The geographical area of this study focuses on the Mackenzie River extending from the western end of Great Slave Lake (61.36∘ N, 118.4∘ W) to the Mackenzie Delta (67.62∘ N, 134.15∘ W) (Fig. 1). The study area encompasses the main channel and confluences of the river, including any smaller rivers that feed the Mackenzie. Currently, only four hydrometric stations measure water level and ice on the main channel (1100 km long) of the Mackenzie River north of Great Slave Lake. A fifth station located at Fort Providence was shutdown in 2010 (Government of Canada, 2010). The MRB forms the second largest basin in North America, extending beyond the Northwest Territories (NWT) at 1.8×106 km2 (Government of Canada, 2007). Approximately 75 % of the MRB lies in the zones of continuous and discontinuous permafrost with many smaller sub-basins adding to flow at different time periods during the break-up season (Abdul Aziz and Burn, 2006). The MRB experiences monthly climatological (1990–2010) averages of -20 to -23 ∘C air temperature between the months of December and February respectively (Dee et al., 2011). Air temperature increases to an average of -5 ∘C in April with the initiation of ice break-up near 61∘ N.

Air temperature plays an important role on the timing of spring freshet (Beltaos and Prowse, 2009; Goulding et al., 2009b; Prowse and Beltaos, 2002) in the MRB. It has therefore been associated with increased flow and the initiation of ice break-up in the basin as a result of snowmelt onset (Abdul Aziz and Burn, 2006). In thermal (over-mature) ice break-up, there is an absence of flow from the drainage basin earlier in the melt season, and the ice remains in place or is entrained in flow until incoming solar radiation disintegrates the river ice increasing water temperatures (Beltaos, 1997). This slow melting process causes a gentle rise in discharge on a hydrograph, with flooding found to be less frequent during that period (Goulding et al., 2009a). In dynamic (premature) ice break-up, the accumulation of snow on the drainage basin is higher and the stream pulse (or spring freshet) from snowmelt is characterized by a high slope on the rising limb of the hydrograph (Goulding et al., 2009b; Woo and Thorne, 2003). In the presence of thick ice downstream, flow can be impeded causing a rise in backwater level and flooding upstream. However, when ice resistance is weak downstream, stress applied on the ice cover can rise with increasing water levels fracturing and dislodging ice from shorelines continuing downstream, eventually disintegrating downstream (Hicks, 2009). This process can continue until certain geometric constraints such as channel bends, narrow sections and islands can stop the ice run causing ice jams (Hicks, 2009). Here, the wide-channel jam is the most common of dynamic events which develops from the flow shear stress and the ice jams' own weight, which is formed by the collapse and shoving of ice floe accumulation and is resisted by the internal strength of the accumulation of ice flows (Beltaos, 2008). As the jam builds with ice rubble, the upstream runoff forces can increase above the downstream resistance, thus releasing the jam and creating a wave downstream that raises water levels and amplifies flow velocities (Beltaos et al., 2012). Observations have shown an initial increase and final decrease in water levels as wave celerity and amplitude attenuates downstream (Beltaos and Carter, 2009). In general, thermal decay and ice break-up processes continue downstream after the ice jam release (Hicks, 2009). MODIS imagery has also shown the timing of spring flood and location of open-water tributaries to have the most impact on ice break-up processes (Pavelsky and Smith, 2004).

A processing chain was developed in order to determine ice presence or absence (open water) on the Mackenzie River. As seen in Fig. 2, MOD/MYD10A1 Level 3 (primary data set, 500 m) was processed in the MODIS Reprojection Toolkit in order to extract specific subsets of scientific data sets (snow, river ice, cloud and open water, as seen in Table 2), perform geographic projections and write output files. Here, the primary data set was resampled using the nearest neighbour method and reprojected to the UTM projection. In the presence of cloud-free images and images where cloud was not covering the MR, open-water observations were recorded on the Mackenzie River. Observations were manually performed along the MR by visual interpretation wherever a cross-section of the river was partially or entirely ice free. When cloud cover was found to be present in the primary data set, which limited ice observations, MOD/MYD02QKM Level 1B (secondary data set, consisting of bands 1 and 2 at 250 m resolution) was used. This MODIS product was processed using the MODIS Conversion Toolkit in the ENVI/IDL package (nearest neighbour method/UTM projection). Finally, in the presence of high cloud cover in the secondary data set no observations were recorded.

