Fluvial adjustments in response to glacier retreat: Skaftafellsjökull, Iceland
Abstract
This study describes changes to the proglacial drainage network of Skaftafellsjökull, Iceland from 1998 to 2011. Proglacial landscapes are highly sensitive to glacier retreat, and the retreat of glaciers around the world has accelerated since the mid-1990s. Skaftafellsjökull has retreated at an average rate of 53 m per year since 1999. From 1999 to 2003, the river incised and formed a sequence of now abandoned channels and fluvial terraces extending ∼1 km downstream from the glacier. Retreat of the glacier from an over-deepened ice-contact slope meant that there was a positive correlation between the distance of glacier retreat and the amount of fluvial incision. Incision was episodic, occurring annually in response to drainage reactivation and reorganization. On an annual basis, the rate of retreat is moderately negatively correlated with the rate of incision. This is partly because the ice-contact slope decreases away from the position of maximum glacier extent, and also because more sediment is released with faster retreat, counteracting the effect of retreat down an ice-contact slope. From 2003 onwards, proximal terrace formation ceased, as a proglacial lake became established. Downstream of the lake outlet further incision deepened the channel, with most change occurring during a flood in 2006, where incision in the upstream confined reach was accompanied by downstream aggradation and terrace formation. These observations indicate that proglacial changes in response to glacier retreat are a result of the interactions of river channel incision and terrace formation, aggradation, lake development, and flooding, which together control river channel changes, sediment redistribution and sandur stratigraphy.
The retreat of glaciers around the world has accelerated since the mid-1990s (Zemp et al. 2008). Downstream of glaciers, the impacts of glacier retreat include short-term increases in discharge owing to increased melt rates (Yao et al. 2007), but also reductions in water supply to downstream communities as glaciers reduce in volume (e.g. Barnett et al. 2005; Vergara et al. 2007). Nearer to glacier margins, attention is focused largely on the development of proglacial lakes, and on the corresponding increase in the risk of catastrophic flooding (Ageta et al. 2000; Kiernan & McConnell 2002; Harrison et al. 2006). It is also expected that reductions in discharge and changes in sediment supply will lead to adjustments of proglacial rivers, for example regarding incision and aggradation (Maizels 1983, 1986; Gurnell et al. 2000).
Despite the potential for large-scale readjustment of the proglacial fluvial system in response to glacier retreat, there are relatively few studies that relate year-on-year retreat to changes in fluvial geomorphology. Some studies use aerial photography at relatively widely spaced intervals to document changes, which means that the process links between retreat and observed changes must be inferred and that the effects of short-term fluctuations cannot be detected (Thompson 1988; Chew & Ashmore 2001; Zah et al. 2001). Maizels (1979) made repeated surveys of the Bossons Glacier from 1968 to 1975, but this coincided with a period of glacier advance.
The likely response of a proglacial river to margin fluctuations is understood in broad terms. In settings where glaciers can advance over sediment, glacier advance usually causes aggradation, leading to an increase in proglacial gradient near the glacier and an increase in braid-bar density in proglacial river channels (Maizels 1979). Glacier retreat leads to an upstream lowering of the long profile, causing incision through the sediment aggraded during the advancing phase (Thompson & Jones 1986; Thompson 1988; Roussel et al. 2008). This process of terrace formation in response to glacier retreat was confirmed by Marren (2002), who observed that terrace formation was rapid, and that ‘average’ rates of terrace formation calculated using the intervals between aerial photographs and depths of incision over the same time period are largely meaningless, as most of the incision is likely to have occurred over one or two years rather than progressively.
Discharges reduce as glaciers decrease in volume, but are highly unsteady (Marren 2005). Sediment supply increases during advancing phases and remains high throughout and after deglaciation (Church & Ryder 1972; Wilkie & Clague 2009). The development of lakes in front of retreating glaciers reduces downstream sediment inputs to a much greater extent than do margin fluctuations. Sediment starvation downstream of glacial lakes leads to a reduction in braiding in the proximal reaches (Chew & Ashmore 2001). The findings of Chew & Ashmore (2001) have been reproduced by Germanoski & Schumm (1993) using a laboratory flume: reductions in sediment supply reduced braiding in the proximal reach but increased braiding in the distal reach. In the flume experiment, a reduction in sediment supply was accompanied by upstream incision and terrace formation and by a transition to a single-thread channel, while aggradation and braiding occurred downstream (Germanoski & Harvey 1993; Germanoski & Schumm 1993). The process by which incision triggers aggradation (and vice versa) is a type of ‘complex response’ (Schumm 1977). To date, complex response processes have only been observed in proglacial systems where sediment starvation has caused incision. Marren (2005) speculated that incision caused by glacier retreat may also produce a type of complex response, but this idea is untested to date.
