In 2004, ultraviolet-B (UVB) and ultraviolet-A (UVA) attenuation were measured in Mackenzie Delta lakes spanning gradients in water renewal rate, dissolved organic carbon (DOC) concentration, and dissolved organic matter (DOM) composition. DOM compositions (ratio of chromophoric DOM (CDOM) to non-chromophoric DOM) in Delta lakes are complex, evolving seasonally via flooding, dilution, macrophyte production, photobleaching, and bacterial metabolism. Attenuation was more strongly related to CDOM absorption coefficients (a330; UVB r2 = 0.69, p < 0.0001; UVA r2 = 0.58, p < 0.0001) than to DOC concentrations due to variable DOM compositions. Attenuation in one set of lakes was well related (linear models) to a330 and total suspended solids (UVB R2 = 0.80, p < 0.0001; UVA R2 = 0.81, p < 0.0001). When these models were applied to other Delta lakes, however, attenuation of UVB was overestimated in 17 of 19 cases and attenuation of UVA was overestimated in all 18 cases. This bias indicates that models are not transferrable among Delta lakes, and likely cannot be applied in other circumpolar delta lakes with similarly complex DOM compositions. Although attenuation is high in Delta lakes (KdUVB 17.1–33.4 m−1; KdUVA 7.7–19.2 m−1), plankton and photoreactive solutes may be exposed to high levels of ultraviolet radiation (UVR) because Delta lakes are shallow, UVB and UVA penetrate the top 19% and 31% of water columns, respectively, and day lengths are extended during open water. Thus, climate change effects on DOM compositions may significantly alter in situ UVR environments in circumpolar delta lakes.
 The amount and spectral composition of in situ ultraviolet radiation (UVR) plays many important regulatory roles contributing to the overall structure, function, and productivity of freshwater ecosystems [Williamson et al., 1999]. Exposure to large doses of the more energetic ultraviolet-B (UVB; 280–320 nm) wave band can cause DNA mutations in aquatic organisms [MacFadyen et al., 2004; Villafañe et al., 2004], disrupt the larval stages of many aquatic species [Vehniainen et al., 2007], and significantly alter nutrient cycles by accelerating the photolytic breakdown of terrigenous chromophoric dissolved organic matter (CDOM) [Osburn et al., 2001; Zhang et al., 2009]. The less energetic ultraviolet-A (UVA; 320–400 nm) wave band may counteract some forms of UVB-induced photodamage through photorepair processes [Quesada et al., 1995] while also contributing most of the total energy driving CDOM photobleaching due to the far higher fluxes of UVA received at the earth's surface [Vincent et al., 1998; Reche et al., 2000] (also see surface irradiance data presented in the auxiliary material). Although UVB is more biologically damaging than UVA wavelengths, it is also more rapidly attenuated in the water column by dissolved and particulate substances, thus mitigating its negative effects on aquatic biota. Within freshwater lakes the majority of UVR is attenuated by CDOM, particulate matter (herein referred to as total suspended solids, or TSS), and photosynthetic chlorophyll a pigment (Chl a) [Kirk, 1994; Scully and Lean, 1994; Häder et al., 1998; Jerome and Bukata, 1998; Belzile et al., 2002; Maloney et al., 2005]. Accurately measuring in situ UVR fluxes and attenuation have become important research objectives in freshwater environments due to the unquestionable role of UVR in many ecological processes, and the observed increases in ground level UVR fluxes resulting from ozone layer depletion.
 Arctic lakes are particularly vulnerable to the effects of elevated UVR for several reasons. First, although instantaneous [Arctic Climate Impact Assessment (ACIA), 2005] and daily [McKenzie et al., 2007] UVR fluxes in the arctic are lower than those at lower latitudes, very long or continuous day lengths during the ice-free season result in long periods of exposure for freshwater organisms. Second, arctic lakes are generally shallow and may not be deep enough to provide a refuge from UVR exposure through water column attenuation, with important implications for non-motile species such as macrophytes or benthic organisms [Häder et al., 1998; Molot et al., 2004]. Third, and perhaps most importantly, increased UVR fluxes at high latitudes are likely to act cumulatively or synergistically with climate change stressors to alter the physical, chemical and biological characteristics of arctic freshwater ecosystems [Wrona et al., 2006].
 Although poorly understood, lakes located in circumpolar river deltas are an important class of world lakes due to their abundance [Emmerton et al., 2007] and their elevated levels of productivity and biodiversity when compared to lakes found at similar latitudes outside of river deltas [Squires et al., 2009]. For instance, recent work has shown that lakes in the Mackenzie Delta, western Canadian arctic, support high levels of bacterial production [Spears and Lesack, 2006; Tank et al., 2009a] and microbial diversity [Galand et al., 2006] through high levels of macrophytic and benthic algal production in all but the most turbid Delta lakes, where phytoplankton production dominates [Squires et al., 2009]. Also, while Delta lakes are rich in dissolved organic carbon (DOC), in situ pCO2 levels are generally under-saturated relative to the atmosphere [Tank et al., 2009b]. Therefore, in contrast to the prevailing worldwide trend [Sobek et al., 2005], Delta lakes act as carbon sinks although long arctic day lengths in combination with abundant CDOM drive substantial rates of photobleaching within the system [Febria et al., 2006; Gareis, 2007].
