**3.2.1 Snow depth data**

We selected the snow depth measurements at the NamCo station over 4700m in altitude, which is located beside the NamCo Lake and the Mt. Nyainqenttanglha (Fig.14, circle). The Institute of Tibetan Plateau Research, Chinese Academy of Sciences, operates a station in the

Fig. 14. The distribution of the selected meteorological station over Tibet Plateau and western China (from China Meteorological Data Sharing Server System, data used in this work) and the geographic location of the Namco station site (30046.44'N, 90059.31'E).

Satellite-Based Snow Cover Analysis and the

the ground snow depth.

dominant role (Pulliainen, 1999).


TB Difference (K)

difference.

**3.3.2 Comparison to the AMSR-E daily SWE product** 

Snow Water Equivalent Retrieval Perspective over China 69

Fig. 16. The brightness temperature gradient between 89.0/36.5GHz and 18.7GHz with different resolution corresponding to the snow measurement time. The solid lines stand for

is more sensitive to the snow evolution than that of low resolution (L2A.res1).

frequency, and the bottom panel in Fig.16 indicates the high resolution (L2A.res3) resample

Fig.17 shows a sample, located beside (20km away from the field measurement site, the time stamp starts from October 1st), it indicates the same trend as the previous Fig.15. During the first 15days, the 89.0GHz shows a good sensitivity to the fresh snow. During the later succeeding 17days (7/11/2006~24/11/2006), the gradient becomes more and more large, but the snow depth decreases (compare to the Fig.15). This can be explained preliminarily by the evolution of the snow grain size and density, which typically increases with time. A more physical explanation needs e.g. the model simulation work on the snow emission of snow grain size, snow depth and snow density, in order to decide which part plays a

Another comparison (Fig.17) has been done by using the AMSR-E daily snow products, which are EASE-Grid with coarse resolution (25km) and the snow depths at Fig.15. From these figures, we can find that the maximum estimated AMSR-E SWE(mm) (using 36.5GHz gradient) over NamCo station at the winter of 2006~2007 occurs around 26/12/2006, which

10-1 10-1110-22 11-1 11-1211-22 12-3 12-1312-24 1-3 1-14 1-24

Fig. 17. The AMSR-E SWE daily products (EASE-GRID) and the corresponding TB

Time:from 2006.10.1 to 2007.1.28

SWE

SWE (mm)

are do not match the snow measurement, with a delay of almost one month.

 36.5H-18.7H 89.0H-18.7H

area. A snow campaign covering the whole winter offseason between 2006.10~2007.2 was conducted. SD records are acquired over three sites around the Namco station. Compared to the AMSR-E/Aqua satellite footprint, these sites are regarded as one site and represent the general situation of the whole area in this work, though this is a fairly inaccurate estimation in mountainous areas. Other time-series SD data in this work is from the winter-time observation (stations at Fig.14) in 2009~2010, when northern China were suffered from vast snowfall.

Fig 15 (left) shows a time series of the measured in situ snow depth values. From 24/10/2006, snow depth increases from 23cm to about 45cm on 8/11/2006, after which the depth decreased to 17cm on 28/1/2007. In this time span, several snowfall events happened on 12/11/2006, 14/11/2006 and 16/1/2007, with 2 cm of new snow in the last case. A relatively large shift appeared on 14/12/2006 because of the change of the observation sites for the surface wind.

Fig. 15. The SDs from the Danxung station (No.55493) (Left) and the SD (cm) field campaign near NamCo station (Right)

#### **3.2.2 AMSR-E L2A swath dataset and processing**

We selected the AMSR-E daily L2A swath brightness temperature (ascending and descending pass, A/D, http://nsidc.org/data/docs/daac/ae\_l2a\_tbs.gd.html) over the experiment site and other western stations in China according to the geographic coordinate, which means that the extracted swath Tbs are in the area of 10km2 around the site. We chose the Tb difference between 89.0/36.5/18.7GHz and 10.7GHz channels for the gradient time series comparison with station snow depth (cm).

