The impacts of changing nutrient load and climate on a deep, eutrophic, monomictic lake

Nutrient availability and climate have substantial effects on the structure and function of lakes. Predicted changes to climate (particularly temperature) over the 21st century are expected to adjust physical lake functions, changing thermal and nutrient use processes. Both increasing anthropogenic nutrient inputs and net reductions following remediation will also drive ecological change. Therefore, there is an increasing necessity to disentangle the effects of nutrient and temperature change on lakes to understand how they might act in additive and antagonistic ways. This study quantified internal and external nutrient loads at Rostherne Mere, U.K., a deep (zₘₐₓ = 30 m), monomictic eutrophic lake (average annual total phosphorus >100 μg/L) that has a long, stable period of stratification (c. 8.5 months). A lake biophysical model (PROTECH) was used to assess the effect of changes in these loads and climate change on lake productivity in a factorial modelling experiment. During the summer, phosphorus released from the sediment is largely restricted to the hypolimnion and phytoplankton production is supported by the external load. On overturn, phosphorus at depth is distributed throughout the water column with the elevated concentration persisting to support algal productivity in the following spring. Consequently, the model showed that internal nutrient loading was the main driver of current and future changes in the concentration of phosphorus (responsible for up to 86% P reduction), phytoplankton chlorophyll a and cyanobacterial blooms. However, although the external phosphorus load had a relatively small influence on annual mean phosphorus concentration, it had a statistically significant effect on chlorophyll a concentration, because it supported algal production during summer stratification. Climate had minimal direct impact, but a substantial indirect impact by altering the timing, depth and length of lake stratification (c. 14 days longer by 2100), and therefore altered nutrient cycling and phosphorus availability. In summary, the recovery trajectory at Rostherne Mere is limited by the annual internal soluble reactive phosphorus load replenishment that realistically is unlikely to change greatly on a shorter time‐scale. Therefore, the external soluble reactive phosphorus load has the potential to play an important role as it can be managed further, but is complicated by the indirect impact of climate changing stratification and flushing patterns.


| INTRODUC TI ON
Nutrient availability is a major factor affecting lake ecosystem functioning including productivity, the development of potentially toxic algal blooms and oxygen depletion (Sas, 1989). Anthropogenic nutrient enrichment has increased nutrient loads, causing many lakes to become eutrophic (Smith, Joye, & Howarth, 2006). Climate also has a major effect on lakes via water temperature and stratification (Hutchinson & Löffler, 1956) and climate change has increasingly been recognised to have a substantial impact on lake structure and function globally (Paerl & Huisman, 2008;Tranvik et al., 2009;Williamson, Saros, Vincent, & Smol, 2009). Long-term incremental increase of air temperature influences the thermal structure (Gauthier, Prairie, & Beisner, 2014;Liu, Bocaniov, Lamb, & Smith, 2014) and timing of stratification in lakes (Izmest'eva et al., 2016;Meis, Thackeray, & Jones, 2009). The shorter-term impact of heatwaves, droughts, and flooding can alter the hydrological balance and ecological structure in many lakes (Bakker & Hilt, 2016;Bertani, Primicerio, & Rossetti, 2016;Wigdahl-Perry et al., 2016). Thus, if future projections of U.K. climatic warming are correct, with annual average temperatures rising by +1.4°C from 2020 to 2060 and +3.4°C from 2020 to 2100 (UKCP09 projections; Murphy et al., 2009), there are likely to be substantial effects on lakes across the U.K. However, the precise direction and magnitude of change is uncertain since multiple stressors may interact in synergistic, antagonistic, or additive ways (Coors & De Meester, 2008). Lakes are differentially sensitive to system stressors (i.e. George, Maberly, & Hewitt, 2004) further complicating the assessment of how an individual lake will respond to environmental change or management intervention.
Rostherne Mere, Cheshire, U.K. is a prime example of a eutrophic lake that has undergone a significant change in trophic status over the last century caused by the direct impact of human activity (Moss et al., 2005). Like many such lakes (Schindler, 2006;Zamparas & Zacharias, 2014), Rostherne Mere has recently undergone catchment-scale management intervention to limit nutrient load and reduce lake nutrient concentrations. While reducing the external phosphorus load improved ecological condition in some lakes , many have only witnessed a slow recovery as a result of internal loading of soluble reactive phosphorus (SRP) derived from historic SRP inputs stored within the upper sediment Schindler, 2006). Rostherne Mere is a small but deep lake permitting a particularly long and strong summer stratification (Radbourne, Ryves, Anderson, & Scott, 2017). In conjunction with high productivity, this drives rapid oxygen depletion at depth following spring stratification (Scott, 2014), favouring the release of large quantities of remobilised P from the sediment into the hypolimnion, potentially slowing the rate of recovery and extending the requirement for management intervention.
This study used observed catchment and lake monitoring data from 2016 to assess contemporary nutrient dynamics at Rostherne Mere. These data are used in the biophysical PROTECH model (Phytoplankton RespOnses To Environmental CHange; Reynolds, Irish, & Elliott, 2001;Elliott, Irish, & Reynolds, 2010) to simulate a range of future nutrient (internal SRP load and external SRP load) and climate (air temperature) scenarios (UKCP09 projections; Murphy et al., 2009). The aim was to determine the importance of internal nutrient loads, external nutrient loads and changing climate on key biophysical lake properties and dynamics. Using a factorial model design, the key drivers were tested to see how they individually and collectively changed the lake properties in either additive or antagonistic ways to better identify future possible recovery pathways.

