Concern exists about future changes to air-temperature gradients along large “northward”-flowing Arctic rivers having the potential to affect the timing and severity of spring river-ice breakup, and associated flooding events. To evaluate the significance of this concern, an analysis was conducted of temporal and spatial changes to the spring 0°C air-temperature isotherm (I0°C), which is also known to be a good index for the timing of spring melt/breakup conditions. Changes in I0°C were analyzed for the downstream 2000-km main-stem reaches of four large Arctic rivers: the Lena, Mackenzie, Ob and Yenisey. Current climatic conditions (1979–2008) were compared to those of two future climatic periods (2041–2070 and 2071–2100) projected by an ensemble of four Global Climate Models. Future projections show I0°C chronology patterns along the rivers that closely parallel current conditions, but with earlier dates varying from an ensemble mean of 7.5 (13.6) to 16.5 (25.5) days for the 2050s (2080s). Results also reveal a progressive downstream increase in warming under future climates. At the time when headwater temperatures reach 0°C, river mouth to headwater temperature differences for the four rivers decrease by an average of 0.8°C (2.4°C) to 2.1°C (3.7°C) for the 2050s (2080s). The implications of such decreases on the severity of spring ice-jam floods are discussed.
 Flood extremes on high-latitude rivers are typically created during the period of spring freshet from a combination of snowmelt and breakup of river ice. For many such rivers, extremes in ice-affected spring water levels exceed those during the open-water period, even though they occur at much lower discharge and typically from spring snowmelt [e.g., de Rham et al., 2008a]. In these environments, a spring flood wave generated in warmer headwaters advances progressively northward into a downstream ice cover that resists the passage of the wave. Generally, the degree of resistance, and hence the level of backwater or flood stage, depends on how strong and intact the ice cover is at the time of the flood-wave arrival [Beltaos, 2008]. In cases where the ice cover has retained much of its winter thickness, mechanical strength and bonds to the bed and banks, the flow resistance and resultant backwater flooding are maximized-often termed “dynamic” events. The contrasting case of a “thermal” event, involves significant thinning, decreases in mechanical strength, and shore detachment of the ice cover. It therefore poses minimal resistance to passage of the rising discharge and produces minimal backwater flooding.
 The timing of spring freshet and associated river-ice breakup is physically linked to the arrival of above-freezing air temperatures. For example, Bonsal and Prowse  found a strong relationship between the timing of the 0°C-isotherm (I0°C) for air temperature and other hydro-cryospheric variables, including the spring freshet and breakup of freshwater ice. The severity of river-ice breakup and whether it will be a dynamic or thermal event is also dependent on the gradient of air temperature along the river. Larger (smaller) gradients are likely to produce more of a dynamic (thermal) event because downstream portions are exposed to less (more) thermally induced reductions in ice resistance prior to arrival of the upstream increase in spring discharge [Beltaos, 2008].
 With changing climatic conditions, concern has been expressed that the dynamics of spring breakup and related flood events will be altered on ice-covered rivers [e.g., Rouse et al., 1997]. Based on an analysis of latitudinal gradients in air temperature for quadrants of the Arctic defined by Arctic Council and International Arctic Science Committee , Prowse et al.  suggested that the projected greater warming at northern compared to mid-latitudes might affect the thermal gradients along major Arctic flowing rivers, such as the Lena, Mackenzie, Ob and Yenisey. Any reduction in air temperature-gradients along rivers is likely to favour the development of more thermal rather than dynamic events and, as a result, reduced backwater formation and related flooding. This would benefit northern communities and infrastructure susceptible to damages from spring flooding, but could also prove detrimental to riparian aquatic ecosystems, the ecological health of which depends on such flooding [e.g., Anisimov et al., 2007; Lesack and Marsh, 2007]. Notably, however, the quadrants analyzed by Prowse et al.  contained broad areas of land and ocean, and thus the gradients were not specific to the river reaches. Hence, the objective of this research was to quantify and assess how air-temperature gradients along the four major Arctic rivers will be altered under future climate regimes, and consider how this might affect the timing and severity of spring breakup.
2. Data and Methods
 The four major Arctic rivers, Lena, Mackenzie, Ob and Yenisey, were selected for this study (Figure 1). All are ice covered for more than one half of the year and experience a downstream advancement of breakup during the late spring period [e.g., Pavelsky and Smith, 2004]. Although small headwaters of these rivers extend well into steep upland and alpine regions, this study focused on the lower, 2000-km main-stem reach of each river (Figure 1). In the case of the Mackenzie River, the main stem was extended up the Liard River because it is the main initiator of spring breakup conditions [Prowse, 1986].
 Given the scarcity of climate stations along these river reaches for evaluating gradients in air temperatures, reanalysis data and in particular, the Japanese Reanalysis Dataset (JRA-25) was employed as the source of current-climate air temperatures. JRA-25 is a recent global meteorological data set created with an advanced numerical assimilation system and is considered a major improvement over other global reanalysis products [Onogi et al., 2007]. Its key advantages for this study include high spatial resolution (1.25° × 1.25°), contemporary nature (1979-2008–herein referred to as current climate) and spatial coverage of the entire Northern Hemisphere.
