The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol.

38, Part II

THE CALCULATION OF TVDI BASED ON THE COMPOSITE TIME OF PIXEL AND

DROUGHT ANALYSIS

Lingkui Meng a, Jiyuan Li a, *, Zidan Chen b, Wenjun Xi e a, Deqing Chen b, Hongwei Duan a
a
School of Remote Sensing and Information Engineering , Wuhan University, 129 Luoyu Road, Wuhan 430079, P. R.
China – lkmeng@[Link], lijiyuan_521@[Link], whuxwj@[Link]
b
Water Resources Information Center, MWR, Beijing 100053, P.R. China – (zdchen, chendq)@[Link]

KEY WORDS: Remote Sensing, Precipitation, Soil Moisture, Temperature-Vegetation Dryness Index, Drought, Time, Composite

ABSTRACT:

Temperature-Vegetation Dryness Index (TVDI) is one of the agriculture drought indexes. This paper presents a data composite
method which improves the calculation of TVDI through taking the time of pixel into consideration, and the adaptability of TVDI in
drought assessment has also enhanced significantly. First, the Normalized Difference Vegetation Index (NDVI) data series are
composed by using maximum value composite (MVC) method, and the Land Surface Temperature (LST) data series are composed
to construct NDVI-Ts feature space. Then, the wet and dry sides of NDVI-Ts feature space are fixed by a number of ways to build
new TVDI, and we note it as T-TVDI, for assessing the drought condition. To verify our proposed method, TVDI in time scale of
ten-days is established for Chongqing region in China, and the results coincide with the actual situation. Finally, the T-TVDI and
TVDI of Chongqing region in 2008 are calculated and compared. The correlations of them and Soil Moisture are analyzed as well as
Precipitation. It shows that T-TVDI has the advantages of stability and high accuracy in the short term. It is feasible to use T-TVDI
to evaluate drought in proper region and reasonable crop growth period.

1. INTRODUCTION
There are a number of conventional composite methods of
The water stress indicator (TVDI) proposed from NDVI-Ts NDVI and LST, such as MVC, CV-MVC and BRDF (These
feature space reflecting the surface soil moisture well, methods will be gave explanatory notes in what follows).
especially in large areas of vegetation coverage, is used to Whichever method is used, composites of NDVI and LST data
assess the drought condition locally. are carried out separately. Although TVDI is the statistical
values of the period, there are no real spatial and temporal
In the NDVI-Ts feature space, the expression of TVDI consistency of geography and time between the vegetation
calculation is as follows: condition and surface temperature which reduces the calculation
accuracy and evaluation effects of drought. This paper analyzes
the principles and characteristics of TVDI firstly, and then
Ts − Ts −min composes the NDVI and LST based on the time of pixel; finally
TVDI = carry out T-TVDI with a good evaluation of drought in the
T s- max − Ts − min (1) region.

Where
Ts − min = minimum of land surface temperature When 2. THE CALCULATION OF T-TVDI
the NDVI is equal to a particular value 2.1 Introduction to the theory of TVDI
Ts −max = maximum of land surface temperature When
the NDVI is equal to a particular value Temperature Vegetation Drought Index uses the relationship
between the surface temperature and soil moisture (relative soil
We can see from the above equation (1), NDVI and LST data moisture) to reflect degree of drought.
are the bases for the TVDI calculation. The NDVI is calculated
from near-infrared and red bands of the multispectral image, TVDI comprehensively considers relation and changes between
and the LST data is able to get by split-window algorithm. the NDVI and LST. From the physical mechanism, it is
Drought is a complex phenomenon, the formation and certainly hysteretic to using NDVI as Water Stress Index.
development of its strength go through a process in gradual Temperature is time-sensitive as indicator of water stress, but is
accumulation which is so slow that it is difficult to detect apt to be affected by vegetation coverage when using
during beginning period. In the process of TVDI calculation, temperature method to monitor soil moisture. TVDI integrates
NDVI and LST data are composed by time-series data vegetation indices and surface temperature to monitor soil
accumulated in a certain observing period. The rationality of moisture with the ability to composite information of visible,
composite algorithm is directly related to the quality and near infrared and thermal infrared bands of light, so it has a
accuracy of TVDI.

