Data Analysis for Orthorectification Process of Formosat-2 Satellite Images

on Surabaya and Yogyakarta

Hepi Hapsari Handayani
Geomatics Department, Sepuluh Nopember Institute of Technology
hepihapsari@gmail.com
parameters. Ground control points play an important role
in increasing accuracy during orbit adjustment and
orthorectification. When the distribution of ground
control points is not sufficient, tie points can take role to
create a good geometry. The experimental results
demonstrate the potential of mosaicked images as a
source of spatial information for Indonesia.

ABSTRACT
Geographical condition of Indonesia presents that
it lies between three earth plates: Pacific, Indo-Australia
and Eurasia. The plates are still active that they move
and interact each other. Therefore, earthquake is
occurred frequently in Indonesia. When earthquake is
occurred, Indonesia government must take disaster
management effectively and efficiently to reduce victim.
The major stages that must do are finding and rescuing
the victim, inventory of damage, rehabilitation and
reconstruction. For doing the disaster management,
government must have accurately spatial information to
make the right decision to mitigate victim
Accurately spatial information can be obtained
from map. Since Indonesia is archipelagic country, it
needs a mapping method which can reach all area of
Indonesia especially the cloistered islands. The method
that is appropriate with this condition is remote sensing
technology using satellite imagery. Because, Indonesia
has large area, there will be many satellite images
covering the area. Therefore, mosaicking for those
images becomes important and orthoimages will be used
in this process.
This research investigates about data analysis for
orthorectification process of Formosat-2 satellite images
on two Indonesias cities that have different terrain
characteristics. The proposed scheme consists of four
major parts: (1) evaluation of digital elevation model
(DEM) accuracy; (2) orbit
adjustment; (3)
orthorectification; and (4) mosaicking. Block adjustment
is developed using direct georeferencing based on the
Shuttle Radar Topography Mission (SRTM). This is used
as the DEM to adjust the satellites orbit. After this least
squares collocation is performed to refine an orbit. The
residual vectors of the ground control points (GCPs) and
tie points are used to collocate the orbit. The indirect
method is used in the process of orthorectification,
projecting 3D object points onto 2D image space. A
simple geometric function is developed to generate seam
lines between two strips in the overlapping area of the
mosaicked images.
The experimental results indicate that the SRTM
can be used to model surface both areas. Block
adjustment using direct georeferencing is a rigorous
method. It shows very promising results to improve the
quality of geometric registration. Least squares
collocation performs well when used to compensate for
local systematic errors and to refine the orbital

KEY WORDS: Data analysis, Orbit adjustment,

Orthorectification, Absolute Accuracy,
Relative Accuracy, Image Mosaick.
INTRODUCTION
Spatial information is important data for country.
They can be obtained by map, satellite imagery, or
another source. Liability of spatial information will
create the crucial problem for planning and development
in many sectors. The information is useful for
government in central level and district as base of
plannings development. Besides, the information will
help government for disaster management. Distribution
of accommodation and relocation of evacuee will
become important issue. Using accurately spatial
information, government can distribute accommodation
for evacuee more easily. Therefore, data analysis itself is
important stage for preparing the accurate spatial
information.
Indonesia is an archipelagic country. Many
Indonesian islands, especially island clusters are not yet
mapped. Some of these are outside the Indonesian region.
The lack of information creates potential problems with
neighboring countries. The maritime boundary of the
country is determined based on the outer islands. Thus,
the spatial information becomes important to define the
exact boundary between Indonesia and neighboring
countries. In addition, geographically, Indonesia lies on
three
different
tectonic
plates:
the
Pacific,
Indo-Australian and Eurasian plates. Conditions are now
still unstable and the terrain may change because of
tectonic movement. In the past few decades, several
earthquakes occurred with a magnitude on the Richter
scale of larger than six, one of which has snatched more
than 6,000 victims and made 350,000 family homeless.
When an earthquake occurs, the Indonesia government
must take steps for disaster management effectively and
efficiently to reduce the number of victims. In order to
do so, the government must have accurate spatial
information.
1

