River-to-lake connectivities, water renewal, and aquatic habitat diversity in the Mackenzie River Delta



[1] Past and ongoing investigations have established that lakes in the Mackenzie River Delta collectively represent gradients in water transparency, nutrient regime, and biotic communities, each strongly linked to the sill elevations of the lakes. Analysis of 40 years of water levels in East Channel of the central delta, in combination with a floodplain geometry model to estimate river water volumes added to lake waters at the annual flood peak, permitted direct estimation of annual river-to-lake connection times, lake water renewal, and interannual variabilities in nine lakes spanning the full range of sill elevations in the delta. Results have revealed a broad range of river-to-lake connectivities and river water renewals that are temporally dynamic and vary considerably among the lakes of this river delta system. Lakes with short and variable connection times plus low and variable river water renewal yield groups of lakes with high degrees of individuality because they are strongly influenced by particular sequences of antecedent years (legacy effects) that may result in lakes simultaneously containing residual waters from multiple river inundation events separated by more than a decade. Lakes with long and less varying connection times plus high river water renewal with multiple possible river water resets per year yield lakes with high degrees of similarity. The full combination of lakes arranged in an intermittently connected continuum, creating variable connectivity for aquatic organisms and water intermixing, may be an important mechanism driving the collectively distinctive habitat productivity and biodiversity of aquatic communities in this system, relative to lakes on the surrounding landscape.

1. Introduction

[2] The striking combination of productivity [Squires et al., 2009, 2002; Squires and Lesack, 2002; Spears and Lesack, 2006; Tank et al., 2009a], biodiversity [Hay et al., 2000, Riedel, 2002; Squires, 2002; Galand et al., 2006], and distinctiveness [Lesack et al., 1998; Ramlal et al., 1991; Fee et al., 1988] of lakes in the Mackenzie Delta relative to lakes on the surrounding landscape parallels major river floodplains worldwide [Thorp et al., 2006; Ward and Tockner, 2001] (e.g., Amazon [Junk, 1997], Orinoco [Lewis et al., 2000], Mississippi [Sparks, 1995], and European rivers [Tockner et al., 2000]). Theoretical models such as the Floodpulse Concept [Junk et al., 1989; Tockner et al., 2000], applications of landscape ecology to riverine landscapes [Wiens, 2002, and references therein], and the recent Riverine Ecosystem Synthesis [Thorp et al., 2006] have been formulated to address such characteristics of large rivers, but many aspects remain poorly understood. Recent work in the Mackenzie Delta [Squires et al., 2009] suggests the productivity can be explained by nutrient-rich river sediments forming the physical setting of the system, rather than being driven by a moving littoral zone as envisioned by Junk et al. [1989]. On the other hand, the biodiversity and the diversity of aquatic productivities, autotrophs, microbial communities, and underwater irradiance for photosynthesis [Squires and Lesack, 2003b] and photochemistry of dissolved organic carbon (DOC) [Gareis et al., 2010; Febria et al., 2006] are poorly understood. A variety of work (reviewed by Thorp et al. [2006]) emphasizes the importance of ecological connectivity [e.g., Hein et al., 2003] and differing types of connectivity, but few papers have quantified such connectivity [e.g., Costelloe et al., 2005; Thoms et al., 2005; Coops et al., 2008] in a useful way over time scales longer than a year or two for large river floodplains.

[3] The issue of hydrological connectivity, more generally, is not yet well understood, though a variety of recent work has sought to clarify the potential roles played by hydrological connectivity from hillslopes to ecosystems and larger landscape units. Examples of such work include how groundwater connectivity controls the nature of hillslope linkage to stream landscape units [Jencso et al., 2009]; how topographic connectivity controls the expansion of saturation overland flow source areas in stream networks [Gomi et al., 2008] and connectivity to important source areas of sediments and nutrients [Lane et al., 2009], and to groundwaters draining deeper soils [Sanderman et al., 2009]; how channel-network connectivity in Canadian Shield environments is controlled by fill-and-spill processes that regulate water storage [Spence and Woo, 2008]; and how spatially dependent dynamics of hydrological connectivity controls how hydrological processes at smaller scales create differing river signatures of nutrients and solutes [Fröhlich et al., 2008] or self-organized vegetation patterns [Hwang et al., 2009] at larger spatial scales.

[4] The circumpolar arctic coast has numerous large lake-rich deltas, and is a region where climate is changing rapidly [Arctic Climate Impact Assessment, 2005; Lesack and Marsh, 2007; Rouse et al., 1997]. The Mackenzie Delta is the second largest river floodplain in this region, and represents an important model ecosystem to advance our understanding of aquatic habitat on large river floodplains. Ongoing investigations have established that lakes in the delta collectively represent strong gradients in water transparency, sediment regime, nutrient regime, and biotic communities, with each gradient strongly linked to the sill elevations of the lakes relative to water levels in the adjacent river channel. Marsh and Hey [1989] established that lake sill elevations directly control the frequency with which lakes are flooded with river water and published return periods in flooding frequency for the range of lakes in their study area [Marsh and Hey, 1989, 1991, 1994]. From a biological-ecological perspective, however, a more useful measure is connection time with the river [e.g., Ward et al., 2002; Amoros and Bornette, 2002; Tockner et al., 2000, and references therein]. Although Marsh and Hey's work explicitly discussed the linkage between frequency of lake flooding and duration of lake flooding (i.e., river-to-lake connection time, hereafter), virtually all of the published results were restricted to flooding frequency statistics. The work also did not address interannual variation in lake behavior [Lesack and Melack, 1995], which is also very important from an aquatic ecology perspective.

[5] Prior work has established that the hydrology of lakes in the Mackenzie Delta, and lakes on the floodplains of large rivers, is complex [Lesack and Marsh, 2007; Marsh and Lesack, 1996; Lesack and Melack, 1995]. River water that enters lakes of the delta in varying amounts, however, plays an important role in replenishing inorganic nutrients [Lesack et al., 1998; Tockner et al., 2000, and references therein] and in cases where the water volume is large, in resetting the nutrient-plankton regime at the beginning of each open water season [Squires and Lesack, 2002; Spears and Lesack, 2006]. While complete water budgets for such lakes are very difficult to work out [Lesack and Melack, 1995; Marsh and Lesack, 1996], in the case of the Mackenzie Delta it is now possible to estimate river water volumes annually added to the lakes following the annual breakup peak via a floodplain geometry approach similar to Emmerton et al. [2007], if the peak water levels each year are known.