Much research on the effect of glacier margin fluctuations on proglacial rivers has been undertaken in SE Iceland, in particular at Skaftafellsjökull and Svínafellsjökull. Thompson & Jones (1986) showed that periods of terrace incision are strongly correlated with periods of rapid glacier recession with a non-linear relationship over decadal time scales, with slower retreat producing greater rates of incision. Thompson & Jones 1986) dated terrace surfaces using lichonometry rather than observing the formation of individual terraces, however, and, as discussed above, this meant that they obtained average rates of incision, which do not reflect the fact that terraces usually form in a single year (Marren 2002). The average incision rates produced by Thompson & Jones (1986) therefore probably underestimate the actual rates of incision. Roussel et al. (2008) carried out a study similar to that of Thompson & Jones (1986), but using data from more glaciers. They observed that initial retreat from the Little Ice Age maximum extent was not accompanied by incision for the first 50 to 100 years. Roussel et al. (2008) suggest that this initial period of stability occurs because paraglacial sediment supply maintains the proglacial sediment budget, and as the paraglacial sediment supply diminishes, incision begins to occur. The work of Roussel et al. (2008) was based on lichonometry, and the average incision rates presented do not reflect the effects of annual retreat on incision rates and the event-based nature of terrace incision.
The aim of this paper is to investigate the links between rates of glacier retreat and proglacial river terrace formation on an annual basis. In doing so, this paper will be the first to provide actual rates of annual terrace incision that can be correlated with annual glacier retreat. This paper builds on the work of Marren (2002), which described changes in the proglacial zone of Skaftafellsjökull for the period 1996 to 2000. It includes observations up to 2011, and considers changes in the proglacial channel further downstream than in Marren (2002) in order to investigate the possibility of complex-response-type behaviour occurring as a result of glacier retreat.
Study area
Skaftafellsjökull is located in SE Iceland, where it descends from the Vatnajökull and Öræfajökull icecaps (Fig. 1A). The terminus of the glacier is ∼120 m above sea level, and is connected to the low-lying coastal plain sandur extending from Skeiðarárjökull to Öræfi. As with many glaciers in SE Iceland, a significant ice-marginal over-deepened basin exists beneath the glacier margin (Tweed et al. 2005). As the glacier retreats, either the resulting ice-marginal slope is incised in the manner suggested by Thompson & Jones (1986), creating fluvial terraces downstream of the glacier margin, or a proglacial lake develops behind the ice-marginal slope. Both responses are common, and in SE Iceland proglacial lakes have increased significantly in area over the past decade, including at Skaftafellsjökull (Schomacker 2010). The margin of Skaftafellsjökull is ∼3 km wide, with major meltwater outlets at both the western and eastern ends of the glacier margin. Retreat of the glacier from the Little Ice Age maximum has left a series of parallel moraine ridges, separated by palaeo-sandur (Fig. 1B; Thompson 1988). The two meltwater rivers meet close to the moraines identified by Thompson as the 1939 moraines, and beyond the moraine belt (and the Highway 1 road bridge) the river forms an unconfined sandur plain extending to the sea.
A. Location map of Skaftafellsjökull in Iceland. B. Map of the proglacial area of Skaftafellsjökull. Shading of the glacier margin indicates the positions in 1997, 2003 and 2008; braided channels in pale grey with black dots are from 1997, in mid-grey from 2003, and in dark grey from 2008; stippled area indicates major moraine belts; lake extent shown for 2008. AWS indicates the location of the Icelandic Meteorological Organisation automatic weather station. Box indicates location of panel C. C. Detail of the location of the cross-sections surveyed in 2003 and resurveyed in 2011. Shading of the braided channel indicates the year as in B.
The fluctuations of Skaftafellsjökull have been described from oral sources, from aerial photography, and since 1932 by annual measurements of the glacier margin position (Thorarinsson 1943, 1956; Thompson 1988; Sigurðsson 1998). The present monitoring site has been used since 1942 and is located close to the centre of the glacier margin (Table 1; Fig. 2; Sigurðsson 1998). In the past there were up to four monitoring sites across the front of Skaftafellsjökull, although they tended to show similar fluctuation patterns (Sigurðsson 1998), and aerial photography indicates that the data from the present location reflects trends across the entire margin (Fig. 1B).
