ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

ISPRS Journal of Photogrammetry and Remote Sensing

journal homepage: www.elsevier.com/locate/isprsjprs

Measurement of ground displacement from optical satellite image

correlation using the free open-source software MicMac
Ana-Maria Rosu a,⇑, Marc Pierrot-Deseilligny b, Arthur Delorme a, Renaud Binet c, Yann Klinger a
a
Laboratoire de tectonique et mécanique de la lithosphère, Institut de Physique du Globe de Paris, France
b
Laboratoire de géomatique appliquée, Ecole Nationale des Sciences Géographiques, Institut National de l’Information Géographique et Forestière, France
c
Laboratoire PS/TIS, Centre National d’Etudes Spatiales, France

a r t i c l e i n f o a b s t r a c t

Article history: Image correlation is one of the most efficient techniques to determine horizontal ground displacements
Received 28 September 2013 due to earthquakes, landslides, ice flows or sand dune migrations. Analyzing these deformations allows a
Received in revised form 4 March 2014 better understanding of the causes and mechanisms of the events. By using sub-pixel correlation on
Accepted 7 March 2014
before- and after-event ortho-images obtained from high resolution satellite images it is possible to
Available online xxxx
compute the displacement field with high planimetric resolution. In this paper, we focus on measuring
the ground displacements due to seismotectonic events. The three sub-pixel correlators used are:
Keywords:
COSI-Corr – developed by Caltech, a free, closed-source correlator, dependent on commercial software
Satellite image
Ground displacement
(ENVI) and widely used by the geoscience community for measuring ground displacement; Medicis –
Image correlation developed by CNES, also a closed-source correlator capable of measuring this type of deformation; and
COSI-Corr MicMac – developed by IGN, the free open-source correlator we study and tune for measuring fine
Medicis ground displacements. We measured horizontal ground deformation using these three correlators on
MicMac SPOT images in three study cases: the 2001 Kokoxili earthquake, the 2005 dyke intrusion in the Afar
depression and the 2008 Yutian earthquake.
Ó 2014 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier
B.V. All rights reserved.

1. Introduction Satellite imagery using synthetic aperture radar (SAR) or optical

satellites can overcome some of the limitations of the techniques
Measuring the fine displacements associated with an mentioned before. The satellite images cover a large area – the
earthquake represents a key issue in seismotectonics, offering fault rupture is partially or totally visible.
information about the geometry of the ruptured fault and the Compared to the optical sensor, the interferometric SAR
energy released by the earthquake (Van Puymbroeck et al., 2000). technique can be used in all weather and nighttime acquisitions
The most common method for measuring these displacements is (Massonnet and Feigl, 1998). However this technique is unable to
based on field measurements. However, the fault area is often hard provide displacement maps in the near-field of the fault because
to access and complex fault ruptures are not easy to detect in the the large amplitudes of the displacements present in this area
field; depending on the extent of the fault slip, it can be measured cause the decorrelation of the interferometric phase, so the
only in a limited number of locations (Leprince et al., 2007a). displacements cannot be estimated. Furthermore, SAR correlation
Another technique for measuring this type of displacement is by gives low resolution planimetric results, and InSAR provides
using permanent Global Navigation Satellite System (GNSS) receiv- mainly the near-vertical component.
ers. This technique can only provide sparse coverage and it is impos- High-resolution optical satellite imagery provides detailed
sible to have measurements if the area is not kept under surveillance. images of the ground and most importantly, it can resolve the
As the area of co-seismic ground displacement is a priori unknown, it near-fault displacement (Van Puymbroeck et al., 2000). The
is not always possible to have measurements before the event. displacement field can be measured using the correlation of images
acquired before and after the event (therefore the temporal base-
line is one of the main issues of the correlation).
⇑ Corresponding author. Address: Laboratoire de tectonique et mécanique de la
Optical satellite imagery has extended archives which allows
lithosphère, Institut de Physique du Globe de Paris, 75238 Paris cedex 05, France.
Tel.: +33 678062697. measurements of old earthquakes and could also provide the be-
E-mail address: am.rosu@laposte.net (A.-M. Rosu). fore-event images when an earthquake occurs. Unlike with InSAR