Through visual interpretation varying land attributes digital number (DN) values (snow, river ice, cloud, open water) in the MOD/MYD02QKM were defined from MOD/MYD10A1 scientific data set (SDS) values of the same land attributes. Observing and comparing the same areas of interest and dates from MOD/MYD10A1 and MOD/MYD02QKM images as seen in Table 2 completed this process. For example, MOD/MYD10A1 images of ice cover at a SDS reading of 100 (river ice) was matched to a DN value ranging from 40 to 110 from the MOD/MYD02QKM images.

In addition to determining instances of ice break-up events with respect to location and time, this study also explored the use of MODIS as a tool for estimating velocity of ice flows. Ice velocity was observed and recorded on stretches of ice debris (> 15 km) where ice and water demarcation was distinguishable. Stretches of ice were defined by the changes in attributes on the Mackenzie River from open water to ice (37–100) on the leading edge of ice and ice to open water (100–37) on the trailing edge of ice. Velocity was estimated by tracking the displacement of ice over time across multiple MODIS L3 and L1B swaths. Displacement estimates over time were made twice daily from Aqua and Terra satellites image captures. It should be noted that the MODIS images do not show displacement within each image capture; therefore the average velocities represent estimates between images. Average velocities were recorded until ice debris could no longer be distinguished as a result of melt or cloud cover. Ice velocities recorded also represent the lower limit of the ice flows, as the ice may not be moving at all times between image acquisitions. Therefore, the average velocities present time periods when the ice could be at rest and, therefore, the velocity measurements represent underestimation of the actual ice velocities. Ice debris movement was also referenced to WSC station provided that an operational station was on the route of the ice run.

Over the 13 years of analysis, the ice break-up period ranged from as early as DOY 115 and lasted as late as DOY 155. Most ice break-up over the 13-year period (2001–2013) began between DOYs 115 and 125 and ended between DOYs 145 and 155. River morphology acted as an important spatial control determining the type of ice break-up process and ice run. Ice break-up processes between years showed different overall patterns with respect to location, and thus temporally the beginning, end and duration of ice break-up varied. For example, the initiation of ice break-up in 2002 (Fort Simpson, 330 km) began 10 days later than the average date when ice would completely clear the river section. Compared to 2007, the initiation of ice break-up began 13 days earlier than the average ice-off date at 270 km (61.57∘ N). As seen in Fig. 3, ice break-up initiates earliest at the headwaters (headwaters at 120 km, 61.43∘ N, to 345 km, 61.92∘ N) between Martin River and Mill Lake, and it proceeds northward towards the Mackenzie Delta (see Fig. 4).

The initiation of the ice break-up period on the Mackenzie River was generally observed at the Liard River (325 km). The beginning and end of ice-off observations were observed to take place sooner near the Liard River than upstream and downstream of this location (Fig. 5). The confluence where the Mackenzie River and Liard River meet (61.82∘ N, 325 km) serves as a point where ice break-up proceeds dynamically northbound and thermodynamically southbound. South of 325 km, ice break-up was observed to be driven by a thermodynamic ice break-up regime (Fig. 6). So, ice break-up advanced opposite to the direction of river flow, southbound towards Great Slave Lake. Interestingly, higher frequencies of observations were observed south of 325 km where thermodynamic ice break-up regime was prevalent. This ice break-up “reverse” to the river flow was observed to continue until it approached Mill Lake, where the ice break-up was simultaneously progressing in the direction of flow. The converging course continued until no ice remained south of Martin River (Fig. 6).

As ice break-up proceeded northbound from the MR–Liard confluence, dynamic ice break-up flushed the ice downstream in a shorter period of time than the thermodynamic ice break-up south of the confluence (Figs. 3, 4). Generally, however, distances above 560 km (63.22∘ N) (Wrigley, NWT) on the Mackenzie River experienced later ice break-up dates over the 13 years studied (Fig. 5).

Between 350 and 682 km (61.96–64∘ N) and north of the Mackenzie River and Liard River confluence, the average ice-off date for the study period was observed at DOY 130. The river width between 350 and 682 km was found to be smaller than reaches upstream (feeding ice into the main river channel) and downstream (letting ice exit the channel) as seen in Fig. 7. Consequently, the movement of ice into this river reach was limited, causing ice entering the channel to jam while ice exiting the channel present from the winter period cleared sooner. There is also the possibility that the release of ice javes (river waves generated from ice jam) at the entrance of the channel could give rise to the rapid clearance of downstream ice over a 1–2-day period over this 230 km stretch of the Mackenzie River (Beltaos et al., 2011).