#### **3.3 Comparison result at Nam Co experiment site**

#### **3.3.1 The AMSR-E swath L2A Tb gradient time series**

We plotted the Tb gradient between 89.0/36.5GHz and 18.7GHz with different resolutions corresponding to the snow measurement time at Nam Co in Fig. 15. Compared to the snow depth (the solid lines), the brightness temperature gradient (traditional algorithm prototype) at Fig. 15 shows a good relationship for the snow depth decreasing period (24/11/2006~26/1/2007) at 89.0GHz (named high frequency) gradient and 36.5GHz (named low frequency) gradient. For this period of time (snow depth are less than 30 cm), we can understand that the high frequency are more sensitive to the snow evolution than the low

area. A snow campaign covering the whole winter offseason between 2006.10~2007.2 was conducted. SD records are acquired over three sites around the Namco station. Compared to the AMSR-E/Aqua satellite footprint, these sites are regarded as one site and represent the general situation of the whole area in this work, though this is a fairly inaccurate estimation in mountainous areas. Other time-series SD data in this work is from the winter-time observation (stations at Fig.14) in 2009~2010, when northern China were suffered from vast

Fig 15 (left) shows a time series of the measured in situ snow depth values. From 24/10/2006, snow depth increases from 23cm to about 45cm on 8/11/2006, after which the depth decreased to 17cm on 28/1/2007. In this time span, several snowfall events happened on 12/11/2006, 14/11/2006 and 16/1/2007, with 2 cm of new snow in the last case. A relatively large shift appeared on 14/12/2006 because of the change of the observation sites

1980-1

10

20

30

Snow Depth (cm)

Fig. 15. The SDs from the Danxung station (No.55493) (Left) and the SD (cm) field campaign

We selected the AMSR-E daily L2A swath brightness temperature (ascending and descending pass, A/D, http://nsidc.org/data/docs/daac/ae\_l2a\_tbs.gd.html) over the experiment site and other western stations in China according to the geographic coordinate, which means that the extracted swath Tbs are in the area of 10km2 around the site. We chose the Tb difference between 89.0/36.5/18.7GHz and 10.7GHz channels for the gradient time

We plotted the Tb gradient between 89.0/36.5GHz and 18.7GHz with different resolutions corresponding to the snow measurement time at Nam Co in Fig. 15. Compared to the snow depth (the solid lines), the brightness temperature gradient (traditional algorithm prototype) at Fig. 15 shows a good relationship for the snow depth decreasing period (24/11/2006~26/1/2007) at 89.0GHz (named high frequency) gradient and 36.5GHz (named low frequency) gradient. For this period of time (snow depth are less than 30 cm), we can understand that the high frequency are more sensitive to the snow evolution than the low

40

50

10-24 11-3 11-14 11-24 12-5 12-15 12-26 1-5 1-16 1-26

Time from 2006.10.24 to 2007.1.28

 West Filed Inside Field Near Lake

Snow Depth (cm)

snowfall.

Snow Depth (cm)

for the surface wind.

DanXung Station

near NamCo station (Right)

1 4 7 10 13 16 19 22 25

Days at the Jan. 1980

**3.2.2 AMSR-E L2A swath dataset and processing** 

series comparison with station snow depth (cm).

**3.3 Comparison result at Nam Co experiment site 3.3.1 The AMSR-E swath L2A Tb gradient time series** 

Fig. 16. The brightness temperature gradient between 89.0/36.5GHz and 18.7GHz with different resolution corresponding to the snow measurement time. The solid lines stand for the ground snow depth.

frequency, and the bottom panel in Fig.16 indicates the high resolution (L2A.res3) resample is more sensitive to the snow evolution than that of low resolution (L2A.res1).

Fig.17 shows a sample, located beside (20km away from the field measurement site, the time stamp starts from October 1st), it indicates the same trend as the previous Fig.15. During the first 15days, the 89.0GHz shows a good sensitivity to the fresh snow. During the later succeeding 17days (7/11/2006~24/11/2006), the gradient becomes more and more large, but the snow depth decreases (compare to the Fig.15). This can be explained preliminarily by the evolution of the snow grain size and density, which typically increases with time. A more physical explanation needs e.g. the model simulation work on the snow emission of snow grain size, snow depth and snow density, in order to decide which part plays a dominant role (Pulliainen, 1999).

