Journal of Geographical Sciences

© 2007 Science in China Press Springer-Verlag

DOI: 10.1007/s11442-007-0259-7

The relationship between NDVI and precipitation

on the Tibetan Plateau
DING Mingjun, *ZHANG Yili, LIU Linshan, ZHANG Wei,
WANG Zhaofeng, BAI Wanqi
Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China

Abstract: The temporal and spatial changes of NDVI on the Tibetan Plateau, as well as the
relationship between NDVI and precipitation, were discussed in this paper, by using 8-km
resolution multi-temporal NOAA AVHRR-NDVI data from 1982 to 1999. Monthly maximum
NDVI and monthly rainfall were used to analyze the seasonal changes, and annual maximum
NDVI, annual effective precipitation and growing season precipitation (from April to August)
were used to discuss the interannual changes. The dynamic change of NDVI and the corre-
lation coefficients between NDVI and rainfall were computed for each pixel. The results are as
follows: (1) The NDVI reached the peak in growing season (from July to September) on the
Tibetan Plateau. In the northern and western parts of the plateau, the growing season was
very short (about two or three months); but in the southern, vegetation grew almost all the
year round. The correlation of monthly maximum NDVI and monthly rainfall varied in different
areas. It was weak in the western, northern and southern parts, but strong in the central and
eastern parts. (2) The spatial distribution of NDVI interannual dynamic change was different
too. The increase areas were mainly distributed in southern Tibet montane shrub-steppe zone,
western part of western Sichuan-eastern Tibet montane coniferous forest zone, western part
of northern slopes of Kunlun montane desert zone and southeastern part of southern slopes
of Himalaya montane evergreen broad-leaved forest zone; the decrease areas were mainly
distributed in the Qaidam montane desert zone, the western and northern parts of eastern
Qinghai-Qilian montane steppe zone, southern Qinghai high cold meadow steppe zone and
Ngari montane desert-steppe and desert zone. The spatial distribution of correlation coeffi-
cient between annual effective rainfall and annual maximum NDVI was similar to the growing
season rainfall and annual maximum NDVI, and there was good relationship between NDVI
and rainfall in the meadow and grassland with medium vegetation cover, and the effect of
rainfall on vegetation was small in the forest and desert area.

Keywords: Tibetan Plateau; land cover change; NDVI; precipitation; correlation

Received: 2007-02-10 Accepted: 2007-03-29

Foundation: National Basic Research Program of China, No.2005CB422006; National Natural Science Foundation of
China, No.40331006; No.90202012
Author: Ding Mingjun (1979–), Ph.D. Candidate, specialized in land-use/land-cover change and physical geography.
E-mail: dingmingjun@igsnrr.ac.cn
*Corresponding author: Zhang Yili, E-mail: zhangyl@igsnrr.ac.cn

www.scichina.com www.springerlink.com
260 Journal of Geographical Sciences

1 Introduction
With the development of remote sensing, it is possible to get more information from
multi-date and multi-spectral remote sensing data, which provide effective methods to study
the vegetation distribution and interannual and seasonal changes. Among all the remote
sensing data, NOAA/AVHRR data are widely used to study vegetation on global and re-
gional scale because of long series, wide extent of observation and short cycle (Defnies et al.,
1994)
The vegetation indexes are very significant to reflect more information of vegetation.
Among them, the normalized difference vegetation index (NDVI, NDVI=(NIR–Red)
/(NIR+Red) NIR denoted the reflection of near infrared, red denoted the reflection of red
light from visible light) is widely used, and it is the perfect indicator of growth status, spatial
density distribution (Sun et al., 1998; Purevdorj et al., 1998; Liu et al., 1999) and phenology
of plant (Defnies et al., 1994; Derrein et al., 1992).
No matter physical or cultivated vegetation, their growth processes are all affected by
temperature, precipitation, etc. (Li et al., 2000; Zhang et al., 2003; Li et al., 2000; Nicholson
et al., 1990; Schmidt et al., 2000). To further study the relationship between vegetation and
climate and to get the spatial variation will boost the research on forecasting of growth status.
With the influence of complicated topography and southwest monsoon, southeast monsoon,
Siberian high, Qinghai-Tibet high etc., the change of vegetation on the Tibetan Plateau is
typical and representative on high altitude region, which is indispensable for the research
about global change. In this research, the NOAA/AVHRR data and precipitation data, with
long series, were used to study the spatio-temporal change characteristics of NDVI on the
Tibetan Plateau, and analyze the relationship between NDVI and precipitation. The study
will offer some information or evidences to exploit the resources on the Tibetan Plateau and
to study the global vegetation change.

