Introduction

The ablation areas of mid-latitude glaciers contribute a disproportionately large volume of meltwater to total summer discharge in proglacial streams. Differential reflection of incident solar radiation by the variety of surfaces encountered on valley glaciers exerts an important temporal and spatial control on meltwater generation. Whilst snow reflectivity has received extensive examination, in terms of both measurement and modelling (see reviews by Reference MellorMellor, 1977; Reference WarrenWarren, 1982; Reference DozierDozier, 1987), exposed glacier ice remains underexplored in this regard. Meteorological investigations on glaciers have included the measurement of ice reflectivity, with considerable variation in the results obtained (Reference AndrewsAndrews, 1964; Reference HolmgrenHolmgren, 1971; Reference Wendler and WellerWendler and Weller, 1974). In light of the sensitivity of Current glacier mass-balance simulations to reflectivity (Reference Oerlemans and HoogendoornOerlemans and Hoogendoorn, 1989; Reference MunroMunro, 1991), estimating a single value from such studies to represent the entire ablation area of a new area will produce erroneous predictions of daily mass loss.

Fieldwork on Peyto Glacier, Alberta, Canada, was designed to quantify the spatial and temporal range of reflectivity in a temperate glacier setting. Snow reflectivity parameterizations in some current global circulation models separate the visible and near-infrared wavebands (Reference Marshall and WarrenMarshall and Warren, 1987) because of their different responses to surface physical properties. Values from Peyto Glacier were obtained for the spectral wavebands of 0.35–0.695 μm (visible) and 0.695–2.8 μm (near-infrared).

Background

Measurements of broad-band reflectivity are often conducted over the short-wave spectrum from 0.35 to 2.8 μm, and are commonly referred to as albedo (ρ). Hemispherical reflectivity over the broad visible and near-infrared spectral bands (ρVIS and ρNIR) is reported in this paper. The terminology adopted here originates from the move precise definitions of Reference Siegel and HowellSiegel and Howell (1972).

Early measurements of ρon mid-latitude glaciers in the Austrian Alps were carried out by Reference Sauberer and DirmhirnSauberer and Dirmhirn (1952), Reference Scheibbner and MahringerScheibbner and Mahringer (1968) and Reference WagnerWagner (1979). Table 1 includes results of these studies in a comparison of the quoted ranges of ρfor different snow and ice surfaces. Reference Grenfell and PerovichGrenfell and Perovich (1984) provided values for spectral reflectivity of glacier ice between 0.38 and 2.5 μm for Athabasca Glacier, Canada. Reference Munro and YoungMunro and Young (1982) made some preliminary investigations of ρ on Peyto Glacier.

The factors influencing reflectivity are those related to the surface itself and those concerning the incident solar radiation. An increase in grain-size results in reduced reflectivity, most effectively in the near-infrared Spectral region (Reference WarrenWarren, 1982). Reference Bohren and BeschtaBohren and Beschta (1979) demonstrated that for a given grain-size ρis, at most, weakly dependent on density. Water alone does not increase radiation absorption, because its refractive index is similar to that of ice (Reference Bergen, Hutchinson, McMillen, Ozment and GottfriedBergen and others, 1983), but the effective grain-size is increased by water in interstices between grains (Reference ColbeckColbeck, 1979), and this reduces ρImpurity content acts mainly on visible reflectivity; larger concentrations reduce values of ρ (Reference WarrenWarren, 1982). Increasing surface roughness, through the development of microtopographical features such as sastrugi, has the same effect. Slope angle and aspect interact with incident solar-radiation geometry to alter its effective angle of approach to the surface, thus altering ρ(Reference MannsteinMannstein, 1985). A dependence of reflectivity on solar zenith angle, θ, has been demonstrated in numerous studies over snow and ice (Reference Sauberer and DirmhirnSauberer and Dirmhirn, 1952; Reference HubleyHubley, 1955; Reference Scheibbner and MahringerScheibbner and Mahringer, 1968; Reference YamanouchiYamanouchi, 1983; Reference McGuffie and Henderson-SellersMcGuffie and Henderson-Sellers, 1985; Reference Wendler and KelleyWendler and Kelley, 1988), where reflectivity increases as θincreases, because of enhanced specular reflection of the direct-beam component of radiation.

