Recent changes in the dynamics of Greenland's marine terminating outlet glaciers indicate a rapid and complex response to external forcing. Despite observed ice front retreat and recent geophysical evidence for accelerated mass loss along Greenland's northwestern margin, it is unclear whether west Greenland glaciers have undergone the synchronous speed-up and subsequent slow-down as observed in southeastern glaciers earlier in the decade. To investigate changes in west Greenland outlet glacier dynamics and the potential controls behind their behavior, we derive time series of front position, surface elevation, and surface slope for 59 marine terminating outlet glaciers and surface speeds for select glaciers in west Greenland from 2000 to 2009. Using these data, we look for relationships between retreat, thinning, acceleration, and geometric parameters to determine the first-order controls on glacier behavior. Our data indicate that changes in front positions and surface elevations were asynchronous on annual time scales, though nearly all glaciers retreated and thinned over the decade. We found no direct relationship between retreat, acceleration, and external forcing applicable to the entire region. In regard to geometry, we found that, following retreat, (1) glaciers with grounded termini experienced more pronounced changes in dynamics than those with floating termini and (2) thinning rates declined more quickly for glaciers with steeper slopes. Overall, glacier geometry should influence outlet glacier dynamics via stress redistribution following perturbations at the front, but our data indicate that the relative importance of geometry as a control of glacier behavior is highly variable throughout west Greenland.
 Recent studies have reported widespread retreat, thinning, and acceleration followed by readvance, thickening, and deceleration of Greenland's marine terminating outlet glaciers within the past decade [e.g., Joughin et al., 2004; Howat et al., 2005, 2007, 2008; Luckman et al., 2006; Rignot and Kanagaratnam, 2006; Moon and Joughin, 2008]. Many studies have focused on large outlet glaciers in southeast Greenland, and Jakobshavn Isbræ (69°10′59.33″N) in the West, because glaciers in these regions together represent nearly half of the total ice discharge from Greenland [e.g., Rignot et al., 2004; Rignot and Kanagaratnam, 2006]. Analysis of changes in ice front positions of Greenland's outlet glaciers and peripheral surface elevations of the Greenland Ice Sheet have been reported by Moon and Joughin  and Sole et al. , respectively. With the exception of these studies, little work has been done to create a spatially and temporally comprehensive assessment of glacier front position and surface elevation change in west Greenland despite observed acceleration of several outlet glaciers between 2000 and 2005/6 [Joughin et al., 2010] and geophysical data showing accelerated mass loss from the western region since 2005 [Khan et al., 2010].
 Observed changes in outlet glacier behavior often can be explained in part by stress perturbations at the glacier front. Driving stress must be opposed by resistance generated by basal and lateral shear stress (i.e., basal and lateral drag) as well as gradients in longitudinal stress [Nick et al., 2009]. The speed of land terminating glaciers is determined primarily by basal drag, so that thinning leads to slowing as the driving stress and basal drag decrease by a proportional amount. In contrast, fast moving, marine terminating outlet glaciers generate a substantially greater resistance to flow from shearing along fjord walls, so that thinning and retreat of the ice front can lead to reduced resistance and accelerated flow. Thinning near the front may bring the glacier close to flotation, causing the breakup and retreat of the floating tongue [Howat et al., 2005, 2007, 2008; Pfeffer, 2007; Joughin et al., 2008] and basal penetration of ocean water to the grounding zone, reducing drag near the front. Decreased resistive stress near the front also causes the glacier surface to steepen at the grounding zone, increasing the driving stress and further increasing flow speed and thinning [Schoof, 2007].
 Ice front disintegration becomes more likely as the glacier approaches flotation due to thinning/stretching (i.e., dynamic thinning) from steepening longitudinal stress gradients caused by front acceleration. Once the glacier reaches flotation, full-thickness calving of large tabular icebergs may occur [Amundson et al., 2010], providing a mechanism for rapid ice front disintegration. Therefore, glaciers with shallow surface slopes inland of the front should be particularly sensitive to dynamic thinning because thinning will bring a greater extent of inland ice close to flotation, potentially triggering ice front and grounding zone retreat, causing further acceleration. Thinning propagates inland as a kinematic wave, so that the rate of diffusion of thinning is determined by the magnitude of the perturbation, the along-flow velocity gradient, the speed of the kinematic wave, and diffusive dampening [Nye, 1960; Price et al., 2001; Hooke, 2005]. The speed of the kinematic wave and the extent of longitudinal diffusion of the wave are inversely proportional to the initial diffusivity and because this term increases with the surface slope, thinning is concentrated near the front for glaciers with steeper slopes. Therefore, thinning of glaciers with steep slopes inland of the front should be concentrated within the lower outlet, making them prone to rapid retreat of ice near flotation without extensive inland migration of the grounding zone.
