Inter-annual variations of snow days over Switzerland from 2000 – 2010 derived from MODIS satellite data

Snow cover plays a vital role in the Swiss Alps and therefore it is of major interest to determine and understand its variability on different spatiotemporal scales. Within the activities of the National Climate Observing System (GCOS Switzerland) inter-annual variations of snow days over Switzerland were derived from 2000 to 2010 based on data from the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra satellite. To minimize the impact of cloud cover on the MODIS snow product MOD10C1, we implemented a post-processing technique based on a forward and backward gap-filling approach. Using the proposed methodology it was possible to determine the total number of annual snow days over Switzerland from 2000 to 2010 (SCD MODIS ). The accuracy of the calculated snow days per year were quantitatively evaluated against three in situ snow observation sites representing different climatological regimes (SCD in_situ ). Various statistical indices were computed and analysed over the entire period. The overall accuracy between SCD MODIS and SCD in_situ on a daily basis over 10 yr is 88% to 94%, depending on the regional characteristics of each validation site. Differences between SCD MODIS and SCD in_situ vary during the snow accumulation period in autumn and smaller differences after spring, in particularly for the Central Alps.


Introduction
Snow cover represents a significant geophysical variable for the climate system through a range of complex interactions and feedback mechanism related to its physical properties (IPCC, 2007).High priority is therefore accorded in the Implementation Plan of Figures

Back Close
Full the Global Climate Observing System (GCOS) to strengthen and maintain snow cover observations, ideally supplemented with other observing systems (WMO, 2006(WMO, , 2010)).The estimation of snow parameters such as snow extent, snow depth, and snow water equivalent plays a vital role in the Swiss Alps for winter tourism as well as for the management of water resources (e.g.Elsasser and Messerli, 2001;Abegg et al., 2007;Uhlmann et al., 2009).Therefore, ground-based monitoring of the snow cover has a long tradition in Switzerland and a number of studies have been published over the last years, focusing on its temporal variability and long-term trend (Beniston, 1997;Laternser and Schneebeli, 2003;Scherrer et al., 2004;Scherrer and Appenzeller, 2006;Marty, 2008).Overall, these results describe a decrease of the alpine snow pack since the mid 1980s especially at lower altitudes which is linked to an increase of local winter temperatures (Scherrer et al., 2011).Most of these studies focused on the spatiotemporal variability of days with snow cover.Beside snow measurements (e.g.new snow depth), the number of snow days is an important climatological indicator serving applications in tourism, transportation, construction, or agriculture.In general, a snow day is defined as a day with a snow depth larger than a certain threshold (WMO, 2009).Different thresholds are published to derive the number of snow days for climatological applications (e.g.Eckel, 1938;Hantel et al., 2000;Laternser and Schneebeli, 2003;Scherrer and Appenzeller, 2006;Marty, 2008).
Compared to ground-based observations with the limitation of being sparse, satellite data give an area-wide and spatially consistent information of the ground or atmosphere.For more than three decades, satellite remote sensing has been used to measure snow properties from drainage-basin to continental scales (Hall and Martinec, 1985).With the continuity of Earth Observation programs of climate relevant sensors, satellite data series will be further extended and the development of new satellite systems will provide the retrieval of critical measurements.Concerning satellite-based snow cover monitoring, climatological studies on a decadal and hemispherical to regional scale were carried out (e.g.Robinson and Frei, 2000;Armstrong and Brodzik, 2001;Bales et al., 2008;Foster et al., 2009;Boi, 2010).For Switzerland, polar-orbiting Introduction

