ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022

XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

RESEARCH ON NDVI NORMALIZATION METHOD BASED ON GF IMAGES

Yuting Tao1, Wenli Huang1*, Wenxia Gan2, Huanfeng Shen1

1School of Resource and Environmental Sciences, Wuhan University, Wuhan, Hubei, 430079, PR China
2School of Civil Engineering and Architecture, Wuhan Institute of Technology

Commission III, WG III/6

KEY WORDS: NDVI products, GF-1, GF-2, Single-scene Global Linear Model, Multi-scene Global Linear Model, Maximum
Value Composite

ABSTRACT:

The existing NDVI products have problems in terms of low spatial resolution and inconsistent values at a large geographical scale.
Based on medium and high-resolution multi-source remote sensing data (GF-1 and GF-2 data), this paper normalized NDVI by
combining absolute radiation normalization with relative radiation normalization. And the existing relative radiation normalization
method, single-scene global linear normalization (SGloLM) method, is improved to adapt to the production of large-range high-
resolution NDVI products. Aiming at the problem of obvious mosaic seams when the SGloLM method is applied to multi-scene
images, it is mainly improved from two aspects. One is to improve the coefficient solution of the SGloLM algorithm and propose a
new method considering the surrounding multi-scene data, the multi-scene global linear model (MGloLM). The other is to
incorporate the Maximum Value Composite (MVC) method to synthesize the maximum value of NDVI at different times in a season,
to represent the optimal situation of vegetation growth in the current season. In this study, combined experiments of different
methods were performed, as well as qualitative and quantitative evaluations. The experimental results show that SGloLM+MVC and
the MGloLM+MVC methods can better eliminate the mosaic seams, and their histogram is most similar to the histogram of standard
data, and all quantitative evaluation indexes of SGloLM+MVC are optimal (CC=0.7804, MAD=0.0643, RMSE=0.1012).

1. INTRODUCTION variation characteristics of vegetation, and provide important

support for the dynamic monitoring of vegetation at the regional
Vegetation plays a vital role in the ecosystem and is closely and global scales. At present, it has been widely used in the
related to natural environment elements such as soil, fields of vegetation growth monitoring, seasonal and
topography, climate, and hydrology, and has a profound impact interannual variation analysis, etc (Gan, 2015).
on the energy balance of the earth-atmosphere system (Zhang,
2009), and has long been of great interest to scientists and However, most of the existing vegetation index products use
governments worldwide. As one of the most effective means of single remote sensing data, which have certain deficiencies in
global vegetation monitoring at present, satellite remote sensing temporal resolution, spatial resolution, accuracy and stability
can be free from the constraints of social and natural conditions. (Ge, 2016), and the integrated use of multi-source remote
And it can quickly obtain large-scale observation data, which sensing data can compensate for these deficiencies to a certain
provides us with conditions for studying and monitoring global extent. Because of the differences in sensors and imaging
or regional vegetation growth changes (Guo, 2003). Based on conditions of multi-source remote sensing data, it is necessary
the vegetation's spectral properties, various vegetation indexes to normalize them. Radiation normalization is divided into
have been generated by linear or nonlinear combination absolute radiation normalization and relative radiation
operations on the visible light and near-infrared bands of remote normalization, and when normalizing multi-temporal data from
sensing data (Guo, 2003; Zhang, 2009). So far, more than one a single sensor, simple relative radiation normalization can
hundred and fifty vegetation index models have been proposed achieve high accuracy. However, in quantitative applications,
in various literature. Among them, the Normalized Difference the absolute radiation normalization should be combined with
Vegetation Index (NDVI) proposed by Deering (1978) is not the relative radiation normalization, and the influence of
only easy to calculate, but also enhances the sensitivity to reflection differences caused by different sensors due to spectral
vegetation by eliminating most of the effects related to sensor response characteristics and atmospheric correction on the
radiometric calibration, topography, atmospheric conditions, normalization results should be considered. And the
and observation angle (Ge, 2016). Hence, it has become one of normalization of multi-source data should be carried out at the
the most widely used vegetation indexes today. As one of the surface reflectance level to truly achieve normalization between
important parameters to describe the characteristics of surface different images from different sensors (Xu, 2019). For a long
vegetation coverage, vegetation index is a simple, effective and time, many research scholars have conducted a lot of research
empirical measure of surface vegetation status (Guo, 2003). It work on the normalization of multi-source remote sensing data.
has been widely used in qualitative and quantitative evaluation However, the previous studies on NDVI normalization of multi-
of vegetation coverage and its growing status. The NDVI source data mostly focus on low and medium-resolution data
obtained based on satellite remote sensing data can such as MODIS and Landsat, and less on high-resolution data.
comprehensively reflect the growth status and seasonal

