Estimation of Continental-Basin-Scale Sublimation in the Lena River Basin , Siberia

The Lena River basin in Siberia produces one of the largest river inflows into the Arctic Ocean. One of the most important sources of runoff to the river is spring snowmelt and therefore snow ablation processes have great importance for this basin. In this study, we simulated these processes with fine resolution at basin scale using MicroMet/SnowModel and SnowAssim. To assimilate snow water equivalent (SWE) data in SnowAssim, we used routine daily snow depth data and Sturm’s method. Following the verification of this method for SWE estimation in the basin, we evaluated the impact of snow data assimilation on basin-scale snow ablation. Through validation against MODIS snow coverage data and in situ snow survey observations, we found that SnowAssim could not improve on the original simulation by MicroMet/SnowModel because of estimation errors within the SWE data. Vegetation and accumulated snowfall control the spatial distribution of sublimation and we established that sublimation has an important effect on snow ablation. We found that the ratio of sublimation to snowfall in forests was around 26% and that interannual variation of sublimation modulated spring river runoff.

Problems exist in relation to snow estimation because of difficulties in satellite-based measurements of snow mass.For instance, Foster et al. [33] reported a large error in radiance retrieval from a Siberian taiga forest because of the forest canopy.Point-scale snow surveys are more accurate than snow models or satellite retrievals; however, it is difficult to extend such surveys to larger scales, such as that of a continental basin.To estimate continental-basin-scale snow distribution, forcing variables for a land surface model are very important, and there are two principal methods by which to obtain such variables over a domain.One introduces independent datasets such as reanalysis data and the second interpolates surface meteorological data across the domain.Both methods have specific advantages and disadvantages.
Liston and Hiemstra [34] evaluated changes of pan-Arctic snow coverage over 30 years using their developed MicroMet/SnowModel and SnowAssim model with 10 km gridded forcing variables from NASA's Modern-Era Retrospective Analysis for Research and Applications (MERRA) reanalysis data [35], and they successfully detected snow mass trends in the Arctic region.Stuefer et al. [36] demonstrated that in situ snow observation was useful for the evaluation of snow distributions over a large domain in Arctic Alaska.

Advances in Meteorology
They used the SnowAssim model with meteorological station data to evaluate these distributions and demonstrated the effectiveness of in situ snow survey data in overcoming the limited meteorological forcing in the Arctic.Their research clearly showed that point-scale snow observations could improve the accuracy of large-scale snow distributions.
Routine measurements of snow water equivalent (SWE) are rarely available at high latitudes, but daily snow depth measurements from weather stations are available instead.Sturm et al. [37] developed empirical methods to determine SWE using snow depth data.Based on their method, Reichle et al. [38] estimated SWE in the Northern Hemisphere to validate their MERRA land products.However, Sturm's method could not be validated against Siberian snow data because the original validation dataset covered only North America and Europe.Thus, it is important to evaluate the applicability of Sturm's method to the vast region of Siberia.
Peterson et al. [39] reported that the annual river runoff for the six largest basins on the Eurasian continent increased by 2.0 ± 0.7 km 3 annually from 1936 to 1999.Rawlins et al. [40] showed that annual river runoff for the three largest Eurasian basins (the Ob, Yenisei, and Lena) correlated well with cold season (October-April) precipitation from 1966 to 1995.In addition, Kitaev et al. [41] demonstrated that the snow mass over the region of the former Soviet Union increased from 1936 to 1995.In the Lena River basin, Iijima et al. [42] showed that an abrupt increase in active layer depth had a positive correlation with precipitation increase.Furthermore, they described the role of snow as an insulator for the permafrost.However, there were very limited observations of snow depth and the study did not reflect basin-scale snow mass change.In the upper Lena River basin, Suzuki et al. [43,44] showed that snow mass strongly affected the river runoff ratio and organic carbon transport.The Lena River is the second largest provider of freshwater from the land to the Arctic Ocean.Freshwater discharge and organic carbon from the river can modulate sea ice production, ocean circulation, and ecosystem.Estimation of snow mass in the Lena River basin is also important for the evaluation of freshwater and carbon transport to the Arctic Ocean.In order to estimate snow mass, Zhang et al. [45] and Suzuki et al. [46] clarified the importance of sublimation in the winter water balance.The ratio of sublimation loss to total winter snowfall ranged from 20% to 50% in the region.However, their research was limited to a small-scale watershed, and thus it is unclear which factors affect the spatial distribution of sublimation on the scale of the Lena River basin.
At high latitudes, meteorological stations are sparse and located mostly at low elevations.Thus, reanalysis data from operational weather centers have uncertainties and relatively large errors in comparison with data from the midlatitudes of the Northern Hemisphere.Estimations using snow data assimilation provide information on changes of both solid precipitation and active layer processes.These changes ultimately modify the water balance components in permafrost regions.Here, through estimation of SWE from daily routine snow depth data from weather stations, we show the effect of simple snow data assimilation on the water balance of the permafrost-dominated Lena River basin.In addition, we clarify the spatial distribution of sublimation and its role in snow ablation processes.

