You are on page 1of 10

Journal of Geographical Sciences

© 2007 Science in China Press Springer-Verlag

DOI: 10.1007/s11442-007-0399-9

Terrain revised model for air temperature


in mountainous area based on DEMs:
A case study in Yaoxian county
YANG Xin1, *TANG Guoan1, XIAO Chenchao1, DENG Fengdong2
1. Key Laboratory of Virtual Geographic Environment of Ministry of Education, Nanjing Normal University,
Nanjing 210046, China;
2. Shaanxi Remote Sensing Information Center for Agriculture, Xi’an 710015, China

Abstract: In mountainous area, spatial interpolation is the traditional method to calculate air
temperature by use of observed temperature data. Due to lack of sufficient observation data
in mountainous areas many precise interpolation methods could give only coarse result which
could not meet the demand of precision agriculture and local climate exploration. Based on
DEMs of 25 m resolution, a reversed model is constructed, with which temperature is simu-
lated to the corresponding slope unit from the solar radiation. Taking Yaoxian county as a test
area, and mean monthly temperature data as basic information sources, which are collected
from 15 weather stations around Yaoxian county in Shaanxi province from the year of 1970 to
2000, a simulation for the solar radiation cell by cell is completed. By simulating solar radia-
tion at each slope and flat cell unit, the terrain revised temperature model could be realized. A
comparison between the simulated temperature and the radiation temperature from TM6
thermal infrared image shows that the terrain improved model gets a finer temperature dis-
tribution at local level. The accuracy of simulated temperature in mountainous area is higher
than it is in flat area.

Keywords: DEM; solar radiation; temperature; simulation; Shaanxi Province

1 Introduction
In mountainous area topography varies greatly, so does the spatial distribution of air tem-
perature accordingly. Air temperature in mountainous area is influenced by many factors
which are composed of both global factors, i.e. altitude, latitude, distance to large water
body, mountain structure, and local factors as slope gradient, aspect, terrain shading and
characteristics of land cover (Fu, 1996). In a certain local region, under the same global cli-

Received: 2007-01-13 Accepted: 2007-03-25


Foundation: National Natural Science Foundation of China, No.40671148, No.40571120; Specialized Research Fund for
the Doctoral Program of Higher Education, No.20050319006; Chair Professor Foundation of Nanjing
Normal University
Author: Yang Xin (1976−), Ph.D, specialized in digital terrain analysis. E-mail: xxinyang@163.com
*Corresponding author: Tang Guoan, E-mail: tangguoan@njnu.edu.cn

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

mate condition, local factors play a key role in the value and variation of surface tempera-
ture to influence temperature at local level (Weng et al., 1990). Virtually, due to the insuffi-
cient climate observation data, the temperature in each slope unit could not be achieved, a
global and comprehensive climate data usually can not meet the demand of precision agri-
culture and local climate exploration.
In rugged areas spatial interpolation is the traditional method for calculating the air tem-
perature (Fu et al., 1996; Weng et al., 1990; Pan et al., 2004). For the insufficient climate
observation sites, those precise interpolation methods such as IDW, SPLINE, KRINGING,
etc. are usually not available and suitable for getting a correct and refined result. Thus, sta-
tistic regression has to play a role in the process (Ouyang et al., 1996; Oleg A, 2001; Pan et
al., 2004). Even so, the results can only reflect the global trend of temperature which varies
with altitude, latitude and longitude. It is still hard to obtain detailed spatial distribution of
temperature on slope unit. In recent years, with the development of digital terrain analysis
methods and multi-scale national DEM database, DEM has been applied to simulate the
surface temperature of rugged area (Zhang et al., 2002; Yang et al., 2006). It is well known
that the surface temperature show a positive correlation with solar radiation. The discrep-
ancy of surface temperature results from a diversity of solar radiation received (Weng, 1997).
The study on the simulation of solar radiation based on DEMs (John P Wilson et al., 2000;
Yang et al., 2004, Zeng et al., 2003) provides experience and methods for reference to the
further study on the simulation of surface temperature more accurately and elaborately.

