Assessing hypolimnetic oxygen concentrations in Canadian Shield lakes: Deriving management benchmarks using two methods

Abstract

The ability to predict hypolimnetic dissolved oxygen concentrations in lakes and to track changes in concentrations over time in response to known environmental stressors is critical for effective lake management. The background concentrations of deepwater oxygen, in particular, provide important management benchmarks for assessing the impact of current and future shoreline residential development on water quality. Background can be defined as the conditions that exist in the absence of, or prior to, human influence. We compare 2 models commonly used to predict end-of-summer, volume-weighted hypolimnetic oxygen (VWHO) concentrations in Canadian Shield lakes. The paleoecological and empirical models are evaluated in their ability to predict present-day VWHO concentrations, and then compared in their predictions of background VWHO concentrations and in predictions of changes in VWHO from background to present-day conditions. The predictive power of the 59-lake paleoecological model (jackknifed r² = 0.51, RMSEP = 2.18 mg/L) is comparable to other models that have used chironomids to predict the degree of hypolimnetic anoxia in lakes but is lower than that produced by the empirical modelling approach (r² = 0.87, SE = 1.04 mg/L). However, this discrepancy may be offset by the enhanced realism of the paleoecological model, including its ability to predict declines in VWHO over time. The combined use of the paleoecological and empirical modelling approaches may allow lake managers to examine changes in deepwater oxygen concentrations in response to a single “targeted” stressor (e.g., residential shoreline development) and to multiple environmental stressors (e.g., climate change, hydrological management).

Key words:

• chironomids
• hypolimnion
• lake management
• lakes
• oxygen
• Precambrian Shield

The background or “natural” water quality of a lake can be defined as the physical and chemical conditions that exist in the absence of, or prior to, human influence (Smol 2008). Background concentrations of phosphorus and deepwater oxygen, well established measures of a lake's trophic status, provide important management benchmarks for assessing the impact of current and future shoreline residential development on water quality. Impairment of these measures may be quantified as a given threshold beyond background conditions (e.g., background concentrations plus 50% for phosphorus; Environment Canada 2004), or as a known physiological boundary that, when crossed, threatens the survival and/or reproduction of a keystone species (e.g., dissolved oxygen concentration < 7 mg/L for lake trout [Salvelinus namaycush]; Evans 2007).

The ability to predict hypolimnetic dissolved oxygen concentrations (DO) in lakes and to track changes in concentrations over time in response to known environmental stressors is critical. Many aquatic organisms require that DO be maintained within narrow physiological ranges (e.g., Clark et al. 2004, Pollock et al. 2007), and oxygen plays an essential role in the geochemical cycling of redox-sensitive elements or compounds (Nealson and Saffarini 1994). Modelling hypolimnetic DO in lakes, however, is not trivial. Spatial variation among lakes in morphometry and mixing regimes, and temporal variability in weather (i.e., air temperature and wind) may affect the strength and duration of stratification and the cycling of oxygen within lakes (Burns 1995). Furthermore, oxygen levels can be both negatively and positively altered by human disturbances, such as the addition of nutrients from shoreline residential development (negatively: Dillon and Molot 1996), changes in water clarity associated with climate change and acid deposition (negatively or positively: Schindler et al. 1996, Keller 2007, Keller et al. 2008) and hydrological management for flood control or navigation (positively: Quinlan and Smol 2002).

Over the past 2 decades, quantitative models have been developed to predict present and historical hypolimnetic DO in lakes. An empirical limnological approach involves modelling the modern-day relationship between hypolimnetic DO and other limnological variables (e.g., total phosphorus concentrations), generating estimates of historical changes in these other variables and inserting these values into the predictive model to determine historical changes in hypolimnetic DO (e.g., Molot et al. 1992). The empirical approach assumes a straightforward, linear relationship between total phosphorus and hypolimnetic oxygen (i.e., if phosphorus concentrations increase, hypolimnetic DO declines; Lind 1978). A paleoecological approach involves examining subfossil assemblages of nonbiting midges (Order Diptera, Family Chironomidae) in a training set of modern-day sediment samples and relating these assemblages to measured values of hypolimnetic DO to generate an inference model for reconstructing historical values. The paleoecological model is then applied to subfossil chironomid assemblages from pre-industrial sediment samples to infer historical data and to evaluate long-term changes in hypolimnetic DO (Little and Smol 2001, Quinlan and Smol 2001). The paleoecological approach assumes that the variable being inferred is the most important variable influencing ecological structure in the indicator community (Birks et al. 1990), an assumption that is rigorously tested (e.g., Quinlan and Smol 2001) before making paleoecological inferences.

