Pan-Arctic rivers strongly affect the Arctic Ocean and their vast lake-rich deltas. Their discharges may be increasing because of an intensifying hydrological cycle driven by warming climate. We show that a previously unexplained trend toward earlier ice breakup in the Mackenzie River Delta is little affected by winter warming during the period of river-ice growth and is unaffected by river discharge, but unexpectedly is strongly related to local spring warming during the period of river-ice melt. These results are statistically linked to declining winter snowfall that was not expected because of an intensifying Arctic hydrological cycle. Earlier ice breakup is expected to cause declining water level peaks that will reduce off-channel flows through the lake-rich delta before river waters enter the ocean. Thus, local spring warming with unexpected snowfall declines, rather than warmer winters, can drive earlier ice breakup in large Arctic rivers and biogeochemical changes in their river-ocean interface.
The great rivers of the pan-Arctic region strongly affect the Arctic Ocean [McClelland et al., 2012], delivering nutrients from the continents [Raymond et al., 2007; Emmerton et al., 2008] to unique delta [Lesack and Marsh, 2010] and ocean ecosystems [Carmack and Macdonald, 2002] and creating warm freshwater plumes that affect ice cover, wave dynamics, and weather patterns for hundreds of kilometers offshore [Mulligan et al., 2010]. Peak annual river levels, driven by peak discharge combined with strong backwater effects of ice breakup [Lesack et al., 2013], replenish waters in the vast lake-rich deltas that function as biological hot spots of this region [Squires et al., 2009], and water passage through these deltas [Emmerton et al., 2007] modify the nutrient content and nature of warm freshwater plumes released to the ocean. The Mackenzie River Delta, for example, is the second largest Arctic delta (Figure 1). Waters released from it as peak levels recede [Emmerton et al., 2007] fuel the productivity of the Beaufort shelf [Emmerton et al., 2008], including an unusual semifreshwater body “floating” within the shelf waters [Galand et al., 2006] that is created as ice ridges at the boundary between mobile sea-ice and land-fast ice constrain outflowing river waters [Carmack and Macdonald, 2002]. The land-ocean interface of the Mackenzie and the other pan-Arctic rivers, in varying degrees, thus function as keystone ecogeosystems, with peak river levels affecting a multitude of interacting biogeochemical processes and associated aquatic resources.
Pan-Arctic river discharges are increasing [Peterson et al., 2002; Arctic Climate Impact Assessment (ACIA), 2005] because of an intensifying hydrological cycle driven by warming global temperatures [Déry et al., 2009; Rawlins et al., 2010]. Warming temperatures and higher discharges are hypothesized to cause earlier spring breakup of river ice and declining water level peaks [Marsh et al., 2002; Lesack and Marsh, 2007] because river ice (downstream resistance to upstream discharge) should become thinner and higher discharge (upstream forcing) should provide greater force and heat to break up the ice [Goulding et al., 2009]. Earlier spring breakup of river ice has been detected in the Mackenzie Delta [Marsh et al., 2002], but the precise cause has been unknown. Lesack et al.  recently investigated various factors that may be driving the earlier peaks in annual river levels in the central Mackenzie Delta and concluded that freshet river discharge associated with the ice breakup period, total annual discharge, and the initiation date of freshet discharge into the delta (i.e., all measures of upstream discharge forcing) have not changed. Among the measures of downstream ice resistance, however, the date of river freezeup, has not changed, but the lag time from freshet discharge initiation into the delta until initial breakage (date tb in Goulding et al. ) of the river-ice sheet has declined sufficiently to account for the earlier water level peaks in the central delta. Moreover, the earlier peaks and ice breakage were found to be weakly related to the loss of snow depth on the ground through the month of April, with these losses increasing over the period of record. These results suggest that earlier ice breakup is a result of either declining winter growth of river ice (thinner end-of-winter ice cover) or more rapid ice melting following initiation of the spring freshet.
The research reported here is part of an International Polar Year investigation of riverine fluxes of water and nutrients through the Mackenzie Delta to the Arctic Ocean [International Polar Year—Study of Canadian Arctic River-delta Fluxes (IPY-SCARF)] and part of longer-term investigations on the hydrology and limnology of this system [Lesack and Marsh, 2010]. Specifically, we assess the roles of declining ice growth versus enhanced ice melting by examining the mean daily air temperatures in the delta as a proxy of available energy to grow or melt ice.