| Site description
Rostherne Mere (53°20′N, 2°24′W) is one of the largest (area 48.7 ha) and deepest lakes, (max depth 31 m; mean depth 13.6 m) of the Shropshire-Cheshire Meres (Carvalho, Beklioglu, & Moss, 1995) with an average water retention time of about 9-10 months. It develops a long, stable summer stratification (thermocline average depth March to November c. 10 m) with an anoxic hypolimnion. With a long history of eutrophication (defined as >100 μg P/L; Carlson, 1977;Radbourne et al., 2017), Rostherne Mere received a substantial nutrient load from upstream sewage treatment works, with effluent discharged directly into the upstream watercourse between the 1930s and June 1991, when it was then diverted out of the 6. In summary, the recovery trajectory at Rostherne Mere is limited by the annual internal soluble reactive phosphorus load replenishment that realistically is unlikely to change greatly on a shorter time-scale. Therefore, the external soluble reactive phosphorus load has the potential to play an important role as it can be managed further, but is complicated by the indirect impact of climate changing stratification and flushing patterns.

K E Y W O R D S
internal loading, multiple stressors, PROTECH modelling, Rostherne Mere, stratification catchment (catchment size: 9 km 2 ). However, the reduced P inputs did not lead to a similar reduction of P export, resulting in the lake becoming a large net source of P following sewage treatment works diversion. This net export is sustained to the present day with a net 552 kg total phosphorus (TP)/year exported during 2016. The net export of P is fed by the internally loaded P from lake sediments during anoxic stratification (Nurnberg, 1984). Internal loading produces high concentrations of SRP in the hypolimnion of Rostherne Mere during stratification (>800 μg/L), which, at autumn overturn, replenishes the whole lake SRP pool and persists into the next spring. This internal loading explains the slow long-term recovery trajectory of Rostherne Mere over the last 25 years (Moss et al., 2005).
The buoy measures various in-lake parameters and lake-surface meteorological variables at high resolution (every 4 min) with the data uploaded in real-time. This study used lake temperature readings and surface wind speed data.

| Collection and analysis of field data
Between January 2016 and January 2017, water samples were collected approximately every 3 weeks from Rostherne Brook (the main inflow, draining 79% of the catchment), Blackburn's Brook Phytoplankton community composition was analysed after concentrating a 1-L sample using a sequence of settling procedures and the gridded cell abundance count method (Brierley, Carvalho, Davies, & Krokowski, 2007). Chlorophyll-a analysis involved a standard spectrophotometer approach, with recordings taken at wavelengths of 630, 645, 665, and 750 nm and an extraction solvent of 80% acetone (Sartory & Grobbelaar, 1984).
Inflow and outflow discharge was determined by calculating a linear discharge relationship between cross-sectional profiles and sectional velocities recorded for a range of stage heights over 2016.
This empirical stage-discharge relationship was then applied to continuous stage height measurements recorded using a Van Essen mini-diver data logger (www.vanessen.com), recording water pressure every 5 min and calibrated to a barometer located at the lake shore, to provide a high-resolution record of inflow and outflow discharge for the whole year.
The standard outflow method for water residence time (WRT) was estimated as the annual average outflow rate against the total lake volume. An adjusted method (WRT m ) was also used, calculated as the monthly average outflow rate against the monthly available mixed lake volume (i.e. during the stratified period from April to November, this is the epilimnion, but during the mixed period from December to March, this is the entire lake).