Conclusions References
Tables Figures
In this study, we focus on the inter-annual variations of snow cover days (SCD) from October 2000 to September 2010 based on data from the Moderate Resolution Imaging Spectrometer (MODIS) on board the Terra platform.The purpose is to examine the spatiotemporal variation of the snow conditions on a yearly basis over Switzerland.A simple cloud gap-filling technique is implemented by post-processing the MOD10C1 product to derive the number of days with snow cover.The multi-year time series of SCD on an annual basis are compared with selected in situ snow observations and further discussed on a monthly and daily resolution to demonstrate the potential of our SCD product.The study reported here is placed in the broader context of the satellite activities within the National Climate Observing System of Switzerland (GCOS Switzerland) (Seiz andFoppa, 2007, 2011;Seiz et al., 2011).(snow, cloud, land, and others) and the total number of land observations mapped into a grid cell of the CMG (Riggs et al., 2006).The MODIS snow cover retrieval is based on the Normalized Difference Snow Index (NDSI) and further criteria tests (Hall et al., 2002;Hall and Riggs, 2007).A number of studies have assessed the accuracy of the MODIS snow cover algorithm and its derived products on a regional scale (e.g.Klein and Barnett, 2003;Tekeli et al., 2005;Parajka and Bl öschl, 2006).A comprehensive overview of the MODIS-based snow products and validation studies is given by Parajka and Bl öschl (2012).

In situ snow observations
In Switzerland, snow variables such as total snow depth and new snow depth are measured mainly by the Federal Office of Meteorology and Climatology MeteoSwiss and the Swiss Federal Institute for Snow and Avalanche Research SLF.The observations consist of either manually or automatically recording stations.The amount of new snow depth and total snow depth at conventional stations is measured twice daily (morning and evening) by an observer on a representative plot in horizontal terrain (Bezzola, 2004).This network of conventional observations covers the entire region of Switzerland.The advantage of manual observations is the longer time series of up to more than 50 yr providing valuable information for climatological studies.Over the past few years, efforts have been made to identify, digitize and explore snow measurements from historical data sources dating back to the second half of the 19th century (W üthrich, 2008).A recently published report by MeteoSwiss (W üthrich et al., 2010) defined a potential basic climatological network for snow, based on the analysis of 160 historical snow measurement series.In our study, three stations have been selected from this so-called National Basic Climatological Network for Snow (NBCN-S) representing main climatological regimes in Switzerland and varying altitudes (Table 1).All in situ observation data have been well quality controlled and verified through various standard quality processing steps (Bezzola, 2004).In our study, we defined snow days with a snow depth of at least 1 cm (SCD in situ ) as used at MeteoSwiss.Introduction

Conclusions References
Tables Figures

Back Close
Full 3 Method for satellite-derived snow days A weakness of optical satellite sensors is that they are not able to see the ground in cloudy conditions in contrast to in situ observations.The spectral specifications of the MODIS sensor do not allow for observations of the Earth's surface under cloudy conditions, which is a major limitation for studies of snow cover.This particularly applies to areas with frequent cloud cover, e.g.mountain regions such as the Alps.The MOD10C1 snow cover product uses the ancillary MODIS cloud mask (i.e.MOD35 product) based on cloud-detection tests indicating a level of confidence that MODIS observes clear skies (Ackerman et al., 1998;Platnick et al., 2003).
Several techniques have been developed to reduce the cloud cover pixels from MODIS products.They include one or a combination of different approaches, such as spatial filtering (Parajka and Bl öschl, 2008;Gafurov and B árdossy, 2009;Xie et al., 2009), temporal filtering, temporal composites of maximum snow cover extent, prior cloud-persistence and combining Terra and Aqua MODIS data (Parajka and Bl öschl, 2008;Gafurov and B árdossy, 2009;Wang et al., 2009;Hall et al., 2010).
To produce a consistent daily MODIS snow cover map at 0.05 degree resolution, we applied a post-processing technique to fill data gaps in the MOD10C1 product caused by clouds.The methodology uses a combination of thresholds in the post-processing of the MODIS product and a subsequent two step gap-filling approach as described below.
In a first step, we re-classified the pixels as follows: if cloud percentage of a CMG cell is >95 % the pixel is re-classified as cloud covered; if snow percentage of a CMG cell is >4 % and cloud percentage ≤95 %, the pixel value is then normalized, resulting in a snow fraction related to the cloud free land observed (confidence index).If the CMG snow cover is less then 4 % and the CMG cloud observation is less than 95 %, the grid Introduction