| Year | Margin change (m) | Cumulative retreat (m) | Year | Margin change (m) | Cumulative retreat (m) |
|---|---|---|---|---|---|
| 1942–1943 | −22 | −22 | 1976–1977 | −15 | −768 |
| 1943–1944 | −63 | −85 | 1977–1978 | +10 | −758 |
| 1944–1945 | −54 | −139 | 1978–1979 | n.m. | −758 |
| 1945–1946 | −7 | −146 | 1979–1980 | +10 | −748 |
| 1946–1947 | n.m. | −146 | 1980–1981 | +26 | −722 |
| 1947–1948 | −32 | −178 | 1981–1982 | +9 | −713 |
| 1948–1949 | −27 | −205 | 1982–1983 | −15 | −728 |
| 1949–1950 | −143 | −348 | 1983–1984 | +5 | −723 |
| 1950–1951 | +94 | −254 | 1984–1985 | +20 | −703 |
| 1951–1952 | +28 | −226 | 1985–1986 | −15 | −718 |
| 1952–1953 | −22 | −248 | 1986–1987 | +5 | −713 |
| 1953–1954 | −64 | −312 | 1987–1988 | +1 | −712 |
| 1954–1955 | −23 | −335 | 1988–1989 | −14 | −726 |
| 1955–1956 | −113 | −448 | 1989–1990 | −2 | −728 |
| 1956–1957 | +10 | −438 | 1990–1991 | −12 | −740 |
| 1957–1958 | −12 | −450 | 1991–1992 | 0 | −740 |
| 1958–1959 | −48 | −498 | 1992–1993 | −6 | −746 |
| 1959–1960 | −63 | −561 | 1993–1994 | −46 | −792 |
| 1960–1961 | −42 | −603 | 1994–1995 | −56 | −848 |
| 1961–1962 | −10 | −613 | 1995–1996 | +51 | −797 |
| 1962–1963 | −55 | −668 | 1996–1997 | +15 | −782 |
| 1963–1964 | −30 | −698 | 1997–1998 | +24 | −758 |
| 1964–1965 | −5 | −703 | 1998–1999 | −40 | −798 |
| 1965–1966 | −15 | −718 | 1999–2000 | −70 | −868 |
| 1966–1967 | −67 | −785 | 2000–2001 | −20 | −888 |
| 1967–1968 | +22 | −763 | 2001–2002 | −48 | −936 |
| 1968–1969 | −33 | −796 | 2002–2003 | −82 | −1018 |
| 1969–1970 | −12 | −808 | 2003–2004 | −30 | −1048 |
| 1970–1971 | +45 | −763 | 2004–2005 | −97 | −1145 |
| 1971–1972 | −33 | −796 | 2005–2006 | −27 | −1172 |
| 1972–1973 | −27 | −823 | 2006–2007 | n.m. | −1172 |
| 1973–1974 | +30 | −793 | 2007–2008 | −89 | −1261 |
| 1974–1975 | +33 | −760 | 2008–2009 | −82 | −1343 |
| 1975–1976 | +7 | −753 | 2009–2010 | −50 | −1393 |
Skaftafellsjökull reached its Little Ice Age maximum extent sometime between 1870 and 1904, and until 1935 Skaftafellsjökull and the adjacent Svínafellsjökull met to form a single piedmont lobe (Thorarinsson 1943). Skaftafellsjökull retreated steadily from 1904 to 1970, with only minor advances in the early 1950s, 1957 and 1968. In total, Skaftafellsjökull retreated ∼900 m from its 1904 position to its 1942 location, and 808 m from 1942 to 1970. The average rate of retreat was 72 m per year from 1934 to 1942, and 30 m per year from 1942 to 1970. From 1971 to 1988 the glacier advanced in most years, resulting in a total advance of 96 m, with the margin in 1988 in a similar location to that in the mid-1960s. From 1989 to 1995 there was 136 m of retreat, and in the following three years the glacier advanced a total of 90 m. This advance resulted in the braided proglacial channel described by Marren (2002). Since 1999, the glacier has been in continuous retreat at an average rate of 53 m per year. The impacts of the first two years of retreat after 1999 were described by Marren (2002).
Methods
Observations of Skaftafellsjökull and associated changes in the proglacial zone have been ongoing since 1996. Photographs taken from the viewpoint Sjónarnípa (Fig. 1) provide a consistent location for observing proglacial changes. The first five years of observations were reported in Marren (2002). Figure 3 shows the changes from 1997 to 2011. Additional photographs from the same viewpoint were obtained from co-workers for years when no visit to the study area was made, and views from 2002, 2007 and 2010 are available in the published literature (Tweed et al. 2005; Schomacker 2010; Bernhardson 2011). In addition, aerial photographs from 1997, 2003 (1:24 000) and 2008 (digital) were rectified in ArcGIS to the WGS84 datum and used to map changes in glacier position, lake extent, and channel type and location.
Oblique views of the margin of Skaftafellsjökull taken from the viewpoint Sjónarnípa. The view is to the south (down-glacier). All nine photographs were taken in July of their respective year except where indicated. The 1997 and 2003 photographs are courtesy of Dr. Matthew Roberts; the 2009 photograph is courtesy of Dr. Russell Drysdale.
In order to investigate changes to the proglacial zone that have occurred since those described by Marren (2002), a long profile along the western bank-top of the western outlet river was surveyed using a total station in July 2003, from the bridge over Highway 1 to where the river emerged from the glacier (Fig. 1B). The bridge was used as a marker for the downstream end of the long profile and as a datum for all subsequent surveys. In addition, 11 cross-sections were surveyed across the full width of the braided channel (Fig. 1C). The cross-sections were located where the confining moraine belt begins to open out, and the channel switched from a single-thread to a braided channel in 2003. Both ends of each cross-section were marked by a wooden stake. The 2003 survey consisted of a total of 531 data points. The 11 cross-sections were located at the transition from a single to a braided channel in order to detect downstream migration of the single channel and aggradation beyond the point of incision. If observed, this transition from upstream erosion to downstream aggradation would indicate that proglacial streams have the potential to develop complex-response-type behaviour. However, this experimental design does not allow further cycles of erosion and aggradation, either downstream or over time, to be detected.