There are many reasons that Indonesia is obliged to

have accurate and updatable spatial information. New
leaps in technology bring access to high resolution
satellite images, a convenient and cost-efficient way to
acquire image data. This familiarity and easy
accessibility of high resolution satellite images leads to a
further increase in the demand. However, since Indonesia
is spread over a large area, many satellite images are
needed to cover the area, and thus orthorectification and
mosaicking of those images become important. Hence,
orbit adjustment is a prerequisite for geometric
correction during the image orthorectification.
Generally, high resolution satellites employ CCD
linear arrays for image acquisition. However, frequently,
with this type sensor, there is no way to satisfy the
demands of large coverage and stereoscopic observation
at the same time. In reality, the major applications of the
remotely sensed images are for the detection of natural
resources including the detection of disaster prone areas
and geoenvironmental monitoring. It is the case that, due
to the satellites small field of view (FOV), two or more
images are often needed to cover the area of interest.
Thus mosaicking becomes a must. For detection usage,
multitemporal images should be registered before
performing the comparison. Simultaneous block
adjustment is the preferred approach to enhance the
quality of this geometric registration. The adjusted
orientation parameters provide a sound foundation for
orthorectification. However, the overlapping area of
satellite images is generally small in order to acquire the
largest possible coverage. In addition, the converging
geometry is commonly poor. Thus, the weak geometry in
the area of intersection will cause a large elevation error.
In the orbit adjustment procedure an elevation control is
needed for these tie points. However, the traditional 3-D
bundle adjustment method is not suitable for direct
orientation modeling. Generally, sensor orientation
modeling may be divided in two categories, namely
rigorous sensor model (RSM) and rational function
model (RFM). Block adjustment needs to be applied to
enhance the geometric consistency among strips in order
to increase accuracy for orientation determination.
The aim of this study is to perform
orthorectification and mosaicking using the block
adjustment method for satellite images of Indonesia with
digital elevation model (DEM) as an elevation control.
Generally, there are three types of block adjustment
approaches, namely, bundle adjustment, direct
georeferencing and the rational function model. Bundle
adjustment uses the collinearity condition as the basis for
formulating the relationship between the image space
and the ground space. In direct georeferencing, orbital
and attitude data are utilized for orientation modeling.
There are increasing amounts of on-board data available
due to advances in global positioning systems (GPS),
inertial navigation systems (INS), and star-tracker
technology. To some extent, the RFM may be interpreted
as another form of direct georeferencing when rational
polynomial coefficients (RPCs) are derived from the
quality on board orientation parameters. Hence, we apply
block adjustment with direct georeferencing observations,
ground control points (GCPs), tie points and DEM as an
elevation control.

METHODOLOGY
The proposed method of this research comprises
four major parts: (1) evaluation of digital elevation
model (DEM) accuracy; (2) orbit adjustment; (3)
orthorectification; and (4) mosaicking.
ORBIT ADJUSMENT & LEAST SQUARES
COLLOCATION
In this process, we obtain orientation parameters
using the attitude information provided from the
on-board ephemeris data. Because the on-board data
include errors to a certain degree, GCPs are needed to
adjust the orbit parameters. Hence, the trend functions of
the orbital parameters are determined. The least squares
collocation performs to refine an orbit. Residual vectors
of GCPs and tie points are used to collocate the orbit.
First we use the orbit parameters and the image
coordinates of tie points to determine observation vectors.
Ray tracing technique is applied to obtain the ground
position of tie points, when a DEM is given. The
procedure is executed for the pair of images. Thus, the
middle of two ground points is treated as a constraint.
Calculating the residual vectors of tie points is the vector
from one ground position to a constrain. Equation of
least squares collocation is presented in eq.1 below