2 Test area and data


2.1 Test area

Yaoxian county, located in the central Shaanxi province, southern part of the Loess Plateau,
is chosen as a test site for its ideal terrain condition and research basis. The geo-coordinate is
between 108.57°–109.1°E and 34.83°–35.33°N, with a total area of 1613 km2. There are
three geomorphic regions: loess low hilly region, loess gully-hilly region, and loess yuan
(loess tableland). Along N-S trending, elevation reduced gradually, ranging from 1713 to
543 m asl, with a mean elevation of 1114 m. The large relative heights in the test area lead to
a corresponding decrease of local temperature from north to south. The slope ranges from 0
to 68º with a mean value of 16.8º. Yaoxian county lies in the warm temperate zone and has a
continental sub-humid, semi-arid monsoon climate, with low rainfall (about 740 mm a year)
and rich solar radiation.
2.2 Experimental data
DEMs: The test area is composed of 10 DEM sheets of map at a scale of 1:50,000 with a
resolution of 25 meters. All of these DEMs are produced from the contours of 1:50,000 re-
lief maps. A Triangulated Irregular Network (TIN) was firstly generated by use of contours
and elevation points, then the grid DEM was interpolated with a resolution of 25 m of the
Xian projected coordinate system.
Climate data: The climate data are collected from 15 weather stations around Yaoxian
county in Shaanxi province. They are Yijun with an elevation of 1395.2 m, Baishui (804.4 m),
YANG Xin et al.: Terrain revised model for air temperature in mountainous area based on DEMs 401

Figure 1 Location of Yaoxian county in Shaanxi province of China and the location of weather stations around
Yaoxian county (a), and topographic hillshade map of Yaoxian county (b)

Yongshou (994.6 m), Pucheng (499.2 m), Dali (368.4 m), Weinan (348.8 m), Gaoling (377.6
m), Jingyang (427.4 m), Qianxian (636 m), Binxian (840.1 m), Xunyi (1325.2 m), Tong-
chuan (978.9 m), Chunhua (1012.7 m), Fuping (470.9 m) and Yaoxian (710 m). The mean
monthly temperature is collected from 1970 to 2000.
Remote Sensing Image: The TM image was obtained on May 28, 1997. The TM6 thermal
infrared image was processed for comparing with the simulated result of the new model. The
procedure includes geometric correction, image matching, resampling and normalization.
The resolution of TM6 is 120 m. In order to match with the result of the new model, the
resolution of the image was converted to 25 m by resampling.

3 Models
3.1 Traditional models

Various interpolation methods were used to calculate spatial distribution of temperature.


Lots of them, such as Inverse Distance Weighting (IDW), Spline and Kriging interpolation
etc. need enough sampling points to acquire exact and refined result. Hence statistic regres-
sion of temperature against altitude, latitude and longitude (formula 1) is usually applied.
T = a0 + a1ϕ + a2 λ + a3 H (1)

where T is temperature (℃), ϕ is latitude (º), λ is longitude (º), and H is altitude (m).
They are the geo-information of the observed points from weather stations. The parameters
a0, a1, a2 and a3 are the regression coefficients. In the model, the latitude, longitude and al-
titude reflect the changes of temperature from equator to two poles, from ocean to inland and
402 Journal of Geographical Sciences

from sea level to high mountains respectively. Although the model could give a comprehen-
sive figure in revealing temperature distribution at regional level, it still fails to reveal the
difference of temperature at local level.

3.2 Improved model

Solar radiation is the main source of surface temperature. Weng (1997) proved that the dis-
crepancy of air temperature could found its reason from the variation of the global solar ra-
diation received. An improved model is put forward as follows:
T ' = T + ΔT (2)

where T′is the air temperature after terrain amend, T is the temperature calculated by for-
mula (1), ΔT denotes the temperature difference between the slope unit and correlated flat
unit. When the astronomical parameters, atmospheric circulation and geographical factors
are similar in certain area, the influence of slope gradient, aspect, and terrain shadings are
relatively obvious. These local factors can be integrated to global radiation model which
synthetically reflects local difference of temperature. As is proved that there exist a close
relation between global radiation and temperature (Schmugge et al., 1998; Qin et al., 2001).
Therefore, the relationship between temperature and global radiation, as well as slope and
flat can be expressed as follows:
Qslope Tslope
= (3)
Q flat T flat
where Qslope and Q flat stands for global radiation amount (MJ/m2) at each slope unit and
flat unit separately. Tslope is the temperature of slope unit. Hence, temperature difference
ΔT between slope and flat is:
Qslope − Q flat
ΔT = T flat (4)
Q flat