519
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

wider range of applicability. Figure 1 shows the simple NDVI- atmospheric correction because of minimizing the selection of
Ts feature space. cloudy and heavy aerosol pixels.

CV-MVC(constraint view angle maximum value composite):

The CV-MVC compares the two highest NDVI values and
Ts selects the observation closest to nadir view to represent the 16-
day composite cycle. This helps to reduce spatial and temporal
discontinuities in the composite product.
dry side

bare T s-max = a + b × NDVI BRDF(bidirectional reflectance distribution function): The

area
BRDF scheme is a more elaborate and constrained technique in
which all bidirectional reflectance observations, of acceptable
Partially cover quality, are utilized to interpolate to their nadir-equivalent band
reflectance values from which the VI is computed and produced.
Completely
cover
wet side T s - min = a , + b , × NDVI NDVI
Because composites of NDVI and LST data are carried out
separately by the methods above, although TVDI is the
statistical values of the period, there are no real spatial and
Figure 1. Simple NDVI-Ts feature space temporal consistency of geography and time between the
vegetation condition and surface temperature which reduces the
Temperature Vegetation Drought Index uses the relationship calculation accuracy and evaluation effects of drought.
between the surface temperature and soil moisture (Relative soil
moisture) to reflect degree of drought. In this paper, MVC and CV-MVC method are applied in
accordance with the following priority sequence of composite
TVDI comprehensively considers relation and changes between methods:
the NDVI and LST. From the physical mechanism, it is 1. CV-MVC within limited perspective: If in the period of
certainly hysteretic to using NDVI as Water Stress Index. composition, the number of days with no cloud is less than
Temperature is time-sensitive as indicator of water stress, but is 30%, and more than 2, choose the maximum of two
apt to be affected by vegetation coverage when using vegetation indexes within smallest perspective.
temperature method to monitor soil moisture. TVDI integrates 2. Calculating the vegetation index directly: If there is only
vegetation indices and surface temperature to monitor soil one day without cloud, choose the vegetation index of the
moisture with the ability to composite information of visible, day directly.
near infrared and thermal infrared bands of light, so it has a 3. MVC: If all days observed is not sunny, choose the
wider range of applicability. maximum of al the vegetation index values in the
composite period.
The NDVI-Ts feature space is able to be simplified to a triangle,
While composing the NDVI, the time information obtained of
T s − max and T s − min are linearly fit at the same time, and result every composite pixel in ten-days are stored in a band. Then
is as follow: choose the corresponding pixel to compose LST from LST data
series referring to the time raster chart.
The specific algorithm is stated as follows:
T s - max = a + b × NDVI (2)
s - min = a + b × NDVI
, ,
T
For x = 0 to Xsize
[Ts − (a + b × NDVI )]
, , For y=0 to Ysize
TVDI = (3)
For i = 0 to dayCount
[(a + b × NDVI ) − (a , + b , × NDVI )]
， ，
If the count radio of sunny day is under 30%:
where a，b，a ，b = coefficients of dry side and wet side If the count of sunny equal one
fitting equation NDVComposite[x,y] = NDVI[i]
Day[i] = the date of NDVI[i]
The range of TVDI is [0, 1], TVDI = 1 on the dry side, TVDI = Else
0 on the wet side. The greater the value of TVDI, the lower the
soil moisture, and the higher the level of drought will be.
NDVComposite[x,y] get the maxNDVI
day[i] = the date of maxNDVI
2.2 The composite of NDVI and LST Endif
Else
The purpose of composite is to choose the best observational NDVComposite[x,y] = maxNDVI of two values
data. This method should be able to ensure spatial and temporal within smallest perspective
consistency of vegetation index values. At present, there are
following common ways of composite:
day[i] = the date of maxNDVI
Endif
MVC(maximum value composite): Select the maximum NDVI EndFor
value of observed pixel as the vegetation value in composite dateComposite[x,y] = day[i]×100
period. It is the best method for information with no LSTComposite[x,y] = LST[i] of day[i]
EndFor