Downstream of 682 km (64∘ N), river sections showed diagonal ice-off observations as seen in Figs. 3 and 4. These patterns are most visible in 2001, 2007–2009 and 2011–2012 observed between 860 and 1460 km (65–67.62∘ N). Observations of these diagonal events were the result of a second channel constriction at the Ramparts (1078 km, 66.19∘ N) as seen in Fig. 7, preventing northerly ice run. Here, the river channel decreased from 4 km to less than 0.5 km in width. It is hypothesised that ice runs downstream of the Ramparts (as a result of an ice jam) gave rise to similar ice-off dates (estimated at the southern ice/water boundaries of these flows) to ice runs towards the Ramparts. It is estimated that ice jams due to sudden decreases in river width as seen in 2001, 2007–2009 and 2011–2012 gave rise to earlier ice-off dates for river sections north of the jam, resulting in impeded ice run which would normally maintain ice-on condition. This phenomenon resulted in a sequence of ice-off observations that occur simultaneously at two different latitudes, north and south of the ice jam. This further outlines the important morphological controls on the Mackenzie River over ice runs.

Based on MODIS imagery, ice break-up began on average between DOYs 115 and 125 and ended between DOYs 145 and 155 (Fig. 5). The standard deviation of estimated ice-off dates decreased with increasing latitude. MODIS-derived dates showed highest deviations across river sections where thermodynamic ice break-up was prevalent. These patterns are similar to those seen from average break-up and standard deviations observed from the WSC. The 13-year average reveals similar ice conditions in the low, mid- and high latitudes of the Mackenzie River from MODIS and WSC data. There was an observed difference of 5 days between ice break-up observed from MODIS imagery and WSC. Also, the respective standard deviations overlap across the similar periods. Ice break-up in general continued in a north to south direction over the ice break-up periods. Near Forth Simpson (330 km, 61.85∘ N), it is worth mentioning that ice break-up was observed earlier than at more southern latitudes as illustrated by MODIS observations. This pattern is likewise visible from the WSC data.

In order to assess the relative timing of ice disappearance in relation to its surrounding sub-basin, the timing of river-ice disappearance was qualitatively compared to the timing of near complete snow disappearance from the surrounding area. MODIS L3 imagery of different years was selected which clearly revealed ice–snow relation with respect to location, where cloud cover was a minimal issue.

Locations where thermodynamic ice disappearance was hypothesised (south of 61.8∘ N, 325 km) corresponded with patterns where ice disappeared much later than snow on land (Fig. 6). For example, DOY 121/2006 (Fig. 8) was observed to be the beginning of the snowmelt period at 290–487 km (61.75–62.5∘ N) and this process ended when the snow had almost completely disappeared by DOY 125. However, DOY 125 corresponded to the initiation of ice break-up. This was not limited to 2006 so that snow generally disappeared sooner from surrounding sub-basins, followed by the initiation of ice break-up.

At reaches north of the MR–Liard River confluence, ice break-up and snowmelt were observed to initiate in sync with one another. As seen in Fig. 9, on DOYs 136–137/2011, ice disappearance on the southern cross-section of the figure is marked by the near simultaneous disappearance of snow. In fact by DOY 140/2011 both ice and snow had completely disappeared analogous to each other. On sections of the Mackenzie River before it enters the Mackenzie Delta, estimated ice break-up and snow disappearance was again observed to occur almost simultaneously (Fig. 12). Over a 6-day period (DOYs 137–142/2007) the ice break-up process continued until ice completely disappeared from the channel (MR). This process ensued sooner relative to complete snowmelt over the surrounding sub-basins. By DOY 142/2007 nearly one-third of the river was completely cleared of ice while most of the snow was still present over the MRB.

Principally, it was concluded that on the upper Mackenzie Basin snow cleared sooner than the initiation of ice break-up. In the mid-Mackenzie Basin (375–860 km, 62–65∘ N), river ice cleared in situ to snow clearance from the surrounding basin. In fact, ice cleared sooner in the mid-basin than the upper Mackenzie Basin. Finally, in the lower Mackenzie Basin, river ice cleared sooner than the snow from the surrounding basin. This could be telling of a river continuum of the build-up of mechanical strength used to clear river ice within the Mackenzie River towards higher latitudes. The Liard River tributary accounts for one-third of the total Mackenzie discharge (Woo and Thorne, 2003), and so a rise in discharge in May initiates earlier ice break-up downstream as a result of increased stress induced on ice by a rise in river stage. Mechanical stress used to shove ice is continually magnified by the addition of small and large tributaries downstream of the Mackenzie River (Great Slave River, Arctic Red River).

Figures 10 and 11 illustrate ice movement from which ice velocities could be estimated over periods of 3–4 days following secondary channel constriction at 66∘ N. Here, ice runs that contained over 15 km of entrained ice were chosen to estimate average ice velocities. Only periods with at least three images with partial or no cloud cover were selected for velocity estimates.