 Given the focus on high-latitudes, future climate projections were obtained from ensemble means derived from the most recent versions of four of the five Global Climate Models (GCMs) used by the Arctic Climate Impact Assessment (ACIA) [Kattsov et al., 2005] (see Table 1). Two future periods were analyzed including 2041–2070 (2050s) and 2071–2100 (2080s) using data for the SRES A2 emission scenario (an upper-level emissions' scenario commonly used by impact studies, including ACIA). The Geophysical Fluid Dynamics Laboratory (GFDL version CM2.1) model was not included in this study as some of its mid-century temperature projections were distinct mid-century outliers. Specifically, the model showed colder than current-climate conditions for the 2050s along the main stems of two of the rivers, followed by a large reversal to pronounced relative warming by the 2080s. All other models projected overall continued warming throughout the 21st century-a trend also reflected in the general body of scientific literature about temperature changes in high latitudes.
 To determine thermal gradients along the lower 2000-km reaches of the four rivers, current and projected surface air-temperature data were regridded to a common 50-km resolution (Figure 1) using inverse distance weighting interpolation. Timing of the spring I0°C was determined following the procedure of Bonsal and Prowse . For each grid, daily temperature data were smoothed initially using a 31-day running mean. The spring I0°C was then defined as the first day of the year when this running mean crossed 0°C. In the rare situation when the 0°C threshold was crossed more than once, I0°C was defined as the date of the last spring crossing.
 Future spring I0°C dates were obtained from time-series data determined by applying GCM-projected (30-year average) temperature changes (i.e., the difference between modeled future and modeled current climate) to the current JRA-25 values (1979–2008). From this procedure, changes to the timing of spring I0°C dates along each river reach could be derived from the 30-year current and future climatological values. Downstream temperature gradients along the rivers were also analyzed. To best reflect the approximate initiation of spring melt conditions on the river main stems, a comparative assessment of these gradients was made when the headwater site (2000 km upstream of the mouth) reached 0°C for current and future conditions. The majority of GCM results are presented as ensemble means, although inter-model variability is assessed where applicable.
 Current and future timing of the I0°C along the main-stem reaches of each river are given in Figure 2. Under current climate, all rivers show a typical south to north seasonal advance of the I0°C which takes on average 35 to 49 days to “travel” the lower 2000 km to the rivers' mouths (Table 2). Non-linearities in the downstream profile are reflective of regional variations in climate. This is most clearly seen for the Ob River, where the reversal in I0°C dates in the middle of the reach is probably linked to warmer western European conditions as the river tends westward towards the Ural Mountains before returning east and northward (Figure 1). Future projections show I0°C chronology patterns that closely track current conditions (Figure 2), but with earlier dates varying from an ensemble mean (averaged over the entire reaches) of 7.5 (13.6) to 16.5 (25.5) days for the 2050s (2080s) (Table 3). Under future climates, the ensemble mean durations for the I0°C advances along the reaches do not appreciably change from current values, or between future time intervals. The largest duration change of approximately 3 days longer is for the Mackenzie River, both for the 2050s and 2080s. However, as summarized in the maximum and minimum columns of Table 2, there is considerable inter-model range in these future duration values for all rivers.
Table 2. Current and Future Duration (in Days) of the Spring I0°C Date Advance Along the Lower 2000-km Main-Stem Reaches of the Lena, Mackenzie, Yenisey, and Ob Riversa
The four-model means and, minimum and maximum of all projected future values, are listed.
Table 3. Projected Changes in the Timing of Spring I0°C (Days) Averaged Along the Lower 2000-km Main-Stem Reaches of the Lena, Mackenzie, Yenisey, and Ob Riversa
Days Earlier 2041–2070
Days Earlier 2071–2100
The four-model mean, minimum and maximum values are listed.
 Although the current and future I0°C dates largely track each other, there are changes in the thermal gradients along the rivers (Figure 3a). Under current climate, when the headwater portion of the river main stems reaches 0°C, differences in air temperature at river-mouth locations range from approximately −9.6 to −15.3°C-the largest being for the Yenisey and the least for the Ob, reflecting differences in the respective severity of river-basin climates. Under future conditions, the overall upstream-downstream air-temperature differences are reduced, although much of the upper ∼600 km from the headwaters of the Lena River indicates a small degree of cooling. Overall, however, the most pronounced changes are in the downstream reaches, potentially reflecting the greater influence of changes in climatic conditions closer to the Arctic Ocean [e.g., Serreze and Francis, 2006].
 The degree of change along each of the rivers is highlighted in Figure 3b, which also provides the ranges in projections among the GCMs. Despite the inter-model variability, all results show a general downstream increase in warming. Some sub-reaches (e.g., the above noted example of the upper portion of the Lena River) reveal a slight downstream cooling, which can likely be ascribed to changes in climate along the river. Overall, however, the ensemble-mean changes in main-stem temperature gradients between the respective headwaters and mouths are reduced by 0.8°C (2.4°C) to 2.1°C (3.7°C) for the 2050s (2080s). The greatest (least) headwater-mouth average decreases are for the Yenisey (Mackenzie) River.