http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
0924-2716/Ó 2014 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
2 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

data, the combination of different archives is possible (Hollings- et al., 2003). Therefore, using images with a ground pixel size of
worth et al., 2012) and is a major advantage of using optical 2.5–10 m, the smallest displacements that may be measured are
satellite images. 0.25–1 m. However, the sub-pixel detection is highly dependent
Numerous studies on sub-pixel correlation applied to high-res- on the quality and the noise level of the data. Important matters
olution satellite images such as SPOT have shown the efficiency of to take into account when doing correlation are the texture of
this technique for measuring ground displacements due to earth- the images and factors causing great changes in aspect between
quakes (Michel and Avouac, 2002; Dominguez et al., 2003; Binet the two images due to diachronism (the period of time between
and Bollinger, 2005; Klinger et al., 2006; Leprince et al., 2007b), the two image acquisitions). Seasonal variations could generate
landslides (Delacourt et al., 2004; Casson et al., 2005), ice flows changes in the landscape, which might cause decorrelation. In or-
(Scambos et al., 1992) and sand dune migrations (Hermas et al., der to avoid parallax problems due to digital elevation model
2012). Analyzing these deformations allows a better understanding imprecision, it is preferable to use satellite images acquired with
of what triggered the event (e.g. analyzing the faults and the dis- very similar incidence angles, close to nadir.
placement field caused by an earthquake can supply extremely The purpose of this article is to present free open-source
important information about the seismic mechanisms). software, MicMac, capable of measuring the two-dimensional dis-
Sub-pixel detection capability is required in order to measure placements by optical image correlation and free from most of the
the displacements which are generally smaller than the image pix- drawbacks of the correlators used up until now (e.g. not open-
el size. In theory, the sub-pixel correlation method using a pair of source and hence not easily adaptable to specific cases, lack of
SPOT panchromatic images could provide fault slip measurements robustness in cases where the time interval between the two
with an accuracy of 0.1 px (Michel and Avouac, 2002; Dominguez image acquisitions is very significant or dependent on commercial
softwares). In this paper we present the sub-pixel correlation re-
sults obtained with MicMac, compared with those obtained using
two other correlators, COSI-Corr and Medicis, both of which are
used by the geoscience community. The horizontal displacements
induced by seismotectonic events were measured using pre-event
and post-event SPOT ortho-images. The three study cases are the
Kokoxili earthquake (M w  7:8, November 2001, Tibetan Plateau),
the September 2005 rifting event of Afar (Ethiopia) and the Yutian
earthquake (M w  7:1, March 2008, Tibetan Plateau).

2. Methodology

2.1. Image correlation

The basic parameters of the correlation are the ‘‘sliding win-

dow’’, the ‘‘search space’’ (the correlator computes a correlation
score for every position given by the ‘‘discretization step’’ in the
search space and keeps the best one; the discretization step defines
Fig. 1. Linear and non-linear correlation cost in MicMac. By default the correlation the sub-pixel precision of the correlation results) and the ‘‘step’’
cost is linear (Cost ¼ 1  Cor, where Cor is the normalized cross-correlation (distance in pixels between two consecutive parallax estimations,
coefficient). When dealing with images with great inhomogeneity, a non-linear defining the correlation image size). The spatial resolution of the
cost is used in order to limit the impact of the noisy signal on the whole
  min
c  displacement is directly related to the size of the correlation win-
ÞC min
measurement: Cost ¼ 1  MaxðCor;C1C min
 ð1  C min Þ, where C min is the corre- dow. The use of a small correlation window implies a result with a
lation threshold (below this value, the correlation has no influence) and c higher spatial resolution, which is extremely important if a fine
determines the influence of the correlation scores (in our studies, C min ¼ 0:5; c ¼ 2). description of the near-fault area is desired. However, the noise