The long profile and cross-sections were resurveyed in June 2011 using a total station. The long profile was extended to the new location of the outlet of the western meltwater channel. Because the channel surveyed in 2003 was abandoned (see below), morphological criteria were used to identify the lowest visible bank-top along the length of the channel (Williams 1978). Two other notable terrace surfaces that ran parallel to the lowest identified channel were also surveyed. All 11 cross-sections were located again, although, because of channel widening and erosion, some cross-sections only had one of their location stakes remaining. Where this occurred the lines were resurveyed along a bearing. Because of the major changes to the western outlet channel that occurred between 2008 and 2011, the long profile of the eastern outlet river channel was also surveyed, from the glacier to its confluence with the western channel just upstream of the Highway 1 bridge (Fig. 1B). Based on repeated visual observations of the glacier margin (Fig. 3), terrace surfaces in the proglacial region were identified and the year of formation determined. A long profile for each of these terrace surfaces was surveyed. Because of the greater number of features surveyed in 2011, the survey consisted of a total of 1453 data points.
Results
Proximal proglacial changes at Skaftafellsjökull, 1996 to 2011
From 1996 to 1998 the western outlet river of Skaftafellsjökull was directly connected to the sandur. During this time, the proximal sandur was characterized by a braided channel pattern that developed immediately downstream of where the meltwater channel exited the glacier (Marren 2002; Fig. 3A,B). The channel remained braided until it was forced into a single channel pattern, where it cut through the 1958 moraine belt. Downstream of the 1958 moraines, the channel remained as a single thread for ∼200 m before exiting the confining moraine belt and developing a braided channel pattern (Fig. 1B). The occurrence of the proximal braided channel from 1996 to 1998 coincided with the three years of advance that occurred in the mid-1990s (Fig. 2).
The retreat of Skaftafellsjökull began in 1999, with 40 m of retreat occurring in that year. As a consequence, the braided terrace surface was abandoned, and drainage was concentrated into a single channel that ran parallel to the ice margin in front of the moraines formed during the 1996–1998 re-advance (Fig. 3C). This channel was occupied during 1996 to 1998 as part of the braided network, but was not the largest channel on the sandur. Where the channel met the 1958 moraine it rejoined the existing single-thread channel that flowed between the moraines. Only a minor adjustment in the channel pattern took place in 2000, despite 70 m of retreat occurring (Fig. 3D). The channel narrowed, and a minor terrace surface was formed. Up to 1.75 m of incision occurred in 1999, and 0.97 m of incision occurred in 2000 (based on the comprehensive re-survey of the terraces carried out in 2011; Fig. 4). With these two years of observations available, Marren (2002) was unable to identify any correlation between rate of retreat and rate of incision. The other major change that occurred in 2000 was the development of a narrow proglacial lake between the glacier margin and the 1998 moraine belt (Fig. 3D).
Long profiles of the proximal terraces and abandoned channels at Skaftafellsjökull, as surveyed in 2011. The full sequence of terraces can be seen in the centre-right of Fig. 3, and the line of the uppermost terrace is shown in Fig. 1. The upper three terraces are equivalent to those shown in Marren (2002: fig. 5) (the datum used by Marren (2002) was arbitrarily selected).
Long profiles of the Skaftafellsá river channel from the Highway 1 road bridge at the downstream end to the glacier margin. The uppermost dashed line was surveyed in 2003; the solid line was surveyed in 2011 but indicates the channel last occupied in 2010; and the lowermost line is the eastern river channel, which became the main drainage outlet from late 2010 onwards (surveyed in 2011).
Further changes occurred between 2000 and 2001 (Fig. 3E). Twenty metres of retreat occurred, along with considerable lowering of the glacier margin. In response, the proglacial channel switched to flowing behind the 1998 moraine belt, abandoning the 1999–2000 channel completely. The presence of parallel belts of abandoned outwash at Skaftafellsjökull was noted by Thompson (1988), who observed that a similar change occurred sometime between 1960 and 1968 (the dates of available aerial photographs). The observations presented here indicate that, as with the abandonment of terraces, the switching of a channel from in front of to behind moraine belts occurs rapidly, over the course of one melt season. The relocation of the channel resulted in incision and in a channel lowering of up to 1.43 m. The rearrangement of the channel resulted in the river occupying the space that in the previous year had been the location of an incipient proglacial lake. In 2001, water could be observed flowing parallel to the glacier for a considerable distance downstream of the outlet before deepening into a lake. As the channel passed through the gap in the moraines, it became a river channel again.
In 2002, 48 m of glacier retreat occurred, and the gap between the glacier margin and the 1998 moraine increased, allowing the proglacial lake to increase in extent (Tweed et al. 2005: fig. 2; Schomacker 2010). A small fluvial terrace developed at the western end of the lake, downstream of the meltwater outlet from the glacier. The development of this terrace was accompanied by incision of up to 0.48 m. The lake remained a similar size throughout 2002 and early 2003, with a possible reduction in size because of winter re-advance (Fig. 3F), although by the summer of 2003 it had enlarged considerably (Fig. 1B). Progressive enlargement of the lake occurred from 2004 to 2010 (Figs 1B, 3G–H; Schomacker 2010). No terraces that could be clearly attributed to a particular year developed over this time. A lowering of the outlet of the lake between 2003 and 2010 is evident from the surveys of the long profile of the channel, from the Highway 1 bridge to the glacier (Fig. 5). At the outlet of the lake, 1.89 m of incision occurred between 2003 and 2010.