[6] From 40 years of water level records now available for East Channel in the central Mackenzie Delta, we provide an explicit calibration of river-to-lake connection times for the sill elevations of 9 lakes that have served as study sites for a variety of prior work in the delta. On the basis of a floodplain geometry model of river water volumes added to lake waters following the flood peak each year, we provide estimates of river water renewal and its long-term interannual patterns among the 9 study lakes. We discuss the above results as the basis for a hypothesis to explain the diversity of organisms and aquatic habitat in this and possibly other river floodplain systems.

2. Study Area

[7] The Mackenzie River is an important representative of large north flowing rivers in the circumpolar Arctic and the Mackenzie Delta is located where the Mackenzie River discharges into the Beaufort Sea (western Canadian arctic, 68°N–69°N, 134°W–137°W). The delta is located within the zone of continuous permafrost and is the second largest in the circumpolar arctic (∼13,000 km2; after the Lena). The open water period in the central delta is from June to November, with peak water levels occurring during spring breakup in response to snowmelt runoff in more southerly parts of the Mackenzie Basin. Peak water levels are partially controlled by the amount of water contained within the winter snowpack, but may be strongly controlled by ice breakup effects. Because the Mackenzie flows in a northerly direction from areas of relative warmth to a colder environment, melt progresses in a downstream direction and the resulting flood wave encounters downstream ice cover which causes extensive ice jams and high water levels [Andres and Doyle, 1984; Prowse, 1986; Goulding et al., 2009a]. Secondary peaks occur during sporadic summer rains, and during latter summer (August onward), storm surges from the Beaufort Sea can also quickly raise water levels by up to 1 m or more [Marsh and Schmidt, 1993].

[8] Whereas major arctic river deltas are recognized as lake-rich systems [Hill et al., 2001], recent work [Emmerton et al., 2007] has established that 45,000 lakes account for 99.9% of the total lake area in the Mackenzie Delta, almost double the number of lakes (25,000) estimated during earlier work [Mackay, 1963; C. P. Lewis, Mackenzie Delta sedimentary environments and processes, unpublished report, Environment Canada, Ottawa, 1988]. The increase in lake numbers is partly related to the abundance of very small water bodies, but the 45,000 cited above are sufficiently significant in area (>0.14 ha) to qualify as small lakes, and this total excludes an additional 5,000 water bodies <0.14 ha in area. The lakes are generally shallow with mean depths ranging from about 0.5 m to 4.5 m, but lake depths are dependent on time of year, whether the year is wet or dry locally, and whether water levels in the delta (not necessarily related to local climate) are higher or lower than average.

[9] Prior work [Marsh and Hey, 1989, 1991, 1994; Marsh et al., 1999] has established that the lakes are perched at a range of elevations above the delta distributary channels, with the lakes being flooded with river water only as water levels in channels rise in response to changes in Mackenzie discharge, river ice breakup effects, and storm surges from the Beaufort Sea. Marsh and Hey [1989] operationally defined “sill elevation” as the highest elevation along the connecting channel thalweg between the lake and river channel. This allowed operational quantification of a simple lake classification system proposed by Mackay [1963]. No-closure lakes remain in connection with main channels for the entire summer and are defined as lakes with sill elevations lower than the 1 year return period for summer low-water levels in the river. Low-closure lakes are flooded each spring but are cut off from the river for some portion of the summer, and are defined by sill elevations higher than no-closure lakes but less than the 1 year return period for spring peak water levels in the river. High-closure lakes are not flooded every spring and never during the summer, and are defined by sill elevations greater than the 1 year return period for spring peak water levels. The 9 lakes representing the focus of this paper (Table 1) are located (Figure 1) in the central eastern delta adjacent to East Channel and the town of Inuvik, which has the longest record of water levels available for this system.

Figure 1.

Map of study area in the Mackenzie Delta.

Table 1. Summary of Descriptive Parameters for the Nine Study Lakesa
ParameterLake 527aLake 520Lake 56Lake 278Lake 280Lake 87Lake 85bLake 80Lake 129
  • a

    Standard deviation is given in parentheses.

  • b

    Connection depth is not included for Lake 129 because it is a no-closure lake that maintains near continuous connection with the river.

Lake area (ha)
Mean depth (m)0.972.231.080.721.641.311.511.521.29
Spring sill (m asl)5.1694.9134.6234.0773.8383.3892.9902.6312.363
Summer sill (m asl)4.9204.5874.2103.5003.1892.6062.0871.6201.272
Mean connection time (d/yr)4.51 (4.03)6.53 (4.74)9.26 (5.27)17.0 (5.64)22.0 (7.36)44.2 (17.0)78.3 (26.2)121.4 (32.0)159.4 (25.7)
Mean river water content, ϕLf0.396 (0.389)0.414 (0.393)0.592 (0.362)0.792 (0.207)0.747 (0.249)0.855 (0.174)0.881 (0.153)0.928 (0.126)0.947 (0.098)
Mean connection depthb (m)0.456 (0.278)0.589 (0.288)0.738 (0.307)0.972 (0.305)1.12 (0.307)1.41 (0.291)1.67 (0.291)1.94 (0.289)
Period of analysis1964–20051964–20051964–20051964–20051964–20051973–20051973–20051973–20051973–2005

3. Methods

3.1. Lake Sill Thresholds and River-to-Lake Connection Times

[10] The general strategy [Marsh and Hey, 1989] to reconstruct the historical frequency with which lakes are flooded was based on measuring the sill elevation of all lakes in the target area, then determining the frequency and duration with which sill elevations are exceeded in the long-term water level records for the adjacent river channel. Lake sill elevations and flooding frequencies for several areas of the delta have been derived via this approach, but these estimates are most precise for the area that contains our 9 study lakes (Figure 1; see also area-A shown in the work by Lesack and Marsh [2007]) because it is fully adjacent to the East Channel gauging station at Inuvik. Flooding frequency estimates are less precise for areas further away from that station [Marsh and Hey, 1991].

[11] For the present analysis, the connection time between the study lakes and the river for each year of water level record was obtained via counting the number days from when the river water level first exceeded the spring sill elevation of each lake until the river water level dropped below the summer sill elevation. If secondary rises in river level occurred following the initial fall below the summer sill elevation, the number of subsequent days the river level exceeded the summer sill of the lake during the open water period were added to the connection time associated with the breakup period to yield a total connection time for each lake in each year.