Conclusions References
Tables Figures

Back Close
Full time, the gaps are filled with the same value provided from the latest cloud-free grid cell.In accordance to this so-called forward gap-filling approach, we applied the same procedure in reverse direction, e.g. 30 September 2001 to 1 October 2000.We assume that through the exactly opposite gap-filling of forward and backward technique, the over-or underestimation in one of the directions is minimized.As we focus on a yearly temporal resolution, we calculate the mean of the total number of snow days derived from the forward and backward gap-filling procedure.Data gaps were treated as cloudy conditions.Furthermore, information from the last or first day of the neighbouring year was not taken into account if the first or last day of the year was cloud covered.The MODIS derived snow cover days (SCD MODIS ) are calculated for the hydrological year starting 1 October of any given year and end on 30 September of the subsequent year.
It should be noted that this procedure is usable for data re-processing only, comparable to nowcast (forward gap-filling) and hindcast (backward gap-filling) modeling.It is the intention to derive the total number of snow days on a yearly time resolution.Figure 2 shows the temporal evolution and variability of the total number of snow days for each year from 2000 to 2010 for the three selected stations based on MODIS (SCD MODIS ) and in situ (SCD in situ ) data.The highest located station Samedan has at least 130 days with snow per year with up to more than 200 for 2008/2009.At the stations Basel and Lugano, some years without snow occurred.The year with a larger number of snow days for Lugano (2008/2009) is reflected in the corresponding maps in Fig. 1 for the region of northern Italy.The year 2006/2007 is characterized with the minimum number of snow days over the 10 yr, except for Basel and Samedan.In general, the inter-annual variability of snow days derived from in situ observations over the 10 yr is captured very well by our post-processed MOD10C1 product.The annual total number of snow days derived from MODIS is higher compared to the in situ defined number of snow days for 7 out of the 10 yr in Basel and for 9 out of the 10 yr in Lugano, respectively.Only at Samedan the in situ based number of snow days is in 6 out of the 10 yr higher than the satellite-derived number of snow days.However, the absolute differences between the SCD in situ and SCD MODIS vary significantly between the three stations with a maximum discrepancy of 33 days (18 %) for Samedan in 2002/2003. At Basel (2002/2003) and Lugano (2008Lugano ( /2009)), largest differences of 16 days (34 %) and 18 days (50 %) are found, respectively.

10-yr snow cover climatology based on SCD
The MODIS derived snow cover days (SCD MODIS ) versus the in situ based calculated snow days (SCD in situ ) for the three stations and over the 10 yr is summarized in Table 2.The mean difference d , the standard deviation of the difference s and the correlation coefficient c is indicated for each station (Table 2).Additionally the annual mean of the total number of snow days derived from in situ measurements is indicated (SCD in situ ) as well as annual mean of the cloud cover based on the MOD10C1 cloud obscuration.The 10 yr average of SCD in situ and cloud cover refers to the regional characteristics of each of the three sites.Although the statistical evidence is based on a small sample, Introduction

Conclusions References
Tables Figures

Back Close
Full a certain tendency of the discrepancy is obvious.Overall, the satellite based total number of days with snow agrees well (c > 0.8) with a slight overestimation of up to 7 days.However, the standard deviation of the difference is twice as high for Samedan as for Basel or Lugano (s = 14.46).

Analysis of monthly SCD
Table 3 summarizes the results of the comparison of SCD MODIS and SCD in situ at monthly resolution for each of the three sites.Additionally, the monthly average of the snow days defined by in situ measurements and of the cloud coverage based on the MOD10C1 cloud mask is indicated.The total number of snow days based on MODIS differs from the in situ observations for most of the month.Outstanding is the month February (Basel and Lugano) and the period January to April at Samedan with a positive bias.Largest negative deviations are found at the beginning of the snow season (October, November), when ephemeral snow fall is frequently observed, especially at high elevation sites such as Samedan.In the Southern Alps at Lugano, highest biases are noted later in winter (December, January).As the snow season advances, the agreement increases quite constantly.Inter-monthly variations of the cloud cover are clearly visible for each station with different cloud cover maxima and minima for each site and region.
Figure 3 shows the month to month variations and differences from October 2000 to September 2010, exemplified for Samedan.Overall, the monthly variations from

