Passive microwave (SSM/I) satellite predictions of valley glacier hydrology, Matanuska Glacier, Alaska

Authors


Abstract

[1] We advance an approach to use satellite passive microwave observations to track valley glacier snowmelt and predict timing of spring snowmelt-induced floods at the terminus. Using 37 V GHz brightness temperatures (Tb) from the Special Sensor Microwave Imager (SSM/I), we monitor snowmelt onset when both Tb and the difference between the ascending and descending overpasses exceed fixed thresholds established for Matanuska Glacier. Melt is confirmed by ground-measured air temperature and snow-wetness, while glacier hydrologic responses are monitored by a stream gauge, suspended-sediment sensors and terminus ice velocity measurements. Accumulation area snowmelt timing is correlated (R2 = 0.61) to timing of the annual snowmelt flood peak and can be predicted within ±5 days.

1. Introduction

[2] Monitoring the extent and duration of glacier melt is important for numerous reasons. High latitude glacier melt is starting earlier and persisting longer [Ramage and Isacks, 2002] making valley glaciers susceptible to rapid ablation. It is critical for scientists to improve remote monitoring techniques for quantifying glacier melt and hydrological responses to climate warming.

[3] Microwave frequency remote sensing has been used to monitor snowmelt, with emphasis on the long-term (1988 to present) passive microwave Special Sensor Microwave Imager (SSM/I) satellite. Multiple-daily polar-orbiting SSM/I overpasses provide nearly complete spatial coverage in high-latitude regions. Frequent overpasses capture daily frozen, melting, and refreezing transitions of snow and ice. Most glacial applications focus on larger ice sheets and icefields where the SSM/I spatial resolution, re-sampled to EASE Grid 25 km scale, provides meaningful maps of melt progression [e.g., Steffen et al., 1993; Ramage and Isacks, 2002; Tedesco, 2007]. Until now, little effort has been made to apply these data to monitor melt over smaller valley and outlet glaciers.

[4] In this study, we demonstrate the utility of SSM/I data for remote valley glacier melt detection and hydrology predictions. Specifically, we apply a SSM/I melt algorithm developed at the Juneau Icefield [Ramage and Isacks, 2002] to monitor conditions at the smaller Matanuska Glacier (61°51′28″N, 147°49′03″W, Figure 1). In addition, we draw predictive links between SSM/I snowmelt and glacial stream flow during the annual snowmelt flood. This SSM/I approach can be a particularly useful tool for remote monitoring of snowmelt and spring runoff from temperate glaciers.

Figure 1.

Index Map. (a) Location of Matanuska Valley in southern Alaska. SSM/I pixels situated over the Matanuska Glacier and (b) the CRREL glacial stream gauge. Higher-elevation EASE Grid cells (m-61 and m-62) cover the accumulation area and icefield, while the lower-elevation mixed m-73 cell is only 15% glacier covered.

2. Background

[5] Microwave data respond to as little as 1% liquid water in snow and are therefore useful for long-term monitoring of snowmelt onset and duration [Ulaby et al., 1986]. At frequencies greater than 10 GHz, passive microwave brightness temperatures (Tb) respond abruptly to snow transitions from dry to wet [Ulaby et al., 1986]; sensors at these frequencies also penetrate through darkness and are only weakly affected by cloud cover. By the Raleigh-Jeans approximation, Tb is the product of a target's physical surface temperature (Ts) and emissivity (E): Tb = ETs at microwave frequencies.

[6] Many passive-microwave snowmelt detection approaches have been proposed using combinations of microwave frequencies and thresholds [e.g., Mote and Anderson, 1995; Drobot and Anderson, 2001]. However, differently from these, the approach of Ramage and Isacks [2002] and Tedesco [2007] use both ascending and descending observations to capture daily snowmelt transitions. By this algorithm (1), wet snow is detected when the 37-V GHz Tb and the difference between ascending and descending overpasses (DAV) exceeds fixed thresholds A and B respectively:

equation image

These thresholds are reevaluated when applied to different regions, terrain conditions, or sensors.