 The width of an outlet glacier may affect glacier retreat due to the inverse relationship between width and lateral drag at the glacier centerline. Lateral stresses may be distributed from the shear margin to the centerline over distances as large as 30 times the ice thickness if basal shear stress is low [Raymond, 1996], as it may be for some Greenland outlet glaciers. As glaciers approach flotation and basal stress near the terminus is reduced, the relative importance of lateral and longitudinal stresses increases [Echelmeyer et al., 1994]. Fjord width and geometry may also affect the availability of pinning points, controlling the location of the grounding zone during retreat and advance [Warren and Glasser, 1992].
 To better understand the dynamic behavior of marine terminating outlet glaciers, we examine changes in surface elevation and front position for 59 large (>1.5 km wide) marine terminating outlet glaciers along the west Greenland coast from ∼76°37′N to ∼61°35′N at seasonal to interannual temporal resolution between the years 2000 and 2009 (Figures 1a and 1b). We also examine time series of surface speed for eight glaciers that exhibited the largest changes in surface elevations and front positions during the study period. We assess these data to determine whether changes in the glaciers have been synchronous on interannual timescales, as was observed in the southeast [Howat et al., 2008], and if changes in glacier dynamics coincide with accelerated mass loss in northwest Greenland since 2003 as observed by GRACE (Gravity Recovery and Climate Experiment) [Khan et al., 2010]. We also investigate potential relationships between front retreat, thinning, speed, and glacier geometry to determine the relative importance of glacier geometry on observed outlet glacier behavior.
 We assembled data from several different sources to compile our time series. Imagery from the visible to near infrared (VNIR) wavelengths (bands 1–3) of the Advanced Spaceborne Thermal Emissivity and Reflection Radiometer (ASTER) were delivered by the Land Processes Distributed Active Archive Center (LP DAAC, https://lpdaac.usgs.gov/). The United States Geological Survey (USGS) Global Visualization Viewer (GLOVIS, http://glovis.usgs.gov/) provided Landsat 7 Enhanced Thematic Mapper Plus (ETM+) images. Ortho-images from the SPOT-5 Stereoscopic Survey of Polar Ice: Reference Images and Topographies (SPIRIT) were provided by the French Space Centre during the International Polar Year (June 2007 to June 2009) [Korona et al., 2009]. Radar mosaics for winter 2000/2001, 2005/2006, and 2006/2007 were derived from data acquired by the Canadian Space Agency's Radar Satellite (RADARSAT) [Moon and Joughin, 2008]. Finally, high-accuracy surface elevation profiles derived from laser-altimeter surveys with NASA's Airborne Topographic Mapper (ATM) were used to supplement gaps in time series when available. For the purpose of this study, our observational record is limited to the years 2000–2009, when Landsat 7 ETM+ and ASTER data are available, in order to create time series of changes in glacier front position, surface slope, and surface elevation with seasonal to interannual temporal resolution.
 Surface elevation profiles were primarily extracted from ASTER DEMs, with SPOT-5 SPIRIT DEMs and NASA ATM data used to supplement the later end of the time series. The ASTER DEMs were used as the primary source of surface elevation data because they provide relatively high temporal coverage and accurate (±5 m) elevations over the several-kilometer scales on low-relief, ice-covered surfaces. SPIRIT DEMs were used to supplement data gaps in 2007 and 2008 because of their vertical accuracy (±10 m) and frequent repeat imaging cycle (1–26 days) [Korona et al., 2009]. NASA ATM surface elevation profiles were used where available because the conically scanning laser altimeter provides very accurate (±20 cm) measures of surface elevation over a narrow swath (∼200 m) [Krabill et al., 1995]. ATM temporal coverage was limited to 2009 for most glaciers included in the study, when data collection occurred throughout the month of April, and we therefore expect a difference between this product and the other DEMs on the order of 1–2 m due to seasonal variability. Locations of glaciers included in the study are shown in Figure 1b.