520
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

EndFor 70 1200
EndFor

Land S urface Tem p eratur e/℃

60 y=50x+66 1000
y=40x+56
NDVComposite[x,y], dateComposite[x,y] and 50 y=30x+46 800
y=20x+36

frequence
LSTComposite[x,y] are separately composite data of NDVI,
date image and LST. Day[i], LST[i] and NDVI[i] respect 40 y=10x+26
600
separately pixels of date image, LST and NDVI series in ten-
days. The figure 2 shows the visualization of this method. 30
400
20
200
y 10
Pixel
0
0 0.2 0.4 0.6 0.8 1.0
Vegetation index
Figure 3. Picture of standard NDVI-Ts feature space

3. EXPERIMENT OF T-TVDI CALCULATION

x NDVI Day LST In this paper, Chongqing region in China is selected as the
Figure 2. Composite of LST referring to the time raster chart experimental area where the typical southern dry farming areas
are with frequently occurrence of drought.
2.3 Fixing of wet and dry sides
The data using in the paper include remote sensing data and
The maximum and minimum temperature corresponding to measured data. Remote sensing data with 1-km resolution is
NDVI as the thermal and cold edges of NDVI-Ts feature space provided by MODIS (Moderate Resolution Imaging
are obtained by method of maximum and minimum. The Spectroradiometer), and measured data (soil moisture and
thermal and cold edges obtained by the method are direct- rainfall) in the experiment are from all the hydrological stations
viewing, clear and easy-to-linear fitting with rapidly handling located in various parts of Chongqing region (Figure 8 & Figure
and showing. But sometimes the pixels of the sides are 9). Soil moisture data are from calculation of measured data in
scattering and shapes are irregular. Then we will consider using the depth of 10cm, 20cm, and 50cm of soil.
the scatter plot to fix the thermal and cold edges.
First of all, calculate the LST through split-window algorithm
Fix the wet side referring to the average of cold edge (base on using NDVI dataset. Then, two methods of TVDI calculation
the cloud removing), water surface temperature and average in are adopted. One is the method using MVC to compose NDVI
the same period of many years in the region. According to the and LST series of 10 day. The other is the methods composing
histogram of image, remove the pixels at the end of the thermal NDVI and LST data by the way mentioned above, extracting
edge with fixing the boundaries at 1% of the total number of the minimum and maximum LST in different climatic zones
pixels, and obtain the parameters of dry side by linear fit. Due and every ten-days under the different conditions of NDVI with
to irregular thermal edge, the parameters may be not good. a small step size.
Then, we set up a standard feature space as background
reference picture according to the relationship between wet and The NDVI-Ts feature spaces from two methods are as follow
dry sides. It can be used to adjust and fix the parameters of wet (data of ten days are acquired in mid-May 2008):
and dry sides. Standard feature space previously divides the
LST in appropriate interval in feature space, and then
establishes the equations of all corresponding dry sides. This
method remains an error. But with the smaller the intervals, the
error is smaller. The parameters of dry side are replaced by
corresponding contour in reference picture of standard feature
space as following figure.

Figure 4. NDVI-Ts feature space based on the Composite Time

of pixel

521
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

and atmospheric. Relative Soil Moisture can be offset the

impact of soil texture to some extent. Figure 7 shows the
relative soil moisture and stations of Chongqing in mid-May
2008. As can be seen, spatial distribution of relative soil
moisture is uneven that in the north is less than in the south.