By late 2010, the proglacial lake extended across the full width of the margin of Skaftafellsjökull (Fig. 3H, I). The lake thus intercepted the eastern meltwater outlet (Fig. 1B), and meltwater from the western outlet channel was able to drain via the lake into the eastern outlet. The eastern outlet is 4.6 m lower than the western outlet (Fig. 5), and the water level of the lake therefore decreased in elevation as a result. The western gap through the 1998 moraine, occupied from 2001 to summer 2010, was therefore abandoned in late 2010 (Bernhardson 2011: fig. 5). In 2011, the western channel was unoccupied at the time of the re-survey, with all meltwater exiting the glacier through the eastern river channel (Fig. 3I).
The new extended data set of terrace incision rates allows a much clearer relationship between glacier retreat and incision to be observed (Table 2; Fig. 6). When treated cumulatively, retreat is strongly positively correlated with terrace incision and channel lowering (correlation coefficient of 0.95, associated P-value of 0.011; Fig. 6A). The cumulative trend indicates that the glacier is consistently retreating down an over-deepened ice-contact slope, such that, over the past decade, retreat has always caused the river channel to incise. The presence and significance of the ice-contact slope is supported by the behaviour of the river channel when it switched from the western to the eastern outlet as the proglacial lake extended across the full length of the glacier in late 2010. The switch in location of the main channel was associated with a drop in lake level of 4.6 m as the river occupied the lower channel. This lower channel already existed, and there was no incision associated with the drainage reorganization, so this is a different type of observation, and the interpretation that the elevation of the eastern outlet reflects the ice-contact slope should be treated with some caution. However, when the change in elevation is compared with the glacier retreat that occurred between 2004 and 2010, the same relationship between retreat and lowering occurs as for the period 1998 to 2003 (Fig. 6A). This observation indicates that the ice-contact slope that the glacier is retreating from extends across the full width of the glacier to a depth greater than that which has been uncovered by the glacier retreat to date.
A. Cumulative incision for the period 1998 to 2003 plotted against cumulative retreat rates (diamonds). The square symbol indicates the switch from the western to the eastern drainage outlet; this lowering was caused by drainage reorganization rather than by incision, but because the lowering is occurring on the same ice-contact slope, the relationship between retreat and channel lowering is maintained. B. Annual retreat rate plotted against annual incision rate. This plot considers only those years in which retreat was associated with incision, and shows that faster rates of retreat show a moderate tendency towards slower rates of incision.
| Year | Retreat (m) | Incision/ lowering (m) | Metres of incision per metre of retreat | Cumulative retreat (m) | Cumulative incision (m) | Cumulative slope |
|---|---|---|---|---|---|---|
| 1998–1999 | 40 | 1.75 | 0.044 | 40 | 1.75 | 0.044 |
| 1999–2000 | 70 | 0.97 | 0.014 | 110 | 2.72 | 0.025 |
| 2000–2001 | 20 | 1.43 | 0.071 | 130 | 4.15 | 0.032 |
| 2001–2002 | 48 | 0.48 | 0.01 | 178 | 4.63 | 0.026 |
| 2002–2003 | 82 | 0.84 | 0.01 | 260 | 5.47 | 0.021 |
| 2003–2004 | 30 | 1.89 | No retreat induced incision in this period | 290 | 7.36 | 0.012 |
| 2004–2005 | 97 | (lake outlet lowering) | 387 | (including lake outlet lowering) | (7.36/585) | |
| 2005–2006 | 27 | 414 | ||||
| 2006–2007 | n.m. | – | ||||
| 2007–2008 | 89 | 503 | ||||
| 2008–2009 | 82 | 585 | ||||
| 2009–2010 | 50 | 4.6 | 0.017 (1.89+4.6)/375 | 635 | 11.96 | 0.018 |
When treated cumulatively, the amount of incision per metre of retreat effectively reveals the gradient of the ice-contact slope at the end of each year of retreat (‘Cumulative slope’ column of Table 2). These data show that the ice-contact slope became progressively flatter with continued retreat, with a maximum slope of 0.044 in 1999, and then from 2001 reduced annually from 0.032 to 0.018.
On an annual basis, there is a more complex relationship between retreat and incision. There is a general tendency for the rate of incision to decline in years with higher rates of retreat (Fig. 6B), so that on an annual basis there is a moderate negative correlation (correlation coefficient of –0.53 and associated P-value of 0.35) between the rate of glacier retreat and rate of terrace incision (Fig. 6B). Because Fig. 6B deals with the rates of the two processes irrespective of when the retreat and incision occurred, it can be seen that this tendency for reduced incision with faster retreat is at least partly independent of the overall reduction in the ice-contact slope described above. This is counterintuitive, as it would be supposed that faster retreat would take the glacier further down the ice-contact slope, and therefore always lead to a greater depth of incision. The fact that this is not the case indicates that other processes must operate in addition to the physical lowering of the meltwater exit from the glacier. In order for the channel to not incise, localized aggradation must be occurring in response to increased rates of sediment release during the periods of faster glacier retreat (Maizels 1979).
