A minimal , statistical model for the surface albedo of Vestfonna ice cap , Svalbard

Introduction Conclusions References


Introduction
Glaciers and ice caps outside Greenland and Antarctica (GIC) contributed 0.028 m (∼ 16 %) to 20th century sea-level rise (Raper and Braithwaite, 2006).In the period 1961-1990 the share of Arctic GIC in this sum was about one fourth (Kaser et al., 2006).Moreover, the Arctic ice masses are located in the region of highest predicted air temperature increase during the coming decades (Rinke and Dethloff, 2008) and can thus be expected to further increase their contribution in the future.The Arctic can therefore be considered as a major source region for present and future GIC induced sea-level rise and knowledge on Arctic glacier mass balance thus emerges as a key factor in understanding current sea level-rise dynamics.
On Arctic glaciers the major source for melt energy is net shortwave radiation (e.g.Arendt, 1999;Winther et al., 2003) and robust albedo parameterizations play thus a key role in calculations of their energy and mass balance.Most calculation schemes for the surface albedo of glaciers run on high temporal resolution considering age, depth, density and temperature of the snow layer or accumulated melt on the glacier surface as input variables (Brock et al., 2000;Essery et al., 2005).These models are inappropriate for application on large Arctic ice caps where snow drift conditions (Sauter et al., 2012) frequently disturb the in situ developed surface-albedo pattern.As, moreover, small scale spatiotemporal distribution and variability of most meteorological parameters are poorly known in Arctic environments, the spatial distribution of surface albedo on Arctic ice caps can only be reliably treated on a less highly resolved scale.
The surface albedo on Vestfonna ice cap shows a characteristic pattern and evolution throughout a mass-balance year.It varies mainly with terrain elevation and thus reflects the combined influences of air temperature and both liquid and solid precipitation on the glacier surface.This fact facilitates the development of a statistical albedo calculation scheme that uses altitudinal profiles of the most easily accessible meteorological variables as input.Due to its empirical basis it thus avoids the drawbacks of more physically oriented modelling approaches that were outlined before.The aim of this study is to describe a new parameterization scheme for the surface albedo of large Arctic ice caps.The presented albedo model has a monthly temporal resolution while its spatial resolution is limited to altitudinal variability only.It is thus especially designed for application in longterm mass-balance studies like future projections where calculations with high spatiotemporal resolution are inappropriate or even impossible due to data limitations.
The model is based on a minimal number of meteorological input variables that reflect both, present weather conditions within each month and a longterm memory since the start of the corresponding mass-balance year.Monthly weather conditions are represented by altitudinal profiles of rain-snow ratio and thus implicitly include information about both air temperature and precipitation.The longterm memory is represented by cumulative positive degree days as well as cumulative snowfall sums since the start of the mass-balance year in September.
The study employs Terra MODIS (Moderate Resolution Imaging Spectroradiometer) derived surface albedo data as well as ERA-Interim based air temperature and precipitation data covering the period 2000-2008 for model setup.These data were provided by M öller et al. (2011a).Calibration and validation of the model is done using crossvalidation techniques following Marzeion et al. (2012).

Study area
Vestfonna ice cap is located on the island Nordaustlandet in the Northeastern Svalbard archipelago (Fig. 1).Its surface area of ∼ 2340 km The climate of the Svalbard region is governed by the contrasting influences of different air masses, cold and dry Arctic air coming from the north and warm and humid air coming from the Northern Atlantic Ocean (Svendsen et al., 2002).The dominating ocean currents in the region also reflect this contrast.The warm West Spitsbergen Current influences the western coastal regions of Spitsbergen (Walczowski and Piechura, 2011) while the eastern parts of the archipelago are mainly under the influence of cold Arctic ocean currents (Loeng, 1991).
On Nordaustlandet the climatic setting is governed by easterly weather systems originating in the Barents Sea region (Taurisano et al., 2007).They provide the major moisture source for precipitation (Førland et al., 1997).This means that Vestfonna is mostly located in the lee of the larger and higher ice cap Austfonna that covers the eastern part of Nordaustlandet.Precipitation sums are thus generally smaller on Vest-than on Austfonna (Hagen et al., 1993).They show considerable variability between different years (Beaudon et al., 2011) while the spatial distribution over the ice cap is almost entirely determined by terrain elevation (M öller et al., 2011b).Air temperatures in the study region show pronounced annual cycles.The mean summer air temperature at 370 m a.s.l. on Vestfonna is ∼ 0 • C with most of the days showing values between −3 • C and +3 • C (M öller et al., 2011b).Hence, melt conditions frequently extent over the entire ice cap (Rotschky et al., 2011).Melting generally starts in late June, reaches its maximum during mid and late July and then declines until end of August (M öller et al., 2011a).Due to the highly maritime setting of the study area, air temperatures show distinctly higher intra-monthly variability during winter.Daily means vary in the range öller et al., 2011b).