[12] The spring sill for a given lake is effectively higher than the summer sill because of the effects of ice and snow along the connecting path from the river to a given lake, but the elevation difference between the spring and summer sill is well defined in this system [Marsh and Hey, 1989]. The spring sill was estimated as being equivalent to the measured water level in the adjacent river channel on the day when river water (distinctively turbid) was first observed entering the given lake. The day of river water entry was identified via aircraft-based observations over the study area (Figure 1; see also area-A shown by Lesack and Marsh [2007]) every other day throughout the spring ice breakup period of 1986 [Marsh and Hey, 1989]. This technique has been employed in subsequent years for other areas of the delta [Marsh and Hey, 1991; Marsh et al., 1999]. Summer sill elevations were established via ground survey on a subset of the lakes where spring sill elevations had been established.

[13] Our present analysis focused on the spring sill range from 2.363 (Lake 129) to 5.169 (Lake 527a) m above sea level (asl) because Lake 129 is a well-studied no-closure lake whereas Lake 527a was among the group of lakes that was the last to flood as water levels peaked during 1986, the year aerial observations were performed to identify its spring sill elevation. Spring sill elevations of the lakes higher than 5.169 m asl (about 20% of the high-closure class, Marsh and Hey [1989]) are not actually known because they did not flood during the 1986 aerial survey. On the basis of ground survey, the highest levee heights in the area are about 6.5 m asl. We also know the difference between spring sill versus summer sill elevations converges to a negligible difference at ∼6.0 m asl on the basis of extrapolating the differences observed among lower-elevation lakes. We thus included in our analysis potential connection times for unspecified lakes with spring sill elevations of 5.5 and 6.0 m asl to assess how connection times would change when sill elevations approached and exceeded the long-term mean peak level of 5.636 m asl (up to 1986 [Marsh and Hey, 1989]).

[14] The water level record for East Channel at Inuvik (Water Survey Canada, http://www.wateroffice.ec.gc.ca/index_e.html) encompasses the full year and ranges from 1973 to present. The record of peak annual water levels during river ice breakup, however, was extended back to 1964 with additional data (i.e., only available for the breakup period) from the Inuvik Research Centre (fully compiled by Marsh and Hey [1988]). Thus, the length of the connection time analysis runs from 1973 to 2005 in the case of lower-elevation lakes that remain connected to the river for some time beyond the annual breakup period (Lakes 129, 80, 85b, and 87; Table 1), but runs from 1964 to 2005 in the higher-elevation lakes (Lakes 280, 278, 56, 520, and 527a; Table 1) that only connect to the river during the breakup period.

3.2. Lake Water Renewal Times Versus River Water Renewal

[15] Because the water level peak associated with spring breakup in the Mackenzie Delta is a single annual event (∼1 month in duration), lake water renewal rates can be estimated via partitioning the volume of river water that is annually added to the lakes at peak levels from the initial volume of water in the lakes prior to the spring breakup. At annual peak water level, the total volume of water in a lake plus its associated floodplain area is

equation image

where vLi is the initial volume of water in the lake prior to the flood, vLf is the volume of flood water added directly to the lake surface, and vFf is the volume of floodwater added to the adjacent floodplain associated with the lake (boxes cefg, bdec, and abc-dhe, respectively, in Figure 2; volumes are in m3). The symbol scheme used above and in subsequent equations consists of a lowercase character for the quantity of interest (e.g., volume, area, elevation), with an upper case character to distinguish conceptual units (e.g., lake, floodplain, levee top), and if necessary, an additional lowercase character to distinguish type of water (i = initial water prior to flooding, f = floodwater, w = water from local catchment). From the above quantities, an annual water renewal time (in years) for a given lake can be expressed as follows:

equation image

where the denominator represents m3/yr of water added to the lake, vLw is the volume of local water yielded, after the annual flooding period, from the net water balance of the lake and its local catchment (i.e., lake evaporation versus all other hydrological inputs and losses) as defined by the local floodplain topography [Marsh and Lesack, 1996; Lesack and Melack, 1995]. The quantity vLw is very complex to estimate in this type of environment. In the delta, it is at most only a minor component of water renewal in lakes where flooding occurs, though it may become important when flooding does not occur [Marsh and Lesack, 1996].

Figure 2.

Geometric model of river water volumes added to lakes of the Mackenzie Delta during the flooding peak associated with spring breakup. The site conditions change from no-closure to high-closure lakes [Marsh and Hey, 1989] because levee heights are lower than the mean water peak in no-closure lakes and higher than the mean peak in high-closure lakes [Marsh and Hey, 1994]. Three reservoirs distinguished are the initial volume of lake water (box cefg), river water added directly to lake surface (box bdec), and river water in the delta floodplain (box abc-dhe) locally associated with each lake. Diagram is drawn to approximate vertical and lateral scales.

[16] Because of high interannual variability of τL and in the relative importance of vLw in this system (i.e., driven by whether or not flooding occurs in a given year), a more intuitive measure is the fraction of the lake water mixture consisting of new river water following the flood peak. We define this postpeak river water content as follows, and hereafter, refer to it as a “river water renewal” coefficient:

equation image
equation image

where aL is the preflooding surface area of the lake and zLi is the mean depth of the lake. Both of these parameters are assumed to be nonvarying among years, given that prior work [Marsh et al., 1999] has shown sediment infilling should be minor relative to lake depths over the time scale of this analysis. Though this calculation is based on a box shape, it is also based on appropriately representative mean lake depths, and thus, does not assume lakes are actually shaped with vertical sides (as in Figure 2):

equation image

where zLf is the depth of the flood water layer (bc in Figure 2) added directly to the lake surface and

equation image

where eP is the elevation of peak water level in a given year and eS is the fixed Spring Sill elevation of the lake:

equation image

where aFf is the area of floodplain associated with the lake that is inundated with floodwater during the peak level of a particular year:

equation image

where aTf is the total area of water associated with the lake plus inundated floodplain during the particular year (plane ah in Figure 2). This was estimated, assuming the total areas of water surface and the lake surface occur in the shape of concentric circles, as

equation image

where rL is the fixed radius of the lake surface (ce in Figure 2) aL. The rate of change in total radius with changing flood water depth can be approximated as

equation image

where rT is the fixed radius of the total area aT of lake plus adjacent floodplain associated with the lake to the boundary of the levee tops (ij in Figure 2). The fixed elevation eE of the levee top at a given lake was obtained from the regression

equation image

based on field measurements of Marsh and Hey [1994]:

equation image

where aF is the total floodplain area associated with a given lake, and was estimated as 2.42 times aL on the basis of the overall ratio of total area of floodplain to total lake area in the Mackenzie Delta [Emmerton et al., 2007].

