In this paper four simple computationally inexpensive, direct insertion data assimilation schemes are presented, and evaluated, to assimilate Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover, which is a binary observation, and Advanced Microwave Scanning Radiometer for Earth Observing System (EOS) (AMSR-E) snow water equivalent (SWE) observations, which are at a coarser resolution than MODIS, into a numerical snowevolutionmodel. The four schemes are 1) assimilate MODISsnow cover on its own with an arbitrary 0.01 m added to the model cells if there is a difference in snow cover; 2) iteratively change the model SWE values to match the AMSR-E equivalent value; 3) AMSR-E scheme with MODISobservations constraining which cells can be changed, when both sets of observations are available; and 4) MODIS-only scheme when the AMSR-Eobservations are not available, otherwise scheme 3. These schemes are used in the winter of 2006/07 over the southeast corner of Colorado and the tri-state area: Wyoming, Colorado, and Nebraska. It is shown that the inclusion of MODIS data enables the model in the north domain to have a 15% improvement in number of days with a less than 10% disagreement with the MODIS obser- vation 24 h later and approximately 5% for the south domain. It is shown that the AMSR-E scheme has more of an impact in the south domain than the north domain. The assimilation results are also compared to station snow-depth data in both domains, where there is up-to-a-factor-of-5 underestimation of snow depth by the assimilation schemes compared with the station data but the snowevolution is fairly consistent.
lation and forecast systems. The assimilation of real MODIS data will be investigated in future works.
Seasonal snowpack modeling is a crucial issue for a large range of applications, including the forecast of natural haz- ards such as avalanches or floods, or the study of climate change (e.g., Durand et al., 1999; Lehning et al., 2006; Bavay et al., 2013). The most sophisticated detailed snowpack mod- els represent the evolution of snow microstructure and the layering of snow physical properties (Brun et al., 1989, 1992; Jordan, 1991; Bartelt and Lehning, 2002; Vionnet et al., 2012) in response to meteorological conditions. Despite con- stant efforts to improve these models, large uncertainties re- main in the representation of the snow physics, as well as in the meteorological forcings (Carpenter and Georgakakos, 2004; Essery et al., 2013; Raleigh et al., 2015). These uncer- tainties are highly amplified when propagated to avalanche hazard models (Vernay et al., 2015). For operational appli- cations, the assimilation of observations can help reduce the impact of the model and forcing uncertainties in the snow- pack simulations (e.g., Dechant and Moradkhani, 2011).
These studies have suggested that, most of the time, assim- ilating snowobservations may be useful to improve snow- pack estimation. SWE or SD assimilation generally out- performs the assimilation of SCF only, except from An- dreadis and Lettenmaier (2006) because of large errors in the AMSR-E SWE products. The assimilation of both com- bined revealed larger benefit by mitigating sensors limita- tions. Recently, Navari et al. (2016) investigated the assim- ilation of (synthetic) ice surface temperature while Dumont et al. (2012) also experimented the assimilation of albedo retrievals, both from optical sensors. Dumont et al. (2012) obtained a mass balance RMSE decrease of up to 40 % as- similating albedo data. However, satellite snow products are derived using retrieval algorithms which are not perfect and, perhaps more importantly, not physically consistent with the snowpack model used for the data assimilation. For this rea- son, and as advocated by Durand et al. (2009) who tested the assimilation of in situ microwave radiance observations, assimilating the original satellite radiance data should be pre- ferred when possible.