Study Domain and Forcing
Variables.The topography and vegetation distribution of the study area are shown in Figures 1(a) and 1(b), respectively.The domain size of about 2,500 × 2,000 km (50 ∘ N-74 ∘ N, 110 ∘ E-165 ∘ E) required about 1.493 million 2 km grid cells for the model simulations.The study area encompasses the Lena River basin, which has an area of about 2,500,000 km 2 and generates one of the largest river inflows to the Arctic Ocean.The Lena River flows 4,400 km from its source in mountainous southeastern Siberia to its mouth in the Arctic Laptev Sea.The mountainous areas in the basin have gentle slopes with the highest elevation of around 1,300 m.Most of the basin is relatively flat from the middle to lower reaches.We used digital elevation data (GTOPO30 [47]) and converted the original 30-second grid cells into 2 km cells using ArcGIS 9.2 software.
We used the GLC2000 dataset to set up the vegetation classification within the domain [48].The original GLC2000 vegetation map had 1 km grid cells, but we converted it into 2 km cells with 18 vegetation classifications, which were required for the model parameters of MicroMet/SnowModel.Figure 1(b) and Table 1 show the vegetation distribution within the domain.The major vegetation types were forest and tundra and forest comprised 80% of the area of the Lena River basin.Within this forest, deciduous trees that were mainly larch were predominant.

Dataset.
Here, we describe the details of the forcing and validation datasets used for the simulations.First, for the forcing variables, we used the baseline meteorological dataset (BMDS) archived by [49] and distributed athttp://www.jamstec.go.jp/acdap/Summary.action?downloadList=A20100830-001& selectFile=A20100830-001.The stations in the BMDS are mostly at low elevations and latitudes, as shown in Figure 1(a).Thus, meteorological data from high elevations and latitudes within the domain were very limited.The number of meteorological stations used for the simulations was 47, and the meteorological data were daily mean air temperature, relative humidity, wind speed, precipitation, and snow depth.Therefore, the time step implemented in the simulations was one day.All data were quality controlled and, in particular, precipitation data were modified carefully based on station-specific parameters of wind-induced precipitation gauge undercatch, which causes underestimation of winter precipitation [50].
Second, to assimilate SWE data in the model, we used in situ snow survey observations from two snow courses in mid-March 2000.In that month, the National Polar Research Institute of Japan performed two snow survey courses, and the locations of the snow observation are shown in Figure 1(a) and Table 2.These once-per-year snapshots of observed snow data included snow depth, snow density, and SWE.Finally, to verify the spatial distributions of model estimates, we used monthly composite MODIS MOD10 snow coverage products with 0.05 ∘ resolution.We used these products to overcome the problem of missing snow coverage information in the eight-day composite data.We converted our simulated results from 2 km to 0.05 ∘ resolution for comparison with the MOD10 snow products.

Sturm's SWE Estimation.
SnowAssim requires SWE data, but the weather stations used in this study routinely only report daily snow depths.Therefore, we estimated SWE from the snow depth data using the following method developed by [37], which has snow depth observation and climate snow classes.The basic equation for estimating SWE can be written as follows: where  ℎ,DOY  is bulk snow density for snow depth ℎ  and day of the year (DOY  );   is water density;  1 and  2 are densification parameters for depth ℎ  and DOY  , respectively; and  max , 0 ,  1 , and  2 are parameters that vary with snow class.
To examine the accuracy of SWE estimated using Sturm's method, we compared it with in situ snow observations (Table 2).Figure 2 shows a comparison of observed and estimated SWE at in situ observation sites.There is a clear linear relationship with a large determination coefficient.The slope of the linear regression is close to 1 and the root mean square error (RMSE) is about 24 mm, lower than that of the original paper [37].Thus, we believe that Sturm's SWE method is applicable to our study domain.However, the RMSE tends to become larger at higher values of SWE, and Sturm's method usually provides an overestimate.