 Additionally, Delta lakes collectively contain a diverse assortment of dissolved organic matter (DOM). DOM composition is strongly affected by the complex hydrology of the Mackenzie Delta, which is driven by annual flooding during the period of river ice breakup [Lesack and Marsh, 2007]. During the annual flood, meltwater from the southern Mackenzie Basin inundates the Delta and raises local water levels, resulting in large volumes of river water moving off-channel into Delta lakes [Emmerton et al., 2007]. Those Delta lakes that are flooded in a given year are, in varying degrees, replenished with nutrient- and carbon-rich DOM carried in river water. Patterns of DOM variation among Delta lakes, however, differ from other lake regions such as northern Wisconsin where interannual patterns of DOC and color [Pace and Cole , plus other factors [Baines et al., 2000], among lakes are to a significant degree synchronous because of strong linkage to interannual variations in runoff. In the Delta, lake water renewal rate (i.e., volumes of river water added to lake water from the prior year) is dependent on lake position along elevational gradients within the Delta floodplain [Gareis, 2007; Lesack and Marsh, 2007; Chateauvert, 2008; Tank, 2009] and lateral distance from river connection points [Squires and Lesack, 2003b]. While DOC concentrations typically increase with decreasing water renewal rates [Febria et al., 2006; Tank et al., 2009b], DOM compositions in Delta lakes are variable and complex mixtures including some or all of the following: (1) terrigenous CDOM delivered via river water [Tank, 2009], (2) non-chromophoric DOM produced in situ by abundant macrophyte communities in lakes with low water renewal rates [Tank et al., 2009a], (3) aged CDOM derived from thermokarst melting of lake beds [Tank, 2009], and (4) non-chromophoric DOM derived from CDOM photobleaching [Febria et al., 2006; Gareis, 2007]. These complex DOM mixtures, created via differing rates of lake water renewal in combination with multiple DOM pools that vary both seasonally and among lakes, may thereby create a complex underwater environment of in situ UVR among Delta lakes.
 Despite the important effects of UVR on the structure, function, and productivity of all freshwater ecosystems [Bothwell et al., 1994], our understanding of in situ UVR in circumpolar delta lakes has thus far been limited to laboratory-based absorption readings of filtered water samples [e.g., Squires and Lesack, 2003b] or empirical models derived in other systems that estimate in situ UVR fluxes from DOC concentrations or CDOM absorption coefficients (Table 1). Herein, we report the first direct measurements of in situ UVR attenuation and photic depths for a set of shallow arctic circumpolar delta lakes spanning gradients of DOC concentration and DOM composition. From these results, we tested the following hypotheses:
Table 1. Summary of Empirical Models of the Relationship Found Between Diffuse Attenuation Coefficient (Kd), DOC, Absorption Coefficient (aλ), and TSS
 H1: Attenuation of UVR wavelengths will be highest in lakes with the highest water renewal rates, and therefore the highest CDOM absorption coefficients (for this study, calculated at 330 nm; a330) and TSS concentrations derived from flood water.
 H2: UVR attenuation will be highest immediately following peak water levels corresponding to the annual flood and will decline over the open water season in response to CDOM photobleaching, reaching the lowest levels in lakes with the lowest water renewal rates.
 H3: In situ UVR attenuation can be statistically accounted for using combined values of a330, TSS, and Chl a to create empirical models.
 H4: The derived models will be sufficiently robust to estimate UVR attenuation in other lakes within the Mackenzie Delta, as well as in other circumpolar deltas.
2.1. Study Site and Sampling
 The lake-rich Mackenzie Delta (Figure 1a; ∼45,000 lakes in 13,000 km2) is flooded each spring when water from southern, upstream tributaries encounters ice cover within the Delta [Emmerton et al., 2007]. This leads to a rapid rise in local water levels that delivers a large pulse of water, sediment, nutrients, and organic matter to Delta lakes over a time interval varying from just a few days to as long as 4 months depending on small differences in lake elevation relative to the proximal river channel. As a result, water renewal rates vary widely among Delta lakes. Lakes perched at elevations above the one-year return period of high-water river levels (defined as high closure) [Marsh and Hey, 1989] have low rates of water renewal and short annual river-to-lake connection (RLC) times of less than 17 d, and are not flooded at all during some years. Lakes perched at lower elevations but above the one-year return period of low-water river levels (low closure) have appreciably higher water renewal rates with longer RLC times averaging 17 to 120 d. Lakes that sit at elevations below the one-year return period of low-water river levels (no closure) experience a near permanent connection to the river throughout the ice-free season (greater than 120 d on average) [Lesack and Marsh, 2007].
 Renewal rates may also be related to the length of the flow path that river water must follow to reach a lake [Squires and Lesack, 2003a]. When a set of lakes, all perched at the same elevation above the proximal river channel, are connected in a sequential chain stretching back from a single connection point, the input of flood water to a lake decreases, as does the water renewal rate, as distance increases.
 Following the annual flood and period of snowmelt, water levels in the Delta generally fall rapidly in concert with river levels because local precipitation is very low and usually less than local evaporation rates [Bigras, 1990]. Permafrost surrounding Delta lakes is up to 100 m thick [Johnston and Brown, 1965], which negates groundwater and subsurface flow into lakes [Marsh, 1986]. Therefore, once flood waters recede and lakes become cut off from distributary river channels, they no longer receive inputs of sediment or CDOM from river water. Some lakes are strongly affected by thermokarst processes, however, and may contribute aged DOM to the water column via melting permafrost along their lake beds [Tank et al., 2009b].