(eq.1)
c [ ] 1
k
k
k
k
where,
k is x,y,z axis
k
is the correction value of the interpolating point
ck is the row covariance matrix of the interpolating
point with respect to GCPs
k
is the covariance matrix for GCPs
k is the residual vectors for GCPs.
The number of GCPs is, in general, insufficient to
characterize the covariance function. Thus, it uses
Gaussian function with some empirical values for the
covariance function
(eq.2)
where,
d is distance between an intersection point and a GCP
dmax is maximum distance to the intersection point
k is variance of the GCP residual
rn is filtering ratio (in this research we use 0.1).
This empirical value, 2.146, is selected so that the
covariance limit is 1% x (1-rn)k when d = dmax.
ORTHORECTIFICATION
The process of orthorectification used is indirect
method which projects the 3D object point on 2D image
space. Once a DEM is given and the orientation
parameters are determined, the corresponding image
position for a ground point will be obtained.
In this indirect method, we assume that the relief
displacements in a small area with moderate terrain
variations are linear and the mapping geometry between
image coordinates and object coordinates may be
expressed by affine transformation when small area is
considered.
2

making them suitable for UTM map projection with

WGS 84 as the ellipsoid. The GDEM has a resolution of
30.7 m and has a pixel size of 3601 by 3601 for both
study areas. The SRTM has a resolution of 90m and a
pixel size of 2704 by 2471 for the first study area and
2695 by 2462 pixels for the second study area. The DEM
generated from the contours and spots height for the first
study area has a resolution of 12.5 m and a pixel size
4407 by 3070 and that for the second study area is 3309
by 2214 pixels.

MOSAICKING
ERDASS 9.2 commercial software is used for
mosaicking process in this research. The geometry of
seamline is used as method to detect the continuity of
feature when two strips combined.
DATA
National Coordinating Surveys and Mapping of
Indonesia provides the topographic map for this research.
It has scale 1:25,000 with Universal Transverse Mercator
(UTM) as map projection. Both of research areas are in
zone 49 south using WGS 84 for ellipsoid. The map
consists of several layers: hidrography (river,lake,etc),
road, contour line, spot height, administrations boundary
(provinces boundary, villages boundary,etc), land use
(residence, vegetation,etc). However, the map is not
provided for whole images in both of areas as shown in
figure 1.

EXPERIMENTAL RESULTS
The experiments include four parts of validation.
The first one is to evaluate the accuracy of SRTM as
DEM for this research. The second one is to check the
accuracy of the determined orientation parameters using
least squares collocation technique. The validation
includes absolute and relative accuracy and compares
them when least squares collocation applies with tie
points and without tie points. The ray tracing technique
is applied to evaluate the absolute accuracy of orbit.
When satellite orientation and image point are given, we
can compute the intersection point of DEM and ray
direction. The third one is to examine the accuracy of the
generated orthoimages. Distance between two ground
positions of tie points can evaluate the orbit relative
accuracy. The last one is checking on the seam lines
between two strips in the overlapping area of the
mosaicked images.
EVALUATION OF DEMS ACCURACY
The DEMs used in this evaluation are GDEM,
SRTM and DEM generated from contours and spots
height. The contours and spots height are provided by the
topographic map. The contours are presented as 12.5 m
intervals. Spot heights are obtained from the elevation of
the certain features such as mountain peaks, valleys, etc.
The DEM is developed by generating contours and spot
heights for triangular irregular networks (TINs). This is
the first step. When TINs are created, they are generated
as raster files with a pixel resolution of 12.5. The results
of DEM generation from contours for the first and
second study areas are shown in figures 2 and 3,
respectively.

(a)

(b)
Figure 1. Topographic map is superimposed with
satellite images (a) for first area, (b) for second area.
Formosat-2 satellite images for two areas have two
strips with overlapping about 70% and 10% for first and
second area respectively. Reference points obtained from
topographic maps are used for GCPs and independent
check points (ICPs). Tie points are from overlapping area
are treated as tie control points (TCPs) and tie check
points (TKPs). The number of points and related
information of test images is presented in table 1.
Table 1. Related information of test images
Location
Number of images
Date
Incidence angle
Number of GCPs
Number of TPs
Test area (km x km)
Elevation range