We evaluated and compared results from 2 models commonly used to predict volume-weighted hypolimnetic oxygen (VWHO) concentrations in Canadian Shield lakes. First, the models were evaluated in their ability to predict present-day hypolimnetic DO by comparing them to measured values of VWHO in 44 Precambrian Shield lakes in Ontario, Canada. Second, the models were compared with their predictions of background VWHO concentrations and to changes in these concentrations from the background to the present-day condition. We discuss the value and limitations of both modelling approaches and present an argument for the combined use of the models in lake management assessments.

Materials and methods

Study region

The Muskoka-Haliburton region of central Ontario (Canada) is located on the southern Precambrian Shield at the southern limit of the Boreal ecozone (44–45°N; 78–79°W). Forest composition in the region ranges from predominantly coniferous (e.g., Pinus strobus, Tsuga canadensis) to mixed hardwood (e.g., Acer saccharum, Fagus grandifolia). Local soils are generally shallow and acidic, overlying granitic (silicate) bedrock, with thicker deposits of clay, sand or gravel occurring locally (Jeffries and Snyder 1983). Mean annual air temperature and total annual precipitation are 4.5°C and 1034 mm, respectively (Rusak et al. 2002). Long-term monitoring and paleoecological data from the study region indicate that many of the study lakes have undergone significant temporal changes in water quality and ecology over the past few decades (Paterson et al. 2008, Quinlan et al. 2008, Yan et al. 2008). These changes are occurring regionally and coherently among lakes, and have been related to regional stressors such as changes in the deposition of strong acids (Watmough and Aherne 2008), climatic forcing (Dillon et al. 1997) and human settlement (Dillon and Molot 1996). The study lakes vary morphometrically and chemically in accordance with the natural distribution of lakes in the region (Table 1).

Table 1 The location (latitude (°N) and longitude (°W)), lake surface areas (SA), maximum depths (Zmax), total phosphorus concentrations from water samples collected at spring or fall overturn (TP), and the mean end-of-summer volume-weighted hypolimnetic dissolved oxygen concentrations (VWHO) for the study lakes used in the development of the paleoecological and empirical models. Dissolved oxygen measurements are based on 1–4 annual profiles collected for each of the study lakes between 15 August and 15 September. VWHO was calculated for each year from profile data and bathymetric maps provided by the Ontario Ministries of Natural Resources and the Environment. TP concentrations were obtained from water samples collected at spring overturn in 1990–1992 or at fall overturn in 1996 or 1998.