The spectral composition of incident solar radiation is altered by the presence of clouds as water and ice preferentially absorb near-infrared radiation, thus increasing the proportion of visible light. This enhances ρ because snow and ice reflect more visible than near-infrared radiation (Reference WarrenWarren, 1982). Additionally, the diffuse fraction of the incident flux increases with cloud cover. The diurnal variation of ρis smoothed as the diffuse fraction increases, with values for θ> 50° reduced but those for θ< 50° increased, a response which has been well documented for water (Reference Nunez, Davies and RobinsonNunez and others, 1972).

Procedures

Five sites in the ablation zone of Peyto Glacier were examined, each for approximately 1 week of the 1990 summer. Three locations were chosen to represent extremes of clean, dirty and rough ice (excluding crevassed areas). The remaining two sites were on melting snow and rock. A detailed assessment of surface properties was carried out for each site (Table 2).

Impurity content was determined by filtering a melted sample of known volume through a pre-weighed quartz membrane filter (1.5 μm pore spacing). The sample was then dried for 24 h at 110°C and reweighed to yield impurity content. Snow density was measured using standard techniques (Reference Østrem and BrugmanØstrem and Brugman, 1991). Values for ice were obtained from a SIPRE-type coring auger (83 mm diameter) penetrating the topmost 0.25–0.30 m of the surface. Each sample was split at the discontinuity between the weathering crust (Reference Müller and KeelerMüller and Keeler, 1969) and the more dense ice below, permitting separate densities to be calculated. Surface roughness at each site was estimated by laying out 5 m long poles on the ice and measuring the distance from pole to surface at 0.1 m intervals. Five hundred values were collected along each of two transects aligned at right-angles to one another, and standard deviation about the mean distance was used as an index of roughness.

Four pyranometers (Kipp and Zonen, Model CM11, Holland) were used to obtain hemispherical reflectivity. These were mounted in two back-to-back pairs on adjacent horizontal supporting arms 1.6 m above the surface. One pair recorded incoming and reflected solar radiation over that part of the spectrum which passes through clear glass (0.35–2.8 μm), thus capturing all but negligible amounts of solar radiation (K) from the full short-wave spectrum. Hemispherical RG695 filters on the second pair permitted only wavelengths greater than 0.695 μmto pass through, thus capturing near-infrared solar radiation (K_NIR ). Then ρVISand ρNIRare estimated by:

where λ refers to wavelength and arrows denote directions of flux.

It was important to maintain constant sensor height in order to eliminate changes in the field of view of the inverted instruments. A “floating” support system was thus adopted in which a vertical mast rested on a small flat rock (0.05 m² in area) on the glacier surface. A bolt embedded in the rock held the pole centred, while three guy wires provided the necessary stability. The system performed well throughout the season, with minimal adjustment. The pyranometers were aligned such that midday shadows were cast north of the centre of the inverted-pyranometer viewing circle.

The percentage of total reflected solar radiation (F_a ) Originating from a circle of radius acan be calculated from instrument height Z₀ , (Reference SchwerdtfegerSchwerdtfeger, 1976):

Sensors mounted at 1.6 m would therefore receive 91% of their signal from within a circle of radius 5 m. Accordingly, observations of surface physical properties were made within this boundary whenever possible.

The pyranometer signals were scanned at 5 s intervals by a data logger (Campbell Scientific, Model 21X, U.S.A.) which recorded 15 min mean values. A number of adjustments were made to the data:

• (i) A filter factor of 1.049 was applied to the sensors fitted with RG695 filters to correct for solar radiation absorbed by the glass.

• (ii) A correction factor of 1.01, based on measurements of shaded area over a range of solar zenith angles, was applied to data from the inverted sensors.

• (iii) Data obtained during precipitation periods were discarded, since the presence of raindrops on the inverted glass hemispheres may have produced erroneous readings.

Consideration of potential errors in pyranometric measurements, treating them as independent and additive, results in a worst-case error of ±5.4% for the reflectivity measurements. Reference McGuffie and Henderson-SellersMcGuffie and Henderson-Sellers (1985) point out that this range is broad enough to mask many of the subtle variations in reflectivity that are sought. This is, however, not thought to be a problem since the errors represent the mean deviation from the correct reading, the direction of which is unlikely to change over the day, or from one day to the next. Experiments performed on the inverted instruments, to investigate possible alteration in calibration constant due to overturning, showed no significant effect upon the calibration.