where T flat is mean monthly temperature, calculated by formula (1). Q flat denotes astro-
nomical radiation of flat unit. It reflects the impact of those global factors on temperature.
Qslope is astronomical radiation of slope unit, which is able to reflect the impact of local fac-
tors on temperature. In rugged area, global radiation is the sum of direct solar radiation, sky
diffuse radiation and reflect radiation, which is part of solar shortwave radiation from terrain
reflection. The reflect radiation is less than 5% of the first two items, and can be ignored.
Under the same climate condition in the loess hilly region, the proportion of the direct radia-
tion is bigger than the diffuse radiation. The characteristics of global radiation are dominated
by direct radiation. Furthermore, direct solar radiation has the same spatial distribution
characteristics as the astronomical radiation.
As a result, the ratio of global radiation from the slope unit to the flat unit is approxi-
mately the same as the ratio of astronomical radiation from the slope unit to the flat. So the
terrain revised model can be expressed as follows:
S0 slope
T'= (a0 + a1ϕ + a2 λ + a3 H ) (5)
S0 flat
YANG Xin et al.: Terrain revised model for air temperature in mountainous area based on DEMs 403

where S0slope is the astronomical radiation in each slope unit, and S 0 flat is the astronomical
radiation in each flat unit.
I TE n
S0 slope = 0 0 ∑ gi ⎡u sin δ (ωs ,i − ωr ,i ) + v cos δ ( sin ωs ,i − sin ωr ,i ) − wcosδ (cosωs,i − cosωr,i ) ⎦⎤ (6)
2π i =1 ⎣

u = sin ϕ cos α − cos ϕ sin α cos β


v = sin ϕ sin α cos β + cos ϕ cos α (7)
w = sin α sin β

24
S0 flat = I 0 E0 (ωs sin ϕ sin δ + cos ϕ cos δ sin ωs ) (8)
π
where E0 is revised factor of earth orbit, I0 is solar constant (4.9212 MJ/m2·h), T is the length
of a day, δ is solar declination, N is the dispersed number of insolation-duration, ω r ,i
and ω s,i are sunrise hour angle and sunset hour angle on the slope in the ith differential time
unit, gi is terrain shading, and ωs is flat sunset hour angle.

4 Methods
There are two major processes. One is statistic regression by climate observed data which in
order to fix on the parameters of temperature models. Formula (9) is the regression equation
of annual average temperature by using the mean monthly temperature information within
30 years collected from 15 weather stations around Yaoxian county.
T = −2.456 − 1.0 × 10−6 X + 4.7 × 10−6 Y − 4.88 × 10−3 H (9)
where X is the distance to the east (m), Y is the distance to the north (m), and H stands for
the elevation (m). Multiple correlation index (R) is 0.97, and passed the significance test
when F is 0.01.
The other process is the simulation of solar radiation over both flat and slope area based
on DEM. Figure 2 shows the procedure for the temperature simulation. Key problems to
realize simulation of air temperature process are the calculation of astronomical radiation. In
rugged area, the exact sunshine hours on each site are most important to the correct simula-

Observed temperature data


Temperature after terrain revision

Spatial distribution of temperature in flat unit

Slope
DEM

Astronomical radiation in slope unit


Aspect

Terrain shadings
Adjust coefficient

Flat astronomical radiation Total flat radiation

Figure 2 The procedure for the temperature simulation


404 Journal of Geographical Sciences

tion (Dozier 1990, Javier et al., 2003). For the convenience of computing, the duration of
insolation in each DEM grid cell is dispersed (n=36). Along with the shift of solar astro-
nomical position, the sunlight reaching in each grid unit is deduced in every differential time
unit with a consideration of terrain factors. The effective sunshine hours of a day in every
single cell has been calculated by summing up all differential time units. When the flat as-
tronomical radiation and slope astronomical radiation are available, the revised coefficients
could be derived and a cell by cell revised surface temperature matrix could be achieved.

5 Result
5.1 Simulation effect evaluation
Annual mean temperature of Yaoxian county was mapped and compared with that from the
traditional model (Figure 3). Experimental results show that new model is more reasonable
in expressing the characteristics of surface temperature. It shows that the spatial distribution
of air temperature has strong relativity with relief. The traditional method (statistic regres-
sion) can present the general trend of temperature spatial distribution. The new method (after
terrain revised) can reveal temperature diversity with terrain in detail. Table 1 shows that the
range of annual mean temperature is getting larger after terrain revised (Figure 4), especially
the temperature in the northern low hilly region drop greatly and the minimum value. It is
because the new model takes into account the influence of terrain. Temperatures decrease in
some shady area which is always with little solar radiation.
A comparison of annual mean temperature histogram (Figure 4) has set up. It shows that