	Latitude (°N) (base 100)	Longitude (°W) (base 100)	SA (ha)	Zmax (m)	TP (μg/L)	VWHO (mg/L)	Included in paleo. model	Included in emp. model
Art	45.05	78.45	126.40	24.70	19.70	2.31	X
Basshaunt	45.12	78.47	47.30	24.00	6.55	3.81	X	X
Beech	45.08	78.70	124.00	26.80	8.00	6.06	X	X
Big Brother	45.18	78.80	78.00	16.00	7.59	1.88	X	X
Bigwind	45.05	79.05	111.00	32.00	5.42	5.98	X	X
Bitter	45.17	78.58	43.00	29.00	9.20	4.93	X
Black	44.82	78.82	68.00	14.00	20.40	0.16	X	X
Blue Chalk	45.20	78.93	52.35	23.00	4.97	6.10	X	X
Boshkung	45.05	78.73	716.00	71.00	5.63	8.67	X	X
Buck	45.37	79.00	40.30	30.00	7.19	4.65	X	X
Canning	44.95	78.63	244.00	22.30	8.95	2.31	X	X
Chub	45.22	78.98	34.41	27.00	8.31	4.14		X
Clear	45.18	78.72	88.40	33.00	3.91	8.58	X	X
Crosson	45.08	79.03	56.74	25.00	10.79	3.71	X	X
Davis	44.78	78.72	93.00	29.90	23.60	2.88	X	X
Dickie	45.15	79.08	93.60	12.00	7.50	1.00	X	X
Eagle	45.13	78.50	235.00	24.40	6.35	4.15	X	X
Esson	45.02	78.27	245.00	32.00	19.20	6.78	X
Frazer Island Bay (L. Joseph)	45.25	79.75	78.15	29.00	3.15	10.75	X
Glamor	44.95	78.37	194.00	22.00	16.00	2.83	X	X
Gooderham	44.87	78.38	87.00	18.90	12.40	3.16	X	X
Gullfeather	45.10	79.02	65.90	13.00	8.13	0.06	X	X
Haliburton	45.22	78.42	1014.00	54.90	7.63	8.65	X	X
Hall's	45.10	78.75	543.00	80.50	3.55	9.39	X	X
Hamer Bay (L. Joseph)	45.22	79.77	119.00	46.90	3.10	10.53	X
Hammel Bay (3 Mile L.)	45.17	79.45	152.75	13.00	24.30	0.29	X
Harp	45.38	79.15	71.38	37.50	6.33	5.14	X	X
Horseshoe	44.98	78.68	290.00	22.80	7.59	3.98	X	X
Hudson	45.07	78.13	74.00	23.00	16.00	6.12	X
Kabakwa	45.12	78.77	115.00	19.50	5.80	4.78	X	X
Kashagawigamog (S. Basin)	45.02	78.57	517.85	35.00	6.65	7.97	X	X
Kelly	45.25	78.62	99.00	31.00	5.03	7.83	X
Kennisis	45.22	78.65	1417.00	68.00	2.70	9.52	X	X
Kushog (S. Basin)	45.10	78.80	444.00	26.00	6.05	5.29	X	X
Leech	45.05	79.10	82.00	13.70	8.30	1.54	X	X
Leonard	45.07	79.45	195.00	15.20	5.68	4.37	X	X
Little Boshkung	45.03	78.73	127.00	14.30	7.59	5.63	X	X
Little Clear	45.40	79.00	11.00	25.00	10.53	1.24	X	X
Little Hawk	45.17	78.70	343.00	93.00	4.30	9.29	X	X
Long	45.05	78.37	88.20	19.80	20.80	1.59	X
Maple	45.10	78.67	336.00	36.60	7.25	6.14	X	X
Margaret	45.13	78.88	60.00	32.30	4.60	7.04	X	X
Marsden (Marsh)	45.20	78.55	230.00	21.00	7.59	4.44	X
McFadden	45.33	78.85	54.00	30.50	10.30	6.20	X
McKay	45.05	79.17	121.50	19.50	10.37	2.11	X	X
Mountain	44.98	78.72	323.00	31.40	5.50	7.29	X	X
Nunikani	45.20	78.73	116.00	24.00	4.60	7.05	X
Pine	45.12	78.58	113.00	20.40	7.59	1.44	X	X
Plastic	45.18	78.83	32.14	16.30	4.81	2.94	X	X
Poker (East)	45.05	78.93	5.40	20.50	44.70	0.50	X	X
Poker (West)	45.05	78.93	15.30	17.50	17.50	0.65	X	X
Red Chalk (East)	45.18	78.93	13.10	19.00	5.20	0.38	X	X
Red Chalk (Main)	45.18	78.93	44.10	38.00	4.21	6.54	X	X
Redstone	45.20	78.55	1130.00	82.40	5.00	9.21	X
Roothog Island Bay (L. of Bays)	45.25	79.00	1675.00	79.25	3.91	10.28	X
Seagull Rock Bay (L. of Bays)	45.25	79.00	902.70	57.20	4.33	9.68	X
Solitaire	45.40	79.00	124.00	31.00	4.59	8.16	X	X
Twelve Mile	45.02	78.72	350.00	25.00	4.73	6.98	X	X
Walker	45.40	79.08	68.20	17.00	5.40	1.31	X	X
Wilbermere	45.00	78.22	44.00	24.00	17.40	2.97	X