Fig. 2. Synthetic displacement field to retrieve by correlation: the slave image was created using the reference image, a QuickBird satellite image, by dividing it into blocks
and moving them along columns and lines, creating successive offsets of 0:1 px per block. The synthetic offsets values go from 0 px to 0:5 px. The correlation results obtained
using COSI-Corr, Medicis and MicMac are in Figs. 3 and 4.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx 3

Fig. 3. The two disparity maps, in columns and lines, obtained with COSI-Corr, Medicis and MicMac on the images from Fig. 2. The sizes of the correlation window used are
32  32 px for COSI-Corr, 33  33 px for Medicis and MicMac. The three correlators retrieve well the synthetic deformation field going from 0 to 0:5 px in columns and lines.
COSI-Corr’s results have very few correlation artifacts, Medicis’ results are slightly noisier than COSI-Corr’s, and MicMac’s results are very close to the theoretical deformation
field.

Fig. 4. The two disparity maps, in columns and lines, obtained with COSI-Corr, Medicis and MicMac on the images from Fig. 2. The sizes of the correlation window used are
8  8 px for COSI-Corr, 9  9 px for Medicis and MicMac. COSI-Corr’s results are not satisfactory, they are very noisy and the deformation field is not well retrieved. Medicis’
results are noisier than the ones obtained using a larger correlation window (Fig. 3), but the deformation field is well retrieved. MicMac’s results retrieve very well the
synthetic deformation field.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
4 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

Table 1 retrieving ground surface deformation from multi-temporal

Correlation parameters for COSI-Corr, Medicis and MicMac used in all three study images. This software is widely used by the geoscience community.
cases; the parameters, specific to each correlator, are described in 2.2.1, 2.2.2 and
2.2.3.
The correlation method implemented in COSI-Corr (described in
Leprince et al. (2007b)) is based on frequency-domain correlation.
Correlators Parameters Values The relative displacement between two patches is estimated from
COSI-Corr Mask threshold 0.9 the phase difference of their Fourier transforms. The correlation
Number of iterations 2 process consists of two steps. The first step determines the
Medicis Initial resemblance threshold 0.6 pixelwise displacement between two patches (windows) to corre-
Final resemblance threshold 0.8 late. The peak correlation method is used to initialize, and then to
Exploration area 73
Sub-pixel precision 0.01
iteratively relocate the patches to compensate for their relative
displacement. The pixelwise displacement being estimated, a final
MicMac C min 0.5 correlation is operated on relocated patches to determine the
c 2 subpixel displacement using a minimization algorithm (Van
Exploration area 55 Puymbroeck et al., 2000; Leprince et al., 2007b).
Number of scanning directions 14
COSI-Corr’s graphical user interface gives the user the possibil-
Regularization term 0.3
Sub-pixel precision 0.05 ity of choosing the size of the ‘‘correlation window’’, the ‘‘step’’
between two consecutive parallax estimations, ‘‘mask threshold’’
(allows the masking of frequencies according to the amplitude of
the log-cross-spectrum for reducing the noise in the measure-
increases with the reduction of the window size (Binet and ments; a value close to one is appropriate in most cases (Ayoub
Bollinger, 2005); therefore, this must be seen as a compromise et al., 2009)), and the ‘‘number of iterations’’ (two to four iterations
between the required resolution and the noise. are adequate in most cases (Leprince et al., 2007b)). There is an
The correlation output consists of two disparity maps with the optional resampling step where the patches to correlate are relo-
relative horizontal displacements in rows and in columns plus the cated from sinc resampling. This option theoretically eliminates
correlation scores image which represents the confidence in the most of the biases at the sub-pixel scale. However, it greatly in-
correlation for each pixel. When working on georeferenced images, creases the processing time (on average by a factor of 10) and it
the two disparity maps contain the east-west and north-south is rarely useful on noisy images (Ayoub et al., 2009). After testing
components of the displacement. this option, we decided not to use it since we did not notice any
improvement in the results.

2.2. Correlators used

2.2.2. MEDICIS (Moyen d’Evaluation de Décalages entre Images,
Sub-pixel horizontal displacement field is computed using three Commun à l