 Eleven sites in two sets of lakes in the east-central Mackenzie Delta were monitored during the summer of 2004. Sites were selected to span the range of water renewal rates as inferred by either elevation or distance from the river connection. One set of lakes consisted of 9 sites that were located within a chain of sequentially connected lakes off the East Channel of the Mackenzie River 15 km north of Inuvik, NT (Figure 1b, herein referred to as the chain set). These sites were chosen because prior work [Squires and Lesack, 2003a] had established that they represent a well resolved gradient of increasing water column transparency (based on photosynthetically active radiation) with increasing distance from the channel connection. The complete lake chain remains connected to the East Channel throughout the open water season with the degree of riverine influence controlled by distance from the river. To facilitate viewing of key changes in the gradient of in situ UVR attenuation and water chemistry, results are highlighted from 3 sites; S1, S5, and S9. The second lake set consisted of 2 discrete lakes 5 km to the southwest of Inuvik (Figure 1c, herein referred to as the sill set). The sill set lakes were end-members of a water renewal gradient inferred from lake elevation. Lake 129 (summer sill elevation = 1.27 m asl) is a no closure lake which experiences a high water renewal rate and a near continuous RLC each year, while Lake 520 (summer sill elevation = 4.59 m asl) is a high closure lake which has a low water renewal rate and an average annual RLC time of only 6.5 d (4 d in 2004; Table 2). All data collected from the eleven sampling sites over the 2004 open water season are available in the auxiliary material.
Table 2. Summary of Site Characteristics and Attenuator Data Measured at Mackenzie Delta Sampling Sites During the Open-Water Period of 2004a
Sill (m asl)
Distance to Channel (km)
DOC (mg L−1)
a*330 (L mg−1 m−1)
TSS (mg L−1)
Chl a (μg L−1)
Average (standard deviation) values are shown for DOC, absorption coefficient (a330, m−1), specific absorptivity (a*330, L mg−1 m−1), TSS, and Chl a. All attenuators were measured 7–9 times at each site during the summer of 2004. LRC = 2004 lake-to-river connection time (i.e., number of days during 2004 that the lake was flooded due to flood heights in excess of lake sill elevation). For the chain set (S1-S9) LRC was continuous for the open water period.
2.2. Spectral Irradiance Scans
 Spectral downwelling UVR was measured in situ at each lake site using a submersible integrating sphere (OL IS-470-WP, Optronics Laboratories, Orlando, FL) attached to a UV-visible scanning spectroradiometer (OL 754-O-PMT, Optronics Laboratories) via a 10 m quartz fiber optic cable. The spectroradiometer was calibrated against a NIST traceable 200 W tungsten-halogen lamp (OL 752–10, Optronics Laboratories). A dual calibration and gain check source module (OL 752–150, Optronics Laboratories) was used to confirm the wavelength and gain accuracy of the instrument immediately prior to each field use.
 The submersible integrating sphere was extended from the port side of a small boat on a 3 m parallelogram swing arm that allowed the submersible sphere to be lowered away from the boat and held at a precise selected depth with the aperture parallel to the water surface [Arts et al., 2000]. The boat was anchored in place with the sun positioned directly off the port side to avoid shading of the integrating sphere aperture. Therefore, only the diffuse component of the global irradiance field was affected by boat shadow.
 Vertical profiles consisted of spectral scans of downwelling UVR (280–400 nm, 2 nm intervals) taken immediately beneath the surface and at several sequential depths in the water column to a maximum depth of 1 m. In most cases several replicate scans, each lasting 1–2 min, were taken at each depth. All scans were taken as close to solar noon (approximately 1500 h Mountain Daylight Time) as possible to reduce changes in incident UVR associated with changing solar elevation. Studies of in situ UVR typically use only measurements taken on days with completely clear sky conditions in order to avoid any effects of changing cloud cover. This is difficult in the western Canadian arctic since summer sky conditions are prone to rapid changes in cloud cover. Therefore, in order to maximize the frequency of in situ scanning, measurements were taken during periods with constant sky conditions (either clear or evenly overcast) as in the work of Laurion et al. .
 Instrument-specific immersion correction factors were applied to raw spectral data to correct for differences in the indices of refraction between air (where the instrument was calibrated) and water (where the instrument was operated). These values ranged from 1.6 to 1.8, similar to values obtained for other underwater spectroradiometers [Kuhn et al., 1999].
 Diffuse attenuation coefficients normalized to 1 m (Kdλ, m−1) were calculated for individual wavelengths between all possible pairs of scans taken at a site on the same day following Kirk :
where E−0λ and Ezλ represent wavelength-specific measurements of irradiance taken at different depths, z (m), in the water column. Kdλ is considered a quasi-inherent optical property of aquatic systems when the majority of surface irradiance is direct [Kirk, 1994]; however, this breaks down as solar zenith angles (SZAs) increase as the sun approaches the horizon, resulting in a high proportion of diffuse surface irradiance requiring correction. Therefore, SZAs were calculated for the midpoint time of each scan using the MIDC SOLPOS calculator (Solar Position and Intensity (SOLPOS), version 2.0, Measurement and Instrumentation Data Center, National Renewable Energy Laboratory, 2005, http://www.nrel.gov/midc/srrl_bms/). The SZAs were then used to calculate the direct and diffuse portions of incident UVR at each wavelength using the SMARTS model developed by C.A. Gueymard (Simple Model of the Atmospheric Radiative Transfer of Sunshine (SMARTS), version 2.9.2, National Renewable Energy Laboratory, 2002, http://www.nrel.gov/rredc/smarts/). Since the SZA for all in situ scans varied from 45 to 75°, attenuation coefficients at each wavelength were further corrected for the geometric condition of the irradiance field by dividing by the wavelength-specific correction factor, D0λ [Gordon, 1989]:
where fλ is the proportion of direct incident irradiance at a wavelength and ν0wλ is the SZA below the water surface, calculated by applying Snell's Law to the SZA above the water surface using wavelength-specific indices of refraction for water given in the work of Segelstein .
 Computed Kdλ values were visually inspected for obvious errors and all negative values were deleted. Since attenuation decreases with increasing wavelength in the UV range, any Kdλ value that was lower than those at longer UV wavelengths were also deleted. Erroneous values were removed across entire wave bands to avoid skewing wave band averages. Values of Kdλ at wavelengths less than 310 nm were especially prone to error, so all Kdλ values below 310 nm were removed from all data sets. Remaining data were used to calculate average wave band diffuse attenuation coefficients for each site and date; KdUVB, herein covering the partial UVB spectrum from 310 to 320 nm, and KdUVA, covering the full UVA spectrum from 320 to 400 nm.