Changes to the distal Skaftafellsá channel 2003 to 2011
Incision along the full length of the Skaftafellsá between the original survey in 2003 and the resurvey in 2011 (Fig. 5). The average incision was 1.5 m, and all of the incision occurred prior to 2010, when the channel was last occupied. Although the channel was not resurveyed in 2004, qualitative observations indicate the nature of the changes that occurred between 2003 and 2004. Some widening occurred through the moraine-confined reach, with pegs being lost on the west bank at cross-sections 2 to 4 (these pegs were re-established during the 2004 field visit). A minor terrace established in this reach indicates that this widening was accompanied by some incision. Downstream of the moraine constriction, the 2003 surface showed signs of fluvial activity, but had been partly abandoned at its upstream end. The point at which the channel switched from incision to aggradation had moved downstream by ∼150 m, from close to cross-section 4 to downstream of the location of cross-section 5. By the time of the 2008 aerial photograph (Fig. 1B), the transition from an incising, single channel to an aggrading, braided channel had moved downstream to approximately the location of cross-section 7.
The nature of the changes between 2003 and 2010 along this reach are illustrated in Figs 7 and 8. The resurveyed cross-sections (Fig. 7) indicate that where the channel flows through the moraines (cross-sections 1 to 4) the channel abandoned the floodplain surface that was active in 2003, and incised a new channel ∼1 to 2 m deeper. Widening of the cross-sections in this region increases in a downstream direction from cross-section 1 through to cross-section 4, but nonetheless the dominant process in this reach was down-cutting of the channel bed, rather than large-scale channel erosion and widening.
Cross-sections across the distal Skaftafellsá river channel. The dashed line is the cross-section in 2003; the solid line is the cross-section in 2011. Locations of the cross-sections are shown in Fig. 1. For explanation see text.
Terraces on the distal Skaftafellsá river channel as surveyed in 2011. The solid black line is the 2010 channel as surveyed in 2011, as in Fig. 5. The uppermost dashed line is the remains of the reworked 2003 surface, resurveyed in 2011. The middle, grey line is a new terrace surface, thought to have been formed by a flood in 2006 (see text).
Downstream of the moraine constriction, where the floodplain broadens and braiding of the channel develops, the pattern of incision and channel change is different. In this proximal braided section (cross-sections 5 to 7) there is still significant incision along the last occupied main channel, but there is evidence for aggradation on the bars flanking the main channel. This aggradation can be seen in the field as a distinctive terrace, which lies above the last occupied channel and below the now discontinuous remains of the riverbank that was surveyed in 2003 (Fig. 8). This new intermediate terrace is also clearly identifiable as a distinctive gravel layer, up to 1 m thick, which in places overlies a soil layer (Fig. 9A). The surface of this overlying gravel layer is characterized by several distinctive ice-block obstacle marks (Fig. 9B), which in this environment are a clear indication that the gravel layer was deposited during a glacial flood (Russell 1993; Marren 2005).
Details of the terrace surface thought to have formed during a flood in 2006. A. Glacifluvial gravels overlying a soil layer. B. Ice-block obstacle mark on the surface of the gravel deposit shown in (A); note the boulders on the stoss-side of the obstacle mark and the scour on the flanks and lee-side. This figure is available in colour at http://www.boreas.dk.
The exact timing and source of this flood are unknown, as the river is un-gauged and there are no observations or records of any significant flooding. Skaftafellsökull has no history of high-magnitude jökulhlaups. At present there are no ice-dammed lakes in the Skaftafellsjökull catchment, and there are no records of volcanically generated jökulhlaups draining through the glacier. It is most likely that the flood was rainfall-generated, and that the ice-block obstacle marks were produced by icebergs that calved from the front of Skaftafellsjökull into the proglacial lake and were then washed downriver by the flood. Data from the Skaftafell automatic weather station, located 1.5 km from the glacier (Fig. 1B), were used to identify candidate rainstorms. Between 2003, when the channel was first surveyed, and 2009–2010, when the channel was abandoned, the most likely time for the flood is between 20th and 25th December 2006, when 326.8 mm of rain was recorded at Skaftafell, with 107.8 mm falling on 20th December alone. For comparison, over the decade 2001 to 2010, the average December rainfall was 194.5 mm, or 159.6 mm if the December 2006 values are excluded, and the December 2006 monthly rainfall of 473.5 mm was the record monthly total for the decade. This period was unseasonably warm, with the first day of rainfall having an overnight low of 6.3°C and a high of 9.0°C. Over the five-day period of heavy rain, average daily temperatures ranged from 2.1 to 7.3°C, and overnight temperatures only just fell below zero on three of the five nights (from the database of the Icelandic Meteorological Office). The average December temperature over the past decade (including December 2006) is 1.1°C, whilst the average temperature for the whole of December 2006 was 2.9°C. Of the other candidate rainfall events that occurred over the period of interest, none had a daily rainfall over 90 mm or occurred as a period of several days with as much rainfall.
Downstream of the zone where both incision and aggradation occur and can be clearly distinguished, the channel shows evidence of channel-wide incision and reworking (cross-sections 8 to 10). Aggradation occurred in this section, as evidenced by the downstream continuation of the intermediate terrace (Fig. 8). However, as incision occurred across the full width of the channel, bar reworking and aggradation are below the level of the 2003 cross-section, and therefore cannot be clearly identified on the cross-sections. Cross-section 11, which is close to the road bridge, was incised across its full width, with significant channel widening also occurring (Fig. 7), but evidence for flood aggradation was not preserved this far downstream, probably because of sediment reworking as the channel narrows to pass through the bridge.