As already pointed out above, the observational uncer- tainty is largest for the point observations, but neither a mean over a snow course can give a “true” mean of the snow conditions within, for example, a 1 × 1 km grid cell. Thus, even with a hypothetically perfect model, no perfect match with observations can be expected. Considerable spa- tial subgrid variability in SWE and SD is caused, among others, by local precipitation patterns and by wind redistri- bution of snow, which again is dependent on the type of vegetation and topography (e.g. Armstrong and Brun, 2008; Clark et al., 2011). As an example of the observational un- certainty, the high-resolution SD measurements across the Hardangervidda mountain plateau (Ragulina et al., 2011; see Sect. 1) showed that a single point measurement of SD has typically an 80 % confidence range of −60 to +70 % in es- timating the mean SD over 1 km subsections of the approx- imately 80 km long transects. By replacing the single point measurement by a mean over a snow course, (by taking a mean SD of 30 samples) the typical 80 % confidence range is reduced to ± 10 %. This example illustrates the generally better accuracy in the HPC-data, which is mostly based on snow courses, than in the point measurements of SD taken at
abstract: One hitherto intractable problem in studying mast seed- ing (synchronous intermittent heavy flowering by a population of perennial plants) is determining the relative roles of weather, plant reserves, and evolutionary selective pressures such as predator sati- ation. We parameterize a mechanistic resource-based model for mast seeding in Chionochloa pallens (Poaceae) using a long-term individ- ually structured data set. Each plant’s energy reserves were recon- structed using annual inputs (growing degree days), outputs (flow- ering), and a novel regression technique. This allowed the estimation of the parameters that control internal plant resource dynamics, and thereby allowed different models for masting to be tested against each other. Models based only on plant size, season degree days, and/ or climatic cues (warm January temperatures) fail to reproduce the pattern of autocovariation in individual flowering and the high levels of flowering synchrony seen in the field. This shows that resource- matching or simple cue-based models cannot account for this ex- ample of mast seeding. In contrast, the resource-based model pulsed by a simple climate cue accurately describes both individual-level and population-level aspects of the data. The fitted resource-based model, in the absence of environmental forcing, has chaotic (but often statistically periodic) dynamics. Environmental forcing syn- chronizes individual reproduction, and the models predict highly variable seed production in close agreement with the data. An evo- lutionary model shows that the chaotic internal resource dynamics, as predicted by the fitted model, is selectively advantageous provided that adult mortality is low and seeds survive for more than 1 yr, both of which are true for C. pallens. Highly variable masting and chaotic dynamics appear to be advantageous in this case because they reduce seed losses to specialist seed predators, while balancing the costs of missed reproductive events.
Abstract. Surface snow density is an important variable for the surface mass balance and energy budget. It evolves ac- cording to meteorological conditions, in particular, snowfall, wind, and temperature, but the physical processes govern- ing atmospheric influence on snow are not fully understood. A reason is that no systematic observation is available on a continental scale. Here, we use the passive microwave obser- vations from AMSR-E satellite to retrieve the surface snow density at Dome C on the East Antarctic Plateau. The re- trieval method is based on the difference of surface reflec- tions between horizontally and vertically polarized bright- ness temperatures at 37 GHz, highlighted by the computa- tion of the polarization ratio, which is related to surface snow density. The relationship has been obtained with a mi- crowave emission radiative transfer model (DMRT-ML). The retrieved density, approximately representative of the top- most 3 cm of the snowpack, compares well with in situ mea- surements. The difference between mean in situ measure- ments and mean retrieved density is 26.2 kg m −3 , which is within typical in situ measurement uncertainties. We apply the retrieval method to derive the time series over the pe- riod 2002–2011. The results show a marked and persistent pluri-annual decrease of about 10 kg m −3 yr −1 , in addition to atmosphere-related seasonal, weekly, and daily density vari- ations. This trend is confirmed by independent active mi- crowave observations from the ENVISAT and QuikSCAT