Model and Data Assimilation.
Here, we describe the MicroMet/SnowModel [51,52] and SnowAssim model [53] that we used to simulate snow distributions and water and energy balances at each grid cell of the domain.MicroMet/SnowModel is a comprehensive snowevolution modeling system that includes blowing snow transport processes [51,52].This system contains all the physics and dynamics required to simulate snow evolution in both forested and nonforested environments.SnowModel incorporates the first-order physics required to simulate snow evolution in each global snow class (i.e., ice, tundra, taiga, alpine/mountain, prairie, maritime, and ephemeral), as  defined by [54].MicroMet and SnowModel have been used to distribute meteorological variables and evolve snow distribution over a wide range of spatial scales (grid increments ranging from 10 m to 10 km and spatial domains from 100 m to pan-Arctic).The models have been used for a variety of complex landscapes, including the mountains of Colorado, Wyoming, Idaho, Arctic Alaska, Svalbard, Central Norway, Greenland, and Japan.SnowModel incorporates four submodels: MicroMet [51], which defines meteorological forcing conditions; EnBal [55], which calculates surface energy exchanges; SnowPack [52], which simulates snow depth and water equivalent evolution; Advances in Meteorology and SnowTran-3D [56], which accounts for snow redistribution by wind.Each submodel was originally developed and tested for nonforested conditions.For application to Japanese forest areas, the submodels have been modified by Suzuki et al. [57], using the model and measurements [58].
Next, we describe the simple snow data assimilation method (SnowAssim) [53].This method requires plausible plot-scale SWE data.Based on the differences between the simulated and these plausible SWE data, SnowAssim modifies either the precipitation or melt from SnowModel.This is performed by calculating the relative contributions (R) of precipitation and melt during each observation interval using the following: where  and  are accumulated precipitation and snowmelt from times  − 1 to , respectively.The greater of these defines whether it is precipitation or melt that is corrected.During periods in which there are no future observations, or in which both the summed precipitation and melt are zero, the correction factor is defined as unity.An additional requirement of our data assimilation system is that it be spatially distributed.

Snow Data Assimilation Experiment.
To evaluate the basin-scale impact of estimated SWE on routine snow depth observations in our model simulation, we performed two simulations targeting the winter of 1999-2000.In this period, both MODIS snow cover products and in situ snow observations were available and we verified the simulated results with those data.Firstly, we executed a control run for September 1, 1999, to August 31, 2000, without snow data assimilation in the model, hereafter, called the CTL9900 run.Secondly, we performed the same simulation, but incorporating snow data assimilation using SWE estimated by Sturm's method, hereafter, called the ANL9900 run.The estimated SWE was based on daily snow depth data observed by BMDS weather stations.For the snow data assimilation, we chose data at 10day intervals from September 1, 1999, to August 31, 2000.For each 10-day period, we estimated the SWE at the BMDS sites from our routine snow depth observations and those SWE data were assimilated into SnowAssim.

Multiyear Snow Process Experiment.
To evaluate the snow ablation processes in different years, we performed multiyear snow simulations for September 1, 1998, to August 31, 1999, and September 1, 1999, to August 31, 2000.The regulation in the latter period was the same as in the CTL9900 run.We refer to the simulation of winter 1998-1999 as the CTL9899 run.Based on these two simulations, we evaluated the variation of the snow processes in different years.Most river runoff occurred during spring (April through June) (Figure 3(a)).This runoff commenced in May of both years and the accumulated river runoff from October through the following June in each year was 122 and 146 mm, respectively; that is, runoff in winter 1999-2000 was 24 mm greater than in 1998-1999.In accordance with this difference in runoff, winter precipitation in 1999-2000 was 20 mm greater than in 1998-1999.Precipitation had clear interannual variations with maxima recorded in summer.Therefore, the difference in river runoff between the two winters was largely caused by the difference in precipitation.Figure 3(b) reveals the extent of the extremely cold winter air temperatures; for example, the monthly mean air temperature for January is around −30 ∘ C. Mean air temperature for October through June in 1998-1999 (−17.5 ∘ C) was lower than in 1999-2000 (−16.2 ∘ C), and the colder winter of 1998-1999 extended the snow period compared with winter 1999-2000.However, maximum snow depth in both years was similar (Figure 3(e)), which suggests that snowfall did not differ much between the two winters.