Modelling approaches

Two modelling approaches were used to predict the mean VWHO of lakes at the end-of-summer period (i.e., 1 Sep). This metric of deepwater oxygen was selected because other oxygen metrics do not adequately describe end-of-summer conditions (e.g., Areal Hypolimnetic Oxygen Deficit; Cornett and Rigler 1979) or describe conditions in only a small portion of the hypolimnion (e.g., “bottom oxygen” 1 m above the sediment-water interface; Quinlan and Smol 2001). In the first approach, a paleoecological model was developed using the fossil remains of chironomids (Diptera: Chironomidae). Chironomids are ubiquitous in freshwater ecosystems and well-preserved in lake sediments. They are sensitive indicators of benthic and lake profundal conditions (Brodersen and Quinlan 2006), as demonstrated in historical reconstructions of hypolimnetic oxygen concentrations (Francis 2001, Quinlan and Smol 2001) and lake-level fluctuations (Korhola et al. 2000).

Sediment cores were collected from the deepest basin of each lake using a Glew (1989) gravity corer fitted with a 6.35-cm i.d. lucite tube. Cores were sectioned using a Glew (1988) extruder at the lakeshore into 2 sediment sections: the top centimeter (representing present-day lake conditions) and a 1-cm interval taken at a minimum core depth of 20 cm (representing pre-industrial lake conditions). Although the precise date of the bottom sample of each core is unknown, radiometric ²¹⁰Pb dating (Appleby 2001) of sediment cores from more than 10 of the study lakes in the region have shown that background conditions are consistently reached at core depths of 15–20 cm (Clerk et al. 2000, Faulkenham et al. 2003).

Approximately 3–52 g (wet weight) of sediment was processed for the identification of chironomid head capsules following Quinlan and Smol (2002). Glass microscope slides of chironomid head capsules were prepared using Euparal© mounting medium, and specimens were identified using a Leitz Dialux compound light microscope at 125–630× magnification. Taxonomy followed Wiederholm (1983) and Walker (1988). A total of 6801 chironomid head capsules were recovered and identified from the surface sediment of 86 study lakes in the study region. Of these sites, 59 were included in the development of the paleoecological model because they were thermally-stratifying lakes (and thus developed hypolimnia) and yielded a suitable number of head capsules to develop meaningful quantitative inferences (Quinlan and Smol 2001). Following data screening and removal of rare taxa, 44 chironomid taxa remained for model development. Further information regarding taxonomic identifications and quality control are found in Quinlan and Smol (2001, 2002).

In the second approach, 2 models were linked to estimate background hypolimnetic oxygen profiles for each lake. First, Ontario's Lakeshore Capacity Model (LCM) was used to calculate the phosphorus budget for each lake based on natural and anthropogenic inputs of phosphorus, hydrologic and morphometric data, assumptions or estimates (Paterson et al. 2006). The model output, the total phosphorus concentration of the lake during the spring overturn period, was subsequently used as an input variable in the oxygen submodel (described later). The background phosphorus concentration of each lake was modelled by setting anthropogenic inputs of phosphorus to zero (e.g., shoreline residential development including dwellings, resort units and camp sites, % watershed as pasture, % watershed as agriculture, % watershed as urban).