2.1. Front Positions
 Summer front positions were mapped using orthorectified Landsat 7 ETM+ images and ASTER VNIR images from 2000 to 2009 and SPIRIT SPOT 5 orthomosaic in 2007 and 2008. RADARSAT amplitude image mosaics [Moon and Joughin, 2008] were used to map winter front positions in 2000/2001, 2005/2006, and 2007/2008. To assess the relative positioning error between images, >20 off-ice control points were selected in several sequential images, giving average errors of ±74 m, ±87 m, and ±25 m for Landsat, RADARSAT, and SPIRIT images respectively relative to ASTER images. These offsets were primarily due to topographic distortion and, therefore, could not be corrected with image coregistration due to their varying spatial distribution throughout each image. Due to this error, as well the effect of seasonality on front position, we only consider variations in front position greater than ±100 m to be significant in our analysis.
 For consistency with recent studies of ice front changes in Greenland [e.g., Howat et al., 2008, 2010; Moon and Joughin, 2008], ice front positions were mapped using the same method as these previous studies. The mean change in front position is determined by differencing polygons formed by straight, parallel lines on each lateral margin of the glacier, a straight line oriented across-flow some arbitrary distance up glacier, and a vector tracing the ice front on the down-glacier side. The front vector is retraced with each sequential image, so that the difference in area between each polygon gives the change in terminus area, which is then divided by the width of the polygon to obtain the mean front position change. Front positions were mapped in this way using all available imagery for each glacier to determine seasonality in front position as well as interannual changes. For each glacier, the annual front position is reported as the midsummer mean front position, typically recorded throughout July, in order to minimize the possible effects of seasonality in the interpretation. Although the timing of the annual front position minima may differ spatially (i.e., between glaciers) and temporally (i.e., between years) based on the onset of the melt season, breakup and movement of sea ice, and other variables, we recorded the annual front position on approximately the same day of year (DOY) for all glaciers over all years, resulting in an unknown magnitude of error in our analysis. Small differences in day of year (DOY) between years due to satellite repeat paths result in additional error of unknown magnitude. We assume that the cumulative front position errors in this study using these methods are on the order of 100 m on annual time scales.
2.2. Surface Elevations and Surface Slopes
 Surface elevations were extracted from ASTER and SPIRIT DEMs either along a central flow line visually identified from flow features in ASTER VNIR images or along the flight path of the NASA ATM data where available. Although flight paths may not follow central flow lines, comparisons of elevation profiles along flow lines and those obtained with NASA ATM flight paths indicate that the differences in the data do not substantially affect our results. Errors caused by cloud coverage along the flow lines were examined on ASTER VNIR images and removed manually, creating data gaps in many elevation profiles. Combining multiple profiles from a single season through averaging minimized the affects of data gaps.
 Changes in surface elevation presented in this paper were measured within 5 km inland from the location of the 2009 grounding zone using the averaged annual elevation profile. The location of the grounding zone was approximated for each glacier based on surface elevation profiles and surface slope. The location of the grounding zone was inferred to coincide with an abrupt increase in surface slope for glaciers terminating as floating ice tongues. Glaciers were assumed to terminate as floating ice tongues if surface elevations were <50 m a.s.l. and surface slopes were approximately zero (horizontal) near the front. The location of the grounding zone coincided with the most-retreated front position for glaciers with grounded fronts. Fronts were considered grounded if surface elevations were >50 m a.s.l. near the front and/or the slope was uniform from the front to ∼10 km inland. These semiquantitative criteria used to distinguish floating ice tongues from grounded termini are imprecise and may incorrectly classify lightly grounded termini as floating ice tongues. Using these criteria, ∼15% of the 59 glaciers we examined had substantial floating ice tongues (>2 km) prior to observed retreat.
 Surface slopes were determined for two different regions of the glacier (near-front and across the entire ablation zone) to encompass variability likely due to changes in ice speed (i.e., dynamic thinning). Changes in slope near the front (hereafter 5 km slopes) were calculated from the front to 5 km up glacier in order to monitor loss of nearly zero-slope floating ice, which is often observed as slope steepening as the region included in the slope calculation shifts inland (Figure 2). Changes in slope in the ablation zone (hereafter snow line slopes) were measured from the inland-most grounding zone to the average summer snow line. The location of the snow line was manually approximated using cloud-free, late-melt season Landsat imagery. This method provides a rough estimate of snow line location due to the sporadic timing of cloud-free satellite imagery and interannual snow line variability, however, for the purpose of this study, our method provides an approximate snow line location used as the demarcation for the inland extent of the region used in snow line slope calculations, which is spatially fixed throughout the time series. Figure 2 illustrates the differences between the regions used when calculating 5 km and snow line slopes for a hypothetical, retreating glacier.
