Introduction

River ice is one of the major components of the cryosphere and affects an extensive portion of the global hydrologic system, particularly in the Northern Hemisphere where major ice covers develop on 29%, and seasonal ice affects 58%, of the total river length (Reference Bennett and ProwseBennett and Prowse, in press). It is also an important modifier of numerous biological, chemical and hydrologic processes (Reference Prowse and AndersonProwse, 2005) and is capable of causing extensive and costly damage to infrastructure (Reference BeltaosBeltaos, 1995). Because the various forms and processes of river ice are directly controlled by atmospheric fluxes, their spatial and temporal trends can be used as indicators of climate variability and change. Given the broad ecological and economic significance of river ice, scientific concern has been expressed about how future changes in climate might affect river-ice regimes (e.g. Reference Anisimov, McCarthy, Canziani, Leary, Dokken and WhiteAnisimov and others, 2001; Reference WalshWalsh and others, 2005; Reference Wrona, Prowse and ReistWrona and others, 2005).

Unfortunately, much of the information regarding river ice is scattered widely through the literature. Therefore, the objective of this review is to consolidate such information by focusing on four major objectives: to describe how variations in meteorological conditions control the timing and hydrologic severity of river-ice events; to summarize trends of river-ice freeze-up and break-up events throughout the Northern Hemisphere cold regions; to identify the principal linkages between river ice and key meteorological/climatic variables, including atmospheric teleconnections; and to review the methods and results for predicting future river-ice regimes under changing meteorological and climatic conditions. A focus is placed on freeze-up and break-up events which form the subject of a majority of the published literature and can be used further to determine other ice-season characteristics, such as ice duration. Notably, virtually all of the available literature focuses on the Northern Hemisphere, and North America in particular.

Timing and Severity of River-Ice Freeze-Up/Break-Up Events

Indications of climate variability are manifested most clearly in river-ice regimes by seasonal and/or interannual variations in ice thickness, ice-cover duration and the timing and hydrologic severity of freeze-up and break-up events. The following two subsections summarize these climate/river-ice relationships.

Historical Trends

Records of river-ice conditions are routinely collected in most cold-regions countries, but by numerous agencies and for a variety of purposes. As a result, the types and methods of observations are not uniform, so observations often differ in length and quality. Despite the importance of the hydrologic severity of river-ice events, large-scale regional analyses of such data are virtually non-existent (Reference de Rhamde Rham, 2006). The same is true for trend analyses of river-ice thickness or composition. By far the most widespread observations are those that relate to the timing of freeze-up and break-up events. Notably, however, there are inter-country and even inter-agency differences in these records. For example, some document the initiation of break-up, while others register the state of ice clearance at the end of break-up. This is an important distinction when comparing trends from different sources, given that the entire break-up process can last for up to 4 weeks at a single site (e.g. Reference de Rhamde Rham, 2006).

There are only a small number of stations in North America that have records longer than 100 years, whereas many Eurasian countries have been archiving river-ice information for up to several centuries. Unfortunately, because of reductions in observing programs beginning around the 1990s in both Russia and Canada (two of the largest cold-regions countries), regional analyses of river-ice trends are statistically more difficult to conduct using the most recent records than they are employing records prior to the end of the 1900s.

The longest records of freshwater-ice dates for the Northern Hemisphere were assembled by Reference MagnusonMagnuson and others (2000). They found that the most dramatic changes in dates have occurred in the last 150 years, during which freeze-up (break-up) became 5.7 days (100 years)^–1 later (6.3 days (100 years)^–1 earlier). Interannual variability in dates also noticeably increased after 1950. The rates of change were noted to correspond to a ~1.2˚C(100 years)^–1 change in mean, hemispheric air temperature. Unfortunately, this long-term dataset was largely comprised of lake observations, and the few river sites are too widely spaced to provide a good indication of regional patterns.

