An experimental test of climate change effects in northern lakes: Increasing allochthonous organic matter and warming alters autumn primary production

Lakes in northern regions are rapidly warming (Niedrist, et al., 2018; Rose et al., 2016). Climate impacts lake gross primary production (GPP) in multiple ways, including direct temperature effects on metabolic rates of autotrophs and grazers (Kraemer et al., 2017; O’Connor et al., 2009; Yvon-Durocher et al., 2010) and indirect effects via terrestrial export of coloured dissolved organic Received: 30 April 2020 | Revised: 4 December 2020 | Accepted: 14 December 2020 DOI: 10.1111/fwb.13679

Warming could directly impact biomass growth and metabolic rates of autotrophs through increases in nutrient uptake and light saturation of photosynthesis (De Senerpont et al., 2014;Tilzer et al., 1986;Yvon-Durocher et al., 2015). Increased cDOM, by contrast, has multiple impacts on lake characteristics, including reduced light availability, but also increased macronutrients like nitrogen (N) and phosphorous (P) bonded to cDOM and increased carbon dioxide (CO 2 ) concentrations from cDOM mineralisation (Carpenter et al., 1998;Creed et al., 2018;Nydahl et al., 2019). Generally, elevated nutrient and CO 2 levels enhance GPP, mainly in the pelagic habitat, while reduced light penetration from brownification decreases GPP, mainly in benthic habitats (Ask et al., 2012;Brown et al., 2020;Genkai-Kato et al., 2012).
Climate change research on lakes and ponds has largely focused on the summer period, although measurements in the winter season are increasing (Denfeld et al., 2018;Hampton et al., 2017). Few studies have tested climate change impacts on ecological processes during autumn-winter transition, although this period is important ecologically and will experience warming-related changes such as increased water temperature and runoff as well as delayed ice cover formation (Bintanja, 2018;Veillette et al., 2010). Longer ice-free periods via warming could increase temperature of the water column (Niedrist et al., 2018) and increased runoff could elevate loading of cDOM to lakes (Klimaszyk & Rzymski, 2015). The combined consequences of warming and cDOM on GPP during autumn are largely unknown. While warming can stimulate metabolism, declining daily light irradiation in autumn may constrain the predicted positive temperature response of autotrophs, causing a mismatch between consumer energy demands and GPP as temperatures decline. This mismatch will probably be enhanced when browning-caused light limitation exceeds nutrient limitation, even when cDOM delivers critical nutrients.
Added to this, the reduced duration and thickness of ice and snow cover will increase light levels compared to a longer duration of a thick ice and snow cover, and will therefore potentially stimulate GPP in lakes (Lenard, 2015;Obertegger et al., 2017). Biogeochemical consequences of changes in ice phenology in spring have received increasing interest and studies have shown impacts on ecosystem dynamics and their contributions to whole lake ecosystem processes (Benson et al., 2012;Bertilsson et al., 2013;Hamdan et al., 2018;Hampton et al., 2017), while substantially less attention has been given to ecosystem processes during the late autumn/early winter transition.
The aim of this study was to experimentally test the direct (warming) and indirect (allochthonous cDOM) climate effects on lake GPP during autumn. We hypothesised that climate change will cause a mismatch between the effects of higher temperature and allochthonous cDOM on autumn GPP via asymmetric responses in GPP across habitats. We specifically predicted that: (1) warming stimulates GPP, but responses depend on availability of light and nutrients; (2) cDOM inputs impact habitat-specific GPP asymmetrically by reducing benthic GPP (due to light limitation) while stimulating pelagic GPP (due to relaxing nutrient limitation); and (3) delay of ice cover formation promotes GPP via increasing light availability. To investigate our predictions, we increased water temperature and applied a gradient of inflow rates of stream water with high cDOM in a large-scale experimental pond system to monitor the response of whole-ecosystem and habitat-specific GPP.

| Study system and experiment design
The study was carried out at the Umeå University experimental ecosystem facility (EXEF) in northern Sweden (63°48ʹN, 20°14ʹE) for 2 months (1 October-4 December 2017). The EXEF is a large-scale experimental system (73 m long, 25 m wide, with an average depth of 1.5 m) divided into 20 experimental enclosures (12.5 × 7.3 m, 136 m 3 ) by dark green PVC-sheet walls ( Figure S1). Each enclosure includes a flat soft bottom benthic habitat and one natural shoreline (7.3 m long), has separate inlets and outlets and the facility allows for manipulation of input water chemistry and water temperature.
