to anthropogenically driven climate change and the repeated observation that littoral production contributes to whole ecosystem function, studies are warranted that describe hydroclimatic 10.1029/2021WR031094 3 of 15 connections to these habitats, and how changing metabolic rates affect energy flow to consumers (Caldwell et al., 2020;Rüegg et al., 2021).
In this study, we analyze how years with different hydroclimates, specifically ice-out dates and SWE, triggered different metabolism responses in the open water and habitats of a subalpine lake. This paper (a) quantifies the absolute and relative metabolic rates in the littoral and open water habitats in years characterized by dry, average, and wet hydroclimates; (b) identifies if littoral and open water habitat-specific metabolism have differential responses to different hydroclimate conditions; (c) discusses the influence of metabolic changes on the invertebrate community and their potential feedbacks to our metabolic calculations, and (d) provides a conceptual model that incorporates our findings which can be tested in other lake ecosystems.

Study Sites
We carried out the study at Castle Lake, a meso-oligotrophic, dimictic, subalpine lake (1,646 m.a.s.l., surface area of 0.2 km 2 , max depth = 35 m, and mean depth = 11.4 m) located in northern California, USA (41°13′ N, 122°22′ W; Figures 1a and 1b). Castle Lake is seasonally ice-covered (typically end of November to May), with an ice-free summer season that averages 135 days. We calculated metabolism (GPP, R, and NEP) during August and September (summer) because productivity reaches its peak in Castle Lake during this period.
We quantified metabolism in a nearshore littoral habitat and in the epilimnion of the open water habitat at an index station at the deepest point of the lake (Figure 1c). The index station is part of a long-term monitoring program whereas the littoral collections have only started in the last half decade. The nearshore littoral habitat depth ranges from 0 to 6 m, and its extension from the coast varies between 10 meters (southwest of the lake) and 100 meters (northeast of the lake). The lake littoral metabolism was measured at the north-east region of the lake at 30 meters from the coast. The littoral habitat contains some submerged macrophytes, woody debris, and sediments composed of clay-silt with 13% organic matter. Alder trees (Alnus tenuifolia Nutt.), the predominant vegetation near this site, provide the lake with nitrogen (Goldman, 1961), which is the main nutrient limiting open water phytoplankton production in this lake . We quantified the influence of different hydroclimatic years (dry 2015; wet 2019; average 2016), on summer metabolic rates. We defined the hydroclimate conditions of dry, wet, and average based on the ice-out date and the snow water equivalent (SWE) at the time of ice-out each year (Table 1). We defined the early, average, and late ice-out date based on Castle Lake's 58-year ice-out date data, which ranged from February 20 to June 27, with a mean ice-out date on May 9. Daily time lapse photographs were used to determine the ice-out (80% of the surface free of ice) date each year. SWE (https://nsidc.org/data/g02158) was recorded when ice break-up started based on the above photographic time lapse records, which typically started one month prior to the lake being completely ice free.