[7] The geophysical responses of Tb and DAV are tied to physical processes active in transitioning snow packs. Water content increases snow permittivity and Tb [Ulaby et al., 1986]. Snow that is frozen or re-frozen at night appears dry to passive-microwave satellites, yielding low Tb. Daytime melting significantly increases Tb. The difference between brightness temperatures from morning and evening overpasses (i.e., high DAV) is large. Most meltwater is retained by the snowpack when liquid water refreezes at night. When nighttime temperatures remain above 0 °C to allow night melting, the differences between morning and evening Tb overpasses are smaller (i.e., low DAV). Snow is ripe and meltwater is released during this phase. Other conditions can impact Tb (e.g., decreased snow crystal grain size, internal snow structures and snowpack heterogeneity, and refrozen snow [Ramage et al., 2006]). Yet the satellite sensor's response to snowmelt geophysically overwhelms the impact of these other factors [Ulaby et al., 1986].

[8] The DAV approach has been applied over several different regions. Ramage and Isacks [2002] mapped melt over the maritime Juneau Icefield using thresholds A = 246 K and B = 10 K; these same thresholds were used over the boreal forests of the Canadian Yukon Territory [Ramage et al., 2006]. Tedesco [2007] applied this approach over the Greenland Ice Sheet using the 37 V GHz channel (A = 258 K, B = 18 K) and the 19 H GHz channel (A = 245 K, B = 25 K). The breadth of these works motivate us to expand this approach to environments such as smaller, heterogeneous, higher-relief mountain-valley glacier basins like the Matanuska Glacier.

3. Data Used in this Study

[9] We use 37 GHz vertically polarized ascending and descending SSM/I data from the F13 satellite (1995–2006) supplied by the National Snow and Ice Data Center through the EOS Data Gateway [Armstrong et al., 1994]. At 37 GHz, SSM/I instantaneous field of view (IFOV) dimensions are approximately 37 × 29 km. Original SSM/I swath data are re-sampled to the Equal Area Scalable Earth-Grid (EASE-Grid), giving each pixel a nominal 25 × 25 km resolution; three EASE grid cells cover the Matanuska basin (Figure 1). Local acquisition times range between 0730–0900 for the AM overpass and 1930–2100 for the PM overpass.

[10] The Matanuska Glacier is relatively small so that each of the three EASE grid cells is fractionally perennially snow covered. The m-61 cell (mean elv. 2179 m) covers most of the upper icefield with 90% snow cover. The m-62 cell (mean elv. 1888 m) includes middle elevation glacier and icefield with 75% snow cover. The m-73 cell (mean elv. 1078 m) contains only 15% perennial snow cover at the glacier terminus. Near-surface temperatures in the Matanuska basin were measured at the glacier terminus (490 m elevation) and on a nunatak just below the glacier equilibrium line in cell m-62 (1400 m elevation).Winter-spring temperatures in the Matanuska basin vary daily by 15 to 20°C, and often reach as low as −30°C at night.

[11] Discharge from the glacier is measured in an armored reach of the river 400 m from the terminus, and is thought to represent 90% of the water flowing from beneath the glacier, which occupies about 57% of the Matanuska basin. Ice velocity was measured daily on the terminus surface during 1996 and 1997 using high-precision laser survey instrumentation. Suspended sediment concentration was measured at hourly intervals from water emerging from subglacially connected conduits exposed at the glacier terminus. Ice velocity and sediment data were originally published by Ensminger et al. [1999], who demonstrated a link between ice motion and glacier hydrology; we use these data to describe ice response to the spring snowmelt flood.