#### **3.3.2 Comparison to the AMSR-E daily SWE product**

Another comparison (Fig.17) has been done by using the AMSR-E daily snow products, which are EASE-Grid with coarse resolution (25km) and the snow depths at Fig.15. From these figures, we can find that the maximum estimated AMSR-E SWE(mm) (using 36.5GHz gradient) over NamCo station at the winter of 2006~2007 occurs around 26/12/2006, which are do not match the snow measurement, with a delay of almost one month.

Fig. 17. The AMSR-E SWE daily products (EASE-GRID) and the corresponding TB difference.

Satellite-Based Snow Cover Analysis and the

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

Xizang Date g) Jiali 30.67N 93.28E

Tb (A/D) and snow depth at any available time.


**3.5 Conclusions** 

the vegetation effect removal.

of Institute of Plateau Meteorology.

Research Letters, 28(19), 3673−3676.

**4. Acknowledgement** 

**5. References** 

TB Difference(K)

Snow Water Equivalent Retrieval Perspective over China 71

Snow Depth(cm)

Fig. 18. The time series Tb difference (89/36.5-18.7GHz and 18.7-10.7GHz) and the ground snow depth (cm). The first 4 figures are plotted only ascending Tb and the corresponding ground measurements (local morning time), the last four figures are plotted with all of the

We get the preliminary analysis result, the Tbs at18.7-10.7GHz are insensible to the snow evaluation except the deep snow depth (a, b and f), although the depression in f is obvious due to the local vegetation influence. Over deep snow (a, b, f and g, continuous accumulation > 20cm), the Tbs at 36.5-18.7GHz are more reliable than that of high frequency, while over the shallow snow (c, d, e and h, discontinuous snow occurrence, < 15cm), the pair 36.5/18.7 is insensitive, but the high frequency pair (89.0/18.7) shows its distinct response. The pair 89.0/18.7 shows its shallow snow retrieval ability in a, b, c, d, e and when the snow depth over 20cm, the signal is more variable and suspect. The pair 89.0/18.7 indicates its sensitive response to the quick presence of the snowfall, and keeps turbulence when the snow depths are unchanged due to the temporal snow physical characteristics and climate factors. The last four figures show that the A/D Tbs act the similar behaviors with difference correlation intensity.

From what we have shown above, it can be argued that the high frequency (89.0GHz) shows its sensitive to the relative shallow snow pack, which suggests that we can develop the shallow snow depth retrieval via the good Tb pair and ground snow depths over the western China. Model simulation work is needed to explain the discrepancy of the snow evolution and brightness temperature gradient at high frequencies, and we should enhance the following aspects, the possible mixed pixel effect, the atmosphere effect elimination, and

This work is now supported by the Chinese "973" Program "Earth Observation for Sensitive Factors of Global Change: Mechanism and Methodologies" (NO. 2009CB723906), the Director Foundation of Center for Earth Observation and Digital Earth Chinese Academy of Sciences, the National Natural Science Funds (Grant: 40901175), and the "Open Foundation"

[1] Armstrong, R. L., & Brodzik, M. J. (2001). Recent Northern Hemisphere snow extent: A

comparison of data derived from visible and microwave sensors. Geophysical


TB Difference(K)

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

Gansu Date

89V-18V 36V-18V 18V-10.7V

Snow Depth(cm)

h) Hezheng 35.42N 103.33E

Snow Depth(cm)

Compared with the gradient figure, from Fig.16, we can find that the 36.5GHz gradient shows a relatively stable value from 1/12/2006~26/12/2006. The discrepancy between the SWE and TB gradients is probably due to the response of 37 GHz saturating for SWE values over 120-140mm, or the mixed pixel by the lake. This requires a more extensive field dataset to acquire the explanation.

If we consider the typical snow density over Tibet area to be approximately 0.239g/cm3, we get a maximum snow depth value of about 75 cm from the AMSR-E observation. This is quite larger compared to the in situ measurements, an indication that the AMSR-E SWE value is overestimated, which is consistent with the result in paper(Pulliainen, 1999).