 In summary, even though inter-model variability exists, the direction and relative magnitude of future air temperature changes along the reaches of the four major Arctic rivers consistently reveal both an earlier arrival of melt conditions and greater warming in the downstream direction. Discussion of these results and their implications for the severity of river-ice floods are provided below.
 The main stem reaches of all four rivers are projected to experience earlier I0°C dates in climate scenarios for the 2050s and 2080s. The relative magnitude of change seems to reflect the differences in coldness among these river systems with the greatest changes being for the more moderate climate of the Ob River and the least for the cold continental-interior climate of the Lena River. The I0°C dates summarized in Figure 2 are not intended to represent actual river-ice breakup dates but rather reflect the timing of melt conditions that precede breakup. As noted by Bonsal and Prowse , however, these dates are related. For the Mackenzie River, the mean I0°C dates of April 15 and May 20 for the headwater and mouth locations are approximately 5–10 days in advance of the mean dates of breakup initiation identified by de Rham et al. [2008b]. Such a temporal lag between melt and breakup initiation dates is physically reasonable given the time required to produce sufficient runoff volume to begin moving river ice. It is also likely that the advances in the timing of I0°C dates would be reflected in the related dates for river ice breakup.
 Typical of most “northward” flowing rivers, these four all have a pronounced air-temperature gradient from their main-stem headwaters to their Arctic Ocean mouths. Projected climate changes by 2050 and 2080 indicate that these gradients will decrease. Although the data presented in Figure 3 reflect conditions at the time when the headwaters reach 0°C, there was a persistence of “northward” decreases in the gradients as the I0°C advanced downstream to the mouth over the full melt season. This means that enhanced warming will probably occur during the pre-breakup phase under future climatic conditions and lead to relatively thinner ice cover at the time of breakup. Warmer winter temperatures also mean that the thickness of the ice cover prior to spring thinning would be less [Beltaos and Prowse, 2009]. Thinner cover should reduce hydraulic obstruction to the passage of the spring freshet and thus produce smaller increases in water levels. Two other factors might complicate this scenario. These are future changes in the mechanical strength of the ice cover and the magnitude of the spring runoff, which affect the breakup resisting and driving “forces”, respectively.
 Changes in mechanical strength are strongly controlled by internal melt, induced by the absorption of shortwave radiation [e.g., Hicks et al., 2008]. Earlier melt dates, however, would occur at lower levels of shortwave radiation because of natural annual cycles. Hence, despite melt and thinning of the ice cover, less radiation would be available reduce ice strength that is also required to cause an overall reduction in ice resistance. As for the magnitude of the snowmelt freshet, it might increase in the future as a result of greater winter precipitation and snow-water equivalents that have been projected for higher latitudes [Bates et al., 2008; Brown and Mote, 2009]. Quantifying the net result on breakup flood potential of such an increase in this component of the breakup driving force along with a reduced ice-cover resistance is complex. It can essentially only be accomplished by using ice breakup models that consider future combined changes to landscape hydrology, instream hydraulics, and ice mechanics-models that require significant additional research to couple and successfully validate on large rivers [e.g., Beltaos and Prowse, 2009].
 Beyond detailed physically based modeling, it might also be possible to use historical analogues for evaluating what may occur under future climate conditions. Although the observational database for high latitudes is limited, the ACIA found it “probable” that there has been polar amplification of air temperatures over the past 50 years [McBean et al., 2005]. Moreover, spatial patterns in decadal trends of gridded air-temperature data [Serreze and Francis, 2006, Figure 10] suggest that some of the large Arctic basins could have experienced more pronounced warming in their downstream than upstream regions over the last few decades (1970–2003). Although the data are spatially coarser (resolution of only 5° × 5° latitude and longitude) than those examined in this analysis, changes appear to have been as large as ∼2.6°C (the difference from similar headwater to river mouth zones ranging approximately from 0.1°C to 0.9°C/decade) for the Mackenzie River over this three-decade interval. Such a gradient change is within the range projected for the large Arctic rivers over this coming century, as found in this study.
 A detailed analysis of breakup phenology of the Mackenzie River has found that a number of breakup characteristics, over almost the same time period as the above noted temperature assessment (1970–2002 [de Rham et al., 2008b]), have become significantly earlier. Breakup severity at selected sites, such as at the Mackenzie River mouth, has been thought to decline in recent decades [e.g., Emmerton et al., 2007; Goulding et al., 2009], although sea-level changes might also play a role at this river-ocean boundary. No assessment, however, has been made of changes in breakup severity along the main stem of the Mackenzie or the three major Russian rivers. Although this would require an extensive bi-national assembly and review of disparate sets of historical data, such an analysis, combined with a more detailed examination of air-temperature gradients, would provide valuable historical analogues for evaluating future conditions.
 This work has been funded by research grants supplied by Environment Canada, NSERC, and IPY Canada. The authors thank the two anonymous reviewers for their comments toward an improved version of the manuscript.