2 The study area

The study area extends from 26º00′12″N to 39º46′50″N and 73º18′52″E to 104º46′59″E
and includes low area of the southeastern Himalaya Mountains (Mêdog, Zayü and Cona
counties), that’s the Tibetan Plateau in China. The total area is about 2.61×106 km2 (Zhang et
al., 2002) and the average altitude is about 4380 m. In winter, the climate of Tibetan Plateau
is controlled by Siberian high and Mongolian high, and it is very cold and arid, and has very
little rain and more strong winds. In summer, the climate is controlled by Indian low there,
and it is warm and moist (Zhao, 1998). All kinds of habitats offer favorable condition for
most plants. From southeast to northwest, there are seven types of vegetation. They are sub-
tropical evergreen broad-leaved forest, deciduous broad-leaved forest, alpine brush, alpine
meadow, alpine grassland and high and cold desert (Chen et al., 1999).

3 Data and methods

3.1 Data

NOAA/AVHRR-NDVI data with an 8-km resolution of every ten days from 1982 to 1999
DING Mingjun et al.: The relationship between NDVI and precipitation on the Tibetan Plateau 261

from NASA was used in this research (Agbu et al., 1994). It is the third generation product
with values ranging from 0 to 254. The formula (NDVI=0.8(I)/254–0.1) can transmit the
value to normalize difference vegetation indexes with a value ranging from –1 to 1. To
eliminate the effect of clouds, the Maximum Value Composite was used in data processing
which assumes that the position of clouds often change in a period of time, such as ten days,
and any point has a day without cloud covering. The value of NDVI in this day at the point
is higher than those of the cloudy days, and this study selects this value as the value of the
point in ten days, then it selects the highest value of the point as value of this month. The
primitive projection of these images is Goode Interrupted Homolsine Projection. In this
study, we convert it into Albers Projection and select the area of the Tibetan Plateau.
Because the annual maximum NDVI value can reflect the vegetation well, so in this study,
we used the Maximum Value Composite method to get the annual maximum NDVI. Mean-
while we count the mean monthly maximum NDVI from 1982 to 1999 to analyze the sea-
sonal change.
The data of temperature and precipitation are provided by Dr. Tao (2005), and his re-
search report utilizes spline function interpolation to obtain spatial interpolation data. In the
process of interpolation, the effects of latitude and longitude were discussed here, mean-
while the DEM data with 0.1 degree was utilized to fit the surface data from the above
method. The data quality is very good through the test. Considering the growing season ends
at the end of September and precipitation after September has no influence on the vegetation
NDVI of the current year, but has effects on vegetation of the next season, so we counted the
amount of precipitation from October of the last year to September of the ensuing year that
we called annual effective precipitation and precipitation from April to August that we called
growing season precipitation respectively.

3.2 Methods

We calculated the linear regression coefficients of annual maximum NDVI by means of

Least Square Method (Micael, 2000). To find the natural fluctuation range of NDVI, 13 ar-
eas were selected–acreage of each is 200 km2–including lakes, frozen grounds, deserts, and
salinas on the Tibetan Plateau (Ding et al., 2006). There was no vegetation in these areas.
Through analysis and statistics, 97% of the coefficients in these areas were distributed be-
tween −0.2 and 0.2, so it was regarded as a range of no vegetation change. Based on the
above analysis, all the coefficients are divided into three grades, decrease: less than -0.2; no
change: −0.2 to 0.2; and increase: more than 0.2.
The change of precipitation was divided by the standard deviation, more than the standard
deviation was considered the precipitation was increasing, but less than the negative stan-
dard deviation was thought the precipitation was decreasing. The correlation between pre-
cipitation and NDVI was adapted to linear correlation. All the processes were done by the
software ARCGIS.
In order to analyze the spatial difference of vegetation seasonal change, we selected 12
areas with 3×3 pixels around the weather stations along longitude and latitude respectively
(Figure 1) and utilized the precipitation from the weather stations and the monthly maximum
NDVI value extracted from NOAA/AVHRR-NDVI by ARCGIS software to analyze the re-
lationship between precipitation and vegetation in one year.
262 Journal of Geographical Sciences