The First Area

The Second Area
1st strip
2nd strip
1st strip
2nd strip
Surabaya Surabaya Yogyakarta Yogyakarta
2
2
3
3
7/5/2007
8/4/2008 7/18/2006 7/12/2007
3.74
3.39
21.71
28.27
40
40
28
27
61
50
26 x 53
54 x 76
0.2 ~ 1533m

(Unit: meter)

Figure 2. Result of DEM generated from contours for the

first study area

1 ~ 3079m

There are three DEMs used in this study: the Global

Digital Elevation Model (GDEM), the Shuttle Radar
Topography Mission (SRTM), and the DEM generated
from contours and spots height. GDEM and SRTM
acquired are within geographic latitude and longitude,
3

Table 3. Evaluation of DEM accuracy for the second

study area
Statistical

(Unit: meter)

Indices
mean (m)
rms (m)
max (m)
range (m)

Figure 3. Result of DEM generated from contours for the

second study area.

DEM Types
DEM generated from
GDEM

contours & spot heights

4.46
14.16
46.25
82.28

2.77
4.06
27.15
41.21

SRTM

2.92
5.18
31.29
40.5

Based on table 2 and table 3, we can conclude that

the accuracy of SRTM is the best for the first study area,
since the terrain of the first study area is flat. For the
second area, the accuracy of DEM generated from
contours & spot heights is the best. But, the range of the
difference for SRTM in the second study area shows that
it is the lowest. Thus, although the result of DEM
generated from contours & spot heights is the best, it
doesnt mean that is the most appropriate DEM. Because
contours, spot heights, and consistency check points are
from the same topographic map, so it is no doubt that
DEM generated from contours will be the best. Actually,
SRTM describes digital surface model. It is derived from
top of features. Thus, it meets the demands of the surface
model that is to be used in orthorectification. Hence, we
will use SRTM as DEM in orbit adjustment and
orthorectification for both areas. The contour
superimposed with spot heights for both areas are
presented in figure 5.

The reference of the evaluation is spot heights of the

topographic map. Then, they will be called consistency
check points. The shortest distance between the two
points is 50m and the longest is 500m. The number of
points in the first area is 1184 and that of the second area
is 2194. Consistency check points that are superimposed
with satellite images for both study areas are presented in
figure 4.

(b)

(a)

Figure 4. Distribution of the consistency check points on

the satellite images (a) the first study area (b) the second
study area.
The evaluation of DEMs accuracy uses the
difference between an elevation of consistency check
points and an elevation of spot heights that are
determined by DEM. Bilinear interpolation is applied for
obtaining an elevation of consistency check points in
DEM. The results of DEMs evaluation are presented in
table 2 and 3.

Figure 5. Contours superimposed with spot heights (a)

the first study area (b) the second area
ORBIT ADJUSMENTS ACCURACY
The results of evaluation for orbit adjusments
accuracy present that the accuracy does not improve
when tie points are included in the adjustment. In fact,
the rmse is little worse when it is used tie points.
Because when tie points apply and topographic map is
not provided in south part of image, there is not elevation
control in that region. Hence, error will contribute to the
result. Table 4 and table 5 illustrate the absolute accuracy
of orbit adjustment for first and second strip respectively
in first area.
Table 4. Absolute accuracy of orbit adjustment for first
strip in first area

Table 2. Evaluation of DEM accuracy for the first study

area
Statistical
Indices

DEM Types
DEM generated from
GDEM

contours & spot heights

SRTM

mean (m)

3.82

1.32

2.10

rms (m)

4.80

3.11

2.40

-25.61
35.63

-10.14
17.22

-12.65
16.25

max (m)
range (m)

(b)

(a)

First strip
RMSE of GCPs
RMSE of ICPs

E (m)

N(m)

E (m)

Without tie points

N(m)

2.95

3.55

9.20

9.64

With tie points

3.40

4.71

9.96

12.44

Table 5. Absolute accuracy of orbit adjustment for

second strip in first area

Table 7. Absolute accuracy of orbit adjustment for

second strip in second area

Second strip
RMSE of GCPs
RMSE of ICPs

Second strip
RMSE of GCPs

E (m)

N(m)

E (m)

N(m)

E (m)

N(m)

Without tie points

2.02

3.20

9.20

9.64

Without tie points

2.40

With tie points

2.69

4.08

9.96

12.44

With tie points

2.46

Relative accuracy of orbit adjustment for first area

improves much when tie points are included in
adjustment. However, it is still remain systematic errors
in south part because there are not GCPs provided in
there. It is shown in figure 6 and figure 7.