2 Study Area and Methods
The Mackenzie Delta is 13,000 km2 of productive aquatic habitat with high biodiversity that was created as a result of the historical regime of high and low water [Lesack and Marsh, 2010]. It extends 200 km upstream from the coast (Figure 1) and contains 45,000 lakes [Emmerton et al., 2007]. Prior work [Marsh and Hey, 1989] has established that these lakes are perched at varying levels above the distributary channels. The extent of water replenishment among the lakes, and biogeochemical transformation of waters while in the lakes and delta floodplain areas, depends on off-channel flooding of river water driven by peak annual water levels in the river. Peak levels are created by freshet river discharge, usually representing the highest discharge of the year, but levels are strongly enhanced by river-ice breakup effects [Lesack et al., 2013]. The water volume released to the coastal shelf from delta lakes and off-channel floodplain, as peak river levels recede, is large [Emmerton et al., 2007] and is rich in organic nutrients [Emmerton et al., 2008].
Water level results are from the analysis of Lesack et al. , based on published records for Mackenzie gauging stations established and monitored by the Water Survey of Canada (WSC) for varying lengths of time (Figure 1). Levels in the central delta are from the gauge on East Channel at Inuvik (WSC station 10LC002), the longest record available for the delta. Levels at the delta inflow are from Tsiigehtchic (10LC014), located just upstream of the delta (115 km upstream from the central delta station 10LC002), and the last point where gauging of total Mackenzie discharge is possible. Levels in the outer delta are from Reindeer Channel at Ellice Island (10MC011), the longest record available for the outer delta. Our analysis of air temperatures and snow depths is based on published records [Environment Canada, National Climate Data and Information Archive] for the long-term weather station near the Inuvik Airport, located near the 10LC002 WSC station on East Channel.
Based on the long-term climate record for Inuvik on the east side of the central delta, we quantified the integrated mean temperature for the hydrologic periods of the Mackenzie Delta established by Lesack et al. . We defined “winter temperature” to represent the temperature associated with the period of river-ice growth. Our best available measure is the integrated mean air temperature over the period of winter river discharge, from initial presence of river ice (day 293 on average at Tsiigehtchic) until the day before the earliest initiation of freshet discharge into the delta (approximately day 110 at Tsiigehtchic). Freshet discharge initiation into the delta represents a key change in the river water thermal regime for initiating local ice decay, because it is driven by warming temperatures and snowmelt occurring well upstream, whereas delta air temperatures at this time are still cold (average on day 110 over the record is −10.7°C). Analogously, we defined “breakup temperature” to represent the temperature associated with the delta ice breakup period. Our best available measure is the mean air temperature over the period from initiation of freshet discharge at the delta inflow point (day 111 at Tsiigehtchic) until the latest date of peak water level (and most typically, the clearance of river ice) in the central delta (day 160 at Inuvik).
The time series analysis in this paper is similar to the exploratory approach of Marsh et al. . Statistical trend significance was based on the Mann-Kendall (MK) test [Hipel and McLeod, 1994]. Based on the approach of Lesack et al. , there is no evidence that a first-order serial correlation (which may affect the MK test) is present in the residual errors of the trends in this paper (Table S1 in the supporting information).
The highly significant trend toward earlier water level peaks in the central delta, despite no change in the initiation date of freshet discharge into the delta, is ongoing (Figure 2a) (results from Lesack et al. ). Among the corresponding patterns in integrated air temperatures (Figure 2b and Table S1 in the supporting information), the statistical trend in the mean winter temperatures is very strong, increasing by ~0.1°C/yr, from −26.5 to −21.2°C (+5.3°C warming) over the 55 year record. The trend in breakup temperatures is weaker, though still highly significant, increasing by ~0.06°C/yr, from −1.7 to +1.5°C (+3.2°C warming) over the same period.
Our key finding is that the interannual pattern in breakup temperatures is strikingly related to the timing of the water level peaks in the central delta (Figure 3a and Table S2 in the supporting information), accounting for 81% of the variance in the dates of these peaks, with a slope of −1.8 d/°C (earlier) increase in temperature. Breakup temperature is also strongly related to the date of initial breakage of the river-ice sheet (Figure 3b and Table S2 in the supporting information), although this record is substantially shorter than in the case of water level peaks. Despite the strong warming trend in winter temperatures, this surprisingly is only very weakly related to the pattern of water level peaks, accounting for only 11% of the variance between these two variables (Figure 3c and Table S2 in the supporting information).