Figure 1 The location of the weather stations and research region

4 Results and analyses

4.1 Spatial distribution pattern of mean annual maximum NDVI value and annual
effective precipitation from 1982 to 1999

Because of controlled by the southwest warm and humid air flow, the southeast monsoon
and Siberian high, the amount of precipitation decreases from southeast to northwest (Figure
2 1 ). Precipitation can reach a range from 700 mm to 800 mm in the southeastern part of the
Tibetan Plateau, from 150 mm to 200 mm in the northwestern part and less than 100 mm in
the Qaidam Basin.
Alike precipitation, the NDVI value decreases gradually from southeast to northwest
(Figure 3). In the eastern and southern parts of the plateau, NDVI reaches the highest, but in
the western and northern, it is very low. From Figures 2 and 3 and previous studies (Chen et

1
Based on the map of system of physico-geographical regions of Qinghai-Tibet Plateau (Zheng, 1996),
we digitalized it (Though the edge of the physico-geographical regions is a transitional belt, in order to
analyze conveniently in this study, we assumed that it was a line and digitalized it. The name of the phys-
ico-geographical regions are as follows: IB1 Golog-Nagqu high-cold shrub-meadow zone; IC1 Southern
Qinghai high-cold meadow steppe zone; IC2 Qangtang high-cold steppe zone; ID1 Kunlun high-cold desert
zone; IIAB1 Western Sichuan-eastern Tibet montane coniferous forest zone; IIC1 Southern Tibet montane
shrub-steppe zone; IIC2 Eastern Qinghai-Qilian montane steppe zone; IID1 Ngari montane desert-steppe
and desert zone; IID2 Qaidam montane desert zone; IID3 Northern slopes of Kunlun montane desert zone;
OA1 Southern slopes of Himalaya montane evergreen broad-leaved forest zone.
DING Mingjun et al.: The relationship between NDVI and precipitation on the Tibetan Plateau 263

al., 1999), we can find that evergreen broad-leaved forest and coniferous forest are distrib-
uted in places with precipitation more than 600 mm and NDVI value more than 200; that
brush and meadow in places with precipitation from 400 mm to 600 mm and NDVI between
180 and 200; meadow grasslands, alpine grassland and montane grassland in places with
precipitation from 200 to 400 and NDVI from 150 to 180; and alpine desert and montane
desert with precipitation lower than 200 mm and NDVI under 150.

Figure 2 Spatial distribution of mean annual pr- Figure 3 Spatial distribution of mean annual
ecipitation on the Tibetan Plateau from 1982 to maximum NDVI on the Tibetan Plateau from 1982 to
1999 1999

4.2 The relationship between monthly maximum NDVI and monthly precipitation

Figure 4 1 shows spatial difference of the rainy season. From north to south along the longi-
tude, the rainy season becomes longer gradually. In the north, there is no apparent rainy
season and only has little rain in July. In the central part, the rainy season may last from the
beginning of April to October, but most of the precipitation is concentrated in July and Au-
gust. In the south, there is no apparent rainy season and two rainfall peaks exist. The grow-
ing season is in accord well with the rainy season (Figure 5). In the north, there is no evident
growing season. In the central part, the growing season begins in May and ends in October,
it lags behind rainy season. In the south, the vegetation has no apparent growing season
and grows all the year round. From west to east along the latitude, extent of rainy season
gets longer (Figure 6 2 ). In the west, there is only a little rain in July, August and Septem-
ber. In the east, the rainy season almost accords with the central part. The growing season
is also in accord well with the rainy season (Figure 7). In the west, the growing season is
very short, but the growing season in the east is long and lasts from mid-April to the end
of October.
Results show that monthly NDVI and monthly precipitation have weak correlation in the
west and strong correlation in the mid-east along the latitude (Table 1). It indicates that pre-
cipitation is a key factor for vegetation growth in the mid-east and not a key factor for vege-
tation growth in the west.