 Values of Kdx (with x designating a wave band) for all sites and dates were converted to 1% photic depths (z1%x, cm), or the depth at which 99% of the incident wave band irradiance had been attenuated, following Molot et al. :
where D0x is the average correction factor across the entire wave band [Gordon, 1989].
2.3. Water Sampling and Analysis
 Water samples were taken at the same time as in situ scans for measurement of CDOM absorption coefficients, TSS, Chl a, and DOC. Dip samples were taken from below the water surface in acid-cleaned and DDI-rinsed (10 times) 1L HDPE bottles, and were kept cool and in the dark until processed in the laboratory (within 12 h). TSS concentrations were determined gravimetrically by filtering sample water onto pre-ashed and pre-weighed Whatman GF/C filters (nominal pore size 1.2 μm) [Emmerton et al., 2008]. Chl a samples were filtered onto Whatman GF/C filters and stored frozen until extraction in 90% acetone followed by fluorometric analysis [Wetzel and Likens, 2000]. No acid correction for phaeophytin was made. CDOM samples were filtered (0.2 μm pore size, Millipore GSWP, Millipore Corporation, Billerica MA) and kept cool and in the dark until spectrophotometric measurement of absorption at 330 nm. Absorption coefficients (a330, m−1) were calculated as indicators of CDOM levels, with higher a330 indicating proportionally higher levels of CDOM, as follows:
where A330 is the raw absorbance measurement at 330 nm and l is the cuvette path length in m [Kirk, 1994; Whitehead et al., 2000]. We report absorption coefficients at 330 nm in this paper for consistency with prior publications on the Mackenzie Delta. DOC samples were filtered using Whatman GF/F filters (nominal pore size 0.7 μm) and kept cool and in the dark until analysis as non-purgeable organic carbon following the high temperature catalytic oxidation method (Shimadzu TOC-V, Shimadzu Corporation, Kyoto, Japan). Specific absorptivity at 330 nm (a*330, L mg−1 m−1) was calculated as a measure of CDOM absorption per unit DOC as follows:
Since UVR in aquatic ecosystems is more strongly absorbed by terrigenous, aromatic CDOM than by non-aromatic, non-chromophoric DOM [Belzile et al., 2002], specific absorptivity is used in this study as an index of the proportion of terrigenous DOM in the total DOC pool. Higher a*330 ratios therefore indicate a greater proportion of terrigenous, aromatic CDOM, while lower ratios indicate a greater proportion of aquatically produced or photobleached non-chromophoric DOM.
2.4. Statistical Analysis and Construction of Attenuation Models
 JMP 6.0.0 statistical software (SAS Institute, Cary, NC) was used to examine a series of simple linear regressions between attenuator (CDOM absorption coefficients reported herein as a330, DOC, TSS, and Chl a) levels and measured Kdx values for each site. Pearson coefficients of determination (r2) were calculated for each regression, and the resulting patterns of significant attenuation (p ≤ 0.05) by individual attenuators were used to examine the influence of water renewal rates on in situ UVR attenuation within Delta lakes. Linear empirical models using only a330 to predict KdUVB and KdUVA values in the chain set were derived from these results.
 Models of UVB and UVA attenuation in the chain set that incorporated multiple attenuators (a330, TSS, and Chl a) as predictor variables were also created. Stepwise multiple linear regression (MLR) analyses were conducted [Vant and Davies-Colley, 1986; Vant, 1990, Squires and Lesack, 2003b] using JMP 6.0.0, with p < 0.15 required for an attenuator to be entered into the model. Levels of attenuators measured in the sill set were then entered into all chain set models to evaluate how accurately the models predicted UVR attenuation in other Delta lakes. Differences between attenuation coefficients directly measured in the sill set (measured Kdx) and attenuation coefficients calculated by entering attenuator concentrations measured in the sill set into model equations (modeled Kdx) were plotted against day of year in order to assess any seasonal trends in model accuracy.
3.1. Patterns of in Situ UVR
 At most sites, attenuation of UVB wavelengths was highest immediately following the spring flood and declined rapidly thereafter (Figure 2a; see also the auxiliary material). In the chain set, the highest initial KdUVB values were found at S5 in the middle of the chain. Decreases in KdUVB over the open water season were generally largest at sites in the middle of the chain set, followed by sites closest to the channel connection. Declines in KdUVB were generally smallest at sites distant from the channel connection, where water renewal rates are lower and the influence of the annual flood was considerably weaker. High KdUVB values were consistently found in Lake 129, a lake that experiences nearly continual water renewal via a long connection to the river channel (RLC of 63 d in 2004; Table 2). Similar to the chain set lakes, KdUVB in Lake 129 was highest immediately following the spring flood and displayed an overall decline for the rest of the open water season. Conversely, Lake 520 (RLC of 4 d in 2004; Table 2) displayed the opposite trend over the open water season, with KdUVB increasing from 15.8 m−1 on 21 June to 19.7 m−1 on 9 July. KdUVB remained relatively high in Lake 520 for the rest of the sampling period.
 At most sites, z1%UVB increased from approximately 10–15 cm in mid-June to 20–25 cm in early August (Figure 2b), indicating increasing penetration of UVB through the water column. Again, the exception to this trend occurred in Lake 520, where z1%UVB was relatively constant over the open water period.