Metabolism Model Data Sampling
The metabolism model used in this study is based on high-frequency measurements of dissolved oxygen (mg L −1 ), water temperature (°C), photosynthetically active radiation (PAR; μmol m −2 s −1 ), wind speed (m s −1 ), and barometric pressure (mbar). Dissolved oxygen and water temperature were measured every 10 min at a depth of 3 m at both the sites using an optical oxygen probe (PME miniDOT Logger). Wind speed and PAR were measured hourly at a weather station (Weatherhawk) on the shore of Castle Lake located 4 m above ground level. Light (PAR; 400-700 nm) and temperature profiles (every 0.5 m) were obtained weekly using a Biospherical Instruments 2104P radiometer at an index station in the open water at the deepest part of the lake. The raw, high-frequency time series of dissolved oxygen and water temperature were filtered to remove outliers by excluding values that were greater than 3 and 5 standard deviations respectively from a 7-day running average (Supporting Information S1). Finally, we averaged the 10-min measurements to an hourly scale to match the data resolution of the weather station. Epilimnetic depth (m) was calculated from water temperature profile using rLakeAnalyzer (Winslow et al., 2019). At the littoral sites, we set epilimnetic depth as a fixed value corresponding to the maximum depth of this habitat (3 m) when epilimnetic depth was larger than 3 m. Light extinction coefficients (m −1 ) were calculated from light profiles (Wetzel, 2009). Epilimnetic depth and extinction coefficients were linearly interpolated at hourly time steps and combined with PAR to estimate the average light levels within the epilimnion of the lake (Phillips, 2020; Staehr et al., 2012). We estimated gas exchange with the atmosphere (m hr −1 ) using the LakeMetabolizer R package (Vachon & Prairie, 2013;Winslow et al., 2016).
We implemented a time-varying ecosystem metabolism model Phillips, 2020) to generate daily estimates (mmol O 2 m −3 d −1 ) of gross primary production (GPP), respiration (R), and net ecosystem production (NEP; NEP = GPP − R), using the data previously described. This model differs from many other models (e.g., LakeMetabolizer R package) in that instead of fitting each day individually, this model uses the entire time series to generate daily metabolism estimates. Model parameters are also constrained to ecologically feasible ranges (i.e., GPP and R must be positive and negative respectively).
The model , uses a photoinhibition P-I curve (Steele, 1962) to describe the relationship between GPP and light: where P I is the production rate at light intensity I, P max is the maximum production rate, and I opt is the optimal light intensity. This photoinhibition model was chosen because recent work by Staehr et al. (2016) found that photoinhibition often occurred in lakes and in situ 14 C production estimates in Castle Lake show evidence of photoinhibition as well (Huovinen et al., 1999). We used the Steele (1962) formulation because it is a simple (two-parameter) model, which increased the ease of fitting over more complex models with a greater number of  (51) April 20 (111) June 1 (152) Snow water equivalent by the date ice-out started 4 mm 743 mm 1,047 mm Table 1 Hydroclimatic Differences in the Dry (2015), Average (2016), and Wet (2019) Years Included in the Study parameters. It is often difficult to identify statistical differences among different model formulations when fit to real data (Aalderink & Jovin, 1997) suggesting that the simple model will perform similarly to its more complex counterparts. We allowed P max , I opt , and the respiration coefficient (see Phillips, 2020 for model formulation) to vary through time at a daily time scale. The degree of autocorrelation in the parameters through time was constrained by hierarchical variance parameters in the random walk components of the model. The random walk variances were selected manually such that the model converged while producing meaningful temporal smoothing in the parameters of the photoinhibition curve .
Observed dissolved oxygen time series were fit to all years (2015, 2016, and 2019) simultaneously. The model was fit via Stan (Stan Development Team, 2020) run in R (R Core Team, 2020) using the rstan package as described in Phillips (2020), and Lottig et al. (2021). Posterior median values were used for daily GPP, R, and NEP values. Model fits were validated by checking effective sample size, ̂ , tree depth, energy Bayesian Fraction of Missing Information, and divergence (Betancourt, 2007).
Metabolic rate offered an avenue to calculate total quantities of whole-lake, epilimnetic metabolism (mmol O 2 d −1 ). We calculated the product of the metabolic rates (mmol O 2 m −3 d −1 ) and the volume (m 3 ) in each habitat (i.e., littoral and open water) separately, and then summed the littoral and open water values. From the bathymetric map of Castle Lake, the littoral (depth ≤6 m) and the open water (depth >6 m) fraction each makes up 50% of the lake area (107,000 m 2 ) (Vander Zanden et al., 2007). We assume that the areal extent of the habitats does not change throughout the period of the study. We used a hypsographic curve of Castle Lake to determine the total water volume of the littoral and open water epilimnetic habitats. We multiplied the area of the littoral-bottom and open water habitats with a time series of the mixed depth, which provided a volume for a given mixed depth. The daily volumetric rates obtained from the model at both sites were multiplied by the volume of each habitat to obtain total habitat metabolism. Here, we present the proportion of total GPP, R, and NEP contributed by each site during the study period (August to September) by summing the daily values of total littoral and open water metabolism.

Physical Variables
We analyzed the daily mean water temperature at the depth the dissolved oxygen sensors were placed (3 m) and epilimnetic PAR to study the relationship of these variables on metabolism estimates. Night values (0 μmol m −2 s −1 ) were excluded when calculating the daily light values.