Two of the most spatially comprehensive, long-term trend analyses of river-ice freeze-up and break-up dates are provided by Reference Ginzburg, Polyakova and SoldatovaGinzburg and others (1992) and Reference SoldatovaSoldatova (1993), respectively, for the former Soviet Union (FSU). Ice records were summarized for similar hydrologic regions of the European and Asian FSU over the period 1893–1985, and modified to account for any effects of water-resource development (e.g. reservoir impoundment) that could influence the climatic signal (Reference SoldatovaSoldatova, 1992). Although appreciable inter-decadal variability was evident, significant long-term spatial patterns and temporal trends of river-ice freeze-up and break-up dates were identified for the almost 100 year period. The most significant regional trend was for later freeze-up in the European FSU (e.g. upper Volga, Oka and Don) and western Siberia (upper Ob and Irtysh), with rivers such as the Danube, Dnieper, Don, lower Volga and those of the Black Sea region exhibiting an average 2–3 week delay. A weaker but still significant trend to earlier freeze-up dates occurred over portions of the Yenisey and Lena in central and eastern Siberia, respectively (Reference Ginzburg, Polyakova and SoldatovaGinzburg and others, 1992). A similar large-scale spatial pattern was observed for break-up dates (Reference SoldatovaSoldatova, 1993). Break-up had advanced an average of 7–10 days (100 years)^–1 in the European FSU and western Siberia. The net effect of this, combined with the delayed freeze-up, was an overall reduction in the ice season by as much as a month (i.e. lower Don River). Again to the contrary, some rivers in central and eastern Siberia (e.g. middle to lower Yenisey and upper Lena) were noted to have an opposing trend: later break-up dates and, hence, an overall expansion of the ice season. Despite the above-noted complexities between the hydroclimatic factors and river-ice event timing, both analyses (Reference Ginzburg, Polyakova and SoldatovaGinzburg and others, 1992; Reference SoldatovaSoldatova, 1993) found that ice-formation and break-up dates correlated well (r ² = 0.6–0.7) with those for air temperature in the preceding autumn and spring months, respectively.

Reference SmithSmith (2000) also conducted a shorter-term (54–71 year) analysis of nine major Russian Arctic/sub-arctic rivers but produced some differing results to those of Reference SoldatovaSoldatova (1993). This included several opposing temporal trends for freeze-up and an absence of any statistically significant shifts in breakup timing. Potential inter-decadal variability in ice-event dates and analytical differences in number/location of station data and record length were offered by Reference Prowse and BonsalProwse and Bonsal (2004) as possible explanations for the differences. Despite finding significant changes in break-up dates, Reference SmithSmith (2000) did determine significant advances for a related break-up observation, ‘timing of melt onset’, which was suggested as indicating a trend toward a longer period of pre-break-up melt. As described earlier, this would tend to favour less severe ‘thermal’ break-ups, which are likely to produce high river stages.

Similar to the results for the western portions of the FSU, earlier break-ups have been documented from long-term records (~100+ year) for rivers in northern Scandinavia (Torneälven and Torino) and the eastern Baltic Sea (Dauga-va) (Reference ZachrissonZachrisson, 1989; Reference Kuusisto and EloKuusisto and Elo, 2000). Again, parallel to the climatic relationships found for the FSU, Reference ZachrissonZachrisson (1989) determined the strongest correlation to break-up dates was with spring (April) air temperatures. For one specific period (1870–1950), it was noted that a 3˚C rise in April temperatures was associated with about a 15 day advance in break-up. Overall, these studies suggest that it is the pre-break-up melt and runoff period that is more important to the timing of break-up, rather than the overall winter severity and maximum ice thickness.

There exist a small number of specific long-term studies of river-ice conditions in North America, all of which show somewhat similar trends over the various hydroclimatic regions. In the temperate maritime climate of southeastern Canada, Reference WilliamsWilliams (1970) discovered that break-up of the Saint John River advanced by approximately the same amount as that of the Torneälven river in Scandinavia over the 1870–1950 period. In the opposite and much colder northwest of North America, studies of the Tanana River in Alaska (1917–2000; Reference Sagarin and MicheliSagarin and Micheli, 2001) and the Yukon River in northwestern Canada (1896–1998; Reference Jasek and ShenJasek, 1999) indicated that the average date of break-up had advanced by approximately 5 days per century, although the long-term trend was characterized by a number of interdecadal cycles. In the central continental portions of the country, Reference RannieRannie (1983) found that the ice season of the Red River was shortened by approximately 2–3 weeks between the 19th and 20th centuries. Specifically, median dates of freeze-up and break-up were 12 and 10 days later and earlier in the 19th than in the 20th century. This corresponded to an increase in autumn and spring temperatures by approximately 2.5˚C between the two centuries.