Analysis of daily SCD
Daily data were analyzed to further examine the discrepancies between both data sets.The results are reported in Fig. 4 for Samedan for the 2002/2003 season.A total of 185 snow days were extracted from the MODIS post-processed product, as opposed to 152 days determined from in situ observations (i.e. a difference of 33 days).Figure 4a illustrates the snow cover fraction (blue) and the frequency and duration of cloud coverage from the daily MOD10C1 product (light brown) and the defined snow free days (green).The longest period of cloud cover occurs at the end of December (8 days) and a maximum cloud free period of up to 20 days in February.Figure 4b and c explains the forward and backward gap-filling approach by artificial adding of additional days with snow and days with no snow, respectively.The complimentary effect of these opposite processes is explained in Fig. 4d: contrary gap-filling results are detected between the middle of October and beginning of November.While the forward approach substitutes cloud cover with snow free land, the backward approach does fill the cloud obscured days with snow.The months of October (+7 days), November (+6 days), and extensively April (+19 days), are causing a larger total number days with snow per year compared to the in situ observations (Fig. 4e).
A confusion matrix (e.g.Pu et al., 2007) was constructed to better assess the overall accuracy of the gap-filling technique.Table 4 presents three confusion matrices for Basel (a), Samedan (b) and Lugano (c), based on a daily comparison between SCD in situ and SCD MODIS from 1 October 2000 to 30 September 2010.The results of the forward (SCD F ) and backward (SCD B ) gap-filling approach are combined for snow covered (e.g.SCD F snow ) and snow free days (e.g.SCD B snowfree ).The overall agreement describes the number of cases where both gap-filling approaches indicate snow days (SCD F snow and SCD B snow ) and agree with the in situ derived snow days (SCD in situ snow ) on snow covered and snow free conditions to the total number of days (Klein and Barnett, 2003).For Basel and Samedan the overall agreement is quite similar and high at 88.4 % and 88.7 %, respectively.At the site Lugano the agreement

Conclusions References
Tables Figures

Back Close
Full is higher at 93.7 %.Both satellite-based gap-filling approaches incorrectly substitute cloudy pixels as snow covered in 1.2 % and 2.0 % of all days (commission errors) for Basel and Lugano, respectively.At Samedan around 3.7 % of all days are incorrectly defined as snow covered by the post-processed MOD10C1 product instead of snow free.In contrary, a wrongly replacement of cloud coverage with snow free observations occurs in about 1 % of all days for all three sites (omission errors).The fraction of inconsistent gap-filling results (e.g.SCD F snow and SCD B snowfree ) on the total number of days varies between 3.8 % for Lugano, 6.1 % for Samedan and a maximum of 9.3 % for Basel.

Discussion
The derivation of the yearly number of snow days from the MOD10C1 product is adequate for describing the inter-annual variability of the snow cover over Switzerland.For example, the relatively small number of snow days particularly in the Swiss lowlands is in agreement with monthly temperatures above average for several months between September and June in the years 2001/2002and 2006/2007(MeteoSwiss, 2001, 2007)).In contrary, completely snow covered Swiss lowlands were observed in the years of 2004/2005 and 2008/2009 due to early snow fall events and winter conditions (MeteoSwiss, 2005(MeteoSwiss, , 2009)).Although the spatial resolution of the MOD10C1 is moderate (0.05 • ) topographic features such as the main alpine valleys are clearly apparent in the satellite-based product.As such, SCD MODIS provides valuable information to complement the sparsely distributed point observations on the ground.Analyses at both annual and monthly levels showed that the number of snow days is overestimated by our post-processed MOD10C1 product.For the lowland stations in the northern (Basel) and southern (Lugano) part of Switzerland, the absolute deviations are smaller compared to Samedan on a yearly basis from 2000-2010.However, for the alpine site Samedan, the relative error (SCD in situ − SCD MODIS ) averaged over the 10 yr is significantly smaller (7 %) than for Basel (45 %) or Lugano (41 %).Introduction