Data from the upward-facing unfiltered pyranometer were used, in conjunction with pyrheliometric measurements of direct radiation (Epply, Model E6, U.S.A.) to find the diffuse radiation, D,by residual. Thus, a statistical relationship was found between global radiation, and the diffuse fraction:

where I₀,the solar constant, is taken to be 1353 Wm⁻² (Reference MonteithMonteith, 1973) and θ is the solar zenith angle. Our data yielded α = 1.171 and β = 1.225, with a coefficient of determination equal to 0.82. It should be noted that the diffuse fraction is assumed to be equal in the visible and near-infrared spectral regions in later data analysis. Commonly the diffuse-radiation peak drifts further into the near-infrared as cloud increases (Reference Weiss and NormanWeiss and Norman, 1985).

Results and Discussion

A strong distinction between ρVISand ρNIRwas observed under clear skies at all three glacier-ice locations (Fig. 1a). The larger values for ρVISagree with previously observed contrasts between the two wavebands over snow surfaces. The large surface impurity content at the dirty-ice site is responsible for the relatively greater difference between ρVISand ρNIRat this site. Reflectivity displayed limited temporal variation, whilst Spatial variation, indicated by comparing the clean-ice and dirty-ice results, was markedly larger in the visible spectral range. The presence of surface roughness elements at the hummocky-ice site, creating microtopographic shading, is apparently less influential than impurity content. Evidence for asymmetry of ρVISand ρNIRaround solar noon, caused by physical changes of the surface over the passage of a day (melting), was only slight, although afternoon values tend to be lower than mid-morning values at the hummocky- and dirty-ice sites.

Snow cover and rock are at the extremes of the range of observed reflectivity behaviour (Fig. 1b). The greater diurnal variation of reflectivities occurred at the snow site.

Fig. 1. Diurnal variation of ρVIS and ρVIS on clear days at the dirty-ice (D.ICE), clean-ice (C.ICE) and hummocked-ice (H.ICE) sites. (b) Diurnal variation of ρVIS and ρNIR on clear days at the snow and rock sites.

As no completely clear days occurred during the snow measurements period, a composite trace was plotted by combining records from two days with overlapping clear periods. Here asymmetry around solar noon is more pronounced, with reduced afternoon values attributable to surface metamorphosis and slope geometry; the slope dips at 9° west (280°), which favours greater reflection in the morning. In contrast to the glacier surfaces, reflection of solar radiation from rock is greater in the near-infrared range. Non-linearities in the traces are probably caused by shading effects from individual rock elements. As expected from an opaque medium, diurnal variation is negligible.

In order to study the influence of cloud cover on ρVIS and ρNIR, reflectivities recorded in the zenith-angle ranges 35° < θ< 45°, and 50° < θ< 60° were plotted against the diffuse fraction of incident solar radiation, D/K↓, an approach used by Reference Nunez, Davies and RobinsonNunez and others (1972) over water. As noted earlier, the presence of cloud should subdue reflectivity for θ> 50°, whilst the converse is expected for θ < 50°.

The results of this analysis (Fig. 2) for dirty ice, in the zenith-angle range 35° < θ < 45°, show that neither waveband is particularly sensitive to increasing diffuse fraction, although there is a rising trend which is quantified by regressing reflectivity against diffuse fraction of incident global radiation (Table 3). For dirty ice in the zenith-angle range 35° < θ< 45°, ρVIS rises by 0.033 as diffuse fraction goes from 0 to 1, whereas ρVIS increases by only 0.011. Reflectivity for all glacier-ice sites shows a similarly weak response to increased diffuse fraction, although of the two wavebands the near-infrared shows the greater sensitivity. It should be noted that the standard error displayed in Table 3 is often >50% of the change in reflectivity from purely direct to purely diffuse conditions.

Fig. 2. The responses of ρVIS (open squares) and ρNIR (closed circles) at the dirty-ice site to increasing diffuse fraction of incident solar radiation for 35° < θ < 45°.

As theory predicts (Reference Choudhury and ChangChoudhury and Chang, 1981), snow reflectivity under clear skies (D/K↓→0) is greater for the larger of the two θ ranges, so there is a more dramatic response of snow reflectivity to cloud (Fig. 3).

Fig. 3. The response of ρVIS (closed squares) and ρNIR (closed circles) at the metamorphosed snow site to increasing diffuse fraction of incident solar radiation for 35° < θ < 45°, and equivalent responses for 50° < θ < 60° (open symbols).

Again ρNIR is the variable more sensitive to increasing diffuse fraction, and ρVIS provides a good example of how the two zenith-angle ranges may converge at D/K↓ = 1 (as suggested by Reference Nunez, Davies and RobinsonNunez and others, 1972). These results indicate that most Of the rise in ρ under increasing cloud is due to the rise of reflectivity in the near-infrared part of the spectrum.