2.3. Surface Speed
 Surface velocities were determined using multiple-image/multiple-chip (MIMC) repeat image feature tracking (RIFT) software [Ahn and Howat, 2009], which is modified from traditional feature tracking methods to provide more complete temporal coverage. MIMC feature tracking was performed using pairs of Landsat 7 ETM+ and ASTER VNIR imagery with acquisition date separation of 10–90 days. The RIFT algorithms use cross correlation of spatial variations in image intensity, such as crevasses, to match images. Cross correlation produces a pixel displacement field, which is processed further to produce a velocity field. Image-pair offsets were minimized by coregistration of stationary control points within both images. Each time series of ice speed was obtained along the centerline or flight path used for surface elevation profiles at a location 3–5 km from the most retreated front position. Availability of image pairs for speed measurement varied depending on the extent of overlapping flight paths near the front and cloudiness. In general, the quantity and quality of image pairs decreased with glacier latitude. Our speed record is therefore currently limited to eight glaciers with large front retreats to demonstrate variability in response to large perturbations at the front.
2.4. Climate Data
 Climate forcing conditions throughout our observational record were examined using air and sea surface temperatures along the west coast of Greenland. Locations for the various climate data are indicated in Figure 1b. Meteorological data was obtained through Goddard Institute for Space Studies (GISS). Monthly and annual average air temperatures from Nuuk/Godthab (64.2°N, 51.8°W) and Aasiaat/Egedesminde (68.7°N, 52.8°W) were analyzed from 2000 to 2009 for comparison with observed glacier behavior.
 Sea surface temperatures (SSTs) were acquired from the Physical Oceanography Distributed Active Archive Center (http://podaac.jpl.nasa.gov/), providing monthly average SSTs during sea ice-free portions of the year (April–October) from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard NASA's Terra and Aqua satellites. Both satellites image the entire Earth surface every 1–2 days at relatively high spatial resolution (4.63 km × 9.26 km) creating a comprehensive record of global SSTs. We use SSTs as a basic measure of changes in ocean temperature forcing throughout the study period although they may not reflect subsurface changes in ocean temperature.
 Changes in front positions for the glaciers included in this study are summarized in Figure 3. Due to the complex nature and large quantity of surface elevation profiles, these data are only presented for a few select glaciers in Figure 4. These glaciers encompass the large range of variability in surface elevation profiles throughout the data set, including different slope configurations (shallow, moderate, steep) and magnitudes of thinning. Interannual front position and surface elevation change relative to the first year with available data for each glacier are presented in Table S1 of the auxiliary material. The data presented in Figures 3 and 4 and Table S1 indicate changes in front positions and surface elevations were asynchronous on annual time scales, and highly variable over the 10 year period from 2000 to 2009 for west Greenland outlet glaciers.
3.1. Changes in Front Position
 Changes in front positions indicate that of the 59 marine terminating outlet glaciers in the study, 49 retreated and 10 maintained stationary front positions (Figure 3), with large variability in the timing and magnitude of retreat (Table S1). Fronts were considered stationary if their position varied by no more than ±0.1 km yr−1 to account for image registration error and seasonal variability (see section 2.1). The majority (39) of the glaciers retreated ≤1.0 km while 20 glaciers experienced front retreats >1.0 km, with a maximum retreat of 12.8 km by Jakobshavn Isbræ (see Figure 1b for location).
3.2. Changes in Surface Elevation
Figure 4 shows the surface elevation profiles for a sample of 6 of the 59 glaciers included in this study (see Table S1 in the auxiliary material for surface elevation change data for all glaciers within the data set). Changes in surface elevation varied from ∼115 m of thinning to ∼20 m of thickening from 2000 to 2009. Nearly all glaciers thinned during a portion of the study period, with the exception of Nansen (75°43′36.76″N), Tuvssaq (73°13′27.42″N), Rink (71°44′8.00″N), and Eqip (69°47′2.96″N) glaciers, though interannual variability in the timing and magnitude of elevation change was large. No obvious latitudinal trends in the magnitude or extent of elevation change were evident.