Zooplankton and Benthic Macroinvertebrates
Zooplankton were collected biweekly from July to September, in mid-day and after dark, using a 12 L Schindler trap from the epilimnion of the open water site. Since zooplankton in Castle Lake express a strong diel vertical migration (Redfield & Goldman, 1978), mean estimates are obtained from the day and night collections. Zooplankton were identified to genus, and biomass (μg L −1 ) was estimated using density-dry mass regressions determined previously for Castle Lake zooplankton (Redfield & Goldman, 1978). Community composition was determined as the proportion of group-specific biomass (Daphnia sp., Bosmina sp., Holopedium sp., Diacyclops sp., Diaptomus sp.).

Statistical Analyses
Differences among hydroclimatic years in air and water temperature, incident PAR and PAR at 3 m depth, wind intensity, GPP, R, and zooplankton biomass were quantified using linear mixed-effects models (LMM), with year modeled as a fixed effect. We considered the specific covariance structure for repeated-measures data in each model, and when models did not meet the assumption of homoscedasticity, we modeled the variance. Tests for significance of the effects of the years in the models were performed via the Wald statistic (Zuur et al., 2009) using the gls function, in the nlme R package (Pinheiro et al., 2020). Multiple comparisons among years were performed with Tukey's HSD post hoc test (emmeans R package; Lenth, 2021). We analyzed the relative contribution of each site to overall GPP, R, and NEP, by summing the daily values of the two sites (hereafter "whole-lake GPP, R or NEP") during the entire study period and calculating the percentage that each of the sites contributed to the total. Permutational analysis of variance (PERMANOVA) using the adonis function, in the vegan R package (Anderson, 2017;Oksanen et al., 2019) was used to compare the zooplankton community composition data. All the analyses were performed in the statistical software R (R Core Team, 2020). Data and metadata for this study are available at Scordo et al. (2021).

Hydroclimatic Conditions
We classified three hydroclimatic year conditions during the study period: dry, average, and wet years (Table 1; Figure 2). The dry year (2015) had an early ice-out (February 20) and low SWE (4 mm). The average year (2016) had a mean ice-out date (April 20), and SWE (743 mm). The wet year (2019) had a late ice-out date (June 1) and high SWE (1,047 mm).

Metabolic Rates
The littoral habitat has substantially higher metabolism (GPP and R) rates than the open water habitat, resulting in a strong relative contribution to whole-lake production ( Figure 4, Table 2, Supporting Information S1). In the average hydroclimatic year, the littoral habitat has significantly higher GPP and R rates than the open water habitat (p < 0.01). GPP ranged from 46.9 to 26.1 mmol O 2 m −3 d −1 in the littoral habitat, and 6.8 to 3.9 mmol O 2 m −3 d −1 in the open water habitat (  (Table 2). NEP did not differ significantly between habitats (p = 0.7); however, the littoral site showed more variability (Figure 4 left panels, Compared to the average hydroclimatic year, the dry and wet years had lower GPP and R rates in both habitats (p < 0.01) (Figure 4 left panels; Table 2, Supporting Information S1). GPP and R reduced the most in the littoral habitat (Figure 4 left panels; Table 2).

Total Habitat Metabolism
When scaled to the volume of open water and littoral habitats, during the study period of the average hydroclimatic year the relative contribution to total GPP was 80% (693 × 10 6 mmol O 2 ) and 20% (175 × 10 6 mmol O 2 ) in the littoral and open water habitats, respectively (Figure 4 right panels; Table 3). In an average year, the relative contribution to total R was 85% (659 × 10 6 mmol O 2 ), and 15% (120 × 10 6 mmol O 2 ) in the littoral and open water habitats, respectively (Figure 4 right panels; Table 3). In the average hydroclimatic year, positive NEP values (GPP higher than R) represented 86% of total NEP (Figure 4 right panels). Negative NEP values represented 14% of total lake NEP and they occurred only in the littoral habitat.
The relative contribution to total lake GPP and R of the littoral habitat decreased, and the contribution of the open water habitat increased during the dry and wet years ( Table 3). The relative contribution to whole lake GPP of the littoral habitat decreased by 5% (dry year) and 16% (wet year), while the relative contribution to whole lake R of the littoral habitat decreased by 7% (dry year) and 9% (wet year) (Figure 4 right panels).
Whole lake NEP behavior differs markedly depending on hydroclimate year (dry vs. wet). During the dry year, positive NEP values represented 96% of total NEP, implying an increase in the relative contribution of GPP to NEP compared to an average year (Figure 4 right panels). During the wet year, positive NEP values represented 61% of total NEP, implying an increase in the relative contribution of R to NEP compared to an average year (Figure 4 right panels). The littoral habitat had mostly negative NEP values, resulting in a high proportion of negative NEP values during the wet year (Figure 4 right panels, Supporting Information S1).