Data preparation
This study requires surface-elevation data and monthly mean albedo fields of Vestfonna ice cap.Meteorological data of air temperature and precipitation as well as local lapse rates are also needed.All data preparation in this study is done in a model domain with Introduction

Conclusions References
Tables Figures

Back Close
Full a regular 500 m grid that serves as a basis for deriving the altitudinal gradients of the input variables.

Terrain data
The outline of Vestfonna ice cap is digitized from a Terra ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) scene dating from 17 August 2000 (EOS Data Gateway Granule ID: SC:ASTL1B 00-08-12:36:0010269001). Surface-elevations are based on the ASTER Global Digital Elevation Model (GDEM).Limited areas of data voids and isolated elevation outliers in the central parts of the ice cap are interpolated on the basis of surrounding grid cells.Glacier outlines and the surface elevation grid are coregistered with the 500 m grid of the model domain using standard resampling techniques.Finally, a digital elevation model (DEM) of the ice-cap surface is created by masking the resampled GDEM to the glacier area using the digitized outlines.

Albedo profiles
The monthly mean albedo profiles are based on the MODIS snow product MOD10A1 version 5 (Hall et al., 2002;Hall and Riggs, 2007)  The accuracy of the daily MODIS albedo data is assessed by comparison with in situ measurements at an automatic weather station (AWS) located on the northwestern slope of Vestfonna (Fig. 1).For the period May 2008 to July 2009 a root mean square error (RMSE) of 0.12 is obtained (M öller et al., 2011a).As considerable parts of this error can be attributed to the fact that the observations at the AWS only cover a very limited part of the corresponding, much larger MOD10A1 grid cell (Stroeve et al., 2006), the MODIS data are assumed to adequately reproduce the surface conditions on the ice cap.Wang and Zender (2010) describe a systematic bias in the MODIS albedo data depending on solar zenith angle.Due to the limited length of in situ albedo measurements no characteristic annual evolution of the bias between MOD10A1 and AWS albedo is identifiable.Hence, no corrections of the MODIS albedo data are carried out.

Meteorological data
All meteorological data used in this study are based on daily ERA-Interim reanalysis data of the grid point located at 79.5 • N 19.5 • E (Fig. 1).Data cover the period September 2000 to October 2008.The original air temperature and precipitation data are statistically downscaled to fit local conditions on the ice cap (M öller et al., 2011a).From these data, altitudinal profiles of monthly means of positive degree days, snowfall and rain-snow ratio are created using lapse rates given by M öller et al. (2011a).

Positive degree days
The original, daily ERA-Interim air temperatures are downscaled according to M öller et al. (2011a) by using variance-inflation techniques (Huth, 1999;Karl et al., 1990;Storch, 1999).From the downscaled daily data monthly means are calculated.The distribution over altitude is done using a constant linear lapse rate of 7.0 K km −1 (M öller et al., 2011a).
Positive degree days were calculated according to Braithwaite (1984) and M öller and i that is defined as with σ T ,m being the standard deviation of air temperature that is characteristic for a specific month m(1, 2, . . ., 12) of the annual cycle (Table 1), T z the air temperature at elevation z and T i ,z the mean air temperature of month i at elevation z.Based on the integral over the positive interval of Eq. ( 1), the profile of positive degree days in month i (Φ pdd,i ) is calculated according to with N i being the number of days of month i and T + i ,z the mean over the positive daily air temperatures of month i at elevation z that is calculated by solving the following equation for T + i ,z Finally, the profile of cumulative positive degree days (Φ cpdd,i ) in month i is calculated as the sum over the monthly profiles of positive degree days since the beginning of the corresponding mass-balance year according to

Conclusions References
Tables Figures

Back Close
Full In this equation, k is the set of individual months between September of the previous year and month i of the present year.