The most comprehensive spatial analyses of river-ice trends in North America have been conducted by Reference Zhang, Harvey, Hogg and YuzykZhang and others (2001) and Reference Lacroix, Prowse, Bonsal, Duguay and MénardLacroix and others (2005) using Canadian stations from approximately the second half of the 20th century. Both studies showed a trend towards earlier break-up dates, the most pronounced changes occurring in western and southwestern Canada, and the degree of change increasing towards the latter portion of the century. Reference Lacroix, Prowse, Bonsal, Duguay and MénardLacroix and others (2005) noted that break-up advanced on average by 1–2 days decade^–1, the more rapid change being for 30 year climatological intervals near the end of the century. Reference Zhang, Harvey, Hogg and YuzykZhang and others (2001) also found that increasing the period of analysis from 30 to 50 years almost doubled the number of stations showing statistically significant changes. Somewhat different results were found for freeze-up dates, with Reference Zhang, Harvey, Hogg and YuzykZhang and others (2001) noting significant trends to earlier freeze-up dates across the country, whereas Reference Lacroix, Prowse, Bonsal, Duguay and MénardLacroix and others (2005) found that changes in freeze-up ranged from 1 day decade^–1 later to 0.1 day decade^–1 earlier. Differences in the results of these two investigations are likely due to the employment of differing intervals and spatial datasets, as noted above for the contrasting FSU results.

As noted by Reference Prowse and BonsalProwse and Bonsal (2004), a first approximation of river-ice response to climatic changes based on the various cold-regions analyses indicates that a long-term mean increase of 2–3˚C in autumn and spring air temperature has produced an approximate 10–15 day delay in freeze-up and advance in break-up, respectively. This agrees with the average 0.2˚Cd^–1 rate of change in phenological data for numerous lakes and a few rivers estimated by Reference MagnusonMagnuson and others (2000) for the Northern Hemisphere.

Most analyses of river-ice break-up assume that it is a spring event, although there have been some observations in North America that suggest mid-winter break-ups may be becoming more common (e.g. Reference BeltaosBeltaos, 2002). Recognizing that most such events are driven by mid-winter warming and are most likely to occur within the southern temperate fringe of the cold regions, Reference Prowse, Bonsal, Lacroix, Beltaos, Squire and LanghorneProwse and others (2002) evaluated the frequency of mid-winter warming events for a temperate ice zone (calculated from freezing degree-days) across North America (shaded area in Fig. 1). They found little evidence of any trend along the southern boundary of the temperate zone, but an increasing trend in the number of events through the 20th century at the northern boundary (Fig. 2). This was particularly evident for the western region that has also experienced the most pronounced increase in winter air temperatures over the last 100 years (Reference Zhang, Vincent, Hogg and NiitsooZhang and others, 2000). Furthermore, while the Atlantic and central subregions of the zone were devoid of mid-winter events prior to 1950, these areas have experienced a subsequent increase.

Fig. 1.

Map showing the current temperate region (shaded area) and the four subregions used in Reference Prowse, Bonsal, Lacroix, Beltaos, Squire and LanghorneProwse and others (2002): western (W), central (C), eastern (E) and Atlantic (A). Projected shifts in the temperate region of North America based on mean winter temperature increases of 2˚C and 6˚C are also provided.

Fig. 2.

Frequency of mid-winter temperature events (per decade) for the northern boundary stations in the four regions in Figure 1 (from Reference Prowse, Bonsal, Lacroix, Beltaos, Squire and LanghorneProwse and others, 2002).

River-Ice Linkages to Large-Scale Climate

Relationships with large-scale circulation patterns

A large portion of the observed trends and variability in Northern Hemisphere temperature is related to large-scale atmospheric and oceanic oscillations referred to as teleconnections. In particular, El Niño–Southern Oscillation (ENSO) (Reference Rasmusson and CarpenterRasmusson and Carpenter, 1982), the Pacific North American (PNA) pattern (Reference Wallace and GutzlerWallace and Gutzler, 1981), the Pacific Decadal Oscillation (PDO) (Reference Mantua, Hare, Zhang, Wallace and FrancisMantua and others, 1997) and the North Pacific (NP) index (Reference Trenberth and HurrellTrenberth and Hurrell, 1994) are closely linked to late-autumn to early-spring temperatures over much of North America, with strongest relationships in the west. This includes warmer temperatures associated with El Niño events and positive phases of the PNA and PDO that are all representative of a deepened Aleutian low (and the converse) (e.g. Reference Wallace, Zhang and RenwickWallace and others, 1995; Reference Bonsal, Shabbar and HiguchiBonsal and others, 2001). A notable feature in the Pacific involved a shift toward increased frequencies of a deeper Aleutian low in the mid-1970s (e.g. Reference TrenberthTrenberth, 1990).