Using the equations presented in Molot et al. (1992), the oxygen submodel was used to calculate hypolimnetic oxygen profiles for each lake at the end-of-summer period. The oxygen model was based on empirical relationships between oxygen concentration at specific depths in the lake, lake morphometry (e.g., the stratum volume/sediment surface area ratio), the total phosphorus concentration of the lake during spring overturn and the initial or spring oxygen concentration (see Molot et al. 1992 for equations). In our calculations, we assumed no change in lake morphometry from background to present-day conditions. When combined with hypolimnetic volumes derived from bathymetric maps, oxygen concentrations were used to calculate the end-of-summer VWHO for each lake. Input data for the empirical models were available for 44 of the study lakes.

Predicted values from both models were compared to measured oxygen concentrations obtained from oxygen/temperature profiles taken at 1- or 2-m intervals using an analog YSI dissolved oxygen meter. One to 4 annual profiles were collected for each of the study lakes between 15 August and 15 September. Measured VWHO was calculated for each year from profile data and bathymetric maps provided by the Ontario Ministries of Natural Resources and the Environment.

Results and discussion

Comparing models: Predicted versus measured VWHO

The paleoecological and empirical approaches performed well in their prediction of mean end-of-summer VWHO, and correlations of predicted versus measured VWHO were significant for both models (Fig. 1). The predictive ability of the 59-lake paleoecological model (jackknifed r² = 0.51, root mean square error of prediction [RMSEP] = 2.18 mg/L; Fig. 1a) is comparable to other models that have used chironomids to predict the degree of hypolimnetic anoxia (i.e., anoxic factor: Nürnberg 1995a, Quinlan et al. 1998; bootstrapped r² = 0.54), and the average summer dissolved oxygen concentration in lakes (Little and Smol 2001; jack-knifed r² = 0.58). In our model, as has been reported in other studies (e.g., Little and Smol 2001), a bias in the prediction of VWHO exists at the ends of the environmental gradient, where measured values are over-predicted at the low end of the gradient and under-predicted at the high end. Referred to as an edge effect, this bias is common to models created using a weighted-averaging regression approach and is caused by a truncation of the species response curves near the ends of the environmental gradient. At the low end, this bias is further created by the zero value, which provides a finite lower limit to the VWHO gradient. However, as described by Little and Smol (2001), changes in the chironomid species assemblage may occur beyond this point as species tolerant to severe anoxia are replaced by littoral taxa.

Figure 1 Regression models of (a) chironomid-inferred mean end-of-summer volume-weighted hypolimnetic oxygen concentrations (VWHO) and (b) VWHO predicted using the empirical modelling approach versus VWHO calculated from measured profile data averaged over at least 3 years. The dotted lines represent 1:1. r² _jack and RMSEP_jack are the jackknifed coefficient of determination and the jackknifed root mean square error of prediction, respectively. SE is the standard error.

The empirical model produced a higher coefficient of prediction and a lower standard error than the paleoecological model in its prediction of mean end-of-summer VWHO (Fig. 1b). With the exception of 5 large deep lakes, in which the empirical model over-predicted the observed VWHO, the predictions were close to the one-to-one line and showed minimal bias. Morphometry (i.e., lake shape and size), an input parameter that can be determined with a high degree of accuracy, accounted for the majority of the explained variance in the predicted end-of-summer values (Molot et al. 1992) and thus may explain the close relationship between measured and predicted VWHO. In contrast to the paleoecological model, where the chironomid assemblage of a given lake will be shaped by many factors including oxygen concentrations, temperature, habitat structure and quality, and top-down trophic interactions (Brodersen and Quinlan 2006), the empirical model was specifically calibrated to predict changes in oxygen concentrations in response to changes in total phosphorus concentrations only. Thus, the empirical model is conceptually much simpler, and a stronger relationship between measured and predicted values was expected. It is also possible that the predictive ability of the paleoecological inference model would be improved by increasing the number of lakes, thus strengthening estimates of species' optima and tolerances to VWHO concentrations, and by improving the taxonomic resolution of taxa, many of which are currently identified to the genus level.