| The PROTECH model
PROTECH is a mechanistic model, i.e. its predictions emerge from the biological processes simulated within the model, which allow a better understanding of potential cause and effect relationships.
It also simulates at the species level, so allows community level simulations to be conducted rather than just a simulation of total chl-a. The model simulates the responses of a number of phytoplankton populations (here representing a genus) distributed in a one-dimensional vertical water column (described by 0.1-m layers, reflecting the bathymetry of the lake) at daily time steps, but also calculates key physical limnological parameters such as thermocline development, stratification pattern, and nutrient concentrations. A full description of the model's equations and concepts has been already published (Elliott et al., 2010;Reynolds et al., 2001) but the main biological component of the model is the daily change in chl-a concentration (ΔX/Δt) attributable to each phytoplankton population: where r′ is the growth rate defined as a proportional increase over 1 day, S is the loss caused by settling out from the water column, G is the loss caused by Daphnia grazing (it is assumed only phytoplankton <50 μm diameter are grazed) and D is the dilution loss caused by hydraulic exchange.
The growth rate (r′) is further refined by: where r ′ ( ,l) is the growth rate at a given water temperature and light intensity and r ′ Si are the growth rates determined by SRP, N, and DSi concentrations below these respective threshold concentrations: <3, 80, and 500 μg/L (Reynolds, 2006). The r′ values are phytoplankton-dependent (e.g. non-diatom taxa are not limited by silica concentrations below 500 μg/L and nitrogen-fixing cyanobacteria are not limited by nitrogen) and, crucially, relate to the morphology of the taxon. The phytoplankton communities used in the model were selected from analysis of the most abundant species observed in the lake (Radbourne, 2018), giving a total of seven phytoplankton taxa: Asterionella, Stephanodiscus (diatoms), Cryptomonas (cryptophyta), Gomphosphaeria, Microcystis, Aphanizomenon, and Dolichospermum (cyanobacteria).
Water temperature and light (i.e. cloud cover and seasonal irradiance) are varied at each time-step throughout the simulated water The model was run and compared to the observed data recorded from the lake during 2016 by using the coefficient of determination (R 2 and RMSE). After the initial simulation, as expected, it became clear that an internal hypolimnetic nutrient supply of SRP needed to be added to simulate the internal loading of P. This being cited previously as an important component of P dynamics at Rostherne Mere (Carvalho et al., 1995;Moss et al., 2005). Therefore, incremental daily amounts of SRP were added to the bottom 15 m of the water column from 1 June for 90 days until the hypolimnion concentration matched those observed from the depth profiles: 7.8 μg SRP L −1 day −1 was found to be optimal. The model assumes the internally loaded SRP is not mixed into the surface water until stratification breaks down, a suitable simplifying assumption in a strongly stratifying lake such as Rostherne Mere (Mackay, Folkard, & Jones, 2014). The internal load period of 90 days was fixed to ensure that the internal load only occurs during anoxic stratification (a feature of Rostherne Mere; Radbourne et al., 2017) for all climate change scenarios, because anoxia promotes the sediment release of iron bound P for replenishment into the water column (Nurnberg, 1984). This P will only be redistributed throughout the water column at stratification overturn due to the strong summer stratification within the lake.

| Future climate scenarios
The calibrated 2016 simulation was taken as a baseline and then rerun through a combination of progressive changes to air tempera-  (Carvalho et al., 1995). In total, this produced 298 scenario combinations (i.e. 11 different temperature models at three time frames, with three external SRP loadings and with internal SRP loadings, plus on increased external load simulation).
Model simulation results were statistically analysed using a single factor ANOVA, testing the significance of the difference between the 11 temperature forecast models with external load change, internal load change, and temperature change.