*Corresponding author. Email: wenli.huang@whu.edu.cn (Wenli Huang).

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 209
 ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022
XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

This study proposes an improved normalization method for the

construction of high-resolution quantitative product datasets to
improve the dynamic monitoring of urban vegetation resources
by normalizing medium and high-resolution multi-source
remote sensing data based on previous experiences and methods.

2. MATERIALS AND METHODS

2.1 Data and study area

This study used DN data from the GaoFen-1 (GF-1) wide field
of view (WFV) sensor and the GF-2 panchromatic multispectral
sensor (PMS). These data were available for free download
from the China High-resolution Earth Observation System
(CHEOS) grid platform (www.cheosgrid.org.cn). The GF-1
satellite carries a 16 m-resolution WFV multispectral camera Figure 2. GF-1 and GF-2 data coverage maps used in the study.
with an imaging width of 800 km and a temporal resolution of 4
days. The GF-2 satellite has a multispectral resolution of 4 m,
an imaging width of 45 km (Wu, 2014), and an absolute Sensor Scene ID Acquisition Pixel
temporal resolution of 69 days. Both of these multispectral Data Size (m)
cameras have four bands, and in this study, band 3 (red band) GF-1 WFV3 2510476 14/06/2016 16
and band 4 (near-infrared band) were used. The spectral 2509145,
response function curves of GF-1 WFV data and GF-2 PMS 14/06/2016 4
2509146
data are shown in Figure 1 (GF-1 WFV1 and GF-2 PMS2 as an 2763984,
example). It can be seen from the figure that the spectral 2763985,
response function curves of the two are very similar, which is GF-2 PMS1/2
2763983,
the basic condition for the relative radiation normalization of 01/09/2016 4
2764226,
the GF-2 PMS data with the GF-1 WFV data as the reference 2764227,
data. 2763986
Table 1. GF-1 and GF-2 data used in the study.

2.2 Data pre-processing

In this study, absolute and relative radiation normalization were

combined. The high-resolution NDVI data to be normalized
were first calculated from surface reflectance data. In addition,
the normalization process requires correspondingly medium-
resolution reference, classification, and cloud mask data.

Figure 1. Spectral response function curves of GF-1 WFV and The pre-processing of GF-1 and GF-2 data mainly includes the
GF-2 PMS. following steps:

The GF-1 WFV data and GF-2 data used in the experiments are Step 1. Orthorectification: It is the process of correcting
located in Wuhan, Hubei Province. Wuhan is the central city in geometric distortions caused by sensors, terrain and other
the central region of China, located in the eastern part of the factors by using ground control points and certain mathematical
Jianghan Plain and the middle reaches of the Yangtze River. models. Here the orthorectification of the image is realized by
The specific geographical location is from 113°41' to 115°05' E using the rational polynomial coefficients (RPC) files that come
and 29°58' to 31°22' N. Wuhan is the central city of the national with the image and the Digital Elevation Model (DEM) that
synthesis coordinated reforms experimental plot of "two types comes with the ENVI software. The RPC file of the high-
society", which is resource-saving and environment-friendly, resolution remote sensing image is a transformation matrix
and pays attention to forestry ecological construction and solved from the orbital parameters of the sensor and some other
resource protection. The typical vegetation in Wuhan is a mixed physical parameters, combined with ground control points
forest consisting of subtropical evergreen broadleaf forest and (Zhuo, 2017).
deciduous broadleaf forest. Details about the data were shown
in Figure 2 and Table 1. Step 2. Radiation calibration: It is the process of converting the
original dimensionless brightness gray value (DN value)
recorded by the sensor into the absolute radiance value with
actual physical significance. This process can eliminate the
errors caused by the sensor, to determine the exact radiation
value of the sensor at the entrance.GF-1 and GF-2 satellite
sensor radiation calibration formula is as follows:

Lc (c )  gain  DN  offset (1)

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 210
 ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022
XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

where Lc (c ) = a band radiation luminance value

2.3 Methods
gain = the calibration coefficient
DN = the original luminance gray value The SGloLM has a significant problem of mosaic seams when
offset = the absolute calibration coefficient offset NDVI normalization is performed on multi-scene high-
resolution images. Therefore, this study has improved it in two
Step 3. Atmospheric correction: It is the process of eliminating aspects. On the one hand, the coefficient solution of the
the radiation errors caused by atmospheric scattering, SGloLM algorithm is improved. On the other hand, the MVC
absorption, and reflection, so as to obtain the true reflectivity of method is added.
surface objects. Here, the fast line-of-sight atmospheric analysis
of spectral hypercubes (FLAASH) atmospheric correction is 2.3.1 Single-scene global linear model: Linear models are
used for the GF-1 reference data. It has been commercially the most widely used typical models for cross-sensor NDVI
applied and is one of the more accurate atmospheric correction conversion. In the earlier reference-based normalization, the
models available. It is an atmospheric correction algorithm global linear model is often applied to the normalization of
based on principal component analysis and uses the procedure reflectance data. Based on previous studies (Miura et al., 2013;
of MODTRAN 4+ radiative transfer model (Wang et al., 2013) Potapov et al., 2020; Gan et al., 2014), it can be found that there
for pixel-based correction, which can eliminate or reduce the is usually a linear correlation between different sensors and
cascading effects caused by diffuse reflections and adjust the multi-temporal images of the same region. In other words, their
spectral smoothing caused by artificial suppression (Hao et al., grayscale values in the same band can be interconverted by a
2008). Its spectral radiation equation is as follows: linear equation. Therefore, the SGloLM method assumes that
the relationship between the data to be normalized and the
A B reference data can be described by a simple linear regression
L( )( )  L (2)
equation as:
1  e S 1  e S
y  a x b (4)
where L = the radiance of the pixel
A, B = the coefficients depending on the atmospheric
and geometric conditions where x = the reflectance value of the experimental image
y = the normalized reflectance value of the
 = the surface reflectance of the pixel
experimental image
e = the average surface reflectance of the pixel and a, b = the slope and intercept of the linear regression
its surrounding area equation
S = the atmospheric albedo
L = the atmospheric backward scattered radiation Based on the above assumptions, after downsampling the high-
resolution data to be normalized to the resolution of the
medium-resolution reference data, the slope and the intercept of
The GF-2 data are processed using the Quick Atmospheric
the linear regression equation are solved according to the
Correction (QUAC) tool in ENVI to balance the processing
medium-resolution sample point pairs. And then the
time, result accuracy, and ease of batch processing. Compared
normalization results of each image element of the data to be
with the FLAASH correction algorithm, the QUAC algorithm
normalized are calculated by this linear relationship element by
can make full use of remote sensing images to obtain spectral
element. The specific processing flow is as follows:
information of different ground objects, and perform fast and
accurate atmospheric correction through empirical values to
Step 1. The high-resolution data to be normalized should be
obtain the real surface reflectance without complex
downsampled to the medium resolution of the reference data.
environmental parameters.
Step 2. The purity of each medium resolution pixel is calculated
Step 4. Calculation of NDVI: After the above absolute radiation
according to the high-resolution classification data (6-10
normalization, the reflectance data of GF-1 and GF-2 are
categories). Then, through a given threshold (0.5-0.6), the pure
obtained. The red light band and the near-infrared band are used
pixel is determined. The calculation formula for pixel purity is
for combined operation. The reference NDVI data were
as follows:
calculated from the reflectance data of GF-1, and the NDVI data
to be normalized is calculated from GF-2 reflectance data. The
calculation formula is as follows: kc (5)
r=
   RED mm
NDVI  NIR (3)
 NIR   RED
where r = pixel purity
kc = the number of pixels belonging to the feature
where  NIR = the near-infrared band
category c, which has the largest proportion in any
RED = the red band medium resolution pixel range
m = the scale ratio between the high-resolution and
Step 5. Prepare auxiliary data: Classification data and cloud the medium-resolution
mask data. The Classification data corresponding to the data to
be normalized is obtained by applying ENVI IsoData Step 3. Based on the pure pixels, the linear relational
Classification to GF-2 reflectance data. And deep learning- coefficients are solved by Huber-type M-estimation according
based cloud detection method named multi-scale convolutional to the following equation:
feature fusion (MSCFF) (Li et al., 2019) is applied to obtain
cloud mask data corresponding to the data to be normalized.