the Interactive Multisensor Snow and Ice Mapping Sys- tem (IMS) produced by the US National Ice Center (NIC; Ramsay, 1998) is the fusion of many kinds of optical and microwave data, including Advanced Very High Res- olution Radiometer (AVHRR) data, SSM/I data, AMSR- E data, Geostationary Operational Environmental Satel- lite (GOES) data, Polar Operational Environmental Satel- lite (POES) data, european geostationary meteorological satellite (METEOSAT) data, Japanese Geostationary Me- teorological Satellite (GMS) data, the National Centers for Environmental Prediction (NCEP) model data, US Air Force (USAF) snow and ice analysis data, and so on. The IMS is also jointly supported by the US National Oceanic and Atmospheric Administration (NOAA), the US Navy, and the US Coast Guard. Regardless of cloud cover, the IMS produces near-real-time products with spatial resolu- tions of ∼ 1, ∼ 4, and ∼ 24 km, which provide the input for atmospheric forecast models (Brown et al., 2014). In addi- tion, the Air Force Weather Agency (AFWA) and National Aeronautics and Space Administration (NASA) snow algo- rithm (ANSA) blends AMSR-E, MODIS, and Quick Scat- terometer (QuikSCAT) data products (Foster et al., 2011). Since ANSA integrates optical, passive, and active mi- crowave data, it can map the snow-covered area (SCA), the fractional snow cover (FSC), the SWE, and the snowmelt area. The combined products draw together the respective advantages of each of the component products to improve the accuracy, quality, and spatio-temporal continuity of snow cover. Several problems encountered when the component products are used alone, including cloud cover and low ac- curacy, have been solved. The combined products provide more information on the state of snow cover than each of the component products. However, the primary disadvantage of the combined products is the poor spatial resolution (> 1 km; Tait et al., 2000).
In this perspective, remote sensing can be useful for re- constructing recent changes of snow cover extent, distribu- tion and duration in wide regions such as northern Italy. Moreover, the use of snow cover maps for hydrological pur- poses is an effective tool, since the combination of large- scale information with local estimations or measurements of snow features makes it possible to estimate the snow wa- ter equivalent (SWE) stored within a river basin (Molotch and Margulis, 2008; Bavera and De Michele, 2009). Mod- erate Resolution Imaging Spectroradiometers (MODIS) em- ployed by Terra and Aqua satellites provide a Snow Cov- ered Area product (SCA) with 500 m and daily resolutions, which consists of binary maps whereby snow is detected at the pixel scale. The accuracy of MODISSnow Cover Prod- ucts depends on region, season, snow condition and land cover type (Klein and Barnett, 2003; Maurer et al., 2003; Simic et al., 2004; Zhou et al., 2005; Tekeli et al., 2005; Ault et al., 2006; Parajka et al., 2006; Hall and Riggs, 2007; Liang et al., 2008). In Europe, Parajka et al. (2006) compared daily MODISsnow maps with in situ data of 754 climate stations over the whole of Austria, reporting an average classification accuracy of 95 % on cloud-free days. The wide and hetero- geneous Austrian territory presents similarities with our case study. Accordingly, we expect a similar quality of MODIS SCA product for northern Italy. However, the primary limi- tation in using MODISsnow products is that no information on ground conditions is available in areas hidden by cloud. During a year, clouds may obscure most of the study area restricting the potential of using such snow cover images. For example Parajka et al. (2006) indicated that, on average, clouds obscured 63 % of Austria in daily snow maps from February 2000 to December 2005. The possibility of benefit- ing from a reliable product with daily temporal resolution is therefore conditioned by the ability to estimate the presence of snow in overcast conditions. With this aim, several proce- dures for MODIS products have been developed and tested in different regions (Parajka and Blöschl, 2008; Gafurov and Bárdossy, 2009; Wang et al., 2009; Hall et al., 2010; Parajka et al., 2010; Paudel and Andersen, 2011). Such methods are based on a spatio-temporal combination of MODIS data and they can generate cloudless images having accuracy compa- rable to that of the source product. Contrary to the case stud- ies in Gafurov and Bárdossy (2009) (Kokcha basin: eleva- tions range from 416 m a.s.l. up to 6383 m a.s.l., about 75 % of the basin area lies above 2000 m a.s.l.) and Paudel and An- dersen (2011) (Trans-Himalayan region: 96 % of the area lies in the elevation zone above 3000 m a.s.l., with 43 % of the area above 5000 m a.s.l.), many European rivers drain basins which cover altitudes from the lowlands up to 4000 m a.s.l. in
