Using repeat-pass satellite feature tracking, the dynamic behaviour of Jakobshavn Isbræ, which drains around 7% of the Greenland ice sheet, is investigated in improved detail through its recent period of thinning, acceleration and retreat. The departure in 1995 from the normal seasonal invariance in velocity near the grounding line suggests a possible change in dynamics well before the subsequent thinning phase. The initiation of the acceleration and retreat is pinpointed to spring 1998 and may have been prompted by the same mechanism that gave rise to the spring 1995 flow increase.
 Jakobshavn Isbræ, Greenland's most active outlet glacier, has received much attention because it drains a significant part of the ice sheet and has exhibited a consistently high and seasonally invariant rate of flow [Echelmeyer and Harrison, 1990]. Its recent thinning, acceleration and retreat has focussed even more attention on the response of the ice sheet to climate warming [Thomas et al., 2003; Joughin et al., 2004]. In this paper we use feature tracking between repeat-pass satellite SAR images to investigate in more detail the dynamics of this outlet glacier through the recent period of thinning and acceleration.
2. Summary of Recent Observations
 The recent nature and behaviour of Jakobshavn Isbræ has hitherto been investigated through borehole measurements [e.g., Iken et al., 1993; Lüthi et al., 2002], spaceborne observations of frontal position [e.g., Sohn et al., 1998], airborne surveys of surface elevation [e.g. Thomas et al., 2003] and remote sensing analyses of surface velocity [e.g., Joughin et al., 2004]. From the 1960's to the 1990's the frontal position showed regular, seasonal fluctuations suggesting that calving flux is controlled either by the restraining influence of winter fjord-ice or by the availability of surface summer meltwater [Sohn et al., 1998]. Sporadic thickening in the lower part of the ice stream was detected between 1991 and 1997, despite locally warm summers. This was followed by a period of thinning which continued for at least the next three years and which must have been dynamic in nature [Thomas et al., 2003]. More recently, the surface velocity of the ice stream was shown to have increased dramatically between 1997 and 2000 and to have continued to escalate into 2003, accompanied by the break-up of the floating tongue [Joughin et al., 2004]. The potentially complex interactions between thinning, retreat and acceleration are interesting in that they may signify a more general response of such outlets to climate warming [Thomas, 2004; Zwally et al., 2002].
 The crevassing associated with the margins of Jakobshavn Isbræ is extensive for a long way upstream [Echelmeyer et al., 1991; Herzfeld and Mayer, 2003]. This study was aimed at exploring the recent surface velocity variations of the ice stream by exploiting the archive of ERS SAR imagery which is both sensitive to such surface features and more frequently available than optical data.
3. Methods and Data
 The use of correlation-based feature-tracking between repeat-pass airborne or spaceborne images is well established both for optical imagery [e.g., Bindschadler and Scambos, 1991; Scambos et al., 1992] and Synthetic Aperture Radar (SAR) imagery [e.g., Lucchitta et al., 1995; Strozzi et al., 2002; Rignot et al., 2004]. Whilst speckle-based tracking and coherence-based tracking may be used to derive high resolution velocity fields from pairs of SAR images possessing phase coherence [Joughin, 2002; Strozzi et al., 2002], tracking of detectable features in SAR backscatter intensity images, such as crevasses, may also yield useful surface flow measurements [Lucchitta et al., 1995; Luckman et al., 2003; H. Pritchard et al., Glacier surge dynamics of Sortebræ, East Greenland, from synthetic aperture radar feature tracking, submitted to Journal of Geophysical Research, 2005]. This technique is more limited in spatial resolution and coverage but allows pairs of images separated by a full satellite orbital cycle to be used where such surface features exist, thereby increasing the periods available for velocity measurement.
 In this study, a large proportion of the available archived 35-day repeat-track pairs of ESA ERS-SAR images of Jakobshavn Isbræ were acquired and used to derive surface velocity fields during the 1990s. These include a single image pair from 1992, nine consecutive images spanning the summer and early winter of 1995 to 1996, and between one and three image pairs for each of the years 1996 to 2000, all from one of two satellite tracks.
 The feature-tracking technique adopted here follows the normal procedure of deriving the field of 2D offsets between each pair of images and removing the flow-independent image-to-image mapping to leave only the displacements due to ice flow. Offsets are determined by finding the peak position of the intensity correlation field between regularly spaced image patches, each covering around 1 km by 1 km of the scene. The flow-independent part is determined in a similar way but by considering only exposed rock features within the images to define a second-order 2D mapping between them. Since rock outcrops occur only beside the final 15 km of Jakobshavn Isbræ, the error in this correction is expected to increase further upstream. Correlation signal-to-noise ratios are used to reject poor matches, resulting in a sequence of patchy grids of 2D velocity information where surface features are moving coherently. These grids are geocoded and terrain-corrected to a UTM projection using a DEM derived from digital map contours.