1
The weather stations along longitude from 1 to 12 are Lenghu, Da Qaidam, Xiaozaohuo, Golmud, Qumarlêb, Zadoi,
Nangqên, Qamdo, Baxoi, Zogang, Dêqên and Gongshan in proper order.
2
The weather stations along latitude from 1 to 12 are Shiquanhe, Gêrzê, Xainza, Amdo, Tuotuoheyan, Zhidoi, Qing-
shuihe, Madoi, Maqên, Golog, Henan and Hezuo in proper order.
264 Journal of Geographical Sciences

Along the longitude, the north and south have weaker correlation between monthly NDVI
and monthly precipitation than that in the central part. At some special weather stations,
such as Lenghu and Gongshan, the correlations are negative. The reason is that the increas-
ing precipitation may lead to reduction of the NDVI value of bare soil with very low cover-
age at Lenghu station, but at Gongshan, the reason was not clear and further study is needed.
From the above analysis, vegetation responded to precipitation very well in the cen-
tral-eastern Tibetan Plateau.

Figure 4 Monthly mean precipitation along longitude Figure 5 Monthly mean NDVI along longitude

Figure 6 Monthly mean precipitation along latitude Figure 7 Monthly mean NDVI along latitude

Table 1 The correlation between monthly NDVI and precipitation

Station along latitude Correlation coefficient Station along longitude Correlation coefficient
Shiquanhe 0.47 Lenghu -0.67
Gêrzê 0.16 Da Qaidam 0.71
Xainza 0.87 Xiaozaohuo 0.94
Amdo 0.89 Golmud 0.55
Tuotuoheyan 0.86 Qumarlêb 0.91
Zhidoi 0.90 Zadoi 0.93
Qingshuihe 0.92 Nangqên 0.90
Madoi 0.90 Qamdo 0.91
Maqên 0.87 Baxoi 0.81
Golog 0.90 Zogang 0.72
Henan 0.93 Dêqên 0.36
Hezuo 0.91 Gongshan -0.24
DING Mingjun et al.: The relationship between NDVI and precipitation on the Tibetan Plateau 265

4.3 The changes of vegetation types

In order to study the effect of precipitation on vegetation, we calculated the correlation bet-
ween annual maximum NDVI and precipitation (annual effective precipitation and growing
season precipitation) (Figure 8a). Significance level (p<0.01, p<0.05 and p<0.1) is 0.5751,
0.4555 and 0.3887, respectively.

Figure 8 Correlation coefficient images between NDVI and precipitation at different periods on the Tibetan
Plateau (a, Correlation coefficient between annual precipitation and annual maximum NDVI; b, Correlation coef-
ficient between precipitation from April to August and annual maximum NDVI)

The correlation between annual maximum NDVI and annual effective precipitation is
rather strong in most parts of southern Tibet montane shrub-steppe zone, the west of
Golog-Nagqu high-cold shrub-meadow zone, the east of southern Qinghai high-cold
meadow steppe zone, the west of eastern Qinghai-Qilian montane steppe zone and some
transitional belts (Figure 8a), which indicates that the growth of meadow and meadow-
grassland mainly distributed in these regions strongly depends on precipitation.
In the area of forest vegetation (such as southern slopes of Himalaya montane evergreen
broad-leaved forest zone and western Sichuan-eastern Tibet montane coniferous forest zone),
Qangtang high-cold steppe zone and Kunlun high-cold desert zone, the correlation of annual
maximum NDVI and annual effective precipitation is weak. On northern slopes of Kunlun
montane desert zone, Qaidam montane desert zone and most parts of Kunlun high-cold de-
sert zone, the correlation displays negative apparently. We can conclude that the correlation
between annual maximum NDVI and growing season precipitation is similar with the corre-
lation between annual maximum NDVI and annual effective precipitation (Figure 8b).
In order to understand the relation of NDVI and precipitation, we compared the interannual
vegetation NDVI change with annual effective precipitation change on the Tibetan Plateau.
Figure 9a displays the annual maximum NDVI change from 1982 to 1999 1 : the area with
NDVI decreasing is mainly distributed in Qaidam montane desert zone, eastern Qing-
hai-Qilian montane steppe zone and southern Qinghai high-cold meadow steppe zone. In
addition, it is distributed in some parts of Ngari montane desert-steppe and desert zone. The

1
Zhang Yili et al. The change characteristics of land cover on Tibet Plateau. Science in China, on proofing.
266 Journal of Geographical Sciences

results are similar to the previous studies (Sha et al., 2001; Zhang et al., 2006; Yang et al.,
2005; Yang et al., 2004). The area with NDVI increasing is mainly distributed in southern
Tibet montane shrub-steppe zone and west of western Sichuan-eastern Tibet montane conif-
erous forest zone. Figure 9b shows that the interannual effective precipitation change from
1982 to 1999 on the Tibetan Plateau that is almost in accord with the results before (Wu et al.,
2005). The spatial distribution pattern is similar to the NDVI interannual change, and that proved
precipitation increasing is very important for the growth of some vegetation types.