In 2008, the open-water/ice boundary (leading edge) was recorded beginning on DOY 143 (Fig. 10). The open-water/ice (northern edge of ice) and ice/open-water (following edge) boundaries were both visible from DOY 144. Finally, the ice/open-water boundary was last observed on DOY 145. The average ice-run velocity between 1063 and 1210 km (66–66.95∘ N) over the 3 days was estimated to be at least 1.21 ms-1. Likewise, in 2010 (Fig. 11), open-water/ice (leading end) and ice/open-water (following end) was observed between DOYs 138 and 141. The leading edge of the ice was first observed on DOY 138 and on DOY 139 when both the leading and following edges are visible. Finally, by DOY 141 the ice run has exited into the Mackenzie Delta. Across the 4-day period average ice-run velocity was estimated to be at least 1.84 ms-1.

Over the 13-year period, the average estimated ice break-up dates were found to range from DOYs 115 to 155 between distances 60 and 1460 km. These estimates from MODIS are in agreement with break-up dates reported by the WSC ground-based network. Previous studies on the MR, between the late 1930s to 2002, have found the initiation of ice break-up to range from DOYs 123 to 140 between 0 and 1217 km (de Rham et al., 2008a). Furthermore, it was reported by de Rham (2008b) that the duration of average ice break-up ranged from 8 to 10 days over the entire basin. With respect to the findings reported in Fig. 5, the observed ice break-up patterns agree, such that the average observed break-up dates over the 13-year period ranged from DOY 128±8 days at 61.5∘ N (260 km) to 145±4 days at 68∘ N (1460 km). Others have reported, using MODIS and AVHRR imagery acquired between 1992 and 2002, that ice break-up ranged from DOYs 120 to 155 (Pavelsky and Smith, 2004). The earliest reports of mean ice break-up dates ranged from 15 May (DOY 135) to 25 May (145) (1946–1955 averages), from Fort Providence to Arctic Red River respectively (MacKay, 1966). Furthermore, others have reported a range of ice break-up dates from 22May (DOY 142) to 31 May (DOY 151) (1927–1974, Fig. 3) from Fort Providence to Fort Good Hope, NWT, respectively (Allen, 1977). A summary of the approaches and results reported in these previous investigations as well as those from this study are compared in Table 4.

In the headwaters of the Mackenzie River, ice break-up initiates the earliest between Mill Lake (120 km, 61.43∘ N) and Martin River (345 km, 61.92∘ N). As seen in Figs. 3–5, ice break-up between 120 and 300 km initiated earlier as compared to other sites on the Mackenzie River, but ice cleared later than other for other river sections downstream. Here ice in the channel remains stagnant for extended periods of time as ice usually freezes to bed and is most susceptible to thermodynamic melt (MacKay and Mackay, 1973).

Furthermore, at the Liard River confluence (325 km, 61.84∘ N) it was found that the seasonal initiation of ice break-up began and cleared earliest at this central location where the Liard River converges into the MR. Others (Pavelsky and Smith, 2004) have noted that at the MR–Liard River confluence flooding is common between years, especially near channel junction. Ice break-up at the Liard River confluence occurs rapidly, as the flow contribution is of greater magnitude than the Mackenzie River (MacKay and Mackay, 1973). This causes a lifting of the river stage, exerting pressure on the ice cover and resulting in ice jam downstream most often attributed to the presence of channel bending (Camsell Bend, 456 km) and channel constriction. Similar processes have also been reported for the Susquehanna River, USA, where an observed increase in discharge downstream fosters earlier ice break-up while sections of the upper river remain ice covered (Chaouch et al., 2012). The severity of ice break-up stage is therefore largely controlled by upstream discharge (Goulding et al., 2009b). Pavelsky and Smith (2004) also observed irregularities in ice break-up timing between years, particularly at 325 km (MR–Liard confluence) on the MR. Here, ice break-up began earlier at distances north of 325 km (61.8∘ N) than river sections south. Postponed ice break-up in the upper Mackenzie can result from the lack of discharge required to initiate ice break-up and so the ice is thermodynamically disintegrated.

Channel morphology is, therefore, a more important control on ice break-up patterns than previously believed. Both Pavelsky and Smith (2004) and de Rham et al. (2008a) alluded to the fact that channel morphology may exert influences on the patterns of ice break-up. De Munck et al. (2011), through the use of geospatial modelling on the Chaudière River identified that channel islands, confluences and channel sinuosity predisposed the Chaudière River to ice jamming events. In this study, it is determined that channel constriction at 350–682 km (61.96–64∘ N) and 1078 km (the Ramparts) is responsible for the delay of ice break-up timings upstream while promoting earlier ice break-up downstream. Upstream of the Liard River junction, river flow is stable. However, excessive discharge supplied by the Liard River causes earlier ice break-up and ice jamming downstream when the channel constricts between 350 and 682 km (MacKay and Mackay, 1973). Furthermore, excess supply of ice cover from the Great Bear River (821 km) into the Mackenzie River causes the development of ice jamming at the Ramparts when the channel width decreases from over 3.5 km to less than 0.6 km (as seen in Fig. 7) (MacKay and Mackay, 1973).