| Calibration and validation of PROTECH
The only adjustment of PROTECH that was required was to include

| Response of SRP concentration to future change
Model simulations of possible future changes altered annual mean lake SRP concentrations. Reduction of the internal loading had the greatest effect on epilimnion SRP concentration (Figure 4), with a 48.9% and 85.6% reduction for high-mid and high-low, respectively (Table 1); all scenario results being statistically significant ( 100 µg SRP L -1 200 µg SRP L -1 300 µg SRP L -1 400 µg SRP L -1 500 µg SRP L -1 600 µg SRP L -1 700 µg SRP L -1 800 µg SRP L -1 Altering external load alone (i.e. maintaining current baseline of high internal load) had a much smaller impact on SRP concentration compared to changing internal load, with a 3.4% and 6.4% fall in concentration from high-mid and high-low, respectively (Table 1). These changes are not statistically significant (Table 2).
However, the relative impact of the external load increased in scenarios where the internal load was lower (under low internal load: 10.2% reduction high-mid and 20.3% reduction high-low external loads;

| Response of chl-a concentration to future change in climate and nutrient load
The concentration of chl-a responded to the alterations in SRP concentration ( Figure 5). Reduced internal loading led to the greatest chl-a reduction (51.0% high-mid, 65.0% high-low; Table 1). All scenarios of internal load change were statistically significant (Table 3). Like SRP, the chl-a reduction is amplified by a reduced external load. However, temperature increased chl-a in the 2060 scenarios, yet is shown to support reduction to 2100 (Table 1). The impact of changed external load on chl-a was statistically significant in all scenarios, with increasing significance in future temperature scenarios and at reduced internal loads (Table 3).
Changes in air temperature increased chl-a concentrations between 2020 and 2060, with the biggest alterations in higher nutrient load scenarios (Table 1). However, from 2060 to 2100, all nutrient scenarios suggested a decline in chl-a concentrations with the largest reduction in the low nutrient scenarios (3.5% to 27.8%; Table 1).
Statistical significance of temperature change was only evident in the high nutrient load scenarios (Table 3).

| Response of phytoplankton to future change in climate and nutrient load
Annual modelled phytoplankton assemblages differed among future scenarios (Figure 6a,b). Cyanobacteria showed a substantial decrease in bloom size (represented as chl-a) and proportional chl-a dominance in reducing internal load scenarios (Figure 6a).
External load reduction had a greater influence by reducing cyanobacterial dominance in lower internal load scenarios, while temperature slightly increased the cyanobacterial dominance in future warmer climates (Figure 6a). The additional simulation of future change with a 10-fold external SRP concentration increase showed an increase in SRP and chl-a, with the predominant increase found in cyanobacterial abundance (up to 88% of total chla; Table 4).
Annual modelled diatom abundance (in terms of chl-a) showed little change under all future change scenarios (Figure 6b). Due to the limited response in chl-a production, the relative contribution to total algal chl-a increased with a decreasing total chl-a ( Figure 5) and cyanobacterial bloom size (Figure 6a).

| Phosphorus legacy of the sediment limiting future recovery
The PROTECH projections of possible future change scenarios clearly show that the main driver of substantial change at Rostherne Mere would be the reduction of the large internal SRP load, producing a 10-fold greater percentage reduction than

| Contribution of internal and external loads to the concentration of chl-a
A reduction in SRP concentration is anticipated to reduce chl-a production, ultimately improving the wider lake ecosystem. It has TA B L E 1 Modelled annual mean values (μg/L) for soluble reactive phosphorus (SRP), chlorophyll a (chl-a) and the ratio between chl-a: SRP for different internal and external SRP loads and climate scenarios TA B L E 2 Significance of external load change, internal load change and temperature change on the final modelled year annual soluble reactive phosphorus for 11 future temperature models been suggested that a reduced external SRP load will starve internal load replenishment, eventually exhausting the standing SRP store (assuming sediment oxygen condition does not change).
However, it may take decades or longer before internal P store is exhausted and full lake recovery occurs (Reddy, Newman, Osborne, White, & Fitz, 2011;Sharpley et al., 2013).
The response between the reduction of chl-a and SRP concentrations was non-linear. The model simulations show that there is a greater proportional reduction of SRP than chl-a with a reduced internal load, whereas a reduced external load alone results in a similar proportional chl-a reduction, suggesting that external load is a key driver in chl-a production, despite its small impact on SRP load. Similar non-linear patterns between chl-a and SRP were evident in all nutrient scenarios. External loading was only significant for SRP when internal loading was low, predominantly due to the increasing relative ratio of load size (i.e. the external load forms a greater proportion of the total nutrient input), or with the influence of future temperatures.
Here, we propose that the non-linear responses of chl-a to SRP The future scenario of a 10-fold increase in external SRP concentration, designed to simulate an unmanaged catchment, showed a substantial increase in SRP and chl-a, driven by a rise in summer algal blooms. This confirms the role of the external load in replenishing the late summer SRP and highlights its importance for catchment management in lakes such as Rostherne Mere. Additionally, whilst not modelled here, the UKCP09 model projections suggest that precipitation will increase with future climate change, especially during the summer (Murphy et al., 2009). Thus, given the diffuse nature of the sources, external SRP (and TP) load is likely to rise (Andersen et al., 2006;Jeppesen et al., 2011), increasing the overall importance of the external load and further highlighting the necessity for continued catchment management.