Predicting background and long-term changes in VWHO

A comparison of paleoecological and empirical model predictions of present-day and background VWHO revealed a significant correlation between the 2 models across the region (Fig. 2). In general, the range of predicted values was smaller using the paleoecological model, again due to the edge effect discussed above. Despite significant agreement between the models, the 2 approaches provided very different interpretations of long-term changes in VWHO (Fig. 3). For the majority of lakes, the chironomid model inferred changes within the RMSEP, with approximately half the lakes showing a rise in VWHO and half showing a decline relative to background (pre-industrial) conditions. In contrast, the empirical model inferred either no change in VWHO (in lakes with no shoreline residential development) or a decrease in VWHO relative to background concentrations (Fig. 3).

Figure 2 A comparison of mean end-of-summer volume-weighted hypolimnetic oxygen concentrations (VWHO) predicted using the paleoecological and empirical modelling approaches. The models are compared in their prediction of present-day (left panel) and background (right panel) VWHO. The dotted lines represent 1:1.

Figure 3 A comparison of the change in mean volume-weighted hypolimnetic oxygen concentrations from background levels, predicted using the paleoecological and empirical modelling approaches.

Model comparisons: Strengths and limitations

The models presented here use very different methods to predict the mean, end-of-summer VWHO concentration of lakes on the Canadian Shield. Although both models are limited by the range of lakes in which they were calibrated, they provide a distinct set of strengths and limitations in their predictions. The paleoecological model, for example, is more expensive and labor-intensive than the empirical approach. It requires the collection of sediment cores, sample preparation and the detailed taxonomic identification of chironomid remains in addition to the collection and synthesis of monitoring data required to test and calibrate the model. However, these limitations are offset by the enhanced realism of the paleoecological model, particularly in the evaluation of changes in VWHO over time or relative to background conditions.

Concentrations of VWHO in lakes are a function of the dissolved oxygen concentrations at depth, and the volume of the hypolimnia (Quinlan et al. 2005). These measures, in turn, are influenced by many environmental factors that, when altered, may positively or negatively affect the end-of-summer VWHO (Quinlan and Smol 2002). For example, oxygen depletion rates (and consequently end-of-summer VWHO) may be affected by hypolimnetic temperature, which influences metabolic rates of bacteria at the sediment-water interface (Charlton 1980). A warmer climate may also result in earlier ice-off in spring, increasing the strength and duration of thermal stratification (Schindler et al. 1996). If oxygen depletion rates are equivalent, a lake experiencing an extended period of thermal stratification would experience lower end-of-summer VWHO. Changes in lake depth may also influence VWHO by changing the volume of the hypolimnion, thus changing the mass of oxygen available for bacterial consumption during the period of stratification (Charlton 1980, Cornett and Rigler 1987). Because chironomids respond to changes in the ambient oxygen levels regardless of the mechanism or direction of change (i.e., increases or decreases over time; Fig. 3), the paleoecological model provides greater realism and a more complete picture of long-term environmental change (Smol 2008).

Considering the typical context of water quality modelling (i.e., to evaluate and maintain ecological integrity in aquatic ecosystems), the enhanced realism of the paleoecological approach is important. Biological communities respond not only to low hypolimnetic DO, but also to the duration of low DO (Nürnberg 1995b, Little and Smol 2001). Thus, different taxa may be similarly affected by brief periods of anoxic or hypoxic conditions but have differing survivability during extended periods of anoxia (Heinis and Davids 1993). Because chironomids are relatively long-lived (1–2 years in the profundal zone of temperate lakes; Oliver 1971), their subfossil assemblages preserved in lake sediments represent not only end-of-summer VWHO but also integrate the effects of the duration of anoxia experienced.