Abstract. A spatial filter (SF) is used to reduce cloud coverage in Moderate Resolution Imaging Spectroradiome- ter (MODIS) 8-day maximum snow cover extent products (MOD10A2) from 2000–2007, which are obtained from MODIS daily snow cover extent products (MOD10A1), to assess the topographic control on snow cover fraction (SCF) and snow cover duration (SCD) in the Quesnel River Basin (QRB) of British Columbia, Canada. Results show that the SF reduces cloud coverage and improves by 2% the accu- racy of snow mapping in the QRB. The new product devel- oped using the SF method shows larger SCF and longer SCD than MOD10A2, with higher altitudes experiencing longer snow cover and perennial snow above 2500 m. The gradient of SCF with elevation (d(SCF)/dz) during the snowmelt sea- son is 8% (100 m) −1 . The average ablation rates of SCF are similar for different 100 m elevation bands at about 5.5% (8 days) −1 for altitudes <1500 m with decreasing values with elevation to near 0% (8 days) −1 for altitudes >2500 m. Dif- ferent combinations of slopes and aspects also affect the SCF with a maximum difference of 20.9% at a given time. Correlation coefficients between SCD and elevation attain 0.96 (p<0.001). Mean gradients of SCD with elevation are 3.8, 4.3, and 11.6 days (100 m) −1 for the snow onset sea- son, snowmelt season, and entire year, respectively. The SF decreases the standard deviations of SCDs compared to MOD10A2 with a maximum difference near 0.6 day, 0.9 day, and 1.0 day for the snow onset season, snowmelt season, and entire year, respectively.
Studies modelling ROS events mostly analyse the daily to weekly timescales and successfully reproduce the temporal evolution of snow water equivalent over several days (Marks et al., 1998, 2001). This suggests that snowpack-related pro- cesses during ROS events are sufficiently understood. Con- sequently, one can estimate rather precisely how much water will be available for snowpack runoff (Marks et al., 1998; Mazurkiewicz et al., 2008), but the temporal dynamics of the release of meltwater on the sub-daily timescales has sel- dom been investigated in detail. This knowledge is essential, however, to estimate the response in streamflow discharge in catchments and to assess flood risks from ROS events. In Rössler et al. (2014), the meteorological circumstances lead- ing to this event have been studied in combination with a hydrological catchment scale model to simulate streamflow discharge in one of the affected areas. To reproduce the rapid peak discharge in the event, considerable recalibration of the hydrological model setup was required. For example, rela- tively simple single layer snow models, which are often used in hydrological model frameworks, were unable to follow the snow cover dynamics without significant calibration of snow-related parameters for this particular situation (Rössler et al., 2014).
The observation data used for gridding have been extracted from both national (Finnish Meteorological Insti- tute, FMI) and international source (European Climate Assessment & Dataset (ECA&D) databases (Klok and Klein Tank, 2009)). Meteorological station data from the neigh- bouring countries (i.e., Sweden, Norway, Russia and Esto- nia) were used to reduce the uncertainty near the border regions of Finland. The snow observation network is rela- tively evenly distributed across the study area (Figure 1). On average, the number of stations per day available for interpo- lation was 351 over the period 1961 –2014. The station den- sity dropped towards the 21st century due to the automation of the measurement stations. In Finland, the transition from manual to automated measurements has been gradually pro- gressing from the end of the 1990s until now, and at the moment, approximately 50% of the snow measuring stations are automated. The interpolation errors remained relatively stable throughout the years (the mean root mean square error [RMSE] over 1961 –2014 was 5.3 cm, excluding months from June to August).