 Similar to the UVB wave band, KdUVA values were highest immediately following the flood at most sites and declined as flood water receded (Figure 2a; see also the auxiliary material). The decline in KdUVA occurred rapidly, with curves flattening out by the middle of the open water season. Again, similar to KdUVB trends, KdUVA was generally highest throughout the open water season at sites in the middle of the chain set, where CDOM levels were also highest (Table 2), while KdUVA was generally lowest at sites furthest from the channel connection, where flood influence was considerably weaker. KdUVA remained low at distal sites throughout the open water season. The exception to the observed trend of declining KdUVA again occurred in Lake 520, where values remained fairly constant throughout the open water season, fluctuating between 7.4 and 9.7 m−1.
 At most sites, z1%UVA increased over the open water season from approximately 20–30 cm in mid-June to 35–50 cm in early August (Figure 2b). Again, the exception to this trend occurred in Lake 520, where z1%UVA decreased over the open water season.
3.2. Specific Absorptivity
 Most Mackenzie Delta sampling sites had a*330 values in the range of 2.5–3.0 L mg−1 m−1 at the beginning of the open water season (Figure 2c; see also the auxiliary material). As the open water season progressed, these sites generally experienced steady declines in a*330. Site S1 in the chain set showed a sharper decline in a*330 in the middle of the open water season (2.36 L mg−1 m−1 on 30 June to 1.04 L mg−1 m−1 on 20 July) than did any other site. In contrast to all other studied Mackenzie Delta sites, a*330 values in Lake 520 remained low and constant over the open water season.
3.3. Contributions of a330, DOC, TSS, and Chl a to UVR Attenuation
 KdUVB and KdUVA were significantly correlated with a330 at all sites (Tables 3a and 3b; p ≤ 0.05 in all cases). Generally, Pearson coefficients of determination (r2) for a330 versus either KdUVB or KdUVA decreased with increasing distance from the channel connection in the chain set lakes. This indicates that other attenuators became increasingly important in determining UVR attenuation as water renewal rates and channel influence decreased. In the sill set, a330 explained more variation in Kdx in Lake 520 than in Lake 129, although both were significant relationships (Tables 3a and 3b). TSS was significantly correlated with UVR attenuation at S1, a site directly adjacent to the East Channel, while Chl a was significantly correlated with UVR attenuation at sites S5, S9 and Lake 129 (Tables 3a and 3b). DOC was significantly correlated with UVR attenuation only at sites with high water renewal rates in close proximity to the channel (Tables 3a and 3b).
Table 3a. Site-by-Site Results of Correlation Analysis Between Measured Values of KdUVB and Absorption Coefficient (a330), DOC, TSS, and Chl aa
In all cases, simple linear regressions were used to assess the correlation between measured Kdx values and levels of individual attenuators. Values displayed in the chart are Pearson coefficients of determination (r2), with results in bold italics indicating significant linear relationships (p ≤ 0.05).
Table 3b. Site-by-Site Results of Correlation Analysis Between Measured Values of KdUVA and Absorption Coefficient (a330), DOC, TSS, and Chl aa
In all cases, simple linear regressions were used to assess the correlation between measured Kdx values and levels of individual attenuators. Values displayed in the chart are Pearson coefficients of determination (r2), with results in bold italics indicating significant linear relationships (p ≤ 0.05).
 In the chain set lakes, more of the variation in both KdUVB and KdUVA values was explained by linear models using a330 as the sole predictor than by power models using either a330 or DOC concentrations (data not shown for power models), in contrast to the relationships found in some other systems (Table 1). Linear models for the chain set lakes are
 Although DOC has often been found to be the strongest predictor of UVR attenuation in aquatic systems (Table 1), DOC was eliminated from MLR analyses in this study because DOC pools contained variable proportions of non-chromophoric DOM (Figure 2c) which reduced the power of DOC to predict variations in Kdx. When MLR analyses were run with DOC included, very little additional explanatory power was gained. The low power of DOC concentrations to predict variation in Kdx is further supported by the narrow range of DOC concentrations measured at chain set sites (4.46–8.99 mg L−1; see the auxiliary material) as compared to the relatively wide range of measured a330 values (3.54–26.72 m−1; see the auxiliary material).
 Therefore, MLR analyses were used to determine the contributions of a330, TSS, and Chl a to UVR attenuation in chain set lakes. Although the simple linear regression between KdUVB and Chl a (r2 = 0.23, p = 0.0002; Figure 3a) across all chain set sites was stronger than that between KdUVB and TSS (r2 = 0.04, p = 0.164; Figure 3b), Chl a did not contribute significantly to KdUVB during MLR analyses. The same patterns of significant correlation across all chain set sites were found for simple linear regressions involving KdUVA values (data not shown). Therefore, only a330 and TSS were found to contribute significantly to UVB and UVA attenuation in the chain set lakes. Model equations for each wave band are
Values of KdUVB and KdUVA calculated by entering a330 and TSS values measured in the sill set into model equations (modeled Kdx) differed widely from measured values (measured Kdx; Figure 4). Modeled Kdx ranged from 96.2 to 141.4% of measured Kdx values, with an average of 112.1%. Although a few modeled Kdx were within 5% of measured Kdx, most were overestimates of the corresponding measured Kdx value. Differences between modeled and measured Kdx values were plotted against day of year in order to determine any seasonal trends in model accuracy; however, no seasonal trends were found for any of the four models.
4.1. Attenuation of UVR in Delta Lakes
 Previous measurements of underwater UVR in arctic lakes are predominantly from the high arctic where CDOM levels are very low due to scarce vegetation in the surrounding catchment. By comparison, Mackenzie Delta lakes have high CDOM levels largely resulting from riverine inputs from the heavily forested southern Mackenzie Basin, with smaller contributions from local drainage following flooding and melting permafrost [Febria et al., 2006; Tank et al., 2009b]. Generally, values of KdUVB and KdUVA measured in Mackenzie Delta lakes (Figure 2a) far exceed those found in high arctic lakes, and are also higher than those found in many subarctic, temperate and alpine lakes (Table 1).