4. Results

4.1. Lake Sill Elevations and River-to-Lake Connection Times

[17] The pattern of peak water levels at the Inuvik station (Water Survey Canada) on East Channel over the past 40 years (Figure 3) is a primary control on river-to-lake connection times (d/yr) and variation in the proportion of lakes in the study area that receive river water inputs. Interannual patterns of connection times differ substantially among lakes of differing sill elevation (Figure 4). Average annual connection time (τC, d/yr) over the full record of water levels presently available ranges from 159 days in Lake 129 to 4.4 days in Lake 527a (Table 1). Given that one goal of this paper is to calibrate connection times to lake sill elevations, an important issue is that, over the period of observation, the annual connection times have significantly lengthened in the lowest-sill lakes (Lake 80 and lower) as a result of sea level rise, with the effect possibly amplified by recently enhanced storm surge magnitudes and interactions with river discharge [Lesack and Marsh, 2007]. Moreover, the connection times may have declined in the highest-sill lakes (Lake 527a and higher) as a result of declining river ice breakup effects and possibly declining water level peaks, though this trend remains uncertain.

Figure 3.

Annual peak water level in the central Mackenzie Delta (East Channel at Inuvik) from 1964 to 2005. Water level is referenced to the pre-1990 Water Survey Canada datum, as in work by Marsh and Hey [1989].

Figure 4.

Temporal variations in annual river-to-lake connection times (d/yr) among the study lakes (full range of spring sill elevations, m asl) from 1964 to 2005.

[18] Among the lowest elevation lakes (Figure 4), the annual connection times of Lake 80 range from 70 to 170 d/yr, but Lesack and Marsh [2007] have shown that the trend of the 5 year running mean lengthens in average connection time from 101 to 138 d/yr from 1973 to 2005. This trend was also significant on the basis of the more conservative (i.e., relative to the running mean result) Mann-Kendall test of the unsmoothed data. Connection times for Lake 129, the only no-closure lake shown here, range from 114 to 227 d/yr, but Lesack and Marsh [2007] have also shown these times have lengthened from 145 to 169 d/yr over the record. The case for Lake 129, and other no-closure lakes, is more complicated than in higher-elevation lakes because the connection times have lengthened significantly into the period of initial ice cover formation. The opportunity for lengthening in Lake 129 is thus limited, compared to Lake 80, because Lake 129 was initially closer to the maximum open water connection time that would be possible in this system.

[19] Among the highest-elevation lakes (Figure 4), Lesack and Marsh [2007] showed that the “connection time” for a lake at a reference elevation of 5.500 m asl (just below the mean peak level from 1964 to 1986 [Marsh and Hey, 1989]) declined from 3.6 to 1.0 d/yr on the basis of the trend of a 5 year running mean from 1964 to 2005. Similarly, Lesack and Marsh [2007] showed that the connection time of Lake 527a may have shortened from 5.5 to 3.5 days. The more conservative Mann-Kendall (MK) test, though, indicates these trends are weak [Lesack and Marsh, 2007], with a significance level of p = 0.15 in the former case and outside the range of statistical significance (p = 0.36) in the latter. If the extreme high water event of 1972 is ignored, and the time is restricted to the period 1973 to 2005 (same as Lake 80 and Lake 129), the significance level of the MK tests are stronger (p = 0.09 and 0.16, respectively). The potential trend in “connection time” for a lake at reference elevation of 6.0 m asl is substantially influenced by the 1972 water peak. The running mean trend that suggests potential shortening of the connection times at that elevation is not significant if 1972 is excluded from the time series. Whereas the statistical case for the shortening of connection times in higher-elevation lakes is thus far not strong, this potential trend is consistent with other work that has documented earlier dates of peak water levels [Marsh et al., 2002] and potentially earlier dates of ice breakup initiation [Goulding et al., 2009b]. This also represents an important constraint on the stability of potential “calibrations” of connection times.

[20] Spring sill elevations of the lakes (eS) from 2.363 (Lake 129) to 5.169 (Lake 527a) m asl are very well related to average annual connection time (τC, d/yr) with the river over the full record of water levels according to the following least squares fit:

equation image

The r2 value is 0.998 (Figure 5, top) and the significance level is very high (p < 0.0001). It is not surprising that this relation is strong given that connection time must in some manner derive from sill elevation. The important result here, however, is the nature and varying precision of this calibration over the range of sill elevations, given that differing processes may drive water levels between the highest levels and the lowest levels. The standard deviation of the annual connection time for each lake plotted on Figure 5 shows how the relative variability progressively increases with lakes of higher sill elevations. Also annotated onto the plot are changes in connection times over the period of record that appear to be statistically significant.

Figure 5.

General relation of Spring Sill elevation versus the means (long-term record) of river-to-lake connection times (d/yr) and versus the means of postpeak river water content ϕLf. The means are shown with horizontal bars of ±1 standard deviation on the basis of the long-term record for each lake. Elevations are referenced to the pre-1990 Water Survey Canada datum, as in work by Marsh and Hey [1989].

4.2. Lake Sills Versus River Water Renewal

[21] Interannual patterns of river water renewal differ substantially among lakes of differing sill elevation (Figure 6). However, the relation with sill elevation is more complicated than the case of connection times. When levee top elevations are exceeded, the lake is effectively flushed by a nearly unlimited volume of water rather than a quantifiable fixed volume that can be estimated when water levels are lower than the levee tops. A measure that can be estimated under these conditions is the content of new river water in the lake water (ϕLf, equation (3), and Figure 6) following the breakup peak. The 9 lakes in this analysis have average contents of new river water ranging from 0.39 to 0.95 (Table 1 and Figure 5). Because the general patterns of river water contents shown in Figure 6 are driven by annual peak water levels, and the peaks may have declined over the period of record, a net decline in the river water content of Lake 527a, analogous to its now shortened connection time, may have occurred. However, this analysis does not account for the secondary rises in water level later in the summer, lower in elevation than the breakup peaks, that have driven the lengthening of connection times in Lake 80 and Lake 129. Such secondary rises cause additional river water renewal in those lakes. Quantification of that effect is a more complicated problem than estimating ϕLf, but will be done in the future.

Figure 6.

Temporal variations in annual river water renewal among the study lakes (full range of spring sill elevations, m asl), expressed as the fraction of lake water consisting of new river water following the annual flood peak (i.e., postpeak river water content ϕLf) from 1964 to 2005.