and Gao et al., 2010) This might be due to differences in evaluation statistics used, the amount of stations with ground truth observations used to calculate the accu- racies and their distribution throughout the watersheds, and/or the fact that the other studies included the entire water year including summer where mapping accuracies are high because there is no snow and therefore omis- sion and commission errors are practically non-existent. 3. The time interpolation method achieves the best over- all accuracy, and provides consistently better evaluation statistics – better even than those of the original images. This is likely due to the seasonal persistence of snow at the SNOTEL sites (e.g., see Fig. 2), and the result may not be representative of mapping accuracy near the snow line. We will return to this point later in the paper. 4. The other three methods have varying results, with LWLR achieving the worst PC, TS, B, and H statis- tics. The lower accuracy results for the LWLR might be caused by the size of the window used for the analysis; a smaller window has better accuracy results but removes less clouds (L´opez-Burgos, 2010). The most appropri- ate tradeoff between these two qualities can be chosen by the user based on the future applications of the fi- nal images. The lower accuracy of the LWLR may also be due to the thresholds used to decide if a pixel has snow or not. One way to improve this would be for the user to give more weight to the minimization of the sum of conditional probabilities of commission errors than omission errors (refer to Sect. 4.4) since this would give more conservative results for snow cover. It is better to plan for less SWE in the watershed (underestimation) and find that there is more usable SWE than to plan for a higher amount of SWE (overestimation) and find out there was actually less. Development of a method that improves this step remains a topic for future work. 5. The sequential approach is second best in terms of ac-
Both T and P are modulated by ENSO in the region (Masiokas et al., 2006; Santos, 2006; Zamboni et al., 2011; Meza, 2013; Valdés-Pineda et al., 2015b). In the tropics, El Niño conditions relate to low precipitation and warmer air temperatures, whereas in higher latitudes El Niño relates to higher precipitation and warmer temperatures (Garreaud et al., 2009). Our results suggest that this ENSO influence on temperature and precipitation most affects snow patterns north of 30 ◦ S. The trends of decreasing P and increasing T we found in these latitudes are generally consistent with previous studies (Vuille and Bradley, 2000; Bradley, 2004; Quintana, 2012; Salzmann et al., 2013; Kluver and Leathers, 2015). South of 35 ◦ S, SAM is the climate index best cor- related with SP. The influence of SAM on South American climate has been documented in other studies (Vera and Sil- vestri, 2009; Fogt et al., 2010), and it has been found to be an important control on radial tree growth (ring width) in the same latitude range where we identified its influence on snow (Villalba et al., 2012). These relations between SAM and both tree rings and snowpack may indicate that SAM is influencing tree growth through its effect on snowpack char- acteristics. More specifically, snowpack can influence grow- ing season length and water availability during the growing season. Given the relatively dry summers of this region, in- creased snowpack would be expected to increase radial tree growth by providing a source of moisture during the grow- ing season. We present the first study relating SAM to snow response in South America, and these relations suggest that tree rings could be a viable method to reconstruct snowpack in the southern Andes (Woodhouse, 2003). However, we note that the short duration of our study time period is insufficient for complete analyses of how snow patterns relate to ENSO and SAM. During the period of our study SAM was mainly positive (Fig. 7b), so a long-term study is required to deter- mine whether different phases of SAM have different snow cover responses.
Quintana-Seguí, P., P. Le Moigne, Y. Durand, E. Martin, F. Habets, et al., 2008: Analysis of near-surface atmospheric variables: validation of the SAFRAN analysis over France.
J. Appl. Meteorol. Climatol., 47, 92-107, doi:10.1175/2007JAMC1636.1.
Robock, A., K.Y. Vinnikov, G. Srinivasan, J.K. Entin, S.E. Hollinger, N.A. Speranskaya, S. Liu, and A. Namkhai, 2000: The global soil moisture data bank. Bull. Am. Meteorol.