Snowfall and rain-snow ratio
The original, daily ERA-Interim precipitation amounts are summed up to monthly values.The precipitation sum in month i is then distributed over altitude using quadratic scaling according to an index function of elevation (z) given by M öller et al. ( 2011a) In this equation P i ,0 is the original ERA-Interim precipitation of month i .It is set to represent sea-level conditions by forcing the intercept of Eq. ( 5) to one.The result (P i ) is the scaled precipitation profile.The corresponding profile of the proportion of snowfall (Φ sf,i ) in month i depends on the probability of negative air temperatures that can be derived from Eq. ( 1).It is calculated as The profile of cumulative snowfall sum (Φ csf,i ) in month i is calculated analogue to Eq. (4) as The profile of rain-snow ratio (Φ rsr,i ) in month i is calculated on the basis of the overall precipitation-sum profile (Eq.5) and the profile of the proportion of snowfall (Eq.6) according to

Conclusions References
Tables Figures

Back Close
Full

Model description
The presented model calculates mean monthly profiles of the surface albedo of Vestfonna ice cap on the basis of different meteorological input variables using multiple, non-linear regression techniques implemented in a two-step procedure.Meteorological variables are given as altitudinal profiles and the fitting parameters of the model as functions of terrain elevation.They therewith represent the specific altitudinal variability that characterises the surface-albedo pattern on the ice cap.
Calibration and implicit validation of the model is done using cross-validation techniques.Uncertainty considerations that serve as a basis for a quality assessment regarding the modelled surface albedo are also derived from the cross validation.

Fundamentals
The albedo of a glacier surface is influenced by a variety of factors that show a complex interaction with each other.Primarily, it depends on the type of the surface, i.e. snow cover or bare glacier ice.In general, snow albedo is more variable than ice albedo.Ageing of the snow cover that involves snow-grain metamorphism results in a continuous decrease of snow albedo (Jordan et al., 2008).This process is amplified by melt conditions and positive air temperatures as well as rainfall can thus be considered to have a major impact on snow albedo.Fresh snowfall on the other hand results in a sudden albedo increase.
To combine the counteracting influences of air temperature, rain-and snowfall into one parameter, an artificial meteorological variable called rain-snow ratio is introduced.Air temperature is assumed to decrease with terrain elevation according to a constant gradient.Hence, the proportion of rainfall in the total precipitation sum shows a decrease with terrain elevation while the proportion of snowfall increases correspondingly.However, precipitation sums in total increase with terrain elevation according to an increasing gradient (Eq.5) and the transition between rain and snow proportions is also non-linear (Eq.6).Taken together, these characteristics thus result in a non-linear relation between rain-snow ratio and snow albedo (Fig. 2).
The temporal evolution of glacier-surface albedo is largely influenced by snow depth, i.e. by the amounts of snowfall during the winter season, as well as by cumulative length and intensity of melt conditions that have already been effective since the beginning of the melt season.These variables mainly control the timing of bare ice exposure.