[22] The effect of years where river levels overtop the local levees, resulting in river water contents of 1.0 (Figure 6), needs to be considered in each (can have differing meaning) of the lakes. In Lake 520 (deeper than Lake 527a), for example, if the years where the lake was reset to full river water are excluded (11 of 42 years), the average river water content is around 0.19 rather than 0.42. However, an important issue in this lake system is legacy effects during sequences of years when complete renewal doesn't occur. A good example of this is Lake 527a, which was completely renewed during 1992, but then not again until 2006. In 1993, 0.204 of the lake water from 1992 was replaced by new river water (i.e., in 1993, ϕLf = 0.204), with 0.796 (= 1 − 0.204) of the lake water now left over from 1992. The lake was then not flooded again with river water until 1996, when 0.177 of the lake water from the prior year (a mixture of 1992 and 1993 waters) was replaced by new river water (i.e., in 1996, ϕLf = 0.177). Of the remaining lake water (assuming it is well mixed) at that point (1 − 0.177 = 0.823), 0.168 was left over from the 1993 event (= 0.204 × 0.823) and 0.655 from 1992 (= 0.796 × 0.823). Via analogous logic, the composition of lake water becomes more complex as river water is added in subsequent years with lengthening time since the last point of full renewal (i.e., ϕLf = 1.0 only where peak water levels fully exceed levee boundaries of the lake). In the case of Lake 527a, full renewal did not occur again after 1992 for 13 years, with the composition in year 13 thereby dependent on not just the prior year, but on each of the prior 12 years.

[23] More generally, for any sequence of years of n duration since the last full renewal with new river water (i.e., sequence of ϕLf values < 1.0) in a given lake, we define the legacy-affected river water content from year j of the n years as

equation image

where values of j range from 0 to n − 1, new river water during year j of the sequence is ϕLfj, and the values of new river water following year j up to year n are ϕLfi. A complete legacy analysis from 1992 through 2005 (i.e., all Λ values for the sequence) for Lake 527a is shown in Table 2, and illustrates how the composition of lake water becomes increasingly complex. During 2003, for example, river water from 4 separate years, from up to 6 years ago, each represent >14% of the overall lake water composition at that point, and 7% of the lake water is still left over from 1992 (11 years prior). If a water quality property is time dependent, such as colored dissolved organic matter (CDOM) (controls underwater UV radiation) and its degree of photobleaching [e.g., Gareis et al., 2010; Febria et al., 2006], the results of the full legacy sequence would be essential to understand the CDOM level in the lake in any given year over that time series. Whereas, the Lake 527a example above is the longest legacy sequence (i.e., where ϕLf < 1.0) among the set of lakes shown here, longer sequences would occur in lakes with higher sill elevations.

Table 2. Legacy-Affected Content of River Water in the Lake Water Composition of Lake 527a Each Year Following 1992, the Last Year the Lake Was Fully Renewed With River Watera
  • a

    The bold values represent new river water added each year (ϕLf values), and the nonbold values represent river water remaining from prior years (Λ values). See equation (14).


[24] Spring sill elevations of the lakes are well related to the average river water content in the lakes following peak water levels (ϕLf, Figure 5, bottom) over the full record of water levels, though the overall relation is neither linear nor log linear. It is possible to fit linear trend lines separately for the group of high-closure lakes and the group of low- and no-closure lakes, but it is not useful to do so because the presence of legacy effects driven by sequences of antecedent conditions means the observation values are not independent of one another.

4.3. Variation in Connection Time and River Water Renewal

[25] An important result in the case of both connection times and river water renewal for the 9 lakes is that the lakes with higher sill elevations are substantially more variable than the lower-sill lakes (Figures 5 and 7). The coefficient of variation (CV = standard deviation of observations/mean) for connection times among the lakes (Table 1 and Figure 7) increases from 0.16 to 0.90, whereas the coefficient of variation in river water content increases from 0.10 to 0.98 from Lake 129 to Lake 527a, respectively. In the case of unspecified lakes with sill elevations higher than Lake 527a, a CV of 1.37 was estimated for connection time in lakes with a spring sill elevation of 5.5 m asl, but CVs of river water renewal could not be specified for the higher-sill lakes because a specific lake volume is needed for the calculation. From the pattern of the CVs with river-to-lake connection time (or sill elevation), there appears to be a rapid increase in variability as connection times decline among the high-closure lakes. In the case of the low- and no-closure lakes, CVs steadily increase with declining connection time, but not to the same extent as the high-closure lakes.

Figure 7.

Coefficient of variation (standard deviation divided by mean) of river lake connection times (diamonds) and of postpeak river water content ϕLf (triangles), versus mean river-to-lake connection times (d/yr).

4.3.1. Connection Time Versus Peak Water Levels

[26] An obvious process that affects river-to-lake connection time is peak water elevations during breakup (Figure 3). However, from Figure 8, it is clear that the importance of peak levels changes with increasing connection time (or declining sill elevations). The connection times of high-closure (e.g., Lake 520) and some of the higher low-closure lakes (e.g., Lake 280) are primarily controlled by the peak elevation and water level pattern of the breakup period. On the other hand, other factors such as summer precipitation patterns, and latter summer storm surge patterns in the Beaufort Sea are more important in most of the low-closure (e.g., Lake 80) and in the no-closure lakes (e.g., Lake 129).

Figure 8.

River-to-lake connection times (d/yr) among five representative lakes versus annual peak water level in East Channel at Invuik. Cases where Lake 520 connection times equal 0 (9 of the years) are not shown in the plot.

4.3.2. Connection Time Versus Water Depths

[27] From an organism access perspective, depth of the water connection path (i.e., not the average depth of the lake) and its possible relation to connection time may also be important. Among the 9 lakes in this analysis, the lower-sill lakes with longer connection times do indeed have substantially deeper connection depths than the higher-sill lakes (Figure 9). On the other hand the higher-sill lakes have enhanced connection depths on the occasions they flood for longer than normal durations.

Figure 9.

River-to-lake connection times (d/yr) among four representative lakes versus the mean depth of water connection. Cases where Lake 520 connection times equal 0 (9 of the years) are not shown in the plot.