Zooplankton and Benthic Macroinvertebrates
In the average hydroclimatic year of 2016, total zooplankton biomass (98 ± 35 mg m −3 ) was higher (p < 0.01) than in the dry (53 ± 46 mg m −3 ) and wet (58 ± 21 mg m −3 ) years (Figure 5c). The mean biomass decreased by 46% during the dry year and 40% during the wet year compared to the average year. Community composition remained similar between the average and wet years (Figure 5d; PERMANOVA, p = 0.34) but differed during the dry year (Figure 5d; PERMANOVA, p = 0.02). Daphnia accounted for 82% and 68% of the total zooplankton biomass in the average and the wet years, respectively. In the dry year by contrast, Daphnia represented 42% of the total zooplankton biomass, and the proportion of the copepod Diacyclops increased substantially. During the dry year, Diacyclops accounted for 30% of the total zooplankton biomass compared to 4% during the average year.
Total benthic macroinvertebrates biomass was also higher in the average hydroclimatic year (4,339 mg m −2 ) than in the dry (3,175 mg m −2 ) and wet (3,158 mg m −2 ) years. The biomass decreased by 27% during the dry and wet years compared to the average year. The FFG composition of the benthic macroinvertebrates remained similar during the average and dry years but differed during the wet year ( Figure 5e). During the dry and average years, collector/gatherers, predators, and shredders dominate the community. Grazers were not present, and collector/filterers represent fewer than 4% of the community composition. In the wet year, collectors/gatherers still dominated the community composition (40%), but grazers (31%) were the second most abundant FFG (Figure 5c).