 The annual open water season in the Delta, lasting from June to October, coincides with the period of continuous or very long day lengths at arctic latitudes. Therefore, aquatic organisms are continuously exposed to UVR, although cumulative daily fluxes are lower than at lower latitudes [McKenzie et al., 2007]. Calculated z1%x values reveal that Delta lakes contain sufficient attenuating material to remove virtually all UVB by 30 cm depth and all UVA by 50 cm depth (Figure 2b). Below these depths, the water column is “sun screened” and organisms are protected from the negative effects of UVR exposure. Delta lakes are generally shallow, however, with average depths of only 1.6 m [Emmerton et al., 2007], which results in the top 19% of the water column being exposed to UVB wavelengths and the top 31% to UVA. Additionally, Delta lakes are well-mixed by wind and do not permanently stratify [Febria et al., 2006], so non-motile planktonic taxa such as phytoplankton and microbes are exposed to UVB for variable periods of time as they are passively moved through exposed surface layers. When wind speeds are low, mixing rates will also be low, and plankton near the surface of Delta lakes may receive fluxes of UVB high enough to cause damage either at the cellular level or, in the case of phytoplankton, through photosynthetic inhibition [Xenopoulos et al., 2000].
4.2. Effects of River Flooding and Water Renewal Rates on in Situ UVR
 Patterns of UVR attenuation observed in Mackenzie Delta lakes are related to river-water renewal rates, but not in a straightforward manner. Within-lake DOM pools, composed of UVR-absorbing CDOM and non-absorbent, non-chromophoric DOM, along with rates of photobleaching and dilution of these pools, are determined both directly and indirectly by water renewal rates within this system.
 CDOM photobleaching, which has been experimentally examined in these lakes [Febria et al., 2006; Gareis, 2007], likely contributed to declines in UVR attenuation at all sites over the open water season. During 2004, measurable declines in a330 were observed in microcosms of lake water from the chain set, Lake 129, and Lake 520, when exposed to just 8 h of full spectrum midday sunlight in Inuvik [Gareis, 2007]. In the current study, a*330 values were used to indicate changes in the proportion of CDOM absorption at 330 nm relative to total DOC concentration. At most sites a*330 declined (Figure 2c), indicating losses in UVR absorption capacity over the open water season that were consistent with the losses of CDOM aromaticity and terrigenous DOM character seen during photobleaching [Belzile et al., 2002], although some of this decline could have also occurred via dilution of the DOM pool with river water low in CDOM.
 Mackenzie Delta sites with high water renewal rates, such as S1, S5, and Lake 129 (1.27 m asl), had more variable Kdx values throughout the open water season (Figure 2a; see also the auxiliary material). Such temporal variability is consistent with prior results seen during continuous logging of PAR attenuation in the chain set [Squires and Lesack, 2003a]. Although large amounts of CDOM and TSS were delivered to lakes with high water renewal rates during the flood, once water levels fell and meltwater from southern tributaries had flushed through the delta, river water became depleted in both substances (Figure 5). Observed declines in CDOM and TSS carried in river water in 2004 co-occurred with declines in a330, TSS, and Kdx at lakes/sites with a strong river linkage, suggesting that river water dilution was important in these locations. Additionally, a sudden decrease in a*330 was observed at S1, located close to the river connection point, in early July (Figure 2c). Such a decline was too rapid to be accounted for by photobleaching, and further indicates that dilution with CDOM-depleted river water contributed to declines in Kdx at sites located near channels because Figure 5 indicates that the sudden decrease in a*330 at S1 was concurrent with the post-flood period of rapid decline in river water a330 values.
 The site with low water renewal rates, Lake 520 (4.59 m asl), had more stable Kdx values throughout the open water season (Figure 2a). Like all lakes perched high above their proximal river channels, Lake 520 experiences little or no flooding during some years, and thus, minimal lake-water replenishment with terrestrial CDOM derived from flood water [Lesack and Marsh, 2007]. In addition to low, near-constant a330 values relative to the other sampled lakes, DOC concentrations in Lake 520 exceeded those measured at any other site and rose steadily over the 2004 open water season (Table 2 and the auxiliary material). Moreover, a*330 values were lower in Lake 520 than at other sites, and remained constant throughout the open water season (Figure 2c), indicating a large and stable proportion of non-chromophoric DOM in the DOC pool. When flood-derived CDOM does not dominate DOC, as was the case in Lake 520, macrophyte- or thermokarst-derived DOM can become important contributors to the total DOC pool [Tank, 2009]. The observed increases in KdUVB in Lake 520 during June and early July 2004 may be due to the aggregation of small macrophytically produced DOM molecules into larger molecules exhibiting properties similar to CDOM [Tranvik, 1993], or aged terrigenous DOM released into the water column as a result of thermokarst processes in the lake bed [Tank et al., 2009b]. Both sources would offset the continual photobleaching of any small amounts of CDOM contributed to Lake 520 through runoff from the surrounding landscape, which would give rise to the constant a330 and a*330 values that were observed.
 Generally, absorption by CDOM alone accounts for the majority of UVR attenuation in freshwater lakes [Morris et al., 1995; Laurion et al., 1997; Gibson et al., 2000]. In Mackenzie Delta lakes, however, CDOM is delivered in varying degrees via river water during the annual spring breakup and flood. The composition of DOM pools is subsequently altered over the course of the open water season via photobleaching [Febria et al., 2006; Gareis, 2007], macrophyte production of DOM exudates [Tank et al., 2009a] and subsequent macrophyte decomposition, thermokarst processes [Tank, 2009], and river water dilution (once flood waters have passed through the Delta). As a result, DOM pools in the Mackenzie Delta are highly complex mixtures that vary widely among lakes, with DOM composition evolving seasonally in response to the various forcing factors. The complex optical environment found in the Mackenzie Delta indicates that lakes in other circumpolar river deltas with similar flood-driven hydrology, coupled with seasonally and spatially variable sources of DOM to individual lakes, are also likely to exhibit similarly complex patterns of in situ UVR attenuation.