5. Discussion

5.1. Calibration of River-to-Lake Connection Times

[28] Our new calibration of average river-to-lake connection time versus lake sill elevation represents a significant advance in our understanding of the Mackenzie Delta because connection time is biologically and biogeochemically more meaningful [e.g., Squires et al., 2009; Tank et al., 2009b; Lesack et al., 1998]. Though connection time at a given sill elevation (Figures 4 and 5) can vary substantially among years, this variation is also biologically meaningful and it can be estimated via its coefficient of variation, as we have done here (Figure 7). A more difficult issue is to what degree such a calibration will remain relatively stable over time, because global change stresses are affecting the study system [e.g., Lesack and Marsh, 2007]. In the temporal plots of Figure 4, the connection times of the two lowest-elevation lakes (Lake 129 and Lake 80) have significantly lengthened over the past 30+ years as a consequence of rising sea level (∼0.0035 m/yr at Tuktoyaktuk since 1961 [Manson and Solomon, 2007]) and possible backwater interactions with the Mackenzie River, whereas the connection times of the highest-elevation lakes (e.g., Lake 527a) may have shortened as a consequence of declining effects of river ice breakup [Lesack and Marsh, 2007]. Unresolved at this point, is how high up the lake sill continuum sea level effects may propagate, and how low down the continuum might the effect of declining ice breakup effects propagate, and whether a longer observation period will be able to confirm the ice breakup trend.

[29] We have also identified a broad suite of lakes, located within the elevation band between the present threshold elevations where changes in river ice breakup and sea level rise are occurring, that show diversity in their patterns of connection times (Figure 4), but thus far no significant changes in their connection times over the period of record. These lakes ought to be monitored carefully for possible expansion of the ice breakup-related and sea level–related effects. The diversity of patterns among these lakes does not mean they are collectively reflecting statistical noise. Because each lake is sitting at a differing elevation, each lake is potentially sensitive to differing signals that may be embedded in the water level record. Three such independent signals are the pattern in peak water levels during spring breakup, the pattern of summer precipitation in the Mackenzie Basin, and the pattern of latter-summer storm surges from the Beaufort Sea. Figure 8 clearly shows that peak flood level, which is controlled by a combination of winter snowpack amounts in the Mackenzie basin plus degree of river ice breakup effects [Goulding et al., 2009a], is very well related to river connection time in the high-sill lakes, but not related to connection time in the lower-sill lakes.

5.2. Interannual Dynamics of Connection Times and River Water Renewal

[30] Our results on annual river water renewal represents a significant advance in our understanding of lake dynamics in the Delta, even though ϕLf (equation (3)) does not represent a complete water balance, because river water generally represents a large (often primary) source of new sediments, inorganic nutrients, and chromophoric (colored) dissolved organic carbon (CDOC) coming into the lakes each year (Table 3). The lakes that receive large amounts of river water in a given water year collectively reset to a near river water composition that is then processed by the biota and biogeochemical processes in the lakes. Such groups of lakes tend to be relatively similar in water composition and aquatic communities, though macrophyte importance can vary, depending on water transparency (function of distance from primary distributary channels and connection type) and lake depths [Squires et al., 2002; Squires and Lesack, 2003a]. Our river water renewal coefficients (ϕLf) for these lakes estimate the degree to which the lake has reset each specific year.

Table 3. Summary of Gradients in the Physical Properties, Nutrient Regimes, and Biotic Communities as a Result of Differing River-to-Lake Connection Times Among Lakes of the Mackenzie Delta
 >120 to >150 d/yr>17 to 120 d/yr<4.5 to 17 d/yr
Physical-Chemical Gradients
TSShigh low
Transparencylow-unstable high-stable
Chromophoric colorhigh low
Total DOClow high
Inorganic nutrientshigh low
Lake sedimentsinorganic organic
Underwater UVRnegligiblelowhigh
River connectionlongshortdiscontinuous
Gradients in Biota
Bacterialow-high low-high
HNANhigh low
Macrophyteslow high
Epiphyteslow high
Zooplanktonlow, small bodiedhigh, small bodiedlow, large bodied

[31] A new finding that has not previously been reported in other major river floodplain we are aware of is the potential importance of legacy effects on lake water renewal. In water years where lakes receive only a limited input, or no input, of river water, the resetting mechanism is reduced or eliminated in those water years, potentially leading to multiyear sequences of legacy-affected river water content (i.e., Λ from equation (14) and Table 2) that go beyond the simpler concept of “antecedent conditions.” A consequence of this is that the water nutrient composition and sediment composition then evolves from the point where it arrived at the end of the preceding year, rather than from a common starting point of near river water. Because each lake typically arrives at a somewhat different endpoint (nutrient regime and community composition) by the end of the open water season, the starting point for the subsequent open water season differs among lakes, and such difference apparently can be enhanced when lakes, even with the same sill elevation, receive small but variable amounts (relative volume contributed to each lake depends on lake depth and area of local floodplain catchment) of river water.

[32] Groups of lakes that have a low ϕLf value, in a given water year, typically are more diverse in their characteristics (nutrient regime, aquatic communities) over the open water season than lakes that have near fully reset (i.e., ϕLf value near 1.0) at the beginning of that year. More generally, lakes with low average ϕLf values (across years) are more diverse in character than lakes with high average ϕLf values, because not only is the chain of antecedent conditions in prior years important (e.g., Dishwater Lake [Lesack et al., 1998]), but lakes with low average ϕLf values can vary across the full range from 0 to 1.00, depending on the year, while the ϕLf range in lakes with long connection times tends to be significantly lower (e.g., 0.80 to 1.00 in Lake 129).

5.2.1. Nonmixing and Flushing While Connected

[33] In cases where two lakes may both have fully reset in a given water year, the lake with the lower average ϕLf value is likely to substantially differ from the lake with a higher average ϕLf value, because antecedent years will generally be more important (i.e., years where ϕLf < 1.0 more frequent) and river connection time will be shorter in the lake with lower average ϕLf (because sill elevation is higher). One consequence of short connection time is that even though a river water layer may be introduced over top of the lake, if the connection time is not long enough for the two layers of water to mix together, the actual amount of river water retained in the lake may be significantly less than ϕLf results presented here. Such an occurrence was well documented in 1989 for Lake 278 [Lesack et al., 1991a, 1991b], where floating lake ice acted a shield from wind mixing and allowed the two water layers to remain unmixed until the river water layer was much smaller than its thickness at peak water level. On the basis of that event, lakes that are flooded with river water prior to significant local melting of lake ice, can take on the order of 10 days for the subsequent floating lake ice to decompose. Such wind shielding events do not occur every year, because the decomposition of floating lake ice occurs more quickly if significant local melting has occurred prior to the river water layer being introduced to the lake.

5.2.2. Multiple Mixing and Flushing While Connected

[34] A consequence of long connection times is that the river water input to lakes may not necessarily consist of a single event derived from the breakup water peak. A number of secondary rises in water levels can occur above the summer sill of the lake, in response to summer rain in the river basin or storm surges from the Beaufort Sea [Marsh and Schmidt, 1993], that will push a secondary layer of river water into the lake and subsequently mix into resident lake water. The true amount of river water in the lake on average over the open water period can thus be higher than indicated by the ϕLf value (if < 1.0, in a particular year), or alternatively (if = 1.0) the reset to river water composition can occur more than once over the summer, rather than only once at peak annual water level during breakup period.