The food web consists of benthic and pelagic primary producers, and a consumer community consisting of zooplankton (phylum Rotifera, Cladocera, and Copepoda) and zoobenthos (mainly larvae of chironomids, ephemeroptera, odonates and snails) (Koizumi et al., in preparation).
During autumn of 2017, we manipulated temperature and cDOM inputs in sixteen enclosures, while the four enclosures in the middle were left as a buffer. Eight enclosures followed the ambient autumn air temperature development and they started to be ice-covered on 30 October. The remaining eight enclosures were heated with the aim to reach 3℃ above ambient water temperature during 1 October until 12 November when they started to become ice-covered (and warming treatment was terminated). Warming was carried out separately for each of the eight enclosures by circulating water through individual land-based heat exchangers and back to same enclosure. Separate temperature sensors in one of the ambient (i.e. natural season-dependent temperature development) and one heated enclosure continuously controls the closed flow system of heated media from an air-source heat pump. The temperature difference (ΔT) between the heating media in the closed flow system and the flow water from each warm enclosure was approximately 3°C.
A gradient of cDOM concentrations was created by varying the volume of naturally cDOM-rich water inputs from a small boreal river, Hörneån (63°57ʹN, 19°25ʹE). The water was collected and transported (three times within a 14-day interval) from the source in a tanker truck to the EXEF facility and kept in a 40-m 3 tank. During the experimental period, all enclosures received the same rate (1.48 L/min) of continuous water inputs of either only clear groundwater (Umeå municipality groundwater source) or only cDOM water or a mix of both. The rate of cDOM water L/min in the total continuous rate of water inputs from low to high was; 0 (two controls), 0.07 and 0.15 (low), 0.30 and 0.59 (medium), or 0.89 and 1.48 (high). Water chemistry and phytoplankton chlorophyll a biomass (Chl a; mean ± 1 SD, n = 3) of the cDOM river water during the experimental period were: dissolved organic carbon (DOC) 25.0 ± 3.0 mg/L, dissolved inorganic carbon (DIC) 262 ± 2.6 µM, pH 6.2 ± 0.1, total nitrogen (TN) 674.2 ± 82.8 µg/L, total phosphorus (TP) 54.5 ± 2.2 µg/L, nitrate (NO 3 ) 35.3 ± 1.8 µg/L, ammonium (NH 4 ) 13.3 ± 3.0 µg/L, phosphate (PO 4 ) 24.6 ± 0.9 µg/L, and Chl a 2.6 ± 0.9 µg/L.
Warming was terminated on 13 November when the warmed enclosures became ice-covered. Clear and cDOM water input to ambient enclosures was terminated on the 31 October in connection to ice cover formation, while it was terminated on the 13 November in warm enclosures. Measures of additional response and explanatory variables were terminated on 4 December.

| Sampling and analyses
We used logging sensors to record high-frequency patterns in light, water temperature, dissolved oxygen, and wind speed. Light intensity (photosynthetic active radiation [PAR, μmol m -2 s -1 ]) was continuously measured every 10 min over the whole study period by light sensors (SQ-110, UT, U.S.A.) which were deployed at 0.8 m depth at each enclosure and recorded with loggers (Delta-T Devices, Cambridge, U.K.). Dissolved oxygen and water temperature were measured at 10-min intervals by logging sensors (PME MiniDOT, CA, U.S.A.), which were deployed as free loggers at 0.5 m below the water surface in the centre of each enclosure, as well as fixed loggers, which were placed inside the incubation chambers during incubations for pelagic and benthic measurements. Water temperature and PAR were calculated as daily average for each enclosure.
Then, we calculated the daily average of water temperature for the eight ambient and heated enclosures respectively, for whole study period. The daily average of PAR was calculated for the two enclosures of each subgroup within the main groups. Wind speed was recorded every 10 min by a climate station positioned on the shoreline of EXEF.
For chemical analyses and Chl a, 1-L water samples were collected in the middle of each enclosure from the upper meter of the water column on four occasions during the study period (every second week). Partial pressure of CO 2 in the enclosures was manually measured in situ at the same four times with a hand-held nondispersive infra-red CO 2 sensor (CARBOCAP GM70 Vaisala Inc., Helsinki, Finland). Subsamples for TP and TN analysis were taken and stored in the freezer (−20℃). Total P was measured according to Murphy and Riley, (1962), by using potassium persulfate for digesting the samples. For DOC, 50 ml of water sample was filtered through burnt (550℃, 4 hr) acid washed 0.45 µm Whatman GF/F filters, acidified with 500 µl 1.2 M HCl, and stored at 4℃ until analyses. For DOC and TN, water samples were analysed using a combustion chamber (IL550 TOC/TN analyser, Duesseldorf, Germany).