Effect of Data Assimilation on Estimated SWE from
Weather Stations.In this section, we assess model performance and the effectiveness of snow data assimilation relative to other independent observations, that is, MODIS satellite-retrieved snow coverage and plot-scale in situ snow observations.First, to demonstrate the basin-scale spatial performance of the simulated snow distribution, we compare the two sets of snow coverage simulated by CTL9900 and ANL9900 with satellite-retrieved monthly composite MODIS snow data [59].Here, we converted the original 2 km gridded simulation results to 0.05 ∘ resolution to match the gridded MODIS data.Figures 4(a)-4(c) show the spatial distributions of monthly snow coverage by MODIS, CTL9900, and ANL9900, respectively.It can be seen that there is no significant difference between these and April 2000, as shown in Figures 4(a), 4(b1), and 4(c1).Figures 4(a1), 4(b1), and 4(c1) show that in April all three datasets indicate that most of the basin was covered by snow.In May, the difference between CTL9900 and ANL9900 is significant, because the latter indicates snow persisted in the high mountain areas of the upper Lena River basin and that Sturm's method [37] for snow data assimilation overestimated the SWE, a trend confirmed by Figure 2. The results showed that the CTL9900 simulation was closer than ANL9900 to the MODIS observations (spatial correlations between MODIS and the CTL9900 and ANL9900 simulations were 0.69 and 0.58, resp.).In June 2000, all three results were similar because much of the  snow coverage had disappeared, although the snow coverage simulated by ANL9900 was larger than that of either MODIS or CTL9900.In Section 2.3, we highlighted that SWE from Sturm's method is an overestimation for higher values of SWE.The overestimation by ANL9900 was caused by the assimilation of overestimated SWE in the upper Lena River basin.In turn, overestimated SWE affected estimated daily precipitation and thus winter precipitation from ANL9900 was poorer than that of the original model run by CTL9900.Overall, we conclude that large errors in the estimation of SWE in the data assimilation lead to poorer simulation results.However, in Section 3.4, we also consider possible problems with the MODIS snow products, because MODIS satellite-retrieved snow coverage includes uncertainties in its retrievals and therefore it requires ground truth.Next, we compare the simulated and observed SWE at point-scale for the in situ snow observation sites shown in Figure 1(b).Figures 5(a) and 5(b) show comparisons of observed and simulated SWE at these sites (listed in Table 1).Figure 5(a) shows a clear linear relationship between the in situ observations and CTL9900 simulation with a slope of 1.03, large determination coefficient of 0.46, and RMSE of 44 mm.In contrast, the relationship between the observations and ANL9900 is unclear and has larger discrepancies.The determination coefficient of 0.22 is smaller and the RMSE of 89 mm is nearly double that of CTL9900.
As shown in Figure 4(c), the snow coverage simulated by ANL9900 remained in the upper Lena River basin.The results shown in Figures 5(a) and 5(b) correspond to those in Figures 4(b) and 4(c).One of the reasons for the large difference between the observations and the SWE simulated by ANL9900 (Figure 5(b)) was the overestimation of SWE by Sturm's method.This is because SnowAssim corrected the precipitation at the BMDS sites, which are mainly located at lower elevations, in order to match the overestimated SWE, and the corrected precipitation was interpolated to the grid points with the inclusion of orographic effects.Therefore, the SWE estimated for higher elevations by ANL9900 was enhanced by orographic effects of precipitation, derived from overestimations of SWE at lower elevations.
Routine snow depth data for SWE were also measured at the weather stations.As mentioned in Section 2.2, the BMDS precipitation was quality controlled for research purposes and calibrated by empirical wind-induced undercatch parameters for each station.Thus, we believe that the quality of the estimated SWE input into SnowAssim was poorer than that of the winter precipitation, owing to the sophisticated BMDS precipitation, which resulted in no improvement in the SnowAssim simulation.
Another consideration is that the BMDS stations were mainly at lower elevations and latitudes.At higher elevations and latitudes, forcing meteorological variables were estimated Advances in Meteorology  by the MicroMet model using Barnes' objective interpolation method [60].Here, when we excluded orographic effects on air temperature, the disappearance of simulated snow coverage was much faster than with the MODIS snow coverage dataset.We also tested the exclusion of orographic effects on precipitation in MicroMet.The simulation results were similar to those of the exclusion of orographic effects on air temperature; snow disappearance was much faster than with the MODIS snow coverage dataset.Thus, despite most of the BMDS being obtained from lower elevations and latitudes, the proper treatment of orographic effects on air temperature and precipitation meant that the results of CTL9900 were closer to the MODIS snow coverage.Overall, the simulated results of CTL9900 were superior to those of ANL9900.This was because of the overestimations of SWE and winter precipitation in ANL9900, especially at high elevations.Thus, we believe that the control run without data assimilation can reproduce snow processes on the basin scale reasonably well.In the subsequent analysis of the winter water balance, we used only the control run for 1998-1999 and 1999-2000.