Conclusions References
Tables Figures

Back Close
Full The overestimation is also shown by a higher commission error rate of SCD MODIS (e.g.3.7 % for Samedan) than the error rate of omission (e.g.1.5 % for Samedan).Thus, the post-processed MOD10C1 product appears to have a tendency to retrieve too many snow days compared to SCD in situ .A small percentage of about 4 % (for Lugano) to 9 % (for Basel) of all days with contradictory results from the forward and backward gap-filling approach (e.g.SCD F snow and SCD B snowfree ) compared to SCD in situ (e.g.SCD in situ snow ) was registered.
The overall positive bias could not be related directly to the amount of cloudy pixels over the whole period (cloud cover Basel: 50.6 %, Lugano: 38.8 %, Samedan: 42.3 %), introducing uncertainty in filling the gaps with either snow or no-snow.The frequency of cloud occurrence and its temporal continuity has probably a higher impact on the accuracy of the number of snow days compared to in situ results.This could be explained by the increase of the uncertainty when the number of consecutive days of cloud obscuration increases.An isolated snow event followed by a period of continuous cloud coverage could lead to a certain overestimation of snow days, which mainly would happen during the transition period in autumn and spring.In contrary, an underestimation might occur when snow falls after a snow free period and cloud coverage remains stable over a region.
The difference between monthly SCD in situ and SCD MODIS shows a seasonal variation with largest differences between October and April.In contrast, differences are smaller during the months with little or no snow (June to September).Parajka and Bl öschl (2006) related their seasonal performance pattern of the MODIS analysis to the overall snow coverage of Austria, and explained above average errors with above average snow coverage.This was not confirmed in our study, given the better agreement for January at Basel (d = −0.30days) even though snow coverage was relatively high (9.2 days).Only in the Southern Alps in Lugano, the largest discrepancy in December (d = −2.10days) and January (d = −3.60)might be related to the larger number of 4.8 and 5.2 snow days, respectively.Introduction

Conclusions References
Tables Figures

Back Close
Full Several previous studies described larger MODIS product (e.g.MOD10A2) errors in autumn and smaller errors in spring (Vikhamar and Solberg, 2003;Simic et al., 2004;Parajka and Bl öschl, 2006), which is also supported by our results.These studies linked the larger winter errors to the MODIS snow algorithm and pointed out the need to correct for tree and surface shading effects in winter when solar zenith angles are sub-optimal.Other studies such as Gao et al. (2010) and Klein and Barnett (2003) found the lowest accuracies in the transitional months and refer to more changeable and patchy snow distribution, detected by the ground station but not necessarily classified in the corresponding image pixel (i.e.sub-pixel effect).This is apparent for Samedan in the Central Alps, where a positive bias is correlated with months of almost total snow cover (December to March) at this altitude.In contrary, distinct deviations occur during the snow accumulation and ablation phase.In mountainous regions, steep elevation gradients are common and one single MOD10C1 grid cell of 0.05 degree resolution may include both valleys and mountain peaks with e.g.elevation differences up to 1200 m such as around Samedan.As a consequence, differences are expected when a MOD10C1 grid cell is compared to in situ point observations in complex topography, which could explain the lower agreement at Samedan (c = 0.82) compared to Basel (c = 0.90) and Lugano (c = 0.93).This overestimation is also underlined by the percentage of MODIS grid cells defined by the forward and backward gap-filling approach as snow covered (3.7 % of all days) instead of snow free.Furthermore, single snow events, which occur mostly during the transition periods at this altitude, are likely to propagate errors in the final gap-filled product due to the gap-filling technique.This particularly applies to periods with a large number of consecutive overcast days.On a daily basis, the good overall agreement of around 88 % (Samedan) to 94 % (Lugano) between SCD in situ and SCD MODIS is comparable with results from North America based on the MOD10A1 snow product which was compared with in situ measurements from the automated Snowpack Telemetry (SNOTEL) network (Klein and Barnett, 2003).