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 211
 ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022
XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

yn = a  xn +b (6) 2.3.2 Multi-scene global linear model: To address the

problem of obvious multi-scene image mosaic seams that occurs
where when the SGloLM method is applied to product production.
xn = the downsampled data to be normalized
Based on the SGloLM method, this study proposes MGloLM.
yn = the corresponding medium-resolution reference The improvements are made to alleviate the problem of obvious
data mosaic seams in a wide range of products. The improvement
a, b = the coefficients of the linear relationship idea is as follows: when processing the data to be normalized in
one scene, read in the adjacent image data of the top, bottom,
Errors in geometric correction, resampling, and unsupervised left, and right four scenes at the same time, and carry out steps 1
classification are bound to bring noise and outliers. However, and 2 in the previous subsection to obtain the pure pixels of
the standard least-squares method is easily affected by these each scene. And then solve the coefficients of the linear
noises and outliers, which leads to errors in coefficient solving. relationship equation based on the pure pixels of these five
Therefore, a more robust Huber type M-estimation is used here: scenes to obtain the linear relationship equation. Finally,
substitute each pixel of the data to be normalized is substituted
euv =a  xuv +b - yuv (7) into the linear relationship equation derived above to obtain its
normalized value.
The coefficient is solved by minimizing
  (e
uv
uv ) , where  is 2.3.3 Maximum Value Composite method: The MVC can
effectively reduce the effects of atmospheric aerosols, cloud
the influence function. There are many influence functions in
shadows, and solar altitude angle, and has been widely used in
the robust estimation, and the Huber type influence function is
the production of low and medium-resolution remote sensing
selected here:
vegetation index products (such as MOD13 and MYD13 based
on EOS/MODIS sensors; PAL, GIMMS, and GVI based on
 2
euv AVHRR sensors; VGT-S10 based on VEGETATION sensors
 , euv  c，
 2 (8) (Ge, 2016)). In this study, the NDVI data of one season was
 (euv )   2 used as the basis to obtain the quarterly maximum NDVI using
c e c , euv  c
 the MVC, which characterizes the best condition of vegetation

uv
2 growth in the current season. The calculation formula is as
follows:
where c = the Huber parameter
NDVI max  max( NDVI1 , NDVI2 ) (11)
In order to facilitate the optimization of the calculation process,
the minimization problem of the following equation (9) can be
equivalently transformed into the iterative reweighted least- where NDVI max = the maximum value of NDVI
squares problem: NDVI1 , NDVI 2 = multiple NDVI values in
different periods of the quarter
min  ( w(euv )euv )
2
(9)
2.3.4 Precision evaluation: For the experimental results of
w(euv ) is the weight of each pixel, equal to  (euv ) . In this
 this study, the reference data resampled to high-resolution was
used as the standard data, and the accuracy was evaluated by
euv
qualitative and quantitative methods. The qualitative evaluation
way, coefficients can be solved by an iterative process. mainly includes the visual effect of the image and the
comparison of its histogram. The quantitative evaluation is
Step 4. Each pixel of the data to be normalized is substituted mainly completed by the following three indicators:
into the linear relationship equation obtained above to obtain its
normalized value. The calculation formula is as follows: ① The Correlation Coefficient (CC), the calculation formula is
as follows:
ym = a  xm +b (10)