Discussion
Over-deepened glacier beds occur in both bedrock and sediment-based glaciers (Alley et al. 2003), and are usually associated with high rates of subglacial erosion by meltwater. Close to the glacier margin, sediment deposition and moraine formation coupled with the overriding of existing sediment mean that the down-glacier slope of many glacier terminal over-deepenings is usually composed of sediment (Alley et al. 2003). Subglacial glaciohydraulic supercooling can cause the down-glacier slope of an over-deepening to steepen by encouraging sedimentation (Alley et al. 2003). Subglacial supercooling has been observed at Skaftafellsjökull (Roberts et al. 2002; Tweed et al. 2005). Where subglacial over-deepenings are entirely formed of bedrock, a lake will form as the glacier retreats, which may persist even when the glacier retreats past the over-deepening. Where a subglacial over-deepening is wholly sediment-based, or is composed of sediment on the down-glacier side, glacier retreat can trigger either lake formation or incision and terrace formation.
At Skaftafellsjökull, retreat of the glacier from the 1998 moraine limit was associated with incision of the proglacial river channel through the moraine, accompanied by terrace formation from 1999 to 2003. Each phase of incision is essentially annual, occurring as the proglacial drainage system reorganizes itself each spring (Marren 2002). The full record of channel changes during this period indicates that the glacier has been consistently retreating back from an ice-contact slope (the down-glacier side of an over-deepening), resulting in a close correlation between the overall retreat of the glacier and the amount of terrace incision. During this time, the gradient of the ice-contact slope reduced from 0.044 to 0.018, which is consistent with studies that have investigated proglacial changes over longer time scales using lichonometric dating of terraces, which found that rates of incision decrease as retreat progresses. This indicates that proglacial channels become less sensitive to retreat as the glacier moves away from the steepest part of the ice-contact slope (Thompson & Jones 1986; Roussel et al. 2008), but this study has shown that reduced rates of incision are caused by a combination of the reduction in the gradient of the ice-contact slope and increased sediment yield during periods of faster retreat.
As the proglacial lake grew from 2004 to 2010, no channel incision occurred. However, the evidence suggests that once the proglacial drainage reorganization took place in late 2010, the proglacial channel was still connected to the ice-contact slope, and the same relationship between glacier retreat and lowering of the proglacial channel held. It is likely that in the future, continued retreat will cause further incision and terrace formation along the newly enlarged eastern drainage outlet. It remains to be seen whether the relationship between retreat and incision will continue to hold, or if the lowest parts of the subglacial over-deepening are now being exposed, so that the incision rate can begin to slow.
A remaining question is why did terrace formation cease in 2003 and lake development take over as the dominant form of proglacial change from then on? The answer is unrelated to the rate of retreat, as this remained more or less constant from 1999 to 2010 at ∼53 m per year. It is likely that there is a limit to how deeply a coarse-grained moraine can be incised before the channel bed becomes too armoured for further incision to take place, thus stabilizing the lake (Clague & Evans 2000). At Skaftafellsjökull, lake development did not commence until the reorganization of the proglacial channel from in front of to behind the 1998 moraine in 2001. Lake development then began, with the lake fully developed and stable by 2003. By the time the lake was stable, the glacier had retreated down the ice-contact slope sufficiently for the gradient to have flattened enough to reduce the potential for terrace incision to occur (see discussion in the previous paragraph). Proglacial moraine-dammed lake development is therefore an outcome of all possible sources of proglacial drainage reorganization having been exhausted. Certainly, around Vatnajökull proglacial lake development is now widespread (Schomacker 2010).
As part of the experimental design it was originally hypothesized that sandur incision near the glacier would progressively extend further downstream as each lower terrace surface was formed. It was also thought that downstream of the point of furthest incision there would be an inflection point, where material eroded upstream would begin to be deposited, in a form of ‘complex response’. Some evidence of this was observed between 2003 and 2004, although this was not quantified by resurveying the channel in 2004. After 2004, a progressive change was not observed because of the development of the stable proglacial lake and the armouring of its outlet during 2003–2004.
However, downstream of the lake outlet as it was configured in 2003 there was up to 1.89 m of incision before the channel was abandoned in 2010. This incision occurred along the full length of the channel, at least as far as the bridge that marked the end of the surveyed reach. Some of this incision may have occurred between 2003 and 2004, but not all of it. There is also some evidence of an inflection point marking a transition from incising to aggrading behaviour in the channel. Where the proglacial river was forced into a single channel through the moraine belt formed in the 1950s, the channel was incised deeply, with no evidence of new deposition (cross-sections 1 to 4). Once the river exited the moraine constriction, there is evidence for both new deposition and incision (cross-sections 5 to 7). However, rather than representing progressive incision and downstream aggradation, it is likely that after 2004 all of the observed changes are a result of the 2006 flood, with the bar deposited in the braided reach during the flood accounting for the aggradation above the 2003 cross-sections and the new terrace. Between the 2003 and the 2008 aerial photographs, the single channel had extended ∼500 m further downstream (Fig. 1B). It is likely that this happened because the newly deposited flood bar acted to partly confine the channel following the flood. Downstream of cross-section 7, a braided channel continued to exist during 2007 to 2009, with sediment reworking occurring until the channel was abandoned in 2010. These observations indicate that, despite the cessation of terrace formation in 2004, the occurrence of the flood in 2006 forced the channel to continue behaving as a coupled incising–aggrading system until it was abandoned three years later.