4. Results

[12] Threshold A is evaluated for each pixel (Figure 1) using the Tb histogram approach described by Ramage and Isacks [2002]. Brightness temperature histograms generate bimodal distributions with two large populations separated by a low-count interval. The geophysical distinction between wet and dry snow falls within this low count interval. At Matanuska, we find the threshold A = 246 K is appropriate for the high- and middle-elevation pixels (m-61 and m-62, respectively) while A = 238 K is appropriate for the lower-elevation cell (m-73). Thresholds were confirmed by ground observations. Ground observations were made in snow pits over the study area. Temperature profiles were measured manually with snow thermometers and snow pack wetness was measured using capacitance meters during Spring 2007. Field data show a transition of snow from ‘dry’ frozen where wetness values were between 0–1% liquid water content, to water contents exceeding 1% rising to 5% after the SSM/I satellite thresholds are met. Field observations reveal that snow cover on the numerous high-relief bedrock cliffs in the m-73 cell is quickly removed by strong winds, generating partially snow-covered conditions.

[13] DAV threshold B was determined using a combination of air temperature measurements and snow wetness observations collected during Spring 2007. The DAV ranges are large, up to 50 K at the low-elevation m-73 cell and up to 40 K at m-61 and m-62. These large DAV values are due to refreezing of snow and daily temperature variations (e.g., pre-melt diurnal temperatures can range from −30 to −10°C, while melting diurnal temperatures can range from −10 to +10°C). Air temperature is a general guide for temporal trends of target temperatures, but it is important to note that it is actually the temperature of the snow (target) that drives measured SSM/I Tb, not necessarily the air temperature, especially when air temperatures are above 0 °C. The threshold B is determined to be 15 K for all three cells.

[14] Melt onset is determined when the condition on both thresholds A and B are met (discussed above). Onset dates are apparent due to the robust signal of the DAV (see Figures 2a, 2b, and 2c). Melt onset within the three cells is depicted by the left-hand edge of the shaded box over the DAV plots in Figure 2. The gray box brackets the snow “melt-refreeze” interval. Melt starts at similar times in the low- and middle-elevation cells (Figures 2a and 2b), typically around Julian day (Jd) 106 and 107, respectively. Melt starts later at higher elevations around Jd 113 (Figure 2c). End of the snow melt-refreeze interval occurs when nightly temperatures are too warm to refreeze snow, causing DAV to drop below 15 K (i.e., threshold B is no longer met). During the 12-yr record, the mean day when melt-refreeze ceases is Jd 151, 156, and 164 at the low-, middle-, and highest-elevation pixels, respectively.

Figure 2.

Snowmelt and glacier responses for 1997. Snow melt onset defined by both Tb and DAV thresholds (a) starts in the low elevation, (b and c) progressing to higher elevations. Extent of the snow “melt-refreeze” interval is indicated graphically by the gray boxes (Figures 2a, 2b, and 2c) and symbolically represented by Δ O +. The streamflow data show (d) the flood starts in late June (streamflow transitions labeled on graph at dashed lines), and is followed by (e) peak of ice velocity and (f) peak subglacial suspended sediment, and prior to the peak flood discharge. The “lag time” is the number of days between the end of snow melt-refreeze and the date of the peak flood discharge.

[15] The annual spring glacier snowmelt flood generates an abrupt increase and peak of streamflow (noted in Figure 2). Snowmelt flood streamflow normally increases from ∼40 to ∼200 m3/s over a short duration. Snowmelt flood peak flows are among the largest of the melt season, typically as significant as large summer storm-flows. Snowmelt floods usually start between Jd 158 and 173 (mean start date Jd 164). Snowmelt floods peak about 1 to 3 weeks later, ranging from Jd 167 to 187, with a mean peak date of Jd 178.