A positive (negative) North Atlantic Oscillation (NAO) is related to colder (warmer) autumn–spring temperatures over eastern North America and warmer (colder) conditions over much of western Europe (Reference HurrellHurrell, 1996). A recent trend toward more positive NAO values is consistent with the winter and spring cooling observed over most of northeastern North America and Greenland. This, along with the Aleutian-low shift in 1976, helps explain the west-to-east gradient in both air temperature and associated river-ice break-up observed over Canada during the last 50 years. The recently identified Arctic Oscillation (AO) has been associated with the significant trend toward warmer winter and spring temperatures over western North America and most of northern Eurasia during the last 20–30 years (Reference Thompson and WallaceThompson and Wallace, 1998).

Twentieth-century trends and variability in spring 0˚C isotherm dates over Canada were found to be related significantly to the NP index over western regions of the country and to the NAO in the east (Reference Bonsal and ProwseBonsal and Prowse, 2003). Reference Shabbar and BonsalShabbar and Bonsal (2004) determined that the frequency and duration of winter warm spells over most of Canada were significantly increased in association with El Niño events as compared to La Niña events. This increase would affect the occurrence of mid-winter thaws and associated break-ups during the winter period.

Given the correspondence between teleconnections and air temperature, a few attempts have been made to relate large-scale oscillations to river-ice durations over various regions of the Northern Hemisphere. Reference Robertson, Ragotzkie and MagnusonRobertson and others (2000) examined the influence of El Niño on 20th-century variations in lake and river-ice cover over the Northern Hemisphere. For break-up, they found that dates were significantly earlier throughout North America, and slightly later (although not significant) over Finland and Russia in association with strong El Niño events. A comprehensive examination of relationships between several teleconnection indices and freshwater-ice break-up/freeze-up dates over Canada was carried out by Reference Bonsal, Prowse, Duguay and LacroixBonsal and others (2006). In the case of rivers, results for the period 1950–99 revealed distinct large-scale spatial patterns in these relationships. The Pacific-related indices (PNA, PDO and NP) had the strongest correlations with river-ice durations over much of Canada. In particular, the positive phases of the PNA and PDO, and negative phases of the SOI and NP, were associated with shorter durations consisting of earlier break-ups and later freeze-ups (and vice versa). The relationships were stronger, and covered more of the country during spring break-up, while the freeze-up response was generally weaker and confined to western Canada (e.g. Fig. 3). The composite ENSO findings were similar to the other Pacific indices, with shorter ice durations during El Niño and longer durations during La Niña over much of Canada.

Fig. 3.

Composite river-ice break-up and freeze-up anomalies (in days) during positive/negative PNA springs/autumns during the period 1950–99. (a) Positive PNA springs and break-up; (b) positive PNA autumns and freeze-up; (c) negative PNA springs and break-up; and (d) negative PNA autumns and freeze-up. Contour interval is 5 days. Negative values are dashed and the 0.0 contour thickened. Black dots denote river-ice locations with composite anomalies significantly from non-positive PNA (non-negative PNA) springs/autumns at the 5% level using a Monte Carlo procedure. Plus and minus signs indicate regions with similar large-scale responses (from Reference Bonsal, Prowse, Duguay and LacroixBonsal and others, 2006).

Future Projections

At site-specific scales, common methods of predicting river-ice break-up initiation stem from a technique originally proposed by Reference ShulyakovskiiShulyakovskii (1966) in which a relationship is established between the rise in stage above the preceding freeze-up level and the total pre-break-up heat input to the ice cover. Overmature break-ups are represented by high heat inputs and low stage levels, while premature break-ups have lower heat inputs and high stage levels. Unfortunately, relationships between these two terms remain empirical and site-specific (Reference Gray, Prowse and MaidmentGray and Prowse, 1993). Therefore, the approach of projecting future changes to river ice taken by most researchers has involved examination of large-scale changes to relevant air-temperature variables such as seasonal temperatures or 0˚C isotherm dates.