Figure 9 Spatial distribution of annual maximum NDVI and annual rainfall changes between the beginning of
the 1980s and the late 1990s on the Tibetan Plateau (a. Spatial distribution of annual maximum NDVI change; b.
Spatial distribution of annual rainfall changes)

5 Conclusions and discussion

In this paper, we have studied the relationship between NDVI and precipitation, as well as
their spatial distribution. The results are as follows:
(1) The NDVI value reaches the maximum from July to September; in the western and
northern plateau, the growing season is rather short with two or three months; in the south-
ern plateau, vegetation grows all the year round.
(2) Correlation between mean monthly maximum NDVI and monthly precipitation from
1982 to 1999 is very weak in the western, northern and southern plateau, and very strong in
the central and eastern plateau.
(3) Interannual change of NDVI on the Tibetan Plateau is spatial variation too. The areas
of NDVI increasing are distributed in southern Tibet montane shrub-steppe zone and west of
western Sichuan-eastern Tibet montane coniferous forest zone. The areas with NDVI de-
creasing are mainly distributed in Qaidam montane desert zone, eastern Qinghai-Qilian
montane steppe zone and southern Qinghai high-cold meadow steppe zone. In addition, it is
distributed in some parts of Ngari montane desert-steppe and desert zone.
(4) Distribution pattern of the correlation between annual maximum NDVI and annual
effective precipitation is similar to the correlation between annual maximum NDVI and
growing season precipitation. The precipitation is very important to vegetation in some
places, especially for the areas covered by meadow and meadow-grassland, but the precipi-
DING Mingjun et al.: The relationship between NDVI and precipitation on the Tibetan Plateau 267

tation is very little effective in the areas distributed by forest and desert.
This study assessed the temporal and spatial changes of vegetation cover on the Tibetan
Plateau, and analyzed the relationship between precipitation and vegetation using 8-km
resolution multi-temporal NOAA AVHRR-NDVI data from 1982 to 1999. The results
proved that this method is a convenient, quick and cheap way to study vegetation on large
scale area. From analyzing the NDVI trend, we came to the conclusion that the area with
increasing NDVI is mainly distributed in the Yarlung Zangbo River Basin, a food production
base for Tibet (Li, 2000). Recently, the cultivation has developed very quickly for chemical
fertilizer application, irrigation practices and afforestation (Li et al., 1994), which enhanced
vegetation cover there. Meanwhile, the warm-humid trend (Wu et al., 2005) also can boost
the vegetation growth.
Recently, the study about driving factors of vegetation change is emphasized at regional
scale, which is favorable for production and ecological restoration, and is significant for the
study of global change. So the other objective of this study is to select interesting area (such
as the Yarlung Zangbo River Basin) for further study (such as driving factors of vegetation
change).

Acknowledgements The authors would like to thank Prof. Shao Xuemei and Prof. Wang Lili from
IGSNRR and Dr Li Shuangcheng from Peking University for the insightful suggestions to the study. Dr.
Tao Bo in IGSNRR provided the climate data.