Figure 3 Comparison of annual mean temperature before and after terrain revision in Yaoxian county, China

Table 1 Comparison of simulated temperature (℃)


Items Max Min Mean Std dev.
Before terrain revision 12.86 7.32 10.22 1.17
After terrain revision 13.08 1.41 9.25 1.52
YANG Xin et al.: Terrain revised model for air temperature in mountainous area based on DEMs 405

the diversity of the temperature distribution is enhanced after terrain revision. The range of
temperature varies from six divisions to thirteen divisions.
Figure 5 shows that temperature drops quickly along with the increase of slope gradient
after the terrain revision process. The trend is not obvious before terrain revision. Figure 6
shows the trend of simulated annual mean temperature varies with slope aspect. After terrain
revision temperature presents in a rather regular pattern. Temperature is higher on sunny
slope than it is on shady slope, and it seems to be quite balanced on the eastern and western
oriented slope. The statistical result shows the mean temperature on sunny exposed slope
(southern, southeastern and southwestern slope) is 9.83℃, 1.44℃ higher than that on shady
slope (northern, northwestern and northeastern slope).

Figure 4 Histogram of annual mean temperature before and after terrain revision

Figure 5 Temperature against slope before and after terrain revision

Figure 6 Temperature against aspect before and after terrain revision


406 Journal of Geographical Sciences

Figure 7 Average temperature simulated in May of Yaoxian county (a), and TM6 thermal infrared image and
location of test section plane in Yaoxian county of Shaanxi province (b)

Temperature after terrain revision becomes lower than that before terrain revision (Table 2,
Figures 5 and 6). It is mainly because of the revised coefficient. The ratio of astronomical
solar radiation between flat and slope units is different from the ratio of total solar radiation
between flat and slope units. In astronomic radiation model the unit shadowed by terrain is
usually with small value, the coefficient gets small accordingly.

5.2 Comparison and validation

In order to validate simulated result, this paper conducts a comparison between the simu-
lated temperature and the radiation temperature from Landsat TM6 thermal infrared image.
Landsat TM6 thermal image (Figure 7b) contains ground surface heat. The gray value of
TM6 image reflects the land surface temperature. The greater the gray value is, the higher
the temperature (Qin et al., 2001). The mean temperature in May (Figure 7a) was simulated,
which corresponded with the nearly same time of TM image product. The comparison is
conducted by the method of normalization. The equation of normalization is as follows:
Z − Z mean
Zi ' = i (10)
Z max − Z min

where Zi is the old value in each grid, Z i ' is the new value after standardization,
Z max , Z min and Z mean are the maximum, minimum and mean values of both simulated tem-
perature and TM6 image grid layer. A few pre-processes, i.e. geometry adjustment, resam-
pling and standardization are done to make TM6 image match with simulated temperate ma-
trix. The simulated temperatures are also normalized for comparison. Several typical profiles
were selected in the test area located in different sections of Yaoxian county (Figure 7b).
Figure 8 shows the result of comparison across three different landforms. Profiles 2, 3 and
4 located in low hilly region, gully-hilly region and loess yuan (loess tableland) respectively.
A general similarity could be found in their distribution pattern, except for some specific
YANG Xin et al.: Terrain revised model for air temperature in mountainous area based on DEMs 407

land cover area, i.e. a reservoir and river could lead to the drop of temperature (see A and D
in Profile 1 and Profile 4), and urban heat island effect could lead to the rise of temperature
(see C in Proflie 2).

Figure 8 A comparison of temperature between DEM simulated temperature and TM6 radiation temperature in
Yaoxian county

The result proves that terrain revised model has more reasonable result than the traditional
one in the low hilly and loess gully-hilly regions (Figure 2). In loess plain, the errors are
relative large owning to the complexity of land use and land cover.