Reference Prowse, Bonsal, Lacroix, Beltaos, Squire and LanghorneProwse and others (2002) made some first-order estimates for river-ice conditions at both northern and temperate latitudes of North America. Relying on the average 0.2˚Cd^–1 rate of change in phenological break-up dates estimated by Reference MagnusonMagnuson and others (2000), they approximated that an increase of 3–7˚C in spring air temperatures by the end of this century (as projected by several General Circulation Models (GCMs); Reference HoughtonHoughton and others, 2001) would result in a 15–35 day advance in river-ice break-up over northern regions of Canada. In another analysis, Bonsal and others (unpublished information) examined future changes to spring and autumn 0˚C isotherm dates over Canada (Fig. 4) based on average temperature projections from seven international GCMs for the 30 year period centred on the 2050s (2040–69). The projections were based on monthly temperature changes (with respect to the 1961–90 baseline period) from seven international GCMs using the A2 emission scenario. For spring, 0˚C isotherm dates were projected to occur considerably earlier over the entire country, with greatest changes near the western and eastern coasts (12–14 days). Changes are less pronounced over northern Canada (approximately 8–10 days). Autumn is associated with a later occurrence of the 0˚C isotherms over the entire country. In contrast to spring, the largest changes occur over the north (10–12 days) as compared to the south (around 8 days). Given the close correspondence between 0˚C isotherm dates and river-ice break-up/freeze-up dates outlined previously, the authors hypothesized that by the middle of this century, river-ice durations over most of Canada will be approximately 20 days shorter with respect to the 1961–90 baseline period.

Fig. 4.

Projected changes to spring (a) and autumn (b) 0˚C isotherm dates over Canada for the 30 year period centred on the 2050s. Contour interval is 2 days, with earlier dates represented by dashed lines. (Adapted from Bonsal and others, unpublished information.)

Whether these projected temporal shifts will produce more or less severe break-up events remains in doubt, largely because of the complicating role of precipitation, which has the potential to control both the driving and resisting forces that affect break-up severity (see Reference Prowse and BeltaosProwse and Beltaos, 2002). Another factor that impacts break-up severity involves the strong south-to-north climatic gradient between the headwaters and downstream reaches on most cold-regions rivers (e.g. Reference Gray, Prowse and MaidmentGray and Prowse, 1993). Future changes to the strength of this climatic gradient will likely alter the severity of break-up. Late-21st-century temperature changes projected for the cold season (October–May) indicate that north-to-south temperature gradients will be reduced over some regions of the Arctic, since higher latitudes will experience greater warming (Reference ProwseProwse and others, 2006; see their fig. 3). This will likely impact the severity of future ice break-up including the occurrence of more thermal and less dynamic events. However, Figure 4a indicates that with respect to spring 0˚C isotherm changes, northern regions of Canada will experience smaller changes as compared to the rest of the country (thus maintaining or increasing the climatic gradient). These discrepancies are likely due to the different temporal resolutions incorporated in the two analyses. The reduction of the temperature gradients in Reference ProwseProwse and others (2006) was based on projected changes to cold-season (October–May) temperatures, while the spring 0˚C-isotherm changes incorporated temperature changes directly related to the timing of the event (which currently varies from March in the south to early June in the north). Since projected temperature changes are forecast to be greater in the cold season (e.g. March) as compared to the warm season (e.g. June), there is a greater projected change to the timing of 0˚C isotherm dates over more southerly locations. This outlines the importance of examining future climate changes and resultant river-ice impacts at temporal scales appropriate to the processes being considered.

With regard to future mid-winter break-ups, Reference Prowse, Bonsal, Lacroix, Beltaos, Squire and LanghorneProwse and others (2002) suggested a significant northward shift of the temperate climate zone that currently experiences these events (see Fig. 1). These shifts are based on a 2˚C and 6˚C warming in winter temperatures over the temperate region as estimated from several GCM projections (Reference HoughtonHoughton and others, 2001). The northward movement to these boundaries is significant since rivers that currently do not experience mid-winter warming events will be susceptible to river-ice break-ups which could produce dramatic impacts on the hydro-ecology of these river systems. A site-specific analysis of future climatic effects (using the Canadian CGCM2 model) on ice-jam occurrence in the Peace–Athabasca Delta of Canada (Reference BeltaosBeltaos and others, 2006a) revealed that by the end of this century the ice season will be reduced by 2–4 weeks and ice covers will be slightly thinner. In addition, the Peace River basin is projected to experience more frequent and sustained mid-winter thaws. This translates into fewer dynamic river-ice break-ups in the future.