Notwithstanding that other geomorphological properties than slope angle influencing snow patterns are important on scales smaller than the grid size of COSERO (see Sect. 1.1), slope was selected as driving force for the model. One has to be aware that this is a simplification and under realistic con- ditions snow might not necessarily be transported only on the steepest route (Bernhardt and Schulz, 2010; Winstral et al., 2002). Also, the response of glaciers might change when finer spatial resolutions are applied. In a study at the Blaueis- ferner glacier, Germany, Bernhardt et al. (2010) found addi- tional snow getting blown on glacier surfaces when they used a 30 m resolution. At the coarser resolution of 300 m this re- sult could not be found, though. However, as indicated in the introduction, the 1 × 1 km scale is often used when hydro- logical models are applied to medium or large catchments of hundreds or thousands of square metres, not only because of a variety of existing input data sets on that resolution but also for performance issues. The results of this study show that the model operating on that scale is able to reproduce the spatial snow distribution patterns in the catchment and prevent the model from accumulating snow over several years.
This melt model is the ﬁ rst to incorporate biological forcing and is predictive, given cell size and pigment con- centration, along with snow physical properties and meteorological data. The model could readily be expanded to incorporate additional algal species or pigment types. We do not currently have suf ﬁ cient empirical data to fully validate the model; however, we note that each section of the model represents an adapted form of a model previously formulated and validated in the literature [Pottier et al., 2005; Libois et al., 2013; Brock and Arnold, 2000]. There are no known empirical data sets that include all the necessary observations required to truly validate our model. We intend to collect our own ﬁ eld data for this speci ﬁ c pur- pose. This paucity of data along with the current interest in bioalbedo highlights a community requirement for standardized biological and glaciological measurements. Nevertheless, there exist several partial data sets we were able to use ( ﬁ lling information gaps with predicted or literature values) that allow us to make com- parisons between ﬁ eld and modeled data. Two data sets were employed for this purpose: Painter et al. ’ s  spectral re ﬂ ectance data from algal snow in California and Lutz et al. ’ s  broadband measure- ments from Mittivakkat Gletscher. Both studies quantify cell concentrations in snow but crucially omit infor- mation regarding the distribution of these cells in the snowpack. Qualitative descriptions in these papers, along with our own ﬁ eld observations, suggest that the majority of the cells are likely to be concentrated within a 1 mm surface layer, but the ﬁ eld samples likely included snow between the surface and several cen- timeter depth. This is important for the model as a given number of cells has a greater impact on albedo when concentrated on the snow surface than when distributed through a depth pro ﬁ le in the snow. Furthermore, these studies only provide qualitative descriptions of the snow physics at their ﬁ eld sites. Since our model is built upon well-known snow radiative transfer equations that have been well tested, we are con ﬁ dent in its ability to simulate snow physics. There is also no quantitative assessment of inorganic impurities such as BC in the snowpack. These unknown parameters had to be estimated within sensible ranges, guided by existing literature. For example, SSA and density of melting snow have been characterized by Domine et al. , Gallet et al. , Yamaguchi et al. , and Matzl and Schneebeli  among others. We therefore present these analyses to demonstrate the utility of our model for both forward and inverse modeling of algal snow albedo and to illustrate the need for a standardized suite of bioalbedo mea- surements in ﬁ eld studies. Full model validation must wait until we have undertaken ﬁ eld work for this speci ﬁ c purpose.