Initial setup
Model setup and initial calibration are done based on monthly albedo profiles of the period 2001-2008 with the months March to October represented in each year.In total, this makes a data basis of 64 individual months.Profiles of meteorological data cover a slightly longer period and show no winter-time data gaps.Coverage comprises the period September 2000 to October 2008.This is done in order to facilitate the calculation of cumulative meteorological data for the entire calibration period.
For model calibration the profiles are represented by a series of static 20 m elevation bins.To cover the entire set of surface elevations present on the ice cap, 31 of these bins are employed.For each bin a mean terrain elevation is calculated on the basis of the DEM.The meteorological data for each bin are then calculated according to these mean elevations.The albedo data for each bin are averaged over the corresponding grid cells of the model domain.
According to the described fundaments, the model initially calculates the albedo profile as a logistic function of the mean monthly profile of rain-snow ratio (Θ i (z)).The profile of remaining residuals is then approximated by a linear function (Ψ i (z)) of profiles of cumulative snowfall and cumulative positive degree days since the beginning of the corresponding mass-balance year, i.e. the previous September.The albedo profile (α i (z)) of month i is thus calculated as Full The logistic function of rain-snow ratio (Θ i (z)) for any of the 31 elevation bins characterized by its mean terrain elevation z is given as It is fitted to the whole set of the 64 monthly data pairs of rain-snow ratio and surface albedo that represent terrain elevation z.Thus, the parameters θ 1−4 are obtained individually for each of the 31 elevation bins.As they show systematic variability with terrain elevation, continuous parameter functions θ 1−4 (z) can be derived by fitting either linear, polynomial or exponential functions of terrain elevation z to these individual values.Figure 3 and Table 2 give an overview of the fitted parameter functions and their coefficients of determination. Figure 2 presents the fitted logistical functions of rain-snow ratio (Eq.10) for three selected bins and therewith illustrates their altitudinal variability.The residuals that remain after the logistic approximation are then fitted using multiple linear regressions based on the independent variables cumulative snowfall and cumulative positive degree days since the beginning of the corresponding massbalance year, i.e. the previous September.The linear function of cumulative snowfall and cumulative positive degree days (Ψ i (z)) for any 20 m bin with mean terrain elevation z is given as It is likewise fitted to the whole set of 64 months of input data and continuous parameter functions ψ 1−3 (z) are derived (Fig. 3, Table 2).The albedo-profile model calibrated in this manner is termed the initial albedo model (IAM).
For assessment of the accuracy of the IAM, modelled albedo profiles of all 64 months are compared to the ones derived from the MOD10A1 data.For each profile the RMSE is calculated over the set of all 31 individual albedo values that correspond to the respective 20 m elevation bins.Figure 4  of the individual RMSE values over the calibration period.Overall, the accuracy assessment yields a mean RMSE of 0.055 ± 0.026 with a slightly lower median of 0.052.The majority of all RMSE values lies in the range 0.03-0.07(Fig. 4).An elevation-dependent bias between modelled and observed albedo values does only exist at terrain elevations above 500 m a.s.l.(Fig. 5).This documents a generally good model performance over most parts of the ice cap.However, the calibration of the parameter functions θ 1−4 (z) and ψ 1−3 (z) of the IAM employs data of all 64 months available, i.e. calibration and application period of the model are identical.It can thus only serve as an optimum reference of model performance as no independent validation is possible.In order to present an albedo model that is applicable not only in the reference period, calibration and implicit validation of the final model is done using a cross validation-based procedure.

Calibration and cross validation
The final albedo model (FAM) is calibrated using leave-one-out cross-validation techniques by repeating the described calibration procedure of θ 1−4 (z) and ψ 1−3 (z) 64 times.Each time, all data of a specific month i are left out and the parameter functions are thus fitted to a reduced set of input data representing 63 instead of 64 months.By doing so, a model is build that provides the possibility for an independent validation of its one-month application period.This is because the model is capable of calculating the albedo profile of month i without using any input data of this month for calibration.
From the resulting 64 individual model calibrations (Fig. 6) the means of each of the 20 individual parameters of the 7 parameter functions (Table 2) are calculated.They serve as the final parameters of the FAM.The ranges of uncertainty of the parameter functions (Fig. 3) that result from the associated standard deviations of the means (Table 2) are discussed as part of the sensitivity studies.
The accuracy assessement of the FAM is again based on the mean RMSE values between modelled and observed albedo profiles along the 31 elevations bins of each profile.In general, the resulting RMSE values for all 64 different months are very similar 993 Introduction