where xm = the high-resolution data to be normalized

 ( yˆ i  yˆ )( yi  y )
(12)
r i

ym = the corresponding normalized value  ( yˆ

i
i  yˆ ) 2
(y
i
i  y) 2

a, b = the coefficients of the linear relationship

obtained in step 3
where r = the correlation coefficient
yˆi = the value of the i th pixel of the normalized
image
ŷ = the average value of the normalized image
yi = the value of the i th pixel of the standard data
y = the mean value of the standard data

② The Mean Absolute Difference (MAD), the calculation

formula is as follows:

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 212
 ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022
XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

1 n 3. RESULTS AND ANALYSIS

MAD =  yˆi  yi
n i 1
(13)
In this study, two scenes of images were selected, normalization
experiments were performed, and their normalization results
where MAD = the mean absolute error value were mosaicked, as well as qualitative and quantitative
n = the number of pixels evaluations.
yˆi = the value of the i th pixel of the normalized
image The experimental results are shown in Figure 3 below. Visually,
yi = the value of the i th pixel of the standard data the spatial pattern and spatial distribution of the images changed
to a certain extent after normalization, with the SGloLM
method showing the worst results, the normalized images of the
③ The Root Mean Squared Error (RMSE), the calculation
SGloLM+MVC method being the most similar to the standard
formula is as follows: data used for evaluation. The mosaic seams of the
SGloLM+MVC and the MGloLM+MVC methods are not
1 n
 ( yˆi  yi )2 (14) obvious, and the improved SGloLM+MVC method can better
RMSE =
n i 1 eliminate the mosaic seams.

The histogram of experimental results of different methods in

where RMSE = the root mean square error
the small area at the lower right corner of Figure 3 is shown in
n = the number of pixels
Figure 4 below (the NDVI shown here is enlarged by 10,000
yˆi = the value of the i th pixel of the normalized times). It can be seen from the figure that the histogram of
image normalized results is closer to the histogram of standard data
yi = the value of the i th pixel of the standard data than before normalization. Among them, the histogram curves
of SGloLM +MVC and MGloLM +MVC methods are closest to
the standard data.

(a) (b) (c)

(d) (e) (f) (g)

Figure 3. (a) Reflectance data before normalization, (b) NDVI data before normalization, (c) standard NDVI data, (d) NDVI data
after SGloLM normalization, (e) NDVI data after SGloLM+MVC normalization, (f) NDVI data after MGloLM normalization, (g)
NDVI data after MGloLM+MVC normalization NDVI data.

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 213
 ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022
XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

Figure 4. Comparison of histograms before and after normalization.

largest CC, followed by MGloLM +MVC, the SGloLM+MVC

The quantitative evaluation results for this small area are shown method has the smallest MAD and RMSE, followed by
in Table 2. Compared with the data before normalization, the MGloLM+MVC. Combining the three indicators, the
normalized data of the three methods are better in all indicators. SGloLM+MVC method has the best quantitative results,
CC improves from 0.6852 to about 0.74, MAD decreases from followed by MGloLM +MVC method, and the MGloLM
0.1598 to about 0.08, and RMSE decreases from 0.1965 to method is the worst.
about 0.11. Among them, the SGloLM+MVC method has the

Evaluation Before After normalization

Indicators normalization SGloLM SGloLM+MVC MGloLM MGloLM+MVC
CC 0.6852 0.7231 0.7804 0.6872 0.7682
MAD 0.1598 0.0855 0.0643 0.1117 0.0659
RMSE 0.1965 0.1150 0.1012 0.1345 0.1035
Table 2. Quantitative evaluation of normalized results.