 Several glaciers (Edvard (76°18′24.26″N), Sverdrups (75°34′18.75″N), Steenstrup (75°16′15.37″N), Upernavik North (73°00′02.33″N), and Umiamako (71°44′03.32″N)) followed similar, though variable, thinning patterns. This pattern began with a period of relatively little surface elevation change followed by a brief period of rapid thinning of up to ∼100 m yr−1. These rapid thinning periods were then followed by several years of relatively constant surface elevations. The period of rapid thinning was not synchronous between glaciers: Steenstrup and Sverdrups thinned from 2002 to 2005, Edvard from 2003 to 2007, Upernavik North from 2005 to 2006, and Umiamako from 2007 to 2008. Periods of rapid thinning occurred during two consecutive summers for Upernavik North and Umiamako glaciers. Thinning rates could not be calculated for consecutive summers for Edvard, Sverdrups, and Steenstrup due to data gaps within the time series for these glaciers. These five glaciers have no distinct shared geometric characteristics (Table S2) that could explain the periods of rapid thinning relative to the more gradual thinning of the other glaciers in the data set. Their front widths range from ∼3–12 km and their preretreat snow line slopes range from 0.016–0.041.
3.3. Changes in Surface Speed
 Glaciers with periods of rapid thinning had coincident speed-ups, including ∼20% increase in speed for Upernavik North Glacier and Steenstrup Glacier and ∼50% increase in speed for Sverdrups Glacier [Joughin et al., 2010]. Additionally, speed-ups occurred on all glaciers with large retreats for which we had speed time series from 2000 to 2009, although we cannot determine if these speed-ups were ubiquitous throughout the study region due to data gaps. We therefore focused our speed analysis on Alison Glacier (74°37′21.31″N) and Umiamako Glacier, which had relatively complete time series for all observed parameters. The speed time series for these glaciers reveal a direct relationship between retreat rate and acceleration. The retreat rate of Alison glacier was most rapid from 2000 to 2005 corresponding to an 80% increase in speed. The front continued to retreat from 2006 to 2009 while the speed stabilized (Figure 5a). Thus, response to perturbation at the front had two phases for Alison Glacier, abrupt thinning and acceleration, followed by a slow stabilization of front position (Figure 5b), speed, and surface elevation (Figure 5c). Umiamako Glacier displays a similar relationship between retreat rate, acceleration, and surface elevation. Following retreat, Umiamako accelerated by 150% with no sign of stabilization at the end of the record (Figure 6a). Front retreat and surface elevation profiles for Umiamako are shown in Figures 6b and 6c for comparison with the speed time series.
3.4. Comparison of Glacier Geometries
 Near-front glacier widths, 5 km surface slopes (preretreat and postretreat), and snow line surface slopes (preretreat and postretreat) are shown in Figures 7a–7e and included in Table S2 for comparison with Figure 3. Widths range from 1.7 km (Avangnardleq (62°12′5.26″N)) to 11.9 km (Steenstrup), however, nearly 2/3 of the glaciers included in the study are between 3 and 5 km wide. Surface slopes were highly variable, with snow line slopes ranging from 0.014 to 0.095, most likely reflecting differences in basal topography. To determine trends in surface slopes due to dynamic thinning near the front, slopes were calculated prior to and following front retreat. Glaciers with a limited time series or multiple low quality profiles were excluded from slope analysis to minimize errors in slope analysis due to poor data quality. Therefore, slopes were calculated for 48 of the 59 glaciers. Glacier geometries were highly variable and had no direct relationship with the magnitude of front retreat or thinning as demonstrated in Figure 8.
3.5. Changes in Air and Ocean Temperature
 Mean monthly air temperatures from the Aasiaat/Egedesminde and Nuuk/Godthab weather stations are shown in Figure 9. The air temperature record is analyzed as mean monthly temperatures to simplify analysis, although taking the mean obscures any single days with abnormally high or low temperatures. Mean monthly air temperatures exceed freezing from May–August every year. There are no significant trends in mean monthly air temperatures from 2000 through 2009. Mean annual air temperatures derived from the mean monthly air temperatures indicate the mean annual air temperature was below freezing at both stations for every year in the record. The difference in melt season mean monthly air temperatures between Aasiaat and Nuuk was <1.5°C throughout the record; the largest difference occurred in 2003, the year with the highest mean annual temperature at Aasiaat.
 Average SSTs calculated from melt season (April–September) ice-free MODIS data are analyzed as anomalies relative to their respective 2000–2009 mean values. These data are grouped into three geographic regions (north, central, south) for visual simplification (Figures 10a–10c). Average May SST data are also analyzed as anomalies relative to their respective 2000–2009 mean values. May SSTs have been correlated to the timing of seasonal front retreat in the Uummannaq district [Howat et al., 2010] but the correlation was not tested for marine terminating outlet glaciers in other regions. There is no single year with remarkably higher or lower average melt season or May SSTs relative to their respective decadal means. Several years contain positive and negative SST anomalies at nearby locations, indicating substantial regional and local variability in SSTs within the data set.
 The majority of marine terminating outlet glaciers in west Greenland retreated and thinned from 2000 to 2009, supporting GRACE observations of accelerated mass loss from northwest Greenland since 2003 [Khan et al., 2010]. There was, however, no distinct relationship between the magnitude of front retreat, surface elevation change, surface slope, and outlet width applicable to the entire data set. Here we discuss relationships between front retreat, thinning, and surface slopes, as well as differences between grounded and floating termini, and contrasting glacier behavior in an attempt to resolve controls of outlet glacier behavior under the observed climate forcing.
 We first examined the data set for the magnitude and timing of front retreat in relation to regional climate forcing. Approximately 2/3 of the glaciers retreated less than 1.0 km, ten of which had stationary fronts (fluctuated by less than ±0.1 km) from 2000 to 2009. In contrast, twenty glaciers retreated ≥1.0 km from 2000 to 2009, nine of which reached maximum retreat rates exceeding 1.0 km yr−1 during a portion of the record. Only Alison Glacier and Jakobshavn Isbræ sustained rapid retreat for several consecutive years. Rapid large front retreats (≥1.0 km yr−1) were asynchronous on annual time scales, occurring on different glaciers between all consecutive summers other than 2000–2001. The magnitude and timing of front retreat were not related to changes in mean monthly air temperatures measured at Aasiaat and Nuuk throughout the melt season. Front retreat, additionally, was not correlated with changes in regional melt season or May SST anomalies as previously observed in the Uummannaq region [Howat et al., 2010]. Many fronts retreated during years with negative melt season SST anomalies, such as Alison Glacier, which had its largest magnitude retreat from 2004 to 2005 when the melt season SST anomaly was ∼0.5°C below average at nearby locations in Melville Bay. Retreat was ubiquitous over the 10 year study period, however, potentially indicating a more uniform sensitivity to long-term climate variability rather than the short-term changes observed in this study.
 The magnitude and timing of thinning was also examined in relation to regional climate forcing data. Differences in mean surface elevations from the 2009 grounding zone to 5 km inland were calculated for three time periods (2000–2003, 2003–2006, 2006–2009) to determine changes in the magnitude and timing of thinning throughout the observational record. Analysis was limited to glaciers included in the slope calculations, with further limitations due to data gaps within surface elevation profiles. The magnitude of thinning increased for 30 of 40 glaciers from 2000 to 2003 to 2003–2006. These data indicate accelerated mass loss from 2003 to 2006, supporting geophysical data from Khan et al.  indicating rapid acceleration in uplift in northwest Greenland in 2005. The magnitude of thinning then decreased for 21 of 37 glaciers from 2003 to 2006 to 2006–2009. Although the magnitude of thinning decreased for the majority of the glaciers from 2003–2006 to 2006–2009, thinning magnitudes during the latter portion of the study were still greater than those during the 2000–2003 period, indicating discharge continued at an elevated rate relative to that for 2000–2003. In the absence of regional subsurface ocean temperature data, we cannot be certain that the observed changes are totally independent of regional climate forcing conditions, but, the climate forcing data set utilized in this study indicates that magnitude and timing of thinning have no clear relationship to the regional air temperature or SST variability.
 In the absence of a direct relationship between front retreat, thinning, and observed regional climate forcing, we examine the data for a relationship between retreat, thinning, and surface slope. Due to the uncertainty in seasonal variability, we focus our analysis on glaciers with large (>1.0 km) retreats. Retreats of this magnitude are also more likely to result in substantial acceleration and thinning [Howat et al., 2008]. For glaciers that retreated >1 km, the magnitude and timing of thinning varied widely, ranging from ∼15 m (Kong Oscar (76°25′26.23″N)) to ∼115 m (Upernavik North). Slopes were highly variable as well. The 5 km slope ranged from 0.006 (Yngvar Nielson (76°19′29.24″N), postretreat) to 0.108 (Perdlerfiup (70°59′28.88″N), preretreat). Snow line slopes were slightly less variable, with a range from 0.014 (Sverdrups, postretreat) to 0.095 (Leven (76°14′14.07″N), postretreat). Snow line slopes steepened for glaciers with retreating floating tongues likely due to reduced buttressing at the front and subsequent acceleration and thinning down glacier.
 Despite the variability in slope for glaciers with large retreats as mentioned above, the presence or absence of a floating ice tongue had no consistent relationship with front retreat (see section 2.2 for floating versus grounded determination). Although several glaciers with large retreats terminated as floating ice tongues (i.e., Alison, Jakobshavn, Edvard), two glaciers with floating tongues retreated <1.0 km (i.e., Yngvar Nielson and Igssuarssuit (76°2′36.37″N)). Glaciers with grounded fronts, such as Umiamako and Sverdrups, also experienced large front retreats, suggesting the presence of floating ice tongues is not a necessary criterion for large retreat. The continued acceleration of Umiamako (Figure 6a) in contrast to Alison's speed stabilization (Figure 5a) may be due to differences in the loss of grounded versus floating ice. If grounded ice is lost at the front, the glacier experiences a large stress perturbation because basal and lateral drag are reduced, causing acceleration and stretching/thinning until a new stable geometry is reached near the front. The loss of a comparable extent of floating ice results in a smaller perturbation because there is less area of contact with the fjord walls and bed [Bamber et al., 2007]. Although the rate of front retreat should be more rapid for glaciers terminating as floating ice through calving of large tabular icebergs and buoyancy-driven positive feedbacks than for grounded glaciers with smaller calving events, as described by Amundson et al. , our data indicate rapid retreat can occur for glaciers with grounded fronts such as Umiamako glacier. Overall, the loss of floating ice may trigger front instability but, our data suggest that glaciers with grounded fronts should be more sensitive to retreat of a comparable extent on short (interannual) time scales.
 To further investigate the behavior of glaciers experiencing large retreat, we examined variability in their response time by noting the time elapsed from the onset of retreat to stabilization. Here we define response time as the time between the largest magnitude front retreat (≥1 km yr−1) and the time at which the thinning rate slowed to <10 m yr−1. The response time following loss of floating ice was relatively short (<3 years) for glaciers with steep snow line slopes because thinning was concentrated at the front, causing slopes to steepen and the location of the grounding zone to stabilize. For example, the speed of Alison Glacier stabilized within 3 years of the initial front perturbation because steep snow line slopes limit inland propagation of dynamic thinning, thus concentrating thinning at the grounding zone. In contrast, response times were longer for glaciers with more shallow slopes, likely because stretching and thinning bring a further inland extent of ice closer to flotation, causing either continued retreat or reduced effective pressure at the bed and faster sliding, as is the case for Umiamako Glacier. The response time of Umiamako Glacier is also slowed by the likely larger stress perturbation at the front due to the loss of grounded ice, rather than a floating ice tongue. The response time of Umiamako, therefore, is yet to be determined due to the continued acceleration and retreat of the glacier in 2009. Limitations of our data set (i.e., data gaps, retreats in 2008/2009, etc.) prevented assessment of response times for all glaciers with maximum retreat rates ≥1 km yr−1 (limited to 6 of 9), however, our current data indicate steeper sloped glaciers reach stabilization following large retreat more quickly than glaciers with more shallow slopes.
 The observed changes in outlet glacier dynamics are highly variable for our data set, and although several glaciers are referenced as examples in this paper, no single glacier can be used as a regional archetype. Our data indicate no uniform, clear relationship between front retreat and observed climate forcing on interannual timescales (Table S1 and Figures 9 and 10a–10c), or between front retreat and geometric parameters applicable to the entire data set (Figure 8), although temporal limitations of our data sets may affect our findings.
 Glaciers with steeper surface slopes should have a shorter inland extent of ice near flotation and higher driving stresses at the grounding zone. If slopes are sufficiently steep, such as for Docker Smith Glacier (76°14′13.71″N) whose snow line slope was 0.081 prior to retreat, steepening due to concentrated thinning at the front causes glaciers to reequilibrate to front perturbations within a few years. Shallower slopes result in faster propagation of dynamic thinning inland from the front, often bringing a larger area of inland ice to flotation, promoting further retreat. Initial thinning causes a positive feedback of lower lateral stress, acceleration, thinning and stretching, steeper slopes, higher driving stress, acceleration, retreat, etc. as was the case for Upernavik North whose snow line slope increased by more than a factor of 10 during the study period. This positive feedback caused by thinning can be further enhanced by buoyancy-driven feedbacks as a glacier retreats through a bathymetric depression, such as has been modeled for the retreat and readvance of Helheim Glacier in southeast Greenland [Nick et al., 2009]. Other models incorporating bathymetric depressions into their subglacial topography have demonstrated rapid front retreat occurring over several years for grounded outlet glaciers [Vieli et al., 2001; Vieli et al., 2002], suggesting the observed rapid front retreats in west Greenland may be at least partially controlled by local subglacial topography. The absence of subglacial topographic data also prevents complex quantitative analysis of stress changes at the front, thus limiting our understanding of glacier response to front perturbations. In order to better resolve the controls of glacier dynamics in the future, we must expand and improve our climate forcing data to include subsurface ocean temperatures. Additionally, we must incorporate subglacial topography data into the observational record of glacier dynamics to improve our current understanding of recent outlet glacier behavior and model predictions of future change.
 The response of west Greenland marine terminating outlet glaciers reveals complex, variable relationships between glacier behavior and changing climate. The synchronous retreat and subsequent readvance and deceleration of outlet glaciers in southeast Greenland in the mid-2000s, coincident with short-term anomalous warming of the ocean and atmosphere in the region [Howat et al., 2008] demonstrated a high sensitivity of outlet glacier dynamics to external forcing on interannual time scales. We examine whether similar behavior is observed in west Greenland due to changes in climate forcing, as suggested by GRACE observations of migration of accelerated mass loss to northwest Greenland since 2003. Warming of subsurface ocean waters beginning in the late 1990s as mapped by Holland et al.  potentially provided a mechanism for synchronous retreat and acceleration of outlet glaciers in west Greenland during this time, further motivating the need for a comprehensive analysis of recent temporal and spatial changes in west Greenland outlet glaciers. With the available data utilized in this study, front positions and surface elevation profiles for 59 marine terminating outlet glaciers reveal asynchronous retreat and thinning on interannual time scales, although all glaciers retreated or maintained stationary front positions throughout the observational record from 2000 to 2009. Although retreat and thinning were ubiquitous throughout the study period, providing support for GRACE mass loss observations, our climate and geometric data sets could not be used to directly explain the variability in the timing and/or magnitude of observed dynamic changes.
 We initially hypothesized that geometric parameters such as the presence of floating ice tongues, steepness of surface slope, and front width are the primary factors controlling variability in front retreat under similar external forcing. We found that the presence of a floating ice at the terminus controlled the magnitude of retreat for several glaciers within the data set, but the relationship was not applicable to all glaciers terminating as floating ice tongues. Several glaciers with grounded termini also experienced large (>1.0 km) retreats during the study period, indicating the presence of floating ice at the front is not a requisite for significant retreat in this region. We additionally found that the retreat of many glaciers in west Greenland were limited by surface slope near the grounding zone for two distinct reasons: (1) steep slopes concentrated thinning near the grounding zone, limiting the inland extent of ice brought closer to flotation, and (2) steep slopes indicate high driving stress and therefore high basal drag near the grounding zone, limiting the importance of a reduction in lateral drag due to thinning. A relationship between surface slope and perturbation response time is evident for a few glaciers, such as Alison and Umiamako glaciers, but this relationship is not applicable to the entire data set. There was no relationship between glacier width and retreat within our observational record (Figure 8), potentially indicating differences in glacier width are relatively unimportant at controlling glacier behavior on short timescales.
 Our results do not provide a technique to predict the likelihood of glacier retreat based on glacier geometry, as was the original objective. More work can be done, however, to investigate a relationship between glacier behavior and geometry in this region. Interpretation of the relationship between slope and front retreat was limited by data gaps in the time series, limiting spatial and temporal coverage for many glaciers and resulting in uncertainty in determination of interannual changes. Future work must continue to utilize NASA ATM elevation profiles when they are available due to their high resolution and minimum data errors. Subglacial topography and subsurface ocean temperature data must also be incorporated into the current geometric analysis where available in order to better understand observed glacier behavior in response to climate forcing.
 This work was funded by NASA grant NNX08AQ83G to I.M.H., B.E.S., and I.J. and a NASA Earth and Space Science Graduate Fellowship to E.M.M.