Discussion
While our findings agree with existing research that shows higher metabolism in the littoral habitat than the open water habitat, we show that the year's hydroclimate strongly influences metabolism (GPP, R, and NEP) in the littoral habitat. Furthermore, invertebrates experienced variation to both their biomass and community structure depending on hydroclimate. Our study has implications for how ecosystem scientists and managers quantify the absolute and relative contributions of the littoral habitat to whole lake production. GPP and R decreased in both the dry and wet years and the open water habitat varied less than the littoral habitat. The difference between littoral and open water response indicates that processes in the bottom of lakes show greater sensitivity to the year's hydroclimate. The relative contribution of the open water habitat to overall GPP and R increased during wet and dry hydroclimatic conditions. While both the dry and wet conditions triggered a reduction in GPP and R, the effect of these hydroclimatic conditions on NEP was substantially different.
The dry year tended toward autotrophic metabolism, while the wet year tended toward heterotrophic metabolism, particularly in the littoral habitat. We also observed a reduction in zooplankton and benthic macroinvertebrate biomass during the dry and wet year.
Hydroclimate accounts for changes in GPP and R in both open water and littoral habitats; however, the littoral habitat responds more dynamically than the open water to different hydroclimates. We suggest that metabolism varies less in the open water habitat due to having longer retention times of water, nutrients, and detritus than the littoral zone (Peters & Lodge, 2009). We developed a conceptual model to indicate the factors and magnitude of changes to different components of metabolism in the open water and the littoral habitat of the lake during the dry, average, and wet years ( Figure 6). The ecological characteristics of each habitat likely dictate the differences in metabolism ( Figure 6). Interannual variations in metabolism in all habitats conformed with the observed changes in water temperature, ice-out date, and snow water equivalent (SWE) accumulated by the beginning of ice break-up ( Figure 6). Other studies quantifying metabolism from lakes indicate that changes in light dynamics influence GPP and R (Phillips, 2020;Sadro et al., 2011;Staehr et al., 2016). Metabolic rate differences between years does not seem to be sensitive to light considering light measured in the epilimnion was similar across years and throughout the study period. However, we were not able to measure light at the littoral sites and there could be subtle variations in light dynamics at these sites. During the wet year, loading of terrestrial particulate material into the lake after snow melt may have reduced the light penetration, in the littoral habitat, resulting in decreased benthic productivity. As expected, we observed higher volumetric rates of GPP and R at the littoral habitat compared to the open water habitat. In deep meso-and oligotrophic lakes, the nearshore littoral habitats are usually more productive than the offshore (Peters & Lodge, 2009). Littoral metabolic rates are high due to the accumulation of nutrients and detritus in this part of the lake and because algae grow faster when attached to a substrate (Peters & Lodge, 2009). In Castle Lake, the alder trees growing on the shore of the lake are among the main sources of nitrogen and   (Goldman, 1961). Soil under alder trees accumulates nitrogen through leaf fall, which is then transported by spring water into the lake (Goldman, 1961). In an experiment, Goldman (1961) showed that productivity doubled after spring water addition that traveled through alder humus.
The observed changes in water temperature and SWE during wet and dry years explain the reduction in R and GPP in both habitats. The water temperature decreased (on average by 1°C-1.2°C) in both the dry and wet years compared to the average hydroclimate year. Temperature variation influences both GPP and R (Solomon et al., 2013;Song et al., 2018) due to the relationship between enzyme kinetics and thermal regimes (Phillips, 2020). Blue-green algae dominate littoral and epilimnetic phytoplankton communities at Castle Lake during August and September Higley et al., 2001;Huovinen et al., 1999). The metabolic rates of blue-green algae substantially decrease with even a 1°C drop when water temperatures reach ∼20°C (Wetzel, 2009;Carey, et al., 2011;Visser et al., 2016). The large amount of SWE and the late ice-out date likely caused the low water temperature in the wet year. Snowpack and lake temperature are strongly correlated in Castle Lake (Caldwell et al., 2020;Goldman et al., 1989;Park et al., 2004;Strub et al., 1985) and mountain lakes in general (Melack et al., 2020;Sadro et al., 2018;Smits et al., 2020). Previous research at Castle Lake demonstrated that water temperature tends to be higher during dry years, resulting in higher primary productivity (Caldwell et al., 2020;Park et al., 2004). However, in our study, water temperature in the dry year was lower than in the average year. While the differences were marginal, mean daily wind velocity increased and solar radiation decreased in the dry year. Both variables could have reduced water temperature (Supporting Information S1). The dry year had similar air temperature to the average year (Supporting Information S1) and lower snowmelt inputs, making them unlikely culprits to lower water temperatures during the dry year.
Variations in SWE likely constrained metabolic rates by reducing the input of nutrients and organic matter during the dry year (low SWE), and by flushing algae and nutrients from the lake and cooling the water during the wet year (high SWE) ( Figure 6). The effect of SWE on Castle Lake's primary productivity depends on the amount of SWE and ice-out timing (Goldman et al., 1989;Park et al., 2004;Strub et al., 1985). Low SWE limits the amount of nutrients delivered to the lake from the surrounding watershed. In Castle Lake, low SWE combined with an early ice-out, such as in the dry year of our study, leads to nutrient depletion earlier in the open water season (Elser et al., 1995;Goldman et al., 1989;Huovinen et al., 1999). Therefore, the shallow primary productivity by August and September was likely nutrient-limited. In contrast, high SWE delivers more nutrients and organic matter to the lake, but the greater influx of snowmelt flushes primary producers from the lake (Goldman et al., 1989). In the wet year of our study, high SWE combined with late ice-out, in this scenario, incident light and water temperature had already begun to decrease by the time the primary producer community started to recover, which constrained metabolic rates. Moreover, high SWE reduces the lake's water temperature, making it more difficult for the algae community to recover (Goldman et al., 1989;Park et al., 2004;Strub et al., 1985).
Wet hydroclimatic conditions favored heterotrophic metabolism (especially within the littoral habitat), while drier conditions favored autotrophic metabolism. During the wet year, NEP decreased at both sites, and became predominantly negative at the littoral habitat. Sadro and Melack (2012) found a shift in NEP toward heterotrophy in alpine Emerald Lake after a large inflow of water that flushed primary producers and increased the amount of allochthonous organic carbon. We suspect that the increase in NEP during the dry year compared to the average hydroclimatic year is due to less allochthonous (out of system) material available for heterotrophic bacteria. These results show that the internal process of organic carbon in Castle Lake varies between habitats and under different hydroclimatic conditions. Therefore, research that focuses on lakes as sources or sinks of organic carbon would benefit from incorporating within-lake spatial and temporal variability.
We detected reciprocal relationships between consumers and GPP in Castle Lake (Figures 5a-5c and 5d). Decreased GPP in the dry and wet years coincided with reduced consumer biomass. We also observed changes in consumer community composition that may have contributed to reducing GPP. We saw no discernable change in zooplankton community composition during the wet year, but the proportion of the copepod Diacyclops increased while Daphnia decreased during the dry year. Diacyclops have a longer period of juvenile development than  Daphnia (Horne & Goldman, 1994). Therefore, it is more likely that the relative number of Diacyclops would increase in a year with earlier ice-out rather than a year with average or late ice-out (Horne & Goldman, 1994). We suggest that an indirect effect related to nutrient recycling favors a reduction in primary productivity when the proportion of Diacyclops increases and the proportion of Daphnia decreases. At Castle Lake, Daphnia plays an important role in regenerating nitrogen (Brett et al., 1994;Elser et al., 1995), which is the main nutrient that limits shallow GPP in the lake . Daphnia recycle nitrogen from the hypolimnion to the epilimnion due to its strong diel vertical migration (Elser et al., 1995). In contrast, Diacyclops do not modify the water nutrient composition at Castle Lake (Brett et al., 1994). Diacyclops are generally considered carnivores rather than grazers; they consume food with elemental ratios similar to their own and excrete few nutrients (Brett et al., 1994). Limited nutrient cycling from increased Diacyclops helps explain nitrogen-limitation for open water algae and the decreased GPP during the dry year of our study. The FFG shifts in the benthic macroinvertebrates that we observed in the littoral habitat may have contributed to reduced GPP during the wet year. Grazers constituted 31% of benthic invertebrate biomass during the wet year in the littoral habitat yet they were absent from our samples during the dry and average years. Previous studies at Castle Lake demonstrate that secondary invertebrate consumers can reduce or enhance productivity, depending on the animal biomass and the community composition (Elser, 1992). However, in this study, we did not evaluate the relative role of consumers in regulating metabolism. Therefore, while we suggest that hydroclimatic year influences metabolism and the production of secondary consumers, future work that elucidates the connections across the algae-herbivore interface and energy flow into the food web will build on our findings.