For NO 3 , PO 4 , and NH 4 , water was filtered (GF/F as above) and stored in freezer until analysis with photometric flow injection analysis method (Gray et al., 2006). Chlorophyll a was extracted (95% ethanol for 24 hr in the dark) from the Whatman GF/F filter and measured with a spectrofluorometer (Perkin Elmer LS-55, MA, U.S.A.) with the excitation wavelength set to 433 nm and emission wavelength to 673 nm.

| Whole-ecosystem and habitat-specific GPP estimates
Sensor oxygen concentrations in each enclosure were used to estimate daily whole-ecosystem GPP over the study period. Gross primary production was calculated with inverse modelling and Bayesian parameter estimation using a similar parameter estimation approach as for diel dissolved oxygen in streams (Hotchkiss & Hall, 2014), but modified for ponded waters as in Hamdan et al., (2018): where mO i is modelled dissolved oxygen at time i (g O 2 /m 3 ) given parameter estimates of GPP and ecosystem respiration (ER; g O 2 m −2 d −1 ).
Because changes in dissolved oxygen are a function of GPP, ER, and gas exchange, we calculated daily air-water oxygen fluxes based on oxygen saturation and the temperature-corrected gas exchange velocity for oxygen (K i , d -1 ). K i was estimated from K 600 derived from wind speed (Vachon & Prairie, 2013). For estimating whole-ecosystem GPP of ice-on period, the emission flux of oxygen (K i O sat,i − mO i−1 Δt) to or from the atmosphere was fixed at zero. z mix was the mean enclosure depth (m) given the enclosures are shallow and no thermal stratification (no shift >0.4°C/m in water temperature) was noticed during the experiment. The metabolism model used a random walk metropolis algorithm and Markov chain Monte Carlo sampling from the metrop function in the mcmc package of the statistical program R (Geyer & Johnson, 2013) to find the best fit between measured and modelled dissolved oxygen data given model estimates of GPP and ER. Each parameter estimate was derived from 10,000 model iterations after removing an initial 1,000 iterations of burn-in from parameter starting values. Then, we checked for convergence of parameter estimates and removed days with negative GPP and with poor fits between measured and modelled dissolved oxygen.
In each enclosure, pelagic incubations were carried out in transparent cylindrical chambers (1.9 L, diameter 85 mm) suspended vertically 0.5 m below the water surface by a metal wire fixed above and in the centre of all enclosures. Benthic incubations were performed in semi-spherical transparent chambers (12 L, diameter 350 mm) equipped with a thin metal frame (3 cm high) allowing them to be (1) October, 6 November, and 20 November). The obtained oxygen data from these incubations were used to estimate habitats specific GPP (pelagic and benthic). Gross primary production was estimated as for whole-ecosystem rates (same model as above), but the emission flux of oxygen (K i O sat,i − mO i−1 Δt) to or from the atmosphere was fixed at zero for closed chambers used to measure pelagic and benthic primary production. The term z mix also was not considered for pelagic and benthic incubations.
For comparison with habitat-specific incubation periods, daily whole-ecosystem GPP estimates were grouped into four periods.
Further, the average of habitat-specific and whole-ecosystem GPP for whole study period was calculated for each enclosure to estimate the relative contribution of habitat-specific GPP (benthic and pelagic) to whole-ecosystem GPP.

| Statistical analyses
Statistics (SPSS 20)  and to a non-mechanistic hump-shaped Gaussian curve Whole-ecosystem and habitat-specific GPP values were log 10 transformed to achieve homogeneity of variances.

| Whole-ecosystem and habitat-specific GPP
Whole-ecosystem GPP decreased with increasing cDOM input in ambient enclosures, while GPP in warm enclosures increased with low and medium cDOM inputs but decreased at high input (Figure 3a,b; Table 2). Increasing cDOM input caused pelagic GPP to increase at low and medium levels but decrease at the high level and the response was substantially stronger with warming (Figure 3d,e; Table 2). Benthic GPP decreased with increasing cDOM input and was not affected by warming (Figure 3g,h; Table 2).  Figure S2a). Pelagic GPP-to-Chl a ratios did not differ between ambient and heated enclosures along the gradient of cDOM input (t-test; df = 7, t = 0.50, p = 0.62; Figure S2b).