RMSE of ICPs
E (m)

N(m)

3.89

7.29

2.69

4.00

6.99

3.01

Relative accuracy of orbit adjustment for second

area improves much when tie points are included in
adjustment. The systematic errors appeared in without tie
points figure can reduce in the with tie point figure. It
means that tie points applied contribute the geometric
improvement for those images. It is presented in figure 8
and figure 9.

Figure 6. Relative accuracy of orbit adjustment for first

area without tie points
Figure 8. Relative accuracy of orbit adjustment for
second area without tie points

Figure 7. Relative accuracy of orbit adjustment for first

area with tie points
Figure 9. Relative accuracy of orbit adjustment for
second area with tie points

Table 6 and 7 illustrate the absolute accuracy of orbit

adjustment for first and second strip respectively in
second area. The result is similar to the first one that the
accuracy does not improve when tie points are included
in the adjustment.
Table 6. Absolute accuracy of orbit adjustment for first
strip in second area

ACCURACY OF ORTHOIMAGES
The results of absolute accuracy of orthoimage for
first area present that the accuracy is not significantly
different between two cases. When we apply tie points, it
will not contribute for reducing the error. It is illustrated
in Table 8.

First strip
RMSE of GCPs

RMSE of ICPs

E (m)

N(m)

E (m)

N(m)

Without tie points

3.34

1.88

4.86

5.05

With tie points

3.80

4.06

4.98

8.07

Table 8. Absolute accuracy of orthoimage for first area

occurred that two images strips will not continue each

other after mosaicking. We can detect it in along seam
lines. For both of areas, the mosaic images are quite
good the gap of feature along seam line between two
strips is around 10-15 meters. The result may be better
when higher DEM is available and the images
resolution is larger. Hence, it can be used directly for
spatial information data directly or we can use it as
source of map by delineating process. Figure 6 and 7
present mosaic image of first area and second area
respectively.

RMSE of ICPs
First strip

Second strip

E (m)

N(m)

E (m)

N(m)

Without tie points

8.26

13.25

7.34

10.33

With tie points

8.50

13.16

8.09

10.54

Table 9 illustrates relative accuracy of orthoimage

for first area. The accuracy improves much when tie
points are included in orthorectification process. Because
of tie points can adjust geometry model for
orthorectification especially for area that doesnt have
GCPs.
Table 9. Relative accuracy of orthoimage for first area
RMSE of TKPs
E (m)

N (m)

Without tie points

31.30

63.46

With tie points

6.28

9.88

Table 10 presents absolute accuracy of orthoimage

for second area. The accuracy is not significantly
improvement, when we apply tie points in adjustment. It
is obvious, because tie points do not have ground
position. Hence, it will not give effect in geometry
adjustment.
Table 10. Absolute accuracy of orthoimage for second
area
RMSE of ICPs
First strip

Second strip

E (m)

N(m)

E (m)

N(m)

Without tie points

7.09

3.05

6.01

6.36

With tie points

6.91

2.60

5.74

7.61

Figure 10. Image mosaic of first area

Table 11 illustrates the relative accuracy Relative

accuracy of orthoimage for second area. The results
show that the accuracy increases much when tie points
employ in orthoimage process. Comparing with relative
accuracy of orthoimage for first area, those of second
area are larger. Since the elevation range of this area
more roughly and the terrain condition is mostly hilly
land, it due to the geometry model becomes weak. The
overlapping area of the second images which is only
10% contributes for the geometry weakness.
Table 11.
area

Relative accuracy of orthoimage for second

RMSE of TKPs
E (m)

N (m)