Conclusions References
Tables Figures

Back Close
Full to the ones resulting from calculations using the IAM (Fig. 4).Small differences with slightly higher RMSE values only occur in the summer months but do never exceed values of 0.01, i.e. one albedo percent.Accordingly, the calibration and cross-validation of the FAM results in a mean RMSE of 0.057±0.028with a slightly lower median of 0.054.The profiles of the remaining biases of IAM and FAM are likewise similar (Fig. 5).The FAM profile indeed shows a slighly wider spread for the values of all individual elevation bins, but the overall bias pattern appears to be the same for both IAM and FAM.
No systematic elevation-dependency is obvious except for the uppermost parts of the ice cap above 500 m a.s.l.(Fig. 5).The systematic bias is thus limited to regions where only very little ablation occurs (M öller et al., 2011a).As the presented albedo model is intended for application in glacier melt models, this can be regarded as an acceptable drawback.Taken together, these results reveal a good and reliable performance of the FAM.

Sensitivity studies
Calibration and cross validation of the FAM revealed a spread of possible model parameters (Fig. 6) that suggests considerable sensitivity towards the choice of which parameter calibration is used.Differences of more than one order of magnitude are evident in the relative variability within the 64-samples sets of different calibrations of each model parameter.This means that the cross-validation procedure resulted in very stable calibrations of some of the model parameters (e.g.Results indicate a minor model sensitivity regarding most parameter functions.The vast majority of albedo deviations lies below 0.01, i.e. one albedo percent, for individual elevation bins.Moreover, for the albedo deviations of most parameter functions no interannual variability or variability over the range of albedo values is evident. The albedo deviations of two parameter functions (θ 2 and ψ 2 ) differ significantly from this overall pattern.θ 2 shows mean deviations of up to 0.07 during the summer months with a distinct increase towards higher albedo values.While even in summer the sensitivities at low albedo values are still well below 0.05, they regularly exceed 0.09 at albedo values higher than 0.8.In spring and autumn θ 2 is, in contrast, in line with the overall pattern and shows only small sensitivities of less than 0.01.
The most extreme sensitivity of the FAM towards its model parameters results from a variation of ψ 2 (Fig. 7).Introduced albedo deviations show a partly comparable pattern as for θ 2 but with distinctly amplified values.In summer, at low albedo values they reach more than 0.05 while at high albedo values they even exceed 0.50.Moreover, the albedo deviations experience a clear increase with time between September and the following August.This is because of the fact that ψ 2 is the regression coefficient associated with the input variable cumulative snowfall sum (Eq. 11).As the values of this input variable constantly increase until August of each year, a variation of its regression coefficient results in likewise increased albedo deviations.However, the extremely high sensitivity of the FAM towards ψ 2 can partly be explained as a model artefact.The parameter function develops very close to zero above ∼ 200 m a.s.l.(Fig. 3).As a result, small perturbations of the parameters (a, b and c) of ψ 2 (Table 2) already result in large relative variations of ψ 2 itself.This, in turn, blows up the variability of modelled albedo after a winter season due to multiplication of ψ 2 with a high value of Φ csf,i (Eq.11).
The profile of overall model sensitivity, i.e. the summed up albedo deviation ∆α all (z), Results (Fig. 8) reflect the predominant influence of albedo deviations introduced by sensitivity towards variations of ψ 2 and thus show a similar spatiotemporal pattern.Albedo deviations increase with terrain elevation, i.e. towards higher albedo values, and show a superimposed increase over the year until August followed by a sudden drop to significantly lower values in September.