One factor driving changes in reflectivity under increasing cloud is diminished alteration of the angle of approach of incident solar radiation. It is therefore interesting to examine the relationship between reflectivity and COS θ under sunny skies (Fig. 4), contrasting the behaviour of ρVIS and ρNIR at the clean-ice and the snow sites (clean ice was chosen because surface impurity content was the closest match with the snow site), Predominantly sunny sky prevails in the selected range of cloud conditions (0–0.2 cloud), thus minimizing the influence of cloud on θ.As expected from the relatively, greater response of the snow surface to cloud, the snow reflectivity displays the greater sensitivity to cos θ.

Fig. 4. The response of ρVIS (squares) and ρNIR (triangles) to increasing solar zenith angle at the sites for clean ice (open symbols) and snow (closed symbols).

Direct comparison of our results with data sets from other glaciers is hindered by inadequate descriptions of surface properties in past investigations of reflectivity. The broad-band nature of our data also makes comparisons difficult because the majority of measurements available are spectral reflectivities for narrow wavebands. Nevertheless, comparison of our data with some spectral reflectivities published for glacier ice (Fig. 5) shows that ρVIS for dirty ice (solar noon value) is slightly greater than values in the visible range reported by Reference Zeng, Cao, Feng, Liang, Chen and ShengZeng and others (1984) for what they term “honey comb glacier ice”. Their surface is interpreted as ice with multiple cryoconite holes containing morainic debris.

Fig. 5. Comparison of broad-band visible and near-infrared reflectivities from clean and dirty glacier ice in this study with spectral reflectivity data from other glacier-ice surfaces. The division between the visible and near-infrared bands is indicated at the top of the figure.

Overall, it seems that the cleanest ice found on Peyto Glacier is surpassed on some glaciers. Comparison of ρVIS from the cleanest part of Peyto Glacier with spectral reflectivities of glacier ice (Fig. 5) suggests that surfaces examined by Reference Grenfell and PerovichGrenfell and Perovich (1984), Reference Zeng, Cao, Feng, Liang, Chen and ShengZeng and others (1984) and Reference Hall, Bindschadler, Foster, Chang and SiddalingaiahHall and others (1990) contained fewer impurities. The values of Zeng and others are for a blue-ice band, whilst Hall and others only describe their site as being in the ablation area.

Comparison of values in the near-infrared band is complicated by the rapid decrease in spectral reflectivities at 1.4 μm.Additionally, the spectral distribution of solar radiation may be slightly weighted towards the shorter wavelengths (Reference MonteithMonteith, 1973; Reference Weiss and NormanWeiss and Norman, 1985). Taking this into consideration, the data of Grenfell and Perovich, when weighted according to the expected distribution of incoming solar near-infrared radiation (Reference MonteithMonteith, 1973), yield a value of 0.20 for the wavelength range equivalent to ρNIR, which compares well with 0.18 for ρNIR at the clean-ice site.

For comparison of snow reflectivities with the literature, values are taken from solar noon under clear-sky conditions (ρVIS = 0.71, ρNIR = 0.42). Snow surface density obtained 1 d before the reflectivity measurements was 536 ± 11 kg m⁻³. Laboratory tests were conducted by Reference O’Brien and MunisO’Brien and Munis (1975) for snow (density 525 kg m⁻³) over the wavelength range 0.6–2.5 μm. The range 0.6–0.695 μ lies in the visible region as defined by the present study. In this range, O’Brien and Munis observed reflectivities of 0.85; substantially higher than 0.71 for ρVIS. This is attributed to impurities in the natural surface on Peyto Glacier, as well as greater roughness. A comparable snow surface is reported by Reference Grenfell, Perovich and OgrenGrenfell and others (1981) with a density of 480 kg m⁻³ and an impurity content < 500 ppm by mass. Visible spectral reflectivities ranged from 0.72 to 0.78. More surface dust concentration (2000 ppm by mass when converted from gl⁻¹ in Table 2), as well as possible influence from glacier ice under the relatively thin snowpack (0.3 m thick), could account for the lower ρVIS value of 0.71 on Peyto Glacier.

A distinctive feature of rock reflectivity at Peyto Glacier base camp was ρNIR > ρVIS, a reversal from the snow and ice surfaces. This behaviour was also observed by Reference Rott and SøgaardRott and Søgaard (1987) during spectral scans of ice-marginal locations in western Greenland. For “gravel and rocks covered in lichens”, their reflectivities ranged from 0.13 to 0.16 in the visible range, and 0.18 to 0.34 in the near-infrared (compared with ρVIS = 0.16 and ρNIR = 0.25 on Peyto Glacier moraine at solar noon).