Given that large-scale teleconnections have been shown to influence air temperature and associated river-ice characteristics over much of the Northern Hemisphere, knowledge of future changes in the frequency and magnitude of these oscillations could provide insight into future river-ice regimes. At present, however, the effects of climate change on large-scale teleconnection patterns remain uncertain due to the lack of agreement concerning the future frequency and structure of atmospheric and oceanic modes among the various climate models (Reference HoughtonHoughton and others, 2001). For example, with respect to ENSO, certain models have suggested that El Niño-type patterns will become more frequent in a warmer climate (e.g. Reference Timmermann, Oberhuber, Bacher, Esch, Latif and RoecknerTimmermann and others, 1999). However, investigations using other GCMs revealed no significant changes to future ENSO frequency or amplitude (Reference Zelle, van Oldenborgh, Burgers and DijkstraZelle and others, 2005), and even a reduction in the amplitude of these events (Reference Meehl, Teng and BranstatorMeehl and others, 2006). Because ENSO significantly affects climate over many regions of the Northern Hemisphere and is often associated with variations in other Pacific-related teleconnections, such as the PDO and PNA, knowledge of its future occurrence would assist in projecting regional changes to future river-ice duration and severity. The majority of climate-change investigations have indicated increased occurrence in the positive phases of AO/NAO in association with future warming (e.g. Reference Rind, Perlwitz and LonerganRind and others, 2005). This would result in continued warming over western North America and northern Eurasia, and cooling over northeastern North America and the North Atlantic.

Summary and Recommendations

In conducting this review of literature dealing with river ice and climate, it became apparent that the analyses of historical trends have often produced confounding spatial and temporal results. In addition to the complications that arise from climate variability (e.g. inter-decadal fluctuations), other factors such as varying study-period intervals, spatial coverage and/or number of stations used for analysis have also contributed to the confusion. Hence, it is recommended that future assessments consider: more explicitly the variations in climate that have occurred over the period of analysis; the use of the longest period of available records with requisite spatial coverage of reporting stations; and non-linear interpretations of the data. In reference to number of stations and length of records, however, it may be that because of extensive closure of observing stations in the 1990s (e.g. Reference Lacroix, Prowse, Bonsal, Duguay and MénardLacroix and others, 2005), the best interval for trend analysis has already passed. This is exceedingly unfortunate given that the rates of change in some river-ice characteristics, such as break-up timing, seem to have accelerated in later years (probably linked to the increased levels of atmospheric warming in cold regions) and are likely to change even more dramatically in the future. Without a robust observing network, it will be difficult to monitor the effects of future climate change/ variability or to validate predictive models. Apart from a reinstatement of observing stations, the only other method for monitoring river-ice changes would be remote sensing. Advancements in sensors offer the possibility of moving beyond the simple detection of trends in event dates to the assessment of severity indices (e.g. extent of flooding).

The intercomparison of river-ice trends from different agencies and countries is also confounded by the type of observation data employed in the analyses (e.g. Reference Catchpole and MoodieCatchpole and Moodie, 1974). Although development of a completely coordinated observation program for river ice is unlikely (see, e.g., Reference Prowse, Beltaos, Gerard, Petryk and ProwseProwse, 1990), subsequent interpretations need to recognize that differing definitions of critical river-ice characteristics could have been used (e.g. timing of breakup, the components of which can span several weeks). Moreover, it is recommended that the trend analysis of climate/river-ice relationships be extended beyond simple dates of occurrence to include intervals between different ‘event’ components, such as the initiation of pre-break-up melt and final ice clearance, which can provide more useful data about conditions that control the hydrologic severity of the event. Furthermore, additional work should focus on trends in ice-cover thickness and duration (the period between freeze-up and break-up) that are so important to many freshwater processes and factors that control the type and availability of transportation in cold-regions countries.

Lastly, it appears from the review that predictions of river-ice response to variations in atmospheric conditions remain empirical and largely site-specific. As a result, sufficiently robust physically based models are not available for predicting large-scale responses to climate change/variability, and modelling of future conditions has had to rely on proxies such as 0˚C isotherms. More research needs to be directed towards the development of physically based river-ice models that effectively couple with hydrologic-runoff and atmospheric-flux models. This work is a prerequisite to being able ultimately to employ data from GCMs/RCMs (Regional Climate Models) for future climate prediction.