4. CONCLUSION REFERENCES

In this study, using GF-1 WFV and GF-2 PMS multispectral China High-resolution Earth Observation System (CHEOS) grid
data, combined with absolute radiometric correction and platform. https://www.cheosgrid.org.cn/.
relative radiometric correction, the NDVI normalization method
was studied. First, a series of pre-processing of the data was Gan, W., Shen, H., Zhang, L., Gong, W., 2014. Normalization
performed to obtain the reflectance data. Then, based on the of NDVI from Different Sensor System using MODIS Products
reflectance data, relative radiation normalization was performed as Reference. IOP Conference Series: Earth and Environmental
and the existing SGloLM method is improved to accommodate Science 17.
the production of high-resolution vegetation index data
products on a large scale. The improvement mainly starts from Gan, W., 2015. Research about the normalization of multi-
two aspects. On the one hand, it is to improve the coefficient source NDVI. Wuhan University.
solution of the SGloLM method, using the data to be
normalized and its adjacent images to jointly participate in the Ge M., 2016. An Algorithm Study on Normalization of
solution of the coefficients of the linear regression equation, Vegetation Index based on Multi-source Remote Sensing Data.
and MGloLM is proposed. On the other hand, it is to add the Chongqing University of Posts and Telecommunications.
MVC method, the maximum value of NDVI at different times
in a season is synthesized to represent the best condition of Guo N., 2003. Vegetation Index and Its Advances. Arid
vegetation growth in the current season. For the improved Meteorology, 71-75.
method, combined experiments were performed and qualitative
and quantitative evaluations were done respectively. The Hao J., Yang W., Li Y., Hao J., 2008. Atmospheric Correction
experimental results show that visually, both SGloLM+MVC of Multi-spectral Imagery ASTER., Remote Sensing
and MGloLM+MVC are more effective, and the Information, 78-81.
SGloLM+MVC method is the best in terms of quantitative
indicators. Li, Z., Shen, H., Cheng, Q., Liu, Y., You, S., He, Z., 2019.
Deep learning based cloud detection for medium and high
resolution remote sensing images of different sensors. ISPRS
Journal of Photogrammetry and Remote Sensing 150.

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 214
 ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022
XXIV ISPRS Congress (2022 edition), 6–11 June 2022, Nice, France

Miura, T., Turner, J.P., Huete, A.R., 2013. Spectral

Compatibility of the NDVI Across VIIRS, MODIS, and
AVHRR: An Analysis of Atmospheric Effects Using EO-1
Hyperion. Geoscience and Remote Sensing, IEEE Transactions
on 51, p.1349-1359.

Potapov, P., Hansen, M.C., Kommareddy, I., Kommareddy, A.,

Turubanova, S., Pickens, A., Adusei, B., Tyukavina, A., Ying,
Q., 2020. Landsat Analysis Ready Data for Global Land Cover
and Land Cover Change Mapping.

Wang J., Ye Q., Lin Y., 2013. Comparing Effects of Different

Atmospheric Correction Algorithms in Remote Sensing
Dynamic Monitoring of Cyano Bacteria Bloom in Inland Lakes.
Remote Sensing Technology and Application 28, 157-165.

Wu Y., 2014. The successful launch of the "GF-2" satellite, the

newspaper interviewed experts related to the special project of
Gaofen--Hotspot interpretation - Dialogue: My country's remote
sensing satellites have entered the sub-meter "Gaofen era".
http://cpc.people.com.cn/n/2014/0820/c83083-25501524.html.

Xu Y., 2019. Study on Surface Reflectance Normalization

based on Multi-source medium-to-high Resolution Remote
Sensing Data. Henan Polytechnic University.

Zhang X., 2009. Introduction of Several Vegetation Indices

Based on EOS/MODIS. Journal of Anhui Agricultural Sciences
37, 12842-12845+12850.

Zhuo N., 2017. Research on the Object-oriented Method of City

Features Extraction with GF-2 Remote-sensing Imagery. China
University of Geosciences (Beijing).

This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper.
https://doi.org/10.5194/isprs-annals-V-3-2022-209-2022 | © Author(s) 2022. CC BY 4.0 License. 215