 Where there are local rock references (i.e. on the floating tongue and as far as a few km upstream of the grounding line) errors in this technique are expected to be around 0.1 pixels over 35 days, or better than 0.1 md−1 [Strozzi et al., 2002]. Further upstream, errors cannot be estimated accurately because there is no way to determine how the flow-independent image-to-image mapping deteriorates away from rock references. However, correspondences in velocity minima between the sequences of velocity maps suggest it is no worse than 0.5 md−1.
4. Observations of Velocity and Frontal Position
 In all, 24 maps of velocity spanning the 1990s were derived, a selected sample of which are presented in Figure 1. Velocity evolution throughout the sequence was extracted at two locations, one just downstream (labelled A) and one about 30 km upstream (labelled B) of the grounding line, chosen because the signal-to-noise ratio was sufficient here to determine the velocity throughout the sequence (Figures 2a and 2b). In addition, the sequence of mid-stream positions of the calving front was measured from the backscatter intensity images alone (Figure 2c). We note that throughout the sequence of images at least the first 20 km of the fjord appeared to be packed with ice. There was no evidence that this was not free to move but there was also no evidence of open water patches within it.
 In 1995, where images were available from every satellite cycle, a seasonal pattern of velocity variation is apparent. The velocity just downstream of the grounding line increases from 16 to 18 md−1 from May through to July, just as meltwater is expected to be available, and then slowly decreases as winter commences (Figure 2a). The magnitude of this increase cannot be explained by any tidal influence on the vertical position of the ice. This seasonal pattern is echoed at the second measurement point 30 km upstream (which is not as far upstream as the borehole sites of previous studies [Iken et al., 1993; Lüthi et al., 2002]) but lower velocities and larger errors make the observations here less conclusive (Figure 2b).
 Between March and April 1997, there is a modest reduction in the velocity just downstream of the grounding line. This is followed, between April and May 1998, by a small increase in velocity, accompanied by a significant early retreat of the front to the normal minimum position of late-summer, and the calving of large (0.5–1.0 km2) tabular icebergs (Figure 1b). Between May 1998 and the next observation in August, there is a further large retreat to an unprecedented position, but not beyond the ‘ice rumple’ (Figure 1c), thought to be a possible pinning point [cf. Joughin et al., 2004]. Note that this retreat and acceleration coincides with the first observation of thinning of Jakobshavn Isbræ between May 1997 and June 1998 [Thomas et al., 2003]. Between August and September 1998, the velocity of the floating tongue increased dramatically to nearly 20 md−1 while at the upstream measurement point there is some evidence of a delayed acceleration. These new data demonstrate more precisely than previously published the first period of significant acceleration in Jakobshavn Isbræ [Joughin et al., 2004]. Subsequent velocity measurements are limited to one in 1999 and a sequence of three in 2000 (e.g., Figure 1d). The pattern is one of gradually increasing velocities, which agree with the observations of Joughin et al. , and frontal fluctuations between the new minimum of 1998 and the previously established summer minimum.
5. Discussion and Conclusion
 In contrast to previous observations, the data presented here show a seasonal variation in surface velocity for Jakobshavn Isbræ during 1995. This might be explained by a reduction in back pressure in the fjord resulting from warmer temperatures and reduced sea-ice, but there is little support in the observed fluctuation in frontal position which, during 1995, seems to follow the established 1962–1996 pattern [Sohn et al., 1998]. Alternatively, surface meltwater, during this anomalously high melt year [Abdalati and Steffen, 2001], may have contributed towards higher velocities, possibly through enhanced basal motion. If so, this may indicate a change of dynamics for Jakobshavn Isbræ three years before any significant acceleration or frontal retreat.
 Aside from this 1995 anomaly, the first major increase in velocity from the long-established norm was observed between March and April 1998, also a high melt year [Abdalati and Steffen, 2001; Steffen and Box, 2001]. The first departure from the normal pattern of frontal position was detected only a month later [cf. Sohn et al., 1998]. It seems likely, therefore, that the modest thinning measured by Thomas et al.  between May 1997 and June 1998 occurred mostly between April and June 1998. This is not unreasonable considering the normal discharge rates and the surface area of fast flowing ice [Echelmeyer et al., 1991]. If this is the case then the perturbation of the ice stream coincided with the onset of spring 1998, is consistent with the idea of a large calving event [Thomas, 2004] and may in part have been triggered by the same changes that gave rise to the seasonal velocity fluctuation in 1995.
 We are very grateful to ESA for provision of SAR data and to A. Shepherd and D. Wingham (Scott Polar Research Institute) for helping to make this available through the VECTRA project. We thank the Geological Survey of Denmark and Greenland for provision of the digital map data. We especially thank the comments of the two reviewers who played a significant part in improving this paper.