| The effect of climatic warming on the concentration of chl-a
Rising air temperatures to 2060 resulted in a slight increase in chl-a concentration, under current nutrient loads and a smaller increase in lower nutrient scenarios. Some studies have found that temperature increase has no effect on chl-a, only increasing the cyanobacterial dominance (Kosten et al., 2012), whereas others found an increase in chl-a, but the mechanistic reason is unknown (Elliott, McElarney, & Allen, 2016;Izmest'eva et al., 2016). However, in Rostherne Mere at the highest temperature, corresponding to the 2100 time period, chl-a concentrations were forecast to be lower than those in 2060. This reduction in chl-a with continued warming is counter to other modelling studies simulating warming, that reported a step change increase with higher warming scenarios in similar deep, stratifying eutrophic lakes (Elliott et al., 2016;Tadonleke, 2010). Although the mechanism linking chl-a change and temperature change is unclear, we suggest that the response at Rostherne Mere is based on the indirect impact on nutrient cycling caused by altered stratification and its effect on the hydrology of the lake, leading to a changing species dominance that produces a variation in chl-a (discussed below).

| Phytoplankton assemblage change driven by climate and nutrient loads
Cyanobacterial blooms responded to environmental change in a similar way to chl-a since they are the dominant algal group, unlike diatoms, which represent a smaller proportion of total chl-a.
Temperature change altered diatom phenology, but not the popu-  A warming climate could potentially lead to a shallower thermocline with enhanced stability in the water column (Butcher, Nover, Johnson, & Clark, 2015). Here, the statistically significant increase in epilimnion depth at Rostherne Mere under future scenarios of a warmer climate, suggests warming at the surface (and greater energy input to the lake) is transferred deeper into the water column deepening the thermocline depth, as has been found in other studies (Gauthier et al., 2014;Liu et al., 2014;Luoto & Nevalainen, 2013 In warmer future climates, the duration of stratification may be increased, as found here and by others (Izmest'eva et al., 2016;Liu et al., 2014). A longer stratification time would lead to a faster flushing rate with a smaller mean lake volume available for flushing (i.e. epilimnion) for a longer period of the year. In this study, the impact of the future projected increased duration of stratification led to a slight decrease in WRT m that would reduce nutrient availability, leading to smaller phytoplankton blooms. However, the decrease is again minor and is not likely to be significant in isolation. The combination of deeper stratification lengthening the WRT m and a longer stratification shortening the WRT m , in the case of Rostherne Mere, is forecast to result in a net offset and so will have a minimal effect.
An earlier onset of stratification would lead to a reduction in nutrient availability in the early part of the year, because the nutrients stored in the hypolimnion become unavailable for the remainder of the water column at an earlier stage. Therefore, the size of summer cyanobacterial blooms will decrease. Evidence of such reductions is seen in the lower chl-a concentrations in the 2100 scenarios compared to the 2060 scenarios. Earlier onset of stratification in the 2100 scenarios leads to earlier nutrient limitation and thus lower productivity for a larger part of the summer period. Single factor ANOVA f-values and significance included as annotation for each plot appears to have a small direct influence on the future of nutrient use and change at Rostherne Mere. However, the indirect impact of changing stratification patterns (i.e. mixing depth, length, and onset timing) and potential for increased internal loading, could decrease the available nutrients in the epilimnion in late summer and autumn, yet increase the internal load contribution for winter replenishment, subsequently adjusting the rate and trajectory of future recovery and potentially altering algal community structure.