The main difficulty in the assimilation of PMW satellite observations in boreal forest areas is the quantification of all the contributions that affect the measured signal. PMW satellite observations have a low spatial resolution ( ∼ 10 × 10 km 2 ) and satellite sensors measure many contributions in addition to the PMW emission from the volume of the snowpack (vegetation canopy, ice crust, frozen/unfrozen soil, lakes, moisture in the snow, topography, etc.) (Kelly et al., 2003; Koenig and Forster, 2004). In boreal areas, the PMW emission from the forest canopy within a pixel can contribute up to half of the PMW signal measured by satellite sensors (Roy et al., 2012, 2016). This contribution does not only de- pend on the fraction of forest cover, but also on the biomass (liquid water content, LWC), the vegetation volume, and the canopy structure (stem, leaf, trunk) (Franklin, 1987). To ad- just snowpack model simulations, several studies suggest us- ing radiative transfer models, coupled to a snowpack model, to take into account the different contributions to the PMW signal at the top of the atmosphere and to directly assimilate PMW satellite observations (Brucker et al., 2011; Durand et al., 2011; Langlois et al., 2012; Roy et al., 2016). How- ever, the assimilation of PMW must be used with care, and
100% snow cover in the North Park of Colorado where surface measurements were made during the Cold Land Processes Experi- ment. Their grain size errors were 5% to 40%. Grain size has also been mapped over Antarctica using a normalized difference approach between the MODIS band 1 (620 –670 nm) and MODIS Band 2 (841– 876) radiances from pure snow pixels ( Scambos et al., 2007 ). The precursor model to MODSCAG using imaging spectrometer data retrieved grain size and albedo with RMS errors of 50 µm for grain size and 0.02 for albedo ( Painter et al., 2003 ). The Version 004 MODIS release contained a snow albedo product ( Klein & Stroeve, 2002 ) that assumes that the mapped pixel has 100% snow cover; it was evaluated on the fairly level Greenland ice sheet ( Stroeve et al., 2005 ) where the assumption that the snow cover is uniform across the pixel is likely correct. The RMS error between the MODIS retrievals and automated weather stations was ~7%. Liang et al. (2005) evaluated a “direct retrieval algorithm ” for snow albedo over Greenland with errors of 4%. Experiments with the Global Imager (GLI) on ADEOS-II show quantitative results for retrieval of snow grain size and impurity content ( Aoki et al., 2007 ). The retrieval algorithms use absolute re ﬂectance values in discrete channels ( Stamnes et al., 2007 ), thereby requiring that the local solar illumination angle on the pixel is known, and they also apply only to pixels with 100% snow cover. Their algorithm has been tested over relatively level areas at four sites in Alaska and Hokkaido.
Using proposed methodology it was possible to estimate the snow cover dynamics of mountainous regions. The original MODISsnow cover product that contains cloud covered pixels can be processed using this methodology, meaning that snow cover data with 500 m spatial and daily resolution can be prepared. MODIS also offers eight day composite snow cover information with little or no cloud cover, but this is the maximum snow cover extent for eight composite days. Yet, a considerable fraction of snow could melt or fall within eight days time period, which may not be enough information for higher temporal resolution modelling purposes. This is why the daily snow cover information is very valuable also for model calibration and for validation purposes. Such information can be extremely helpful when modeling available water resources in moun- tainous areas where snowmelt in spring or summer becomes a valuable raw material for energy production, agriculture and for drinking purposes in lowland areas.
Optically shallow snowpacks (SD < 0.14 m) in New Hampshire occur predominantly at the beginning and end of the snow season and occasionally during mid-winter thaws. The albedo decay process of optically shallow snowpacks is more complex compared to deeper snowpacks due to the additional influence of underlying terrain, and accurately parameterizing albedo evolution in shallow snow presents a challenge (Slater, 1998). Both shallow and deep snowpacks are subject to surface albedo degradation via snow metamorphism and particulate accumulation. However, albedo reduction of thin snow cover accelerates when solar radiation penetrates through the snow pack and the lower albedo of the underlying substrate increases absorption of solar radiation. As snow cover decreases, surface albedo approaches that of the substrate until snow cover disappears altogether (Gray and Landine, 1987). Additionally, the presence of adjacent bare ground further accelerates snow melt where snow remains. In the CoCoRaHS- Albedo dataset, correlation between snow depth and surface albedo for shallow snowpacks implies snow depth represents some of the complex influences on snow surface albedo brought about by visible ground beneath the snowpack and adjacent bare ground. In addition, the large scatter and lower albedo values (albedo < 0.45) in the residuals plot for shallow snowpacks (Figure 8a) are indicative of the influence of the underlying substrate.