#### **3.4 Time series analysis between Tb and snow depth**

We selected several observations over western China (Fig.14) for the qualitative analysis, which include the Xinjiang and Neimenggu deep snow and Gansu, Qinghai and Tibet shallow snow depth situations.

All of the figures in Fig.18 are plotted with the three gradients (Tb difference at 89.0-18.7, 36.5-18.7 and 18.7-10.7) and the corresponding snow depths (see Fig.18).

Fig. 18. The time series Tb difference (89/36.5-18.7GHz and 18.7-10.7GHz) and the ground snow depth (cm). The first 4 figures are plotted only ascending Tb and the corresponding ground measurements (local morning time), the last four figures are plotted with all of the Tb (A/D) and snow depth at any available time.

We get the preliminary analysis result, the Tbs at18.7-10.7GHz are insensible to the snow evaluation except the deep snow depth (a, b and f), although the depression in f is obvious due to the local vegetation influence. Over deep snow (a, b, f and g, continuous accumulation > 20cm), the Tbs at 36.5-18.7GHz are more reliable than that of high frequency, while over the shallow snow (c, d, e and h, discontinuous snow occurrence, < 15cm), the pair 36.5/18.7 is insensitive, but the high frequency pair (89.0/18.7) shows its distinct response. The pair 89.0/18.7 shows its shallow snow retrieval ability in a, b, c, d, e and when the snow depth over 20cm, the signal is more variable and suspect. The pair 89.0/18.7 indicates its sensitive response to the quick presence of the snowfall, and keeps turbulence when the snow depths are unchanged due to the temporal snow physical characteristics and climate factors. The last four figures show that the A/D Tbs act the similar behaviors with difference correlation intensity.

### **3.5 Conclusions**

70 Remote Sensing of Planet Earth

Compared with the gradient figure, from Fig.16, we can find that the 36.5GHz gradient shows a relatively stable value from 1/12/2006~26/12/2006. The discrepancy between the SWE and TB gradients is probably due to the response of 37 GHz saturating for SWE values over 120-140mm, or the mixed pixel by the lake. This requires a more extensive field dataset

If we consider the typical snow density over Tibet area to be approximately 0.239g/cm3, we get a maximum snow depth value of about 75 cm from the AMSR-E observation. This is quite larger compared to the in situ measurements, an indication that the AMSR-E SWE

We selected several observations over western China (Fig.14) for the qualitative analysis, which include the Xinjiang and Neimenggu deep snow and Gansu, Qinghai and Tibet

All of the figures in Fig.18 are plotted with the three gradients (Tb difference at 89.0-18.7,

36.5-18.7 and 18.7-10.7) and the corresponding snow depths (see Fig.18).

Snow Depth(cm)

Snow Depth(cm)

TB Difference(K)

Snow Depth(cm)



> -40 -30 -20 -10 0

TB Difference(K)

TB Difference(K)

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

Neimenggu Date f) Tulihe 50.48N 121.68E

Qinghai Date d) Guoruo 34.47N 100.25E

Xinjiang Date b) Jimunai 47.43N 85.87E

Snow Depth(cm)

Snow Depth(cm)

Snow Depth(cm)

value is overestimated, which is consistent with the result in paper(Pulliainen, 1999).

**3.4 Time series analysis between Tb and snow depth** 

to acquire the explanation.

shallow snow depth situations.

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

Xinjiang Date a) FuYun 46.98N 89.52E

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

2009-9-1 2009-11-1 2010-1-1 2010-3-1 2010-5-1

Neimenggu Date e) Wushengzhao 39.10N 109.03E

Qinghai Date c) Jiuzhi 33.43N 101.48E



> -40 -30 -20 -10 0 10

TB Difference(K)

TB Difference(K)

TB Difference(K)

From what we have shown above, it can be argued that the high frequency (89.0GHz) shows its sensitive to the relative shallow snow pack, which suggests that we can develop the shallow snow depth retrieval via the good Tb pair and ground snow depths over the western China. Model simulation work is needed to explain the discrepancy of the snow evolution and brightness temperature gradient at high frequencies, and we should enhance the following aspects, the possible mixed pixel effect, the atmosphere effect elimination, and the vegetation effect removal.