Breakup temperatures are also strikingly related to the dates of annual water level peaks at two other locations that span the 200 km north-south extent of the delta (Figure 1). These locations are Tsiigehtchic, where the river enters the delta (accounts for 73% of the variance between these variables) (Figure S1a and Table S2 in the supporting information), and Reindeer Channel in the outer delta, near the ocean interface (accounts for 74% of the variance) (Figure S1b and Table S2 in the supporting information). These relations may be slightly weaker than the relation for the central delta because the records are shorter (37 years and 28 years, respectively, compared to 47 years); nevertheless, they are statistically strong. Such relations with breakup temperatures at these other locations indicate that it is likely that earlier water level peaks are occurring throughout the delta, even though the central delta is the only location where earlier peaks have thus far been statistically detected. Earlier peaks at the other locations may not yet be statistically detectable, because the available records are shorter than in the central delta.
A key question emerging from our results is how mean air temperature during the ice breakup period (i.e., breakup temperature) can so strongly account for the trend toward earlier river breakup dates, given that it is a rather simple and imprecise measure of energy available to melt river ice. For example, mean air temperature does not generally account for snowpack ablation in studies of snow hydrology [Davison and Pietroniro, 2005; Hock, 2003]. We postulate that the explanation lies with combined trends of increasing breakup temperature and declining winter snow depths in this region. The long-term record of snow on the ground in the central delta at Inuvik (Figure 4) strikingly shows that depths during 1986–2012 were substantially lower, and the snow was disappearing earlier than observed during the period 1957–1985. We know that river-ice albedo is strongly reduced by the melt of snow cover on the ice [Gray and Prowse, 1993]. Thus, the decay of river ice should be much stronger per degree of warming through the breakup period if the loss of snow cover occurs more rapidly because of a general decline in snow depths through the winter. Indeed, melt period loss of snow depth (during April) by itself can account for 21% of the variance (p = 0.001) in the dates of water level peaks [Lesack et al., 2013]. Breakup temperature by itself (Figure 3a) though accounts for such a large portion of the variance in the peak dates (81%) that when added together with the loss of snow depth in a multiple-regression model, only breakup temperature remains significant. This is not surprising because these variables are intercorrelated. Declining winter snow depths may also explain why the strong warming trend in winter temperature (Figure 2b) is only weakly related to the earlier river-ice breakup (Figure 3c). River-ice growth indeed depends on cold air temperatures, but is limited by the insulating effect of the winter snow cover that retards loss of heat through the ice [Ashton, 1986; Prowse and Beltaos, 2002]. With a trend toward significantly thinner snow accumulation through the winter, river ice may be growing near equally thick despite the warmer winter temperatures, with the result that ice thickness has had little effect on the earlier breakup dates.
Although one might expect that warming air temperatures should lead to earlier breakup of river ice, our results here showing the importance of warmer springs versus warmer winters are counterintuitive to the effects expected in the Arctic from an intensifying hydrological cycle, driven by warming temperatures combined with the positive water vapor feedback in the global climate system [ACIA, 2005; Rawlins et al., 2010]. Warmer climate should generally mean more water vapor in the atmosphere, which should translate to higher precipitation and higher river discharges. At smaller spatial scales, the evaporated moisture can be redistributed by atmospheric transport in ways that may not necessarily lead to localized increases in precipitation. In the pan-Arctic case, increasing discharge in the great Russian rivers has indeed been observed [Peterson et al., 2002], and this trend has been supported by climatological results confirming increasing precipitation rates through the region [Min et al., 2008; Déry et al., 2009; Rawlins et al., 2010]. However, the combination of increasing Arctic precipitation with a decline in the amount of winter snow, as may be occurring in North America [Callaghan et al., 2011], was not expected. The broad Arctic phenomena of unexpectedly rapid declines in the extent of terrestrial spring snow cover [Déry and Brown, 2007; Derksen and Brown, 2012] and sea ice [Brown et al., 2010] have recently been linked via remote sensing data sets to warming surface air temperatures that appear to be amplified in the Arctic because of a positive feedback from regionally reduced albedo [Groisman et al., 1994]. Our results may represent one of the first field-based cases that show how localized declines in Arctic snowfall and subsequent melt period snow albedo, when coupled with amplified Arctic temperatures, can lead to unexpected effects such as enhancing the sensitivity of river-ice breakup to the warming spring temperatures. The recent unexpected collapses of several Canadian Arctic ice shelves [Mueller et al., 2003; Copland et al., 2007] have been linked to warming air temperatures. These strong responses for the degree of warming however may warrant further investigation into whether a reduced end-of-winter snow depth prior to spring melting might also be contributing to these responses, as in our Mackenzie Delta case.