Figure 5. NDVI-Ts feature space based on MVC

As can be seen from the figures above, angle of triangle in the

second feature space is bigger than the first one, because most
of LST concentrate in high-temperature region, especially in the
wet side. It is difficult to fix the boundaries of LST. Slope of
dry side is less than 0, indicating that LST is decreasing with
the vegetation coverage increase. On the contrary, the slope of
wet side is greater than 0, indicating that there is increasing
trend for LST while vegetation coverage decreases. Figure 7. Relative soil moisture and stations of Chongqing in
mid-May 2008
According to scatter gram of the first feature space, fit the wet
and dry sides as follows: Precipitation is the main factors affecting drought situation, the
amount of precipitation reflects the weather conditions basically.
Standard Precipitation Index (SPI) is easy to compute and
T s - max = 5.6 + 34.5 × NDVI (4)
access to necessary information, and because of not involving a
T = 13.5 - 40.5 × NDVI specific mechanism of drought situation, the space-time
s - min
adaptability is more suitable. The figure 8 shows standard
precipitation index and stations of Chongqing in mid-May 2008.
The graph shows that the rainfall in most parts of the province
Finally the T-TVDI result from calculation shows that heavy- ten days is less than normal. The west of Chongqing seriously
dry appeared in the middle, southeast and small part of east of lack of rainfall.
Chongqing region. RSM(relative soil moisture) of northern
areas is suitable for the growth of crops, and light-dry appeared
in south-western regions. From the experimental results,
detection is generally consistent with the reality.

The figure 6 shows the result of T-TVDI calculation:

Figure 8. Standard precipitation index and stations of

Chongqing in mid-May 2008
Figure 6. T-TVDI of Chongqing in mid-May 2008
The scatter plot of T-TVDI and RSM from Chongqing in mid-
May 2008 is given in the following, as well as the scatter plot of
T-TVDI and RSM.
4. DROUGHT ANALYSIS

Soil moisture and rainfall are two normal indexes to evaluate

soil drought situation, they can better reflect the intensity and
duration of drought situation. Soil moisture is the field soil
moisture and the corresponding crop water status, as one of the
indicators of drought situation. It plays an important role in the
exchange between the water and energy exchange of surface

522
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

T-TVDI--RSM Scatter Plot

Figure 10. Scatter plot of T-TVDI and SPI in mid-May 2008 of
1.2 Chongqing
1
As can be seen from figure 9 and figure 10, it is obvious that
0.8
scattered points distribute nearby on both sides of a straight line
RSM

0.6 and show a good linear distribution that indicating a good linear
0.4
y = -0.9474x + 1.1587
model. The correlation arrives 0.62 and the linear model has
0.2 R2 = 0.3877 passed the significance level F test with reliability of 0.01. At
the same time, T-TVDI also has a good correlation with SPI,
0
0 0.2 0.4 0.6 0.8 indicating T-TVDI has a good effect in evaluating the situation
T-TVDI of the regional drought.
Figure 9. Scatter plot of T-TVDI and RSM in mid-May 2008 of
Chongqing To carry out further analysis and evaluation of drought situation,
this paper uses two kinds of TVDI data calculating from
T-TVDI--SPI Scatter Plot
MODIS data covering Chongqing region in every ten-days to
1 analyze the adaptive real-time indices. The analysis of adaptive
real-time indices is respect that doing correlation analysis
0.8
between the index and RSM in the ten-days. It emphasizes
0.6 adaptability of index at the time to RSM, and provides a basis
SPI

for choosing real-time index for drought monitoring. As a

0.4
reference, two types of indexes and the precipitation index are
0.2 y = -0.544x + 0.8611 analyzed in the same way.
R2 = 0.2419
0
0 0.2 0.4 0.6 0.8 1 1.2
T-TVDI