Conclusions References
Tables Figures

Back Close
Full This study presented inter-annual variations of snow cover days (SCD) over Switzerland for the period 2000 to 2010 based on the MOD10C1 snow product (SCD MODIS ).The total annual SCD were retrieved by applying a gap-filling approach to reduce cloud cover obscuration.The gap-filling methodology proposed substitutes cloud covered pixels accounting for around 30 % (Lugano) to 50 % (Samedan) of all pixels over the whole period.Through the presented method, annual snow day maps were derived from 2000 to 2010 over Switzerland.The MOD10C1 product with a moderate grid resolution of 0.05 degree was shown to be adequate to describe the temporal and spatial variation of snow days over Switzerland on a yearly basis.
An accuracy assessment based on three in situ observation sites representing different climatological regimes in Switzerland (Basel, Samedan and Lugano) demonstrated the potential of the SCD MODIS .The correlation between the satellite-derived annual number of snow days and those calculated from in situ observations (SCD in situ ) is high (0.88) with a mean absolute difference of −6.6 days for the three stations combined.To better understand the annual differences in detail, investigations of SCD MODIS on the monthly and daily resolution have been conducted.On a monthly basis, the discrepancies between SCD in situ and SCD MODIS vary with higher mean absolute differences during the snow accumulation period in autumn and smaller differences after spring.These findings are in line with other studies and are clearly apparent for the alpine site of Samedan.
The impact of the gap-filling approach was visualized on a daily basis for all three stations over the entire period from 2000 to 2010.The day to day comparison of both forward and backward gap-filling against SCD in situ indicates an overall agreement of 94 % and 88 % for the Southern Alps and the Central Alps, respectively.In 1.2 % of all days the MODIS gap-filling approach has erroneously classified snow free days as snow covered for the Swiss Plateau and in 3.7 % of all days for the Central Alps, respectively.

Conclusions References
Tables Figures

Back Close
Full For further discussing the spatiotemporal variability of snow days on a regional to local scale, other higher spatial resolution products from MODIS or other sensors might be used.The longer record of satellite data such as from NOAA AVHRR might provide more significant long term information on the change and variability of snow days over Switzerland although other challenges would arise.Additionally, the use of high temporal resolution data such as from geostationary satellites (e.g.Meteosat) could be of interest for the development of a combined gap-filling approach.Introduction

Conclusions References
Tables Figures

Back Close
Full  Full  Full  Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Within this study the MODIS/Terra Snow Cover Daily L3 Global 0.05 degree Climate Modeling Grid (CMG) Version 5 (MOD10C1) product was used, obtained from the National Snow and Ice Data Center (NSIDC).The data set spans the period from the 1 October 2000 to 30 September 2010.Version 5 (V005), also known as Collection 5, is the most current version of data available from NSIDC.In the MOD10C1 product, a binning algorithm maps MOD10A1 daily snow cover data at 500 m resolution into a grid of a 0.05 degree (CMG) and calculates snow and cloud percentages, Quality Assessment (QA), and a confidence index based on the mapping results.The algorithm generates these parameters based on the total number of observations of a class Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Figure 1
Figure 1 shows the SCD MODIS maps binned into 15 classes on a 25-days interval basis.The SCD MODIS maps highlight the number of snow days for the period of 1 October to 30 September for the years 2000-2010, representing different snow conditions.It is clearly visible that the SCD MODIS product reflects the topography of Switzerland with larger number of snow days in higher altitudes and less days with snow cover in the Swiss midlands and in Southern Switzerland.Regions with less than 25 days of snow cover per year are mostly found in the Rhein Valley, in France and in the plain of the Po River in Northern Italy.The inter-annual spatial variations are most apparent on the Swiss Plateau whereas in the Central Alps the year to year variability is marginal.The years 2001/2002 and 2006/2007 are outstanding with a relatively small number Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | MODIS and in situ correspond quite well.Periods with complete snow cover (December, January, February, March) are generally captured by SCD MODIS in every year.A somewhat lower agreement is found during the transition periods in fall (October, November) and spring (April, May).Differences of more than 5 days per month are registered in November (e.g.2002), December (15 days in 2001), April (19 days in 2002) and May (2003).Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Fig. 3 .Fig. 4 .
Fig. 3. Temporal variability of the total yearly number of days with snow from 2000-2010 for Samedan.

Table 1 .
Location of the selected in situ snow stations.

Table 2 .
SCD MODIS versus SCD in situ on a yearly basis (d = mean difference of SCD in situ − SCD MODIS , s = standard deviation, c = coefficient of correlation).SCD in situ indicates the averaged number of days with snow over the 10 yr and CC is the averaged annual cloud cover in percent derived from the MOD10C1 product.

Table 4 .
Confusion matrix for SCD MODIS compared with in situ derived snow days (SCD in situ ) from 1 October 2000 to 30 September 2010 on a daily basis for all three validation sites.