Sublimation in the
Lena River Basin.Suzuki et al. [57] and Zhang et al. [45] showed that sublimation was important for point-scale water balance during a winter in the upper Lena River basin.Here, we show the distribution of sublimation across the basin and the main spatially controlling factors.
First, we show spatial variations in each winter's water balance components from September 1 to the end of July the following year.Figures 6(a (snowfall, sublimation, snowmelt, and ratio of sublimation to snowfall) in CTL9900 from September 1, 1999, to July 31, 2000.The range of accumulated snowfall was about 50 mm in the southern part of the upper Lena basin to more than 300 mm at higher elevations.In contrast, the range of accumulated sublimation in the basin was smaller (about 50-80 mm).Therefore, the ratio of accumulated sublimation to accumulated snowfall was of the order of about 0.1 to >0.5.Thus, in some parts of the basin, more than half the snowfall was lost to sublimation; these parts had little snowfall.At high elevations, snowfall was heavy and sublimation slight; therefore, the ratio of sublimation loss to snowfall was insignificant and melt dominated the snow ablation.This concurs with the result of Ma et al. [61], who demonstrated that the southern mountain area was the main source of river runoff in the basin.Next, we examined which factors affected the basinscale spatial distribution of sublimation.Table 3 presents the spatial correlation between accumulated sublimation from CTL9900 and various factors, which shows strong correlation between vegetation density and snowfall.Table 4 lists snow ablation components for different vegetation types (where tundra is classified as grass).It is clear that forest had less snowpack, strong sublimation, and less snowfall.Thus, net snow accumulation on the ground was smaller in comparison with other vegetation types.For those types, winter sublimation was 5-7 mm and the ratio of sublimation to snowfall was only 2-3%.Thus, sublimation is important to the winter water balance in forest regions.In other words, forest can limit the amount of snowpack because of the greater sublimation loss there.We believe that the forest distribution can be determined by the amount of snowpack, because deeper snowpack produces a surface water layer during snowmelt due to the frozen ground in permafrost  areas.In addition, wet conditions might cause respiration difficulty for tree roots and thus forest would not survive under deeper snowpack in the permafrost region.Next, we investigated how interannual snow ablation processes in the two winters affected the water balance.As shown in Section 3.1, total precipitation in the winter of 1999-2000 was 20 mm greater than that of 1998-2000.Table 5 lists the water balance components for the two winters.Consistent with this table, snowfall from CTL9899 was about 15 mm greater than that of CTL9900 because of the colder climate in CTL9899, as shown in Figure 3(b).This colder climate caused a longer period of snow, as shown in Figure 7(a) compared with Figure 7(b).However, basin-scale average SWE did not differ between CTL9899 and CTL9900, although the accumulated sublimation from CTL9899 was 5 mm greater than that of CTL9900.Thus, the ratio of sublimation to snowfall from the former run was 16.2%, compared with 14.8% from the latter.This difference could explain the discrepancy in river runoff indicated in Section 3.1, because the difference of total precipitation alone could not explain the 5 mm water equivalent difference in river runoff.Thus, sublimation is important for estimating river runoff accurately, especially that which occurs in spring.Therefore, river runoff can be affected by interannual variation of sublimation within the basin.