Concluding Remarks

The results from three extreme types of ice in this study (clean, dirty and hummocked) clarify the fact that impurity content is the dominant influence on spatial variability of hemispherical reflectivity. The relatively minor difference in values between smooth and hummocked ice surfaces further emphasizes the weaker influence of roughness compared with impurity concentration. Assuming that the three selected sites do represent extreme cases of clean, dirty, rough and smooth (excluding crevassed areas), the data suggest that near-infrared reflectivities vary only slightly over the entire ablation area of Peyto Glacier. Thus, most of the spatial variation in meltwaler generation will be due to absorption of visible solar radiation. On the other hand, the overall higher absorption in the near-infrared band provides the stronger contribution to total meltwater generation by radiative melt.

Differences between clean ice (smoothest) and hummocked ice (roughest) indicate that the development of microtopography during the ablation season may reduce reflectivities. The 1 m hummock relief develops over a number of melt seasons. Lowering of reflectivities by up to 0.1 due to increased shading therefore represents a conservative upper constraint for a single summer.

The diurnal variability of reflectivity at all ice sites was small in comparison to the variation between ice types. In contrast to the greater fluctuation of ρVIS for different locations, it was ρNIR which displayed larger responses to temporal changes in atmospheric and surface conditions (also true at the snow site). This bias has also been noted in broad-band parameterizations of snow-reflectivity carried out by Reference Marshall and WarrenMarshall and Warren (1987) and Reference Marks and DozierMarks and Dozier (1992).

The reduced reflectivities during intensive surface melt on Julian day 218 (Fig. 6), illustrate the contribution of meltwater to the low temporal variability of glacier-ice reflectivities on Peyto Glacier. Melting ice crystals at the surface have less ability to specularly reflect incident solar radiation than frozen crystals. Additionally, water-filled interstices between crystals increase the density of the surface layer, and thus reduce ρNIR.

Fig. 6. The impact of high-melt conditions for Julian day 218 (light plot) on reflectivity, illustrated by comparison with the following day of lower melt (bold plot). Sky conditions on day 218 were predominantly clear, with 0.3 cumulus cloud at all times. The next day was clear throughout.

The timing of the peak meltwater discharge at a glacier tongue will be influenced by the diurnal variation of reflectivity. Our results suggest that peak meltwater generation in the ablation area should coincide with solar noon, although other factors such as increased afternoon air temperatures, cloud presence and katabatic winds will complicate the situation.

The measurements from Peyto Glacier pose a problem for mass-balance modellers in that spatial heterogeneity of reflectivity in the visible range makes it difficult to establish correct “representative” reflectivities for altitudinal zones. Reference OerlemansOerlemans (1990) demonstrated the possibility of a 111 m rise in equilibrium-line altitude due to a decrease of only 0.03 in short-wave reflectivity. Yet differences of up to 0.15 were seen between the dirty- and clean-ice locations, sites with less than 5 m elevation difference which would thus be grouped within the same altitudinal zone. One method of alleviating this problem might be to adopt a gridded model, whereby areas of homogeneous reflectivity are incorporated into single elements. Clearly, this approach would require detailed knowledge of reflectivity over the entire glacier at regular intervals, a feat which is practical with satellite imagery, if it is suitably processed to account for bi-directional reflectance.

The dominant influence of impurity content on spatial variation of ρVIS raises a question regarding the impact of impurity accumulation over time, and its effect on mass balance. The cleanestpatch of ice available possessed lower ρVIS than found on some glaciers, which suggests that the depositional climate may play a role in influencing mass balance. Negative mass balances have become increasingly common on Peyto Glacier in recent years: seven negative years during the 1975–85 period (Reference Letréguilly and ReynaudLetréguilly and Reynaud, 1989). These have previously been associated with increasing temperatures alone, but the possibility that accumulating impurities have lowered the reflectivity should also be considered.

Whilst the lack of diurnal variation in reflectivity at the ice and moraine sites is in itself unexciting, this behaviour is useful from the standpoint of remote sensing. Within a 6 h window centred on solar noon, suitably processed satellite imagery is likely to be a reliable source of reflectivity estimates for ice, as these are within 2–3% of daily mean values, though a 4 h window is more appropriate for snow. Such values can be used as a base to which such factors as solar zenith-angle dependence can be added in modelling the net short-wave radiation regime of the glacier surface.