During the period when ambient enclosures were ice-covered and heated enclosures ice-free (third sampling period), whole-ecosystem GPP showed similar responses to the treatments as the two former sampling periods (significantly affected by cDOM input and warming), while, for habitat-specific GPP, only the effects of cDOM input were evident (Figure 3; Table 2). When both ambient and heated enclosures were ice-covered, whole-ecosystem and habitat-specific GPP were affected by cDOM input (Figure 3; Table 2).

| D ISCUSS I ON
Our results show that autumn GPP was affected by both cDOM input and warming. Increased cDOM input decreased whole-ecosystem GPP at ambient conditions by reducing benthic GPP more than stimulating pelagic GPP, while warming increased wholeecosystem GPP across all levels of cDOM due to the stimulation of mainly pelagic GPP. Differences in benthic and pelagic responses to warming and cDOM input emphasise the need to study the impacts of multiple environmental changes at both habitat-specific and the whole-ecosystem level to understand climate change effects on lake productivity.
The increase in pelagic GPP at low and medium cDOM input suggests that phytoplankton growth was nutrient limited and that the macronutrients bonded to the cDOM promoted phytoplankton growth. The fact that PO 4 did not increase with cDOM input, despite the high TP content in DOM, indicates high phytoplankton demand for phosphorus. The increased CO 2 concentration with cDOM input could potentially also have contributed to the increase F I G U R E 2 Water chemistry in ambient (a, c, e, g, i, k, and m) and warm (b, d, f, h, j, l, and n) enclosures at different allochthonous coloured dissolved organic matter (cDOM) inputs over the course of the experiment. All variables abbreviations are reported. Note the different scales of the y-axis among panels. DOC, dissolved organic carbon; TN, total nitrogen; TP, total phosphorous in pelagic GPP (Hamdan et al., 2018;Jansson et al., 2012) due to its important role as a carbon source for photosynthetic enzymes (Badger et al., 1998). In enclosures with the highest cDOM input, and highest nutrient and CO 2 levels, pelagic GPP decreased to levels similar to clearwater enclosures. We attribute this to be mainly an effect of a shift to light limitation, which prohibits further positive effects of increasing nutrients and CO 2 on pelagic GPP. We acknowledge that the responses in pelagic GPP to our treatments could have been influenced by changes in phytoplankton community composition (Flöder et al., 2002;Kissman et al., 2013). However, although it was not designed to reveal and conceptualise the exact mechanisms for the observed patterns, our study provides a strong experimental support for shifts from nutrient-to light-limited pelagic GPP with increasing cDOM inputs (Bergström & Karlsson, 2019;Kelly et al., 2018;Vasconcelos et al., 2018).
In contrast to the hump-shaped response in pelagic GPP, benthic GPP declined with increasing cDOM, suggesting light availability was the primary control on benthic GPP in the enclosures.
Soft bottom-dwelling benthic algae in general, including those in the experimental ponds (Vasconcelos et al., 2018), have access to nutrients from the sediment and sediment-water interface but also experience lower light conditions compared to phytoplankton (Bonilla et al., 2005;Daniels et al., 2015). Thus, benthic algae are not expected to benefit from additional nutrient inputs in most soft bottom-dominated lake ecosystems (Bonilla et al., 2005). While we do not have nutrient data from the sediment, we cannot exclude the possibility of nutrient availability to also constrain benthic GPP at least when light is readily available.
The combined asymmetric effects of nutrient and light limitation on habitat-specific GPP resulted in a shift in the relative contributions of the habitats to whole-ecosystem GPP, moving from benthic dominance in clearwater ponds to increasing pelagic contributions with higher cDOM input. The decline in benthic GPP was countered by an increase in pelagic GPP at low and medium levels of cDOM. In contrast, at high cDOM input when light-limited benthic GPP, light limitation exceeded the positive effects of increased nutrients on pelagic GPP and whole-ecosystem GPP declined. Taken together our study provides experimental support for the suggested mechanism behind observed patterns in whole-ecosystem GPP with increased DOC concentrations (Finstad et al., 2014;Karlsson et al., 2009;Seekell, Lapierre, Ask, et al., 2015).