References
Agbu P A, James M E, 1994. The NOAA/NASA Pathfinder AVHRR Land Data Set User’s Manaul. Goddard
Distributed Active Archive Center, NASA, Goddard Space Flight Center, Greenbelt.
Chen Linzhi, Wang Zuwang, 1999. The Impact of Human Alteration on Ecosystem Diverstiy. Hangzhou: Zhejiang
Science and Technology Press, 267–277. (in Chinese)
Defries R S, Townsend J R G, 1994. NDVI-derived land cover classification at a global scale. Int. J. Remote
Sensing, 15: 3567–3586.
Derrein M, Farki B, Legleau H et al., 1992. Vegetation cover mapping over France using NOAA-11/AVHRR. Int.
J. Remote Sensing, 13: 2743–2749.
Ding Mingjun, Zhang Yili, Shen Zhenxi et al., 2006. Land cover change along the Qinghai–Tibet highway and
railway from 1981 to 2001. Journal of Geographical Sciences, 16(4): 387–395.
Li Bengang, Tao Shu, 2000. Correlation between AVHRR-NDVI and climate factors. Acta Ecologica Sinica,
20(5): 898–902. (in Chinese)
Li Sen, Dong Guangrong, Dong Yuxiang et al., 1994. Desertification control objective, measures and engineering
layout in the central “One-River-Two-Tributaries” basin in Tibet. Journal of Desert Research, 14(2): 56–63.
(in Chinese)
Li Xiaobing, Shi Peijun, 2000. Sensitivity analysis of variation in NDVI, temperature and precipitation in typical
vegetation types across China. Acta Phytoecologica Sinica, 24(3): 379–382. (in Chinese)
Li Yuan, 2000. Land Resources of China. Beijing: China Dadi Press. (in Chinese)
Liu Shuzhen, Zhou Lin, Chou Chongshan et al., 1999. Studies on Grassland Degradation and Desertification of
Naqu Prefecture in Tibet Autonomous Region. Lhasa: Tibet People’s Press, 73–79. (in Chinese)
Micael C, RunnstrÖm, 2000. Is northern China winning the battle against desertification?: Satellite remote sens-
ing as a tool to study biomass trends on the Ordos Plateau in semiarid China. AMBIO, 29(8): 468–476.
Nicholson S E, Davenport M L, Malo A R A, 1990. Comparison of the vegetation response to rainfall in the Sahel
and East Africa, using NDVI from NOAA AVHRR. Climatic Change, 17: 209–241.
268 Journal of Geographical Sciences

Purevdorj T S, Tateishi R, Ishiyam T et al., 1998. Relationships between percent vegetation cover and vegetation
indices. Int. J. Remote Sensing, 19: 3519–3535.
Schmidt H, Gitelson A, 2000. Temporal and spatial vegetation cover changes in lsraeli transition zone:
AVHRR-based assessment of rainfall impact. Int. J. Remote Sensing, 21: 997–1010.
Sha Zhanjiang, Ma Haizhou, Chen Kelong et al., 2001. Study on desertification of Yellow River source area based
on GIS and RS. Journal of Salt Lake Research, 9(1): 68–70. (in Chinese)
Sun Hongyu, Wang Changyao, Niu Zheng et al., 1998, Analysis of the vegetation cover change and the relation-
ship between NDVI and environmental factors by using NOAA time series data. Journal of Remote Sensing,
2(3): 205–210. (in Chinese)
Tao Bo, 2005. Preliminary study on spatio-temporal patterns of terrestrial carbon budgets driven by climatic
change in China. Postdoctoral Research Report. (in Chinese)
Wu Shaohong, Yin Yunhe, Zheng Du et al., 2005. Climate changes in the Tibetan Plateau during the last three
decades. Acta Geographica Sinica, 60(1): 1–11. (in Chinese)
Yang Jianping, Ding Yongjian, Chen Rensheng et al., 2004. Permafrost change and its effect on eco-environment
in the source region of the Yangtze and Yellow rivers. Journal of Mountain Science, 22(3): 278–285. (in Chi-
nese)
Yang Jianping, Ding Yongjian, Chen Rensheng, 2005. NDVI reflection of alpine vegetation changes in the source
regions of the Yangtze and Yellow rivers. Acta Geographica Sinica, 60(3): 468–478. (in Chinese)
Zhang Yili, Li Bingyuan, Zheng Du, 2002. The area and boundary of Qinghai-Tibet Plateau. Geographical Re-
search, 21(1): 1–8. (in Chinese)
Zhang Yili, Liu Linshan, Bai Wanqi et al., 2006. Grassland degradation in the source region of the Yellow River.
Acta Geographica Sinica, 61(1): 4–12. (in Chinese)
Zhang Yuandong, Xu Yingtao, Gu Fengxue et al., 2003. Correlation analysis of NDVI climate and hydrological
factors in oasis and desert. Acta Phytoecologica Sinica, 27(6): 816–821. (in Chinese)
Zhao Ji, 1998. Physical Geography of China. Beijing: Higher Education Press. (in Chinese)
Zheng Du, 1996. The system of physico-geographical regions of the Qinghai-Xizang (Tibet) Plateau. Science in
China (Series D), 39(4): 410–417.