[16] Predictive statistical correlations are made between the timing of the end of snow melt-refreeze and start of the peak of the snowmelt flood (Figure 3). During the 12 yr record, the mean number of days (lag time) between the two events at m-62 pixel was 22 days (Figure 3), with a correlation value of R2 = 0.61. At the high pixel, mean lag was 14 days, with correlation R2 = 0.54. These r-squared values are based on only 12 data points (limited by the length of the Matanuska Glacier streamflow database), thus quantitative significance testing is not proposed. A predictive link between end of SSM/I-derived snow melt-refreeze at m-62 and snowmelt flood timing is made by the equation y = 0.42× + 112.41 (Figure 3), where y is day of flood onset and x is the day snow melt-refreeze terminates.

Figure 3.

Scatter Plots. End of snow “melt-refreeze” and peak of the snowmelt flood timing are correlated for Matanuska. Only data for the accumulation area (m-62) are plotted in the left graph (R2 = 0.61); the right plot has data for each of the three pixels plotted together. No data points fall below the diagonal lines in both plots, demonstrating that snow melt-refreeze always concludes before the peak snowmelt subglacial discharge.

[17] In 1997, ice velocity immediately preceding the snowmelt flood was 13.1 cm/d. Velocity began increasing on Jd 167, peaked at 24.3 cm/d on Jd 178 (Figure 2e), and dropped to 12.2 cm/d on Jd 187. This peak was the seasonal maximum, matched only by late-season fast-flow in August. Ice flow patterns in 1996 were similar. Ice was flowing at 24.1 cm/d just before the snowmelt flood, began increasing Jd 172 to peak at the seasonal maximum of 31.4 cm/d on Jd 176, then decreased to 23.1 cm/d on Jd 182.

[18] Subglacial suspended-sediment concentrations during 1997 also fluctuated at the time of the snowmelt flood. The daily mean immediately before flooding was 0.66 g/L, ranging prior from 0.25 to 1.5 g/L. Sediment began increasing Jd 172, peaking to 1.7 g/L on Jd 175 (Figure 2f), and dropping to pre-snowmelt-flood conditions only 3 days later. This peak concentration was the seasonal maximum. Spring 1996 produced a similar sediment-concentration pattern. The mean immediately before the snowmelt flood was 1.1 g/L, ranging prior from 0.9 to 1.4 g/L. Concentrations began rising Jd 165, peaked at 1.6 g/L on Jd 180, and decreased to 1.3 g/L on Jd 195.

5. Interpretation

[19] The relatively consistent time for snow pack meltwater to reach the glacier terminus suggests that conditions or processes operating within the glacial environment control migration of the water to the terminus. Meltwater from the snowpack both on and adjacent to the glacier probably contribute to the snowmelt flood signal at the terminus, but the exact routing of that water is currently in doubt. Most of the meltwater could be routed supraglacially within the remaining snowpack or on the ice surface in supraglacial channels to the terminus stream. Alternatively the meltwater may move into the glacier within moulins and crevasses linked to englacial and subglacial conduits.

[20] SSM/I satellite data show that snow melt-refreeze concludes before the peak snowmelt flood, supporting the notion that snowmelt drives the snowmelt flood. The Figure 3 (right) scatter plot depicts this graphically; no data points fall below the diagonal line drawn through the plot area. Snowmelt timing is best correlated to snowmelt flood timing at the m-62 cell (R2 = 0.61), which includes the portion of the accumulation area over the glacier trunk. Snowmelt trends over (m-61), the highest most distal part of the icefield, are slightly less correlated (R2 = 0.54). The lower, more proximal m-62 is likely more collocated with the glacier plumbing at the terminus. This notion is supported by field observations at South Cascade Glacier where Fountain [1989] demonstrated daily forcing links between accumulation area melt and subglacial hydraulics. We interpret the data to argue that snowmelt over the accumulation area and ice field of the Matanuska Glacier drives the annual spring snowmelt flood at the glacier terminus.