Discussion and error assessment
The altitudinal variability of the model parameters as it is represented in the seven parameter functions (Fig. 3) indicates that the surface-albedo pattern of Vestfonna is not only governed by the set of meteorological variables employed in this study.If these were the exclusive predictors of surface albedo, the parameters should be constant over terrain elevation.Hence, additional driving forces for surface-albedo variations must exist.
It IAM in terms of parameter functions (Fig. 3, Table 2) and the IAM is assumed to be the optimum reference of model performance (cf.Sect.4.2).Thus, the albedo profiles modelled by the FAM are considered as reliable despite the high model sensitivities at high albedo values.The accuracy of the albedo values modelled by the FAM shows strong variability with terrain elevation (Fig. 5).This is also reflected in the associated RMSE profile (Fig. 9) that is calculated on the basis of the bias profile of the FAM (Fig. 5).In comparison with the area-altitude distribution of Vestfonna (Fig. 9) this pattern of altitudinal variability of the RMSE reveals that the area-weighted mean RMSE (0.053) is smaller as the nonweighted one shown in Fig. 4.This is because lowest RMSE values along the profile are associated with the interval of most frequent terrain elevations (400-550 m a.s.l.).Highest RMSE values, in contrast, are limited to terrain elevations covering distinctly smaller areas.Terrain elevations below 50 m a.s.l. are only reached at the lowermost parts of the outlet-glacier tongues while elevations higher than 550 m a.s.l.only exist along the main ridges of the ice cap.
For any further usage of the FAM, e.g. in mass balance-modelling studies, an exactly quantifiable error range needs to be defined.Owing to the facts discussed before, this error range (E (z)) is expressed as a function of terrain elevation according to the models' RMSE profile (Fig. 9) rather than as a constant value.Therefore, RMSE values for each elevation bin are calculated on the basis of the bias profile of the FAM (Fig. 5).
Afterwards they are fitted by a fourth-order polynomial according to The parameter values 1−5 are given in Table 3.The fitted function allows for an almost perfect reproduction of the RMSE profile (R 2 = 0.98) and is thus regarded as a reliable expression of model error.Accordingly, the overall formulation of the FAM that is suitable for further usage in broader modelling applications is given as 997 Introduction