5.3. Interpretation Constraints on River Water Renewal

[35] Whereas lakes of the delta do not necessarily flush while connected to the river (discussed above), it is also important to consider that the ϕLf results in this paper do not account for water renewal via local precipitation and runoff in these lakes. Prior work [Marsh, 1986, Marsh and Lesack, 1996] has established that local precipitation and runoff play a relatively small role in the water balance of lakes with long river-to-lake connection times, but can play a significant role in lakes with short connection times [Marsh and Lesack, 1996]. In the absence of flooding, net water levels in a typical lake may decline on average by about 0.033 m/yr [Marsh and Lesack, 1996], mostly as a consequence of the average balance between local precipitation onto the lake surface (∼0.195 m/yr, most as snow) versus open water evaporation (∼0.230 m/yr) [Marsh and Bigras, 1988]. Freezeup generally occurs when local soils are relatively dry, and consequently, the soils typically have high infiltration capacities during snowmelt of the following spring [Marsh, 1988]. In such years little snowmelt is converted into local runoff because the water infiltrates the soils and refreezes until the active layer develops over the summer and most of the soil water is then transpired by vegetation [Marsh and Bigras, 1988]. By contrast, when a wet cold freezeup period is followed by a larger than normal winter snowpack, modest though significant amounts of local runoff can be produced. Such runoff can be variable in inorganic nutrients, but quite high in organic forms of nutrients and CDOC [Tank et al., 2009b; L. F. W. Lesack et al., unpublished data, 1989]. By comparison, snow accumulating on the lake surface that mostly balances the open water evaporation is quite low in nutrients and DOC. True water renewal is thus a combination of river water plus a small but variable amount of local water that is added to the lake water carried over from the prior year.

[36] There are several uncertainties associated with the model used to derive the ϕLf values. A single general ratio of floodplain area to lake surface area for the full Mackenzie Delta [Emmerton et al., 2007] was used to derive the area of local floodplain that captures water for each of the study lakes. Whereas the general interannual pattern of how lakes with differing sill elevations vary in their river water content should be correct, the true river water contents are not fully accurate because of the above floodplain area approximation. Future work will directly map the true floodplain catchment area for each of the lakes and allow these values to be corrected.

[37] There is some uncertainty about the potential effect of sediment infilling on the mean depths and areas of lakes over the 30–40 year observation period. Prior work [Marsh et al., 1999] has estimated average sedimentation rates of 4.4 mm/yr, 3.5 mm/yr, and 2.6 mm/yr for no-, low-, and high-closure lakes in the central delta, respectively. Over 30 years, this could correspond to a reduction in mean lake depth of 0.132 m and 0.105 m in no- and low-closure lakes, respectively, and over 40 years, to a reduction of 0.104 m in high-closure lakes. These are relatively minor reductions in comparison to the mean depths of our 9 study lakes (overall average 1.36 m) and to the mean depths [Emmerton et al., 2007] of no-, low-, and high-closure lakes in the central delta (average depths of 1.66 m, 1.37 m, and 1.84 m, respectively). However, we do not presently know the effect of thermokarst deepening of these lakes that to some degree may offset such infilling over the same period. Prior work [Emmerton et al., 2007; Marsh et al., 1999] has qualitatively compared samples of lake areas from topographic maps (derived from 1950s aerial photos) of the delta relative to modern photos and satellite imagery. Conspicuous changes to some lakes have been noted, though such changes were not common (not detected among the 9 lakes of our present analysis). Assessing the extent of such changes is a goal of our ongoing work in this system.

[38] In cases where river water levels rise to the threshold elevation of the spring sill for a given lake but not much higher, river water may not necessarily enter the lake. Conversely, in cases where a spring sill threshold has not quite been reached, river water may nevertheless enter the lake. The reason for such behavior is that the summer sill elevation is fixed by the delta topography, whereas the spring sill is determined by the summer sill plus the depth of ice and snow along the connecting path from the river channel that needs to be overcome to gain entry to the lake (still typically frozen to the nearshore lake bed at the time of river water entry) [Marsh and Hey, 1989]. Because ice and snow depths vary to some degree from year to year, the spring sill elevation is subject to some minor interannual variation.

5.4. Implications for Diversity of Habitat and Biodiversity

[39] The results of our water renewal and connection time analysis, the first we are aware of for a large lake-rich floodplain over a long time scale, illustrates the extent to which lake disconnection versus river-to-lake connection occurs in this system and provides a framework for understanding how biodiversity may be generated within it. We postulate that the Delta may generate enhanced biodiversity somewhat similarly to the rain forest refugia hypothesis proposed by Haffer [1969], but where water renewal variability drives divergence of aquatic communities in disconnected lakes located toward the elevational periphery of the system, and episodic interconnection of all lakes during high-magnitude floods disperses and intermixes the aquatic communities. Divergence of communities among lakes may be enhanced by at least four mechanisms that are a consequence this complex lake connectivity gradient.

5.4.1. Variable Nutrients and Light

[40] The variable supply of nutrients and in situ light across the connectivity gradient translates, for example, into diverse Diatom assemblages. The 77 lakes characterized by Hay et al. [2000] were dominated by a diverse benthic microflora, and most taxa did not dominate in more than one or two lakes. The Diatom community was sensitive to river-to-lake connection times, and suggested the community composition can provide indication of hydrological variability within the lakes [Hay et al., 1997; Michelutti et al., 2001].

5.4.2. Intermittent Fish Presence

[41] Zooplankton communities in Delta lakes change very distinctively from small-bodied to large-bodied taxa [Riedel, 2002] (Table 3), as would be expected from lakes where planktivorous fish are present to lakes where they are absent. Because of the complex connectivity gradient, however, this translates into a planktivory gradient ranging from very intermittent to full-time presence of fish that is fundamentally different from other lake districts where fish presence is typically all or nothing because of presence or absence of physical access barriers [e.g., Brooks and Dodson, 1965]. In cases where fish find access to high- and low-closure lakes, their presence can only last until the winter ice cover kills them off via termination of their oxygen supply [Pipke, 1996], whereas no-closure lakes usually sustain adequate oxygen levels for fish survival through the winter. We postulate that comparable gradients of intermittent fish presence ought to be present in many other river floodplain systems and may effectively operate in a manner comparable to disturbance gradients where intermediate disturbance [Connell, 1978] facilitates biodiversity.