0.8

0.6

0.4 T-TVDI
0.2 TVDI

0
0 5 10 15 20 25 30 35 40
-0.2

Figure 11. Correlations gram between RSM and two indexes of Chongqing in 2008

0.6
0.4
0.2 T-TVDI
0 TVDI
-0.2 0 5 10 15 20 25 30 35 40

-0.4

Figure 12. Correlations gram between SPI and two indexes of Chongqing in 2008

The figure 11 shows that in the winter the two indexes are not
too high correlating with RSM, this is because of the low 5. CONCLUSION
vegetation cover in winter, the TVDI sensitivity is not high for This paper uses two different composite methods to obtain
the ground. In the crop growing period, the correlation is NDVI and Ts in ten-day scales from MODIS data in 2008, and
increasing with the crop growing,. The correlation of T-TVDI then constructs the NDVI-Ts feature space. On the basis of this
and RSM is better than TVDI. NDVI-Ts feature space, TVDI and T-TDVI that reflecting the
drought situation are created and used to evaluate the drought
The figure 12 shows that T-TVDI and TVDI are not good situation. Comparing with the surface soil moisture data and
correlating with SPI, but the trend of T-TVDI was relatively precipitation data getting from hydrological observation sites ,
stable, so it is an acceptable range of fluctuation for the the result shows that , TVDI basing on the composite time of
evaluation of drought situation. pixel can better reflects the correlation with soil moisture, more
suitable for monitoring and evaluating drought situation.

523
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

TVDI only adapts to a relatively high vegetation coverage area, Gengming Jiang, Zheng Niu, Weili Ruan, Changyao Wang,.
the higher of vegetation coverage, and the greater of the Cloud-free composition of MODIS data and algorithm
correlation with the RSM. Because of the significant difference realization. Remote sensing for land and resource, No.2, Jun,
of wet and dry side in different regions, the evaluated range 2004, pp. 11-15.
should not be too big by using TVDI.
Members of the MODIS Characterization Support Team,
Because satellite view affects the amount of information MODIS Level 1B Product User’s Guide, NASA/Goddard Space
received by sensors, thus vegetation growth status shows in Flight Center, December 1,2003, pp. 15-50.
images further. This study does not take into account the
impact of satellite view to the NDVI and LST.

6. ACKNOWLEDGEMENT
The work presented in this paper is supported by National 948
project of China (200610) which is named Hydraulic
information service of grid system, armed at introducing
advanced foreign technology. Thanks for the Leadership and
colleagues in Water Information Center of The Ministry of
Water Resources in China who provided the MODIS L1B data,
thanks all the same to the Leadership and colleagues in Beijing
Golden-Water Info-Tech Ltd. who supply the necessary
experimental equipment and the environment.

7. REFERENCES
Kogan F N. 1995. Application of vegetation index and
brightness temperature for drought detection. Adv. Space Res.
15(11), pp. 91-100.

Cuizhen Wang, Jiaguo Qi ， Susan Moran ， Robin Marsett.

2004. Soil moisture estimation in a semiarid rangeland using
ERS-2 and TM imagery Preliminary results. Remote Sensing of
Environment，90, pp. 178-189.

Sandholt et al. 2002. Rasmussen K, Andersen J.A Simple

Interpretation of the Surface Temperature/Vegetation Index
Space for Assessment of Surface Moisture Status. Remote
Sensing of Environment. 79, pp. 213-224.

Njoku, E.G. Li, L. 1982. Retrieval of Lnad Surface Parameters

Using passive microwave measurements at 6-18GHz. IEEE
Transanctions on Geoscience and Remote Sensing， 20(4), pp.
468-475

England AW, Galantowkz J F. 1992. [Link]

Radiobrightness Themral Inertia Measure of Soil Moistuer.
IEEE Trans Geosci. Remote Sensing, 30(1), pp. 132-139.

Carlson T N, et al. 1990. Remote sensing estimation of Soil

moisture availability and fractional vegetation cover for
agricultural fields. Agri.& Fore. Meteo. 52, pp. 45-69.

Nemani, R.R., et al. 1993. Developing satellite derived

estimates of surface moisture status. Journal of Applied
Meteorology, 32, pp. 548-557.

Lingkui Meng, Liang Tao, Jiyuan Li and Chunxiang Wang. A

System for Automatic Processing of MODIS L1B Data.
Proceeding of the 8th International Symposium on Spatial
Accuracy Assessment in National Resources and
Environmental Sciences, Shanghai, [Link], June 25-27,
2008, pp.335-343.

524