4.3. Influence of a330 and TSS on in Situ UVR Attenuation
 At most sampling sites in this study, a330 more strongly predicted UVR attenuation than did DOC concentration (Tables 3a and 3b). This may be a consequence of DOC concentrations having only moderate ranges of seasonal and within-lake variation when compared to other measured attenuators (see the auxiliary material). Values of a330 were significantly correlated with both KdUVB and KdUVA at every site monitored during 2004, while DOC was significantly correlated with KdUVB and KdUVA only at those chain set sites located closest to the channel connection. This may indicate that, at greater distances from the channel connection, CDOM levels were too low for the DOC pool to contribute significantly to UVR attenuation (i.e., a large proportion of the DOC pool was composed of non-chromophoric DOM). Prior work in the Delta indicates that at sites distant from a source of river water, two mechanisms may account for the dominance of non-chromophoric DOM in the DOC pool; either distant sites receive only small fluxes of CDOM via flood water, which are rapidly photobleached [Febria et al., 2006], or high rates of macrophyte-derived, non-chromophoric DOM exudates will dominate the DOC pool [Tank, 2009]. In either instance, DOC will primarily be composed of non-aromatic carbon compounds that absorb considerably less strongly in the UVR wavelength range.
 TSS, which is carried into lakes via flood and river water, was significantly correlated with KdUVB and KdUVA only at the site closest to the channel connection (S1) (Tables 3a and 3b). An examination of average open water TSS values at all chain set sites indicates that TSS largely settles out of the water column within 1 km of the river channel (Table 2 and the auxiliary material). The contribution of TSS to total attenuation in MLR models may be due to the very high concentrations of TSS found in Delta lakes immediately post-flood (auxiliary material). Chl a was not found to contribute significantly to linear attenuation models in the chain set in spite of its stronger relationship to KdUVB (r2 = 0.228, p = 0.0002) as compared to TSS (r2 = 0.036, p = 0.164; Figure 3). This is may be due to the low levels of phytoplankton productivity, and high levels of macrophytic and benthic algal productivity, found in Delta lakes that are isolated from channel connection points by either sill elevation or lateral distance [Squires et al., 2009].
 While over 50% of the variation in either KdUVB or KdUVA in the chain set lakes could be explained by a330 alone, substantial improvements in model fit were found when TSS concentrations were included. When chain set models were used to model in situ UVR attenuation in the sill set lakes located 15 km to the southwest, however, modeled Kdx ranged from 96.2–141.4% of measured Kdx (Figure 4), with most modeled Kdx overestimating in situ attenuation. This bias toward overestimation of modeled Kdx indicates that empirical models derived in the chain set were not transferable to other Delta lakes. This is likely due to high inter-lake variability in DOM composition over the open water season, even among Delta lakes perched at similar elevations above their proximal channel, resulting in variable water column transparency to UVR. For example, both Lake 129 (sill elevation of 1.27 m asl) and the chain set lakes (sill elevation of 1.50 m asl) are no closure lakes with high water renewal rates. Although this implies that empirical models derived in the chain set should more accurately predict UVR attenuation in Lake 129 than in Lake 520 (high closure lake with sill elevation of 4.59 m asl), this was not the case.
 It seems unlikely that models derived in 2004 could be accurately applied within the chain set in other years. In addition to high interannual variability in flooding hydrology and the resultant variability in DOM compositions among lakes, TSS concentrations can vary widely over time since they are determined by riverine inputs from the proximal channel. For instance, during the open water season storm surges originating in the Beaufort Sea can temporarily raise local water levels within the Delta [Marsh and Schmidt, 1993; Manson and Solomon, 2007] which pushes river water into lakes in close proximity to a channel connection. A storm surge was observed during early August 2004, and was confirmed by discharge and TSS measurements in the East Channel (Figure 5) as well as by concurrent increases in attenuation (particularly KdUVA) at chain set sites close to the channel connection (see Site S1; Figure 2a and auxiliary material). There was no observed concurrent increase in a330 either in the East Channel or at Site S1, indicating that CDOM levels in river water were indeed depleted by this point in the open water season (Figure 5). Storm surges further complicate the prediction of in situ attenuation using standard limnological measurements and limit the transferability of empirical models to other years, other Mackenzie Delta lakes, or other lake systems. Any attempt to apply empirical models relating in situ UVR attenuation to attenuator concentrations should therefore be approached cautiously.
 Synchrony, interpreted here as the concurrent variation in the time series pattern of water chemistry constituents among lakes [Pace and Cole, 2002], is unlikely to be present among DOM compositions of Mackenzie Delta lakes except in a minority of lakes with near-continuous river flow-through. Although the lakes studied herein exhibit clear gradients in several constituents as well as UVR attenuation, the underlying DOM compositions driving UVR attenuation are variable both seasonally and spatially. Therefore, attempts to develop comprehensive empirical models that would estimate in situ UVR attenuation in Delta lakes with reasonable accuracy using only a few simple water chemistry measurements have been unsuccessful. Other lake-rich circumpolar river deltas with similarly complex flooding hydrology and DOM compositions are also likely to present similar problems when modeling UVR attenuation.