5.4.3. Variable Predictability of Aquatic Food Supply

[42] In lakes with continuous connection to the river, the composition and supply of food for aquatic organisms (e.g., zooplankton) should be somewhat stable and predictable. At the other end of the connectivity gradient, however, food supply could switch quite dramatically and unpredictably over time. For example, a legacy-influenced sequence of years where variable but modest amounts of river water (or none) and inorganic nutrients were delivered during the flood period, would yield variable autotrophic and microbial communities (i.e., food for consumer organisms) that substantially differ from those derived from the intermittent years where the lake basin was fully reset to river water composition. We postulate such food supply switching should enhance biodiversity among the lakes. This view differs, for example, from the Flood Pulse Concept [Junk et al., 1989] where the issue of habitat and food predictability is addressed, but in the context of a continuum from small streams to large rivers. We postulate here such a continuum is fully present within the Delta and likely is present in floodplain systems of other large rivers.

5.4.4. Variable UV Risk

[43] Being in a location with 24 h summer day lengths, underwater UV irradiance is potentially a significant but variable risk factor [Bothwell et al., 1994, Williamson, 1995] for aquatic organisms, depending on CDOC levels in the water. CDOC is abundant in river water and acts as a UV sunscreen, but over time it is broken down via photobleaching (Table 3) [Gareis et al., 2010, Gareis, 2007]. In lakes with continuous connection to the river, aquatic organisms should be relatively free of UV risk. At the other end of the connectivity gradient, though, organisms can experience a legacy-influenced sequence of years where variable but modest amounts of river water and CDOC are delivered during the breakup period, with the CDOC then subjected to varying degrees of photobleaching, or alternatively during variably intermittent years, the lake is fully reset to river water composition with high CDOC. Conspicuous macrozooplankton pigmentation has been observed in lakes with lower levels of CDOC [Riedel, 2002], providing evidence of not only physiological responses to UV, but enhanced predation risk to the organisms if planktivorous fish gained access to the lake in a particular year. Such variable UV risk is created via variable river water renewal, and likely is a strong factor in other river floodplain systems as well.

5.5. A Habitat Connectivity Continuum From Highly Individual to Highly Similar Lakes

[44] A variety of publications on shallow water lakes have argued such lakes often exist in alternative states [e.g., Bayley and Prather, 2003; Scheffer, 2001]. A natural inclination in the case of lakes on river floodplains is to postulate that such lakes may behave similarly: one stable state when regularly reset via river water and another quasi-stable state when river flooding doesn't occur. Our work here suggests such a model does not fit with the Mackenzie Delta, and likely would not fit with most other river floodplains that experience an unregulated flow and water level regime. We postulate here that lakes in large river floodplains may behave as a habitat variability continuum, where lakes with short and variable connection times plus low and variable river water renewal will yield groups of lakes (high in number, though may be modest in collective area) with a high degree of individuality (i.e., in analogy, a snowflake class endpoint), whereas lakes with long and less varying connection times plus high river water renewal with multiple possible river water resets per year will yield lakes with higher degree of similarity (a cookie class endpoint). This hypothesis differs from tenet 14 of the Riverine Ecosystem Synthesis, which postulated that biocomplexity should peak at intermediate levels of connectivity between the main channel and lateral aquatic habitats [Thorp et al., 2006]. Further work is needed to resolve whether the “habitat variability continuum” we have postulated here (i.e., a connectivity continuum ranging from highly individual lakes to highly similar lakes) may be more fully developed in the Mackenzie Delta than in lower-latitude floodplains, given that the Delta is particularly lake rich because of thermokarst subsidence effects.

6. Conclusions

[45] Analysis of 40 years of water levels in the central Mackenzie Delta, in combination with quantified hydrologic locations (sill elevations) of lakes in the aquatic landscape and a floodplain geometry model to estimate river water volumes added to lake waters at the annual flood peak, has revealed a broad range of river-to-lake connectivities and river water renewals that are temporally dynamic among the lakes of this river delta system. In combination with other results from a 20 year case study of the Mackenzie Delta, we postulate here that lakes in arctic deltas behave as a habitat variability continuum. More specifically, lakes with short and variable connection times plus low and variable river water renewal yield groups of lakes strongly influenced by particular sequences of antecedent years (legacy effects) and with a high degree of individuality (in analogy, a snowflake class endpoint), whereas lakes with long and less varying connection times plus high river water renewal with multiple possible river water resets per year yield lakes with higher degree of similarity (a cookie class endpoint). The full combination of lakes arranged in an intermittently connected continuum may be an important mechanism driving the collectively striking productivity, habitat diversity, biodiversity, and distinctiveness of aquatic communities in this system, and other river floodplains, relative to lakes on the surrounding landscape. Our results here also represent an important addition to other recent work on hydrological connectivity (see Introduction) and, in particular, work [e.g., Hwang et al., 2009] that expands our understanding of how hydrology can drive ecosystem patterns and dynamics.

[46] A possible trend toward reduced annual peak water levels in the Mackenzie Delta, in combination with rising arctic sea level and sea ice recession, may cause a decline in abundance of snowflake class lakes and expansion of cookie class lakes, respectively, that ultimately may reduce the biodiversity in this system. Other circumpolar arctic river deltas ought to be assessed for analogous potential changes.


[47] We appreciate the variety of discussions with current and former students and research assistants that have motivated us to take on this analysis. This includes Merv Hey, Mary Ferguson, Cuyler Onclin, Tanya Schmidt, Craig Emmerton, Maggie Squires, Suzanne Tank, Jolie Gareis, Adam Chateauvert, Catherine Febria, Andrea Riedel, Bryan Spears, Chris Teichreb, and Katherine Pipke. Logistical, technical, and in-kind support has been provided over a number of years by Andrew Applejohn, Pippa Seccombe-Hett, Sharon Katz, and Les Kutny of the Inuvik Research Centre/Aurora Research Institute. Financial support was received from NSERC (DGP and NRS programs to L.F.W.L.), from the Polar Continental Shelf Project (helicopter support to L.F.W.L. and P.M.), from the Northern Scientific Training Program (numerous graduate and undergraduate students), Department of Indian and Northern Affairs, and facilities use in Inuvik has been subsidized by the Aurora Research Institute. Helpful discussion on the complexity of river ice jamming has been provided by Faye Hicks (University of Alberta) and Spyros Beltaos (National Water Research Institute, Environment Canada). Comments by P. Kumar and two anonymous reviewers improved this paper.