Our findings are based on statistical analysis, but are not yet accompanied by a physically based model that can precisely link the degree of warming in breakup temperatures and the degree of decline in end-of-winter snowpack with the rate of change observed in the dates of river-ice breakup (see Supplementary Text in the supporting information). A strength of our findings is that they are accompanied by the extensive hypothesis-based analysis of Lesack et al. , that has statistically eliminated all the other mechanisms we are aware of that could explain the earlier breakups. Most importantly, the timing and the magnitude of river discharge through the breakup period have not changed, as it surely would have done if the river was transporting increasing amounts of thermal energy and causing warming spring temperatures to be the symptom of the earlier breakups rather than a cause. Warming spring temperatures and declining end-of-winter snow depths are what we are left with as the potential root cause, and they directly connect to a well understood mechanism of river-ice decay driven by reduction of albedo as the snow cover melts off [Hicks et al., 2009] (see Supplementary Text in the supporting information). This combination of supporting information strengthens our conclusions. Subsequent to narrowing the viable possible explanations to either declining winter growth of river ice (driven by warmer winters) or more rapid ice melting following initiation of the spring freshet (driven by warmer springs) [see Lesack et al., 2013, Table 2] or possibly both mechanisms cocontributing, it seems clear from our results here that local spring warming is driving the earlier ice breakup rather than warmer winters. The breakup temperatures and snow depth relations by themselves represent important findings, and these results should assist other ongoing investigations on the changing Arctic.
Our results demonstrate how the water level regime in one of the great Arctic rivers [Lesack and Marsh, 2010] is being altered via changing climate in a way that was not expected. This represents a valuable starting point for investigating the broader consequences of an intensifying Arctic hydrological cycle, and in particular, field-based investigations of local air temperatures, changing snow depths, and their possible role in the river-ice breakup of other pan-Arctic rivers. The Mackenzie represents a valuable reference system because, in addition to the results here, Lesack et al.  have specifically shown both that discharge over its full period of record has not thus far changed and that discharge is not related to the earlier breakups, as it surely would be if it had increased. Thus, it is not presently a complicating factor in attempting to understand how the ice breakup regime is changing. We recognize that Mackenzie discharge may increase in the future, following the trend of the other great Arctic rivers.
Not yet fully resolved is whether earlier river-ice breakup will lead to declining annual water level peaks, as expected, in the Mackenzie or in other pan-Arctic rivers. If breakup simply occurs earlier without changing the water level regime, the biogeochemical effect on the delta ecosystem and nutrient transport to the ocean may be minor. If annual water level peaks decline, however, the biogeochemical effects will likely be strong and complex. Our results here indicate that the earlier breakup dates are delta wide (Figure S1 in the supporting information). Because our other work shows that ice breakup has such a large amplification effect on water levels [Lesack et al., 2013], our overall findings indicating weakening ice resistance, with no change in discharge forcing, suggest that the earlier breakup dates could indeed lead to declining water level peaks and substantial changes in the nature of the river-ocean interface [Lesack and Marsh, 2007] at this vast complex delta. Declining water level peaks would change the ecohydrology of the lake-rich delta itself [Lesack and Marsh, 2010], in particular, by reducing the off-channel water storage in the delta. Such storage presently occurs as a shallow water layer, spread out over 11,000 km2 of productive aquatic habitat, where it undergoes biogeochemical transformations during the Arctic summer solstice prior to release to the coastal ocean. Coupled with rising sea level, which is raising low water levels in this delta [Lesack and Marsh, 2007], such changes would affect the 45,000 lakes of productive [Squires et al., 2009] and biodiverse [Galand et al., 2006] habitat in the Mackenzie Delta, and the coastal shelf ecosystem of the Beaufort Sea [Carmack and Macdonald, 2002], in ways that are not yet understood.
Financial support was received from the International Polar Year—Canadian Federal Program, the Natural Sciences and Engineering Research Council, Natural Resources Canada (Office of Energy Research and Development, Geological Survey of Canada, Polar Continental Shelf Project), Environment Canada, the Northern Scientific Training Program Canada, the Aurora Research Institute, ArcticNet, and the Networks of Centres of Excellence Canada. Original data on water levels, temperatures, and snow depths in the Mackenzie River Delta are available from public data sources as specified in the Methods section. We thank numerous colleagues and the scientific team of IPY-SCARF for the discussions that shaped this research. This paper was improved by the reviewer comments.
The Editor thanks an anonymous reviewer for assisting in the evaluation of this paper.