4.4. Implications of Research
 Because CDOM plays a major role in determining the extent of UVR penetration through the water column of Mackenzie Delta lakes, any processes resulting in changes in total DOC concentration or DOM composition will influence in situ UVR fluxes. These changes may result from a number of anticipated climate change scenarios. Annual average temperature increases in the western Canadian arctic are expected to exceed the global average [ACIA, 2005]. This may result in increased permafrost melting, increased thermokarst slumping, and increased growth of terrestrial vegetation in the Mackenzie Basin. Each of these changes could result in increased CDOM levels, and thus, decreased UVR penetration in Delta lakes. There is also evidence that peak annual flood heights in the Delta may have declined since 1964 due to declining ice breakup effects, while annual low-water levels in Delta channels, and in lakes connected to them, may have increased due to sea level rise [Lesack and Marsh, 2007]. If this trend continues, a shift toward lower annual flood heights would reduce the volume of water entering all Delta lakes each spring, decreasing water renewal rates in lakes that are either perched high above their proximal river channels or are distantly connected. Both would result in lower levels of CDOM delivered to such lakes, thereby altering the composition of the DOM pool. Proportionally greater DOM contributions from macrophyte production, in particular, should increase fluxes of in situ UVR. Rising low-water levels should expand the number of lakes affected by coastal storm surges, which temporarily increase channel levels and deliver river water into connected lakes [Lesack and Marsh, 2007]. UVR attenuation may either increase or decrease in affected lakes, depending on the levels of attenuating constituents in the river water delivered to them. It seems likely that changes in DOM composition resulting from climate change effects on water levels in the Delta will be significantly stronger than those resulting from direct temperature effects on bacterial processing of DOM or aquatic primary production in the Delta.
 Changing CDOM levels will likely have a greater effect on UVR exposure of aquatic communities than will increases in incident UVB resulting from ozone depletion [Schindler et al., 1996; Vincent and Pienitz, 1996; Molot et al., 2004], although current anthropogenic emissions of nitrous oxide, an ozone-depleting compound released from agricultural fertilizers, by no means preclude further ozone layer depletion [Ravishankara et al., 2009]. Other lake-rich circumpolar river deltas with forested basins will likely undergo similar changes in the quantity and spectral composition of underwater UVR fluxes in response to both changing CDOM levels (resulting from climate change) and surface UVB fluxes (resulting from ozone layer depletion, which is more severe at polar latitudes). Because circumpolar delta lakes are abundant and serve as critical habitat for many species of fish, mammals, and waterfowl [Rouse et al., 1997; Prowse et al., 2009; Squires et al., 2009], potential changes in underwater UVR fluxes will likely be important within this region. Moreover, because many bird and fish species utilizing arctic deltas are migratory, the effects of habitat alteration may be felt well beyond delta boundaries.
 Our findings are generally inconsistent with our four a priori hypotheses regarding in situ UVR behavior in the Mackenzie Delta, and indicate that UVR plays a complex and important role in the ecology of circumpolar delta lakes.
 DOC concentration was poorly related to UVR attenuation since in situ DOM mixtures in the Mackenzie Delta are complex and evolve seasonally via flooding, dilution, photobleaching, in situ production of macrophyte DOM, and bacterial metabolism. This leads to considerable among-lake variations in the proportion of the DOC pool that is composed of terrigenous CDOM, the UVR absorbing component of DOC pool. UVR attenuation, however, was well related to indices of CDOM absorption (a330), confirming the strong effect of annual river water delivery on the in situ UVR environment of Delta lakes. Water renewal is an important control on UVR attenuation in the Delta, but its role varies seasonally. Higher water renewal rates enhance Kdx early in the open water season by delivering high levels of CDOM to flooded lakes, but later decrease Kdx by diluting CDOM levels in lakes that lie in close proximity to channels. In spite of the importance of river flooding, TSS was only weakly related to Kdx values over the open water season on a site-by-site basis, possibly due to the rapid settling of TSS out of the water column once flood water receded.
 Kdx values measured in a set of chain-linked lakes were well related to a330, as well as to a330 plus TSS. When these models were used to estimate UVR attenuation in two other Delta lakes, however, modeled Kdx varied from 96.2–141.4% of measured Kdx. The bias toward overestimation indicates that the derived models cannot be reliably transferred to other Mackenzie Delta lakes, and likely cannot be applied to other circumpolar delta systems with similarly complex mixtures of DOM. Prior work on regional DOC patterns and dynamics in other non-arctic lake systems has suggested that, to some degree, DOM levels and underwater UVR may be predictable from patterns of runoff. Arctic deltas may be a counterexample to such observations, representing a lake-class where the in situ UVR environment is complex and where empirical relations that work in other systems readily break down.
 Our in situ measurements have documented relatively high Kdx values in the Mackenzie Delta as compared to those found in other aquatic ecosystems. Despite such high values, potential exposure of planktonic organisms and photo-reactive solutes to UVR is surprisingly high because Delta lakes are shallow and well-mixed, with average 1% photic depths encompassing the top 19% (UVB) and 31% (UVA) of the average water column depth. Additionally, Delta lakes experience prolonged or continual day lengths during the open water season. Therefore, despite indices of CDOM absorption suggesting that these lakes should be well shielded from UVR, our findings suggest that climate change effects on DOM composition, in concert with ozone depletion, may act to significantly alter underwater UVR exposure in this important class of circumpolar lakes.
 The authors would like to acknowledge Craig Emmerton, Catherine Febria, and Shannon Turvey for their help in the lab and field, as well as Les Kutny and the Inuvik Research Centre/Aurora Research Institute for technical and logistical support. Financial support was received from NSERC (DGP and NRS programs to Lesack, PGSA grant to Gareis), the Polar Continental Shelf Project (helicopter support to Lesack), and the Northern Scientific Training Program (to Emmerton, Febria, Gareis, and Turvey). Use of facilities in Inuvik was subsidized by the Aurora Research Institute. We appreciate the loan of field equipment from Vijay Tumber of NHRI Saskatoon. Donovan Lynch provided valuable spectroradiometer training and support. We also wish to thank three anonymous reviewers for their comments and suggestions on an earlier draft of the manuscript.