In our study, warming increased whole-ecosystem GPP at all levels of cDOM input, probably due to the direct stimulating effect of warming on physiological processes involved in photosynthesis (Porter et al., 1996;Yvon-Durocher & Allen, 2012). Due to the counteracting effects on limiting resource availability by cDOM (positive for nutrients and negative for light), whole-ecosystem GPP increased with warming when nutrients increased at low and medium levels of cDOM input, but this response was moderate because of decreased light availability with higher cDOM. In fact, warming shifted the overall negative effects of increased cDOM on whole-ecosystem GPP to a hump-shaped response with increasing cDOM. This response was in turn mainly driven by the increase in pelagic GPP. Although we do not have the data to test the underlying mechanisms, the positive response in pelagic GPP with warming concomitant with an increased nutrient supply from the cDOM water treatment may be because warming enhances nutrient uptake rates (Malik & Saros, 2016;Reay et al., 1999;Rhee & Gothan, 1981).
Warming did not affect benthic GPP along the gradient of cDOM input, probably primarily due to light availability limiting benthic GPP and especially so at high cDOM input. At the highest cDOM levels, light limitation probably also became strong enough to counteract any positive temperature response in GPP even in the pelagic habitats. Still, the lack of a significant temperature response in benthic GPP in the clearwater enclosures is somewhat surprising. However, nutrient levels in the sediments of our ponds may be fairly low compared to many lakes (Vasconcelos et al., 2016), which explains the lack of temperature response in benthic GPP of the clearwater enclosures. This may suggest that nutrient availability in benthic systems could be an additional factor causing modest GPP responses in clearwater systems (McCormick et al., 2019). Variation in thermal stratification could be another factor that influences interpretations and mechanistic explanations of our results. However, the absence of thermal stratification in our enclosures, probably due to their shallow nature, gives at hand that our results cannot be explained by differences between treatments in stratification properties. Still, we acknowledge that in deeper systems, increased cDOM can cause changes in thermal stratification (Read & Rose, 2013) and impact GPP via for instance affecting nutrient movement across the water column (Staehr et al., 2012).
Warming also extended the duration of the ice-free period by tw2 weeks and during this period GPP was higher in the heated ice-free compared to ambient ice-covered enclosures. Similar to the preceding ice-free period we attribute the differences in habitat-specific and whole-ecosystem GPP to increased photosynthetic rates following higher temperatures in ice-free versus ice-covered enclosures during this period. Surprisingly, there was no significant difference between ambient and heated enclosures in incoming light intensity, presumably because the ice cover was clear and transparent (i.e. no snow cover) with low ability to intercept incoming light (Bolsenga & Vanderploeg, 1992). Hence, it is likely that with thicker and more opaque ice or snow cover than in our experiment, decreased light conditions would result in additional negative impacts on GPP.
The results of this study advance our understanding of how climate change affects aquatic productivity. This is important since lakes and ponds in northern regions are experiencing rapid and ongoing climate change (Finstad et al., 2016;Niedrist et al., 2018;Rose et al., 2016).
Our findings suggest that both cDOM inputs, presumably via associated changes in availability in light, nutrients, and CO 2 as well as warming have strong effects on lake GPP and provide experimental evidence for the predicted hump-shape response of GPP to increasing cDOM inputs (Hanson et al., 2003;Solomon et al., 2015). Warming in autumn is also likely to have positive effect on whole-ecosystem GPP. Hence, we conclude that climate change may F I G U R E 3 Mean (±1 SD) gross primary production (GPP) at the different levels of allochthonous coloured dissolved organic matter (cDOM) inputs over time in ambient (a, d, and g) and warm enclosures (b, e, and h) for whole-ecosystem, pelagic, and benthic habitat, respectively, as mean time-integrated GPP of whole-ecosystem (c), pelagic (f), and benthic habitat (i) in both ambient and heated enclosures along the gradient of cDOM inputs. Dashed line indicates transition between sampling periods of ice-free to ice-on conditions alter patterns of autumn whole-ecosystem GPP through asymmetric responses of habitat-specific GPP to warming and increasing cDOM inputs.
TA B L E 2 Results of two-way repeated-measures ANOVA (for whole experiment period) and two-way ANOVAs (for the third and fourth sampling periods) for whole-ecosystem, pelagic, and benthic gross primary production showing effect of time, allochthonous coloured dissolved organic matter (cDOM) input, warming, and their interactions