These observations are compatible with the work of Germanoski & Harvey (1993), in that upstream degradation causes aggradation in the downstream reach. Furthermore, there is some evidence of an inflection point between incision and aggradation, and of the downstream migration of the inflection point. What is striking, though, is that, apart from possibly during the period 2003 to 2004, the downstream migration of the inflection point is not caused by glacier retreat, as supposed by Marren (2005). Instead, aggradation and terrace formation during the 2006 flood seem to have forced the extension of the confined single channel in a downstream direction. This is largely because flood deposition is topographically higher and coarser than non-flood deposition, forcing the post-flood channel to abandon much of the older terrace surface (Marren et al. 2002). At Skaftafellsjökull, the downstream extent of flood terraces would be limited by the upstream supply of sediment from the incising reach, and by the rapid expansion of the unconfined reach, causing flows to quickly shallow in a downstream direction, leading to downstream thinning of the resulting flood deposit.
Marren (2005) considers the likely stratigraphic response of various magnitude and frequency regimes under conditions of either glacier advance or glacier retreat. The likely outcomes are summarized in Marren (2005: fig. 18). In situations where sandur incision is occurring, owing to glacier retreat, the primary landscape response is terrace formation. Where high-magnitude, low-frequency flows dominate the landscape, high relief flood terraces will dominate the landscape, but with the potential for between-flood reworking and the deposition of low-magnitude, high-frequency ‘normal’ deposits as well. The proglacial zone of Skaftefellsjökull downstream of the lake outlet provides an illustration of these circumstances. In particular, the progressive changes that occurred before and after the flood provide greater insights into the reworking and adjustments of the proglacial landscape as it is progressively disconnected from the glacier.
Conclusions
This study describes changes to the proglacial landscape of Skaftafellsjökull from 1999 to 2011, a period of sustained glacier retreat. In the immediate proximal zone, terrace formation occurred from 1999 to 2003, coupled with minor readjustments of the proglacial drainage system that prevented the exit point of the river from the glacier becoming armoured and stable. From 2004 to 2010 a more stable configuration existed, and proximal terrace formation ceased. Instead, continued glacier retreat was accompanied by proglacial lake expansion. Once the lake extended across the full width of the glacier in 2010, a lower secondary outlet channel was able to be occupied, and the entire drainage network was reconfigured again. Throughout the period 1999 to 2011, the glacier has been retreating back from an ice-contact slope which has a gradually decreasing gradient. Because of this there has been a constant positive relationship between the cumulative rate of glacier retreat and the cumulative rate of sandur incision. However, on an annual basis there is a moderate negative correlation between the rate of retreat and the rate of incision. This is partly because the amount of incision possible reduces as the glacier retreats down the increasingly shallow slope. Furthermore, a faster retreat makes a greater quantity of sediment available, which counteracts the effect of retreat down an ice-contact slope. Terrace formation is always a rapid event occurring with the annual re-establishment of the drainage network each spring.
Downstream of what became the lake outlet in 2003–2004, the expected channel changes did not occur because of the development of the lake and the cessation of terrace formation. Instead, a similar set of changes were triggered by a significant flood, which is thought to have occurred in late 2006 (based on rainfall records). These observations indicate that proglacial channels have a tendency to behave as coupled incising–aggrading channels along their length for a variety of reasons, including upstream incision and upstream sediment starvation because of sediment trapping in lakes. In the case of Skaftafellsjökull, significant downstream aggradation was confined to the flood terrace. Incision during the flood occurred because of the presence of a moraine constriction, with aggradation occurring downstream of the constriction. Following the flood, incision and the conversion of the channel from a braided to a single channel for an additional 500 m downstream occurred because of the confining effects of the flood terrace.
This paper provides an illustration of the types of sandur geomorphology and sedimentology that occur where low-frequency flooding interacts with a retreating glacier. The insights provided by the retreat of Skaftafellsjökull over the past decade have relevance for understanding the potential for channel change and sediment redistribution in front of retreating glaciers today, and also for unravelling the history of former landscapes that have undergone deglaciation.
Acknowledgements
Dr. Matthew Roberts of the Icelandic Meteorological Office is thanked for providing meteorological data and photographs. Oddur Sigurðsson provided information on glacier fluctuations from the database of the Icelandic Glaciological Society. Dr. Russell Drysdale is thanked for the 2009 photograph. Fieldwork in 2003 was supported by a grant from Earthwatch, awarded to Dr. Andrew J. Russell, and Earthwatch volunteers from 2003 are thanked for their assistance with the data collection. Richard Marren and Louisa Davis are thanked for assistance with fieldwork in 2011, and Mandy Winstanley for assistance in 2004. We thank B. Johnson and an anonymous reviewer, whose comments helped to improve this paper.