Ice jamming from channel width decreases gave rise to similar sequences of ice-off observations, which occurred in tandem at two different latitudes, north and south of the ice jam (as seen at the Ramparts). Ice jams are therefore favourable where morphological features impede downstream ice passage (Beltaos, 1997). These ice jams are caused by channel constriction resulting from mid-channel islands and narrow reaches (Terroux et al., 1981). Channel braiding, constriction and changes in slope have also been reported to be important factors influencing ice break-up and flow regimes (de Rham et al., 2008a). In the context of our study, it was found that channel constrictions and bends represented locations where ice runs were impeded. Hicks (2009) also reported that running ice may be stalled when geometric constraints such as tight bends, narrow sections and islands are present in rivers. In fact, it has been shown that ice debris flow drop to a velocity of 0 in the presence of flow depths near channels islands and bars (Kääb et al., 2013). Lastly, Kääb and Prowse (2011), using ALOS PRISM stereo imagery on the Mackenzie River determined that ice velocities decrease to 0 in the presence of bars.

The estimated ice-run events illustrated in Figs. 7 and 10 may have been caused by ice jam releases (javes) initiated at the Ramparts (1078 km, 66.19∘ N). Such processes may also be the reason why ice was estimated to be cleared at higher latitudes before the end of the snowmelt period. Accumulated stress with the rise of water levels behind the jam can result in sufficient kinetic energy to clear river ice downstream before the complete snowmelt overlying the surrounding sub-basins.

Ice-run velocities are believed to be the highest where the ice is minimally effected by channel morphology, unconnected from incoming tributaries, and channel splitting which causes the formation of islands (Kääb et al., 2013). Amongst the variety of ice runs observed over 13 years, ice velocities could be quantified in 2008 and 2010. Over 3–4-day periods, average ice velocities were estimated to be 1.21 ms-1 (2008) and 1.84 ms-1 (2010). More importantly, it is believed that the evolution of such velocities is the product of javes. Our measurements of ice-run velocity in 2008 coincidently synchronise with other independent satellite- and ground-based ice measurements. Extensive measurements of ice runs in 2008 around MR–Arctic Red River junction are believed to be generated by waves released from released ice jams (Beltaos, 2013). This aligns with ice jams which may form at the Ramparts (1078 km, 66.19∘ N) as a result of channel constriction. The evolution of ice runs north of the Ramparts (flowing past the Arctic Red River) observed over DOYs 143–146/2008 (22–25 May/1.21 ms-1) matches similar ground measurements (1.7 ms-1) made by Beltaos et al. (2012). Across the same cross-section of the MR, Kääb and Prowse (with imagery acquired 1–2 days earlier in 2008) estimated a preceding ice run ranging from 0 to 3.2 ms-1. The highest flow velocities were outlined where ice debris flow was most concentrated on the outside turn of the river bend. Finally, in another independent study, Beltaos and Kääb (2014) found ice debris velocities to range between 1 and 2 ms-1 using ALOS PRISM imagery in 2010. Again these high-resolution (2.5 m) image measurements compare quite well with our estimates from relatively coarse spatial resolution (250–500 m) MODIS imagery. Additionally, early investigations have reported that ice can clear at velocities of 0.27 and 0.44 ms-1 at Fort Simpson and Fort Good Hope respectively during the ice break-up season (Terroux et al., 1981).

MODIS is shown to be a viable tool for estimating river-ice velocities. This study finds that in order to monitor ice cover the river width needs to at least 0.5–1 km wide. Furthermore, to quantify river-ice velocities, the river width needs to be at least 1 km wide. With respect to the MR, ice velocities were only quantifiable above the Ramparts. The presence of morphological controls and therefore river width shortening leading to impeded ice run prevented quantifying velocities, as leading river-ice demarcations were difficult to locate. However, it was possible to estimate the overall velocity by observing ice/open-water boundaries. Lastly, it was determined that in order to measure ice-run velocities without major disturbance with impeded flows with respect to river morphology, estimates with MODIS should be made north of the Ramparts. North of the Ramparts, river widths were generally observed to be largest with respect to other parts of the MR.