Advances in Meteorology
3.4.Discussion.Firstly, our study period was limited to two years.Therefore, the contribution of sublimation to snow ablation could vary in response to different extreme climate conditions.However, our results are adequate for indicating the importance of sublimation in ablation processes within the region.Secondly, MODIS snow coverage is based on satellitebased retrievals.Thus, the product itself has uncertainty and requires ground-truth observations to evaluate its accuracy.For example, Klein and Barnett [62] showed that MODIS typically misses snow coverage when its depth is less than 4 cm.Thus, another reason for the overestimation of snow coverage in CTL9900 and ANL9900 for May 2000 (Figures 4(b) and 4(c)) could be ascribed to the underestimation of snow coverage by the MODIS snow product, because the Lena River basin was largely covered by shallow snowpack due to the limited snowfall.
Thirdly, the representativeness of the BMDS stations is difficult to address because their locations are mostly at lower elevations and latitudes.Thus, our analysis would be improved if we had routine snow depth data from high mountain areas.Through the estimation of SWE with snow depth data, we believe that SnowAssim likely improved the simulations of snow mass and snow coverage over MicroMet/SnowModel.Therefore, our conclusion regarding SnowAssim might be limited to situations where there are routine snow depth and meteorological datasets available at the same locations.
Finally, we excluded blowing snow events in the simulations because each grid cell was 2 × 2 km, which means that blowing snow at that scale can be canceled by subgrid depositional and erosional effects.Nevertheless, blowing snow does cause sublimation [63] and it changes snow deposition on the ground.However, at high elevations, we saw no evidence of strong sublimation.With strong winds, the inclusion of blowing snow could enhance the sublimation and heterogeneity of the snow distribution within a grid cell.Thus, fine-scale phenomena such as blowing snow at subgrid scale could alter the contribution of sublimation.Further research should focus on the impact of blowing snow sublimation in windy areas of the Lena River basin.

Conclusions
Overall, we demonstrated the applicability of MicroMet/ SnowModel and SnowAssim in poorly observed regions.The methods we employed are not restricted to the study area but could be applied to other poorly observed cold regions such as the Arctic, North Canada, Greenland, the Alps, or the Tibetan Plateau.In addition, our approach has benefits regarding the understanding of long-term trends of river runoff in large northern river basins.
In this study, we used MicroMet/SnowModel and SnowAssim in the Lena River basin area.In addition, we calculated SWE with routine snow depth observations from weather stations using Sturm's method [37] and assimilated the results in SnowAssim.We were able to draw the following conclusions.
(1) Sturm's method can be applied to the Lena basin area of eastern Siberia, within its original estimation error.
(2) Simulated snow coverage and distribution were verified against in situ observed snow data.SnowAssim did not improve the simulated results of MicroMet/ SnowModel, because the SWE estimation error in SnowAssim was greater than the model error.Moreover, the BMDS precipitation was of good quality, in contrast to the quality of the estimated SWE.
(3) Basin-scale sublimation was important for components of the winter water balance.The spatial distribution of sublimation was determined primarily by vegetation and accumulated snowfall.
(4) Interannual variation of sublimation affected the amount of spring river runoff.Greater sublimation reduced river runoff because of reduced SWE on the ground.

Figure 1 :
Figure 1: Location of study site with (a) topography and (b) vegetation classification.Blue circles in panel (a) denote locations of BMDS sites.Red circles in panel (b) show locations of in situ observation sites of the National Polar Research Institute of Japan.

Figure 2 :
Figure 2: Comparison of SWE by Sturm's estimation with in situ observation by the National Polar Research Institute of Japan.
Winters.We show here the hydroclimatic conditions in the Lena River basin during the study periods of the two winters of 1998-1999 and 1999-2000.Figures3(a)-3(d) show the monthly variations of hydrometeorological conditions using the average of all the BMDS stations shown in Figure1(a), from September 1998 to August 2000.Vertical error bars denote standard deviations of each factor among the BMDS sites.

Figure 3 :
Figure 3: Monthly variations of hydrometeorological conditions.(a) River runoff at mouth of Lena River basin, (b) air temperature, (c) wind speed, (d) precipitation, and (e) snow depth.

Figure 4 :
Figure 4: Validation of spatial distribution of basin snow coverage with monthly MODIS products.(a1-a3) MODIS snow products in April, May, and June 2000, respectively.(b1-b3) Simulated results of CTL9900 in the same months as (a).(c1-c3) Simulated results of ANL9900 in the same months as (a).

Figure 5 :
Figure 5: Point-scale validation of SWE with in situ snow observations and simulated results of (a) CTL9900 and (b) ANL9900.

Table 1 :
Vegetation types in study area and river basin.

Table 2 :
Snow observation sites and observed data (NA indicates no data).

Table 3 :
Spatial correlations of various factors with accumulated sublimation in CTL9900.

Table 4 :
Vegetation classes and mean accumulated sublimation in CTL9900.

Table 5 :
Winter water balance and meteorological components in the basin in CTL9899 and CTL9900.