A rise in phosphorus concentrations with increased shoreline residential development, resulting in a decline in VWHO relative to background conditions, is one of many possible outcomes for lake ecosystems, but it is the only possible outcome predicted by the empirical model. A scenario may exist, for example, where current VWHO concentrations are higher than background levels (Fig. 4, Scenario 1). As discussed earlier, this could be caused by the construction of a dam for hydrological management that increased the volume of the hypolimnion (Quinlan and Smol 2002), or by long-term declines in total phosphorus concentrations caused by drought (Schindler et al. 1996), reforestation, and/or soil acidification (Hall and Smol 1996, Quinlan et al. 2008). In this scenario, it is also possible that an increase in shoreline residential development resulted in a decline in VWHO not detectable relative to the increase in VWHO resulting from other environmental stressors (Fig. 4, Scenario 1). Indeed, a long-term monitoring record from Chub Lake in central Ontario suggests that this scenario is possible (Fig. 5). Despite an increase in shoreline residential development in this lake since the late-1970s, total phosphorus concentrations have declined and VWHO has increased over time (Fig. 5). This outcome cannot be predicted using the empirical model.

Figure 4 Conceptual diagrams showing the theoretical changes in mean end-of-summer volume-weighted hypolimnetic oxygen concentrations (VWHO) from background to current (present-day) levels. In Scenario 1 (left panel), a net increase in VWHO is caused by a multiple stressor effect resulting in an increase in VWHO (e.g., a rise in water levels that increases the volume of the hypolimnion) minus a decrease in VWHO associated with increased shoreline residential development (i.e., single stressor effect). In Scenario 2 (right panel), a net decline in VWHO is caused by a multiple stressor effect resulting in a decrease in VWHO (e.g., climate change resulting in increased strength and duration of stratification) and an increase in shoreline residential development (i.e., single stressor effect).

Figure 5 Long-term monitoring data (1976–2005) from Chub Lake in central Ontario showing a long-term decline in total phosphorus concentrations, concurrent with a long-term increase in mean end-of-summer volume-weighted hypolimnetic oxygen concentrations (bottom panel). These changes have occurred despite an increase in the number of shoreline residential developments from 1978 to 1997 (top right).

A second scenario may exist where climatic forcing has resulted in a shorter period of spring turnover, resulting in lower initial (spring) oxygen concentrations and a lower end-of-summer VWHO. Shoreline residential development may also have contributed to this decline, but because the net decrease in VWHO is caused by multiple factors it may be difficult to attribute cause (e.g., shoreline residential development) to the observed effect (i.e., decline in VWHO relative to background conditions; Fig. 4, Scenario 2).

In these examples, the empirical model can be used to isolate the impact of a single stressor (e.g., shoreline residential development) on mean end-of-summer VWHO, a distinct advantage for lake managers. Furthermore, the empirical model allows resource managers to predict possible future scenarios resulting from increased shoreline residential development or the conversion of forested catchments to intensive agriculture or urban land (Paterson et al. 2006). Even more powerful, perhaps, the combined use of the empirical and paleoecological models allows lake managers to examine changes in deepwater oxygen concentrations in response to both a single targeted stressor and multiple environmental stressors, thus encapsulating the hypothetical scenarios presented above (Fig. 4). Moreover, when combined with other models, additional environmental indicators (e.g., diatoms in sediment cores to reconstruct phosphorus concentrations in lakes; Smol 2008) and experimental work, resource managers may begin to tease apart the relative importance of the many environmental factors that affect oxygen concentrations in lakes and their impacts on biological communities. These efforts are particularly important when one considers that the protection of coldwater fisheries communities, typical keystone species in many aquatic ecosystems, is an integral part of ecosystem management (Gunn and Pitblado 2004).

Acknowledgments

We thank J. Glew, R. Hall, K. Neill, J. Havelock and S. Clerk for assistance with fieldwork. We also thank P. Dillon, N. Hutchinson, N. Yan and D. Galloway for advice on lake selection and logistical support. Measured oxygen and temperature profile data were provided by the Ontario Ministries of the Environment and Natural Resources. J. Little, J Kao, S. Clerk and K. Neill assisted with the sorting of samples for chironomid subfossils. Elements of this project were funded by an Ontario Ministry of the Environment Research Advisory Council Grant, NSERC Grants to J. Smol and R. Quinlan, and an Ontario Graduate Scholarship to R. Quinlan. We thank G. Nürnberg and an anonymous reviewer for valuable comments on this manuscript.