Conclusions References
Tables Figures

Back Close
Full  Polar Res., 30, 5963, doi:10.3402/polar.v30i0.5963, 2011. 985 Sauter, T., M öller, M., Finkelnburg, R., Grabiecz, M., Scherer, D., and Schneider, C.: A snowdrift model for Vestfonna ice cap, Northeastern Svalbard, in preparation, 2012.983, 996 Storch, H. V.: On the use of "inflation" in statistical downscaling, J. Clim., 12, 3505-3506, 1999.987 Stroeve, J., Box, J., and Haran, T.: Evaluation of the MODIS (MOD10A1) daily snow albedo product over the Greenland ice sheet, Remote Sens. Environ., 105, 155-171, 2006. 987 Svendsen, H., Beszczynska-Møller, A., Hagen, J., Lefauconnier, B., Tverberg, V., Gerland, S , 114, 563-575, 2010. 987 Winther, J.-G., Bruland, O., Sand, K., Gerland, S., Marechal, D., Ivanov, B., Głowacki, P., andK önig, M.: Snow research in Svalbard -an overview, Polar Res., 22, 125-144, 2003. 983  Full Discussion Paper | Discussion Paper | Discussion Paper | 2 in 2005 (Braun et al., 2011) that covers elevations between sea level and ∼ 630 m a.s.l.makes it one of the largest ice masses of the Eurasian Arctic.The relief of the generally flat surface of the ice cap is dominated by two main ridges, one of them stretching W-E and the other N-S.Its highest point is located close to the conjunction of the two ridges in the eastern central part of the ice cap (Fig. 1).In between these ridges, Vestfonna is dominated by large outlet glacier basins and land-terminating ice lobes.Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | of the period 2001-2008.From the original, daily albedo fields with a spatial resolution of 500 m monthly mean albedo fields are calculated.Pixels not holding any albedo information due to cloud cover or non-classifiable characteristics are left out during the averaging procedure.This results in a mean update frequency of albedo information of 2.9 days over the ice cap (M öller et al., 2011a).No MOD10A1 datasets are available for the study area during the period of polar night.Accordingly, albedo fields are only created for the eight-months period March to October.Finally, a total of 64 mean altitudinal profiles of monthly surface albedo along a set of 31 individual 20 m elevation bins are calculated over the ice-cap DEM. Discussion Paper | Discussion Paper | Discussion Paper | Schneider (2010) based on the probability density function of air temperature of month Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | presents an overview of the temporal distribution Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | θ 1 b, θ 3 a or ψ 1 a) while others show distinctly weaker calibrations (e.g.θ 3 c, θ 4 c or θ 4 d ).The influence of the parameter spreads on modelled albedo is assessed within the range of one standard deviation of each of the seven parameter functions shown in Figure 3.The FAM is run two times for each parameter function (mean plus one standard deviation and mean minus one standard deviation).For each of the 31 20 m elevation bins, the resulting maximum deviation from the albedo values modelled by the unchanged FAM is then taken as model sensitivity.Figure 7 presents an overview of the individual model sensitivities regarding all seven parameter functions.Discussion Paper | Discussion Paper | Discussion Paper | is calculated from the seven individual sets of albedo deviations (∆α p (z)) as presented Discussion Paper | Discussion Paper | Discussion Paper | in Fig. 7 using error propagation rules according to ∆α all (z) = p ∆α p (z) 2 with p = θ 1−4 and ψ 1−3 (12) is suggested that the altitudinal variability of the model parameters represents a superimposition of in situ conditions (as they are created by the considered meteorological variables) by snowdrift influences.According to Sauter et al. (2012) radial snowdrift trajectories are a common pattern on Vestfonna due to katabatic winds.Hence, the disturbance of in situ developed surface-albedo conditions due to snowdrift should also show a strong altitude dependency.The fact that the remaining bias of modelled albedo values is largest throughout the uppermost parts of the profiles further supports this interpretation as snowdrift-induced erosion and thus disturbance of in situ snowpack evolution are most intense along the main ridges of the ice cap.These upper parts of the ice cap where high albedo values prevail throughout the entire year are also the regions where critical model sensitivities are reached.However, the final calibration of the FAM shows almost no differences to that one of the Introduction Discussion Paper | Discussion Paper | Discussion Paper | statistical albedo model is presented in this study.The model is developed and applied at Vestfonna ice cap in Northeastern Svalbard and intended for further usage within broader, especially longterm energy or mass-balance calculations.It calculates the mean altitudinal albedo profile of the ice cap on a monthly resolution using a minimal set of meteorological variables as input.The surface albedo fields used for calibration and cross-validation purposes are taken from the daily MODIS snow product MOD10A1.The meteorological record used is based on daily ERA-Interim data of a grid point located closely south of the ice cap.Modelling is done for the years 2001-2008 excluding the winter months, i.e. in the period March to October, due to missing MODIS data during polar night.The model architecture is based on a calculation scheme that combines a logistic function of rain-snow ratio with a linear function of cumulative snowfall and cumulative positive degree days.The artificial quantity rainsnow ratio was developed in order to combine the varying influences of air temperature, rainfall and snowfall on the actual snow cover into one meaningful variable.Validation of the albedo model reveals a good model performance over large parts of the altitudinal profile of the ice cap.Modelled and observed albedo values along the profile differ with an RMSE of 0.057 ± 0.028 (mean ± one standard deviation) and an area weighted mean of 0.053.Terrain elevations that show higher RMSE values (0.07-0.09) are limited to the lower-and uppermost parts of the ice cap and thus to regions that either cover very limited areas or are hardly affected by surface melt.Throughout terrain elevations that are most frequent on Vestfonna the RMSE even drops to values well below 0.05.In the period of peak melting, i.e. in July(M öller et al., 2011a), the RMSE is also below 0.05.Hence, the calculated albedo profiles are regarded as reliable reproduction of in situ conditions in the context of further model application in broader, ice cap-wide energy or mass-balance studies.The presented albedo model is therefore suggested to be fully suitable for this purpose.Discussion Paper | Discussion Paper | Discussion Paper | QuikSCAT/SeaWinds scatterometry,

Fig. 1 .Fig. 2 .Fig. 3 .Fig. 4 .Fig. 5 .Fig. 6 .Fig. 7 .Fig. 8 .Fig. 9 .
Fig. 1.Overview map of the study area.Coordinates refer to UTM (Universal Transverse Mercator) zone 34 N. Contour spacing on the ice cap is 100 m starting at sea level.The inset on the upper left shows the location of Vestfonna ice cap in the Svalbard Archipelago.The circle (ERA) marks the location of the ERA-Interim grid point.