Without tie points

89.06

79.02

With tie points

12.72

21.93

Figure 11. Image mosaic of second area

IMAGE MOSAICK
Image mosaicks determined are images having tie points
for both areas. The tie points will be useful in the
mosaicking to collocate the geometry discrepancies
between two strips. Because, it will be frequently
6

EVALUATION OF GEOMETRIC DISCREPANCIES

IN THE MOSAICKED IMAGES

Second Area

First Area

Figure 13.The locations of four tests for evaluation of

geometric discrepancies in the mosaicked images of the
second area
There are some disparities between the mosaicked
images. The parallaxes of the disparity in the test sets of
images for the first area are listed in table 12 below.

Figure 12.The locations of four tests for evaluation of

geometric discrepancies in the mosaicked images of the
first area

Table 12. Parallaxes of the disparity in the test sets of the

mosaicked images of the entire area in the first study
area

There are some disparities between the mosaicked

images. The parallaxes of the disparity in the test sets of
images for the first area are listed in table 11 below.
Table 11. Parallaxes of the disparity in the test sets of the
mosaicked images of the entire area in the first study
area
Parralaxes
Without TPs
With TPs
E(m) N(m) E(m)
N(m)
1
6
4
0
0
2
4
8
0
0
3
8
14
2
4
4
24
30
4
6

1
2
3
4

Parralaxes
Without TPs
With TPs
E(m) N(m) E(m)
N(m)
48
150
6
18
6
24
4
6
4
6
2
2
0
0
0
0

CONCLUSSIONS
SRTM This study proposes a feasible scheme for
orthorectification and mosaicking of satellite images.
The experimental results demonstrate the potential of
mosaicked images as a source of spatial information.
This study takes advantage of block adjustment and least
squares collocation for orbit adjustment and patch back
projection for orthorectification to generate the
mosaicked images.
The experimental results are summarized as
follows.
1. Block adjustment using direct georeferencing is a
rigorous method and performs well in this study. It is
very promising means to improve the quality of
geometric registration.
2. The accuracy of SRTM is the best among the
available DEMs. It can be used as elevation control in
orbit adjustment and to model the surface in
orthorectification.

3. The least squares collocation method performs well,

compensating the local systematic errors to refine the
orbital parameters. Relative accuracy can be
significantly improved by applying tie points.
4. Ground control points play an important role in
improving the accuracy of orbit adjustment and
orthorectification. When the distribution of ground
control points is in sufficient, tie points can assume
role in order to create a good geometry.
LIMITATIONS
The limitations in this study are stated below.
1. Since we use DEM for elevation control in orbit
adjustment and orthorectification, the accuracy of the
DEM will affect the result. Higher accuracy may be
acquired when better DEM is available in for
adjustment and orthorectification.
2. Since we use the topographic map as the horizontal
data source for GCPs for orbit adjustment and
orthorectification, the accuracy of the topographic
map contributes to the accuracy of adjustment and
orthorectification. Higher accuracy may be attained
when large scale and higher accuracy topographic
maps become available for adjustment and
orthorectification.
REFERENCES
Chen, L.C., Teo, T.A., and Rau, J.Y., 2005, Adaptive
Patch Projection for Generation of Orthophotos
from
Satellite
Images,
Photogrammetric
Engineering & Remote Sensing, 71(11):
1321-1327.
Hall, O., Falorni, G., and Bras R.L., Characterization and
Quantification of Data Voids in the Shuttle Radar
Topography Mission Data, IEEE Geoscience and
Remote Sensing Letter, 2(2): 177-181.
Mikhail, E.M., and Ackermann, F., 1982, Observation
and Least Squares, University Press of America,
New York, 393-426.
Vassilopoulou, S., Hurni, L., Dietrich, V., Baltsavias, E.,
Peterki, M., Lagios, E., Parcharidis, I.,
Orthophoto generation using IKONOS imagery
and high-resolution DEM: a case study on
volcanic hazard monitoring of Nisyros Island
(Greece), ISPRS Journal of Photogrammetry and
Remote Sensing, 57, 24-38.