Conclusion
Our study demonstrates that the metabolism of the littoral habitat at Castle Lake, particularly GPP and R, shows greater sensitivity to hydroclimate changes than the open water habitat. We also showed that the site-specific NEP can change from autotrophic to heterotrophic, depending on hydroclimate conditions. This study must be understood in the context of climate variation (Dai, 2013;Min, et al., 2011) and its complex interaction with ecosystems. Not only do ecosystems respond to variations in climate forcing, but within-lake habitats do not have equivalent responses. The complex interactions between climate forcing and ecosystem and habitat-specific responses extends to habitat and resource use by consumers (Caldwell et al., 2020;Rosenblatt & Schmitz, 2016;Rüegg et al., 2021).
Our research amplifies the need for studies across multiple years with a similar hydroclimate to test the consistency of our findings. We recognize other concerns such as lake-specific differences in the littoral habitat composition. Castle Lake's woody debris-dominated habitat may have a different response when compared to metabolism changes in rock-or macrophyte-dominated littoral habitats under different hydroclimate conditions. An integration of such studies across lakes to test effects based on trophic conditions (ultraoligotrophic to eutrophic) would add valuable context to our findings from a subalpine lake. Additionally, many top consumers are non-open water fish, which rely on littoral resources in lakes (Vander Zanden et al., 2011). It would be interesting to analyze the coupling and decoupling effect that climate-driven changes produce across the different trophic levels and habitats of lakes. and open water habitats during dry, wet, and average years. The littoral habitat has the highest values of GPP and R due to receiving the highest loads of allochthonous inputs (nutrients and organic matter) and is the only site to shift to a negative NEP (during the wet year) due to R of allochthonous organic matter, despite GPP being low. The open water habitat presents the lower values of GPP, R, and NEP and the lowest variability within years, implying a high resilience. The height of the bars corresponds to the mean daily rates of GPP, R, and NEP in each habitat