6 Conclusions and discussion


The terrain revised temperature model based on DEMs can present more detailed informa-
tion of temperature distribution with terrain. In hilly and mountainous areas, the terrain re-
vised model is valid and refined to spatial calculating of air temperature, which is of espe-
cially significance in those areas where observed data is not available. Taking solar radiation
as adjust coefficient of temperature, the model can overcome the difficulty of integrating
terrain factors into statistic regression model, which can also improve the precision of tem-
perature spatial calculation.
In the model, the revised coefficient, the ratio of astronomical radiation from slope to flat
area, is different from the ratio of global radiation from slope to flat area virtually. In Loess
Plateau area, the proportion of diffuse radiation in global radiation is relative lower for the
408 Journal of Geographical Sciences

rarely appeared clouds; therefore, the error from the replacement of global radiation by as-
tronomical radiation could be ignored.
New model involves local terrain factors such as slope gradient, aspect and terrain shad-
ings without dealing with the obstruct effect of terrain to atmosphere advection. In some
high mountains, wind is blocked off by terrain and can cause great difference of temperature
between windward slope and leeward slope, which should be taken into consideration in the
corresponding temperature models as well.
By comparing with TM6 images, the new model can achieve a reasonable result in low
hilly and gully-hilly region, but less ideal result in loess tableland. The main possible reason
lies in different land covers. Hence, further works should focus on the impact of land cover
in order to achieve a more perfect result.

References
Collins F C, Bolstad P V, 1996. A comparison of spatial interpolation techniques in temperature estimation. In:
Proceedings of the Third International Conference/Workshop on Integrating GIS and Environmental Modeling,
Santa Fe, New Mexico, January 21−25. Santa Barbara, California: National Center for Geographic Information
Analysis.
Dozier J, Frew J, 1990. Rapid calculation of terrain parameters for radiation modeling from digital elevation data.
IEEE Transaction on Geoscience and Remote Sensing, 28(5): 963−969.
Fu Baopu, Yu Jingming, Lu Qirao, 1996. Mountains Climatic Resources and Its Application. Nanjing: Nanjing
University Press. (in Chinese)
Guglielmina M, Aldighierib B, 2003. PERMACLIM: Testa. A model for the distribution of mountain permafrost,
based on climatic observations. Geomorphology, 51: 245−257.
Javier G Corripio, 2003. Vectorial algebra algorithms for calculating terrain parameters from DEMS and solar
radiation modeling in mountainous terrain. Int. J. Geographic Information Science, 17(1): 1−23.
John P Wilson, John C Gallant, 2000. Terrain Analysis: Principles and Applications. John Wiley & Sons, Inc.
Li Xin, Cheng Guodong, Chen Xianzhang, 1999. Optimization of solar radiation model in any situation. Chinese
Science Bulletin, 44(9): 993−998. (in Chinese)
Oleg A, 2001. Anisimov prediction patterns of near-surface temperature using empirical data. Climatic Change,
50: 297−315.
Ouyang Zongji, Zhao Xinping, Zhao Youzhong, 1996. A small net-method for local climate calculation in moun-
tain area. Transaction of the CASE, 12(3): 144−148.
Pan Yaozhong, Gong Daoyi, Deng Lei et al., 2004. Smart distance searching-based and DEM-informed interpola-
tion of surface air temperature in China. Acta Geographica Sinica, 59(3): 366−374. (in Chinese)
Qin Zhihao, Zhang Minghua, Arnon Karnieli, 2001. Mono-window algorithm for retrieving land surface tem-
perature from landsat TM6 data. Acta Geographica Sinica, 56(4): 456−466. (in Chinese)
Weng Duming, 1997. Radiation Climate in China. Beijing: China Meteorological Press. (in Chinese)
Weng Duming, Luo Zhexian, 1990. Mountains Topographical Climate. Beijing: China Meteorological Press. (in
Chinese)
Yang Qing, Shi Yuguang, Yuan Yujiang et al., 2006. Temperature and precipitation series in the Tianshan Moun-
tains calculated based on DEM. Journal of Glaciology and Geocryology, 28(3): 337−342. (in Chinese)
Yang Xin, Tang Guoan, Wang Lei, 2004. Modeling of global radiation over rugged areas based on DEM. Geog-
raphy and Geo-information Science, 20(5): 41−44. (in Chinese)
Zeng Yan, Qiu Xinfa, Liu Changming et al., 2003. Study on astronomical solar radiation distribution over Yellow
River basin based on DEM data. Acta Geographica Sinica, 58(6): 810−816. (in Chinese)
Zhang Hongliang, Ni Shaoxiang, Deng Ziwang, 2002. A method of spatial simulating of temperature based Digi-
tal Elevation Model (DEM) in mountain area. Journal of Mountain Science, 20(3): 360−364. (in Chinese)
Zuo Dakang, Zhou Yunhua, 1991. Research on Earth Surface Radiation. Beijing: Science Press. (in Chinese)

You might also like