[21] Suspended-sediment concentrations and ice-flow measurements show strong responses to the spring flood events consistent with subglacial routing of the snowmelt floodwaters (Figure 2). Sediment content of subglacial waters increased during the snowmelt flood by 50% in 1996 and by 100% in 1997. Similarly, rate of ice flow just upglacier of the terminus increases during the snowmelt flood by 60% in 1996 and by 85% in 1997.

[22] Subglacial routing of the annual snowmelt flood leads to certain implications. Previous studies at Haut Glacier d'Arolla and other glaciers identified a “spring event,” which refers to a hydrologically triggered reorganization of the subglacial drainage system accompanied by increases in ice velocity and suspended sediment concentration. Spring events can be induced by snowmelt, precipitation or high melt rates, although snowmelt is also discussed as a catalyst [Röthlisberger and Lang, 1987; Nienow et al., 1998; Campbell et al., 2006]. Current thinking suggests that a rapid influx of liquid water to low-capacity poorly-developed plumbing systems generates hydraulic and mechanical instabilities [Kavanaugh and Clarke, 2001] lasting hours to days [Röthlisberger and Lang, 1987] as the plumbing reorganizes [Stone and Clarke, 1996; Kavanaugh and Clarke, 2001]. We suggest that the responses of ice flow and suspended sediment, which appear to be linked to the snowmelt flood event at Matanuska, are consistent with the style of spring events. Our datasets were not originally collected to study spring events, so we cannot prove they exist at Matanuska Glacier. However, the comparable style of ice and drainage responses leads us to speculate that our passive microwave approach may be detecting timing of annual hydrological processes impacting the subglacial plumbing regime of a valley glacier.

6. Conclusions

[23] The SSM/I 37 GHz vertically polarized brightness temperatures from both ascending and descending overpasses have been applied to monitor snowmelt over a valley glacier and predict hydrological response at the glacier terminus. Our algorithm captures daily transitions of snow between frozen, melting-refreezing, and melting continuously. In EASE grid cells largely covered by glacier ice (perennial snow), the snow is wet and melting when Tb exceeds 246 K and the DAV exceeds 15 K. The low-elevation pixel covering the terminus region has a largely patchy snow cover and only 15% of the cell includes terrain within the glacier basin, thus correlations over this pixel are not meaningful.

[24] Brightness temperatures monitor the progression of melt from the terminus region through the accumulation area and into the upper reaches of the icefield. Concurrent increases in the suspended sediment in subglacially derived discharge and in the rate of surface ice flow in the terminus region during the annual snowmelt flood at the Matanuska Glacier provide indirect evidence that snowmelt waters are routed subglacially.

[25] This SSM/I remote-sensing approach could be used to remotely monitor timing of the snowmelt and predict annual snowmelt flood events at glaciers. The timing of the annual snowmelt flood peak can be predicted within ±5 days using satellite observations of snow melt-refreeze over the Matanuska Glacier accumulation area. One of our driving interests regarding snowmelt processes over temperate glaciers is whether meltwater migrates primarily along the ice surface, or mainly through and below the glacier. Realistically, both pathways are likely used, but which pathway routes more water remains unknown. Our current dataset does not permit us to show a direct physical connection between the spring snowmelt flood event and subglacial drainage, but it does show similar attributes to studies of “spring events” at other temperate glaciers. The SSM/I approach can be applied to smaller glaciers in the near future when adapted to NASA's new (2002–present) higher resolution (∼12.5 km) advanced microwave scanning radiometer for EOS (AMSR-E) passive microwave satellite.

Acknowledgments

[26] We thank the National Snow and Ice Data Center for EASE-Grid SSM/I satellite data and William Stevenson of Alaska Outfitting for glacier and high-alpine guide support. Funding was provided by the NASA Graduate Student Earth System Science grant (NNX06AG08H), NASA Terrestrial Hydrology grant (NNG04GR31G), US Army Dissertation LTT Research Fellowship, and Lehigh University. This manuscript was improved by helpful discussions with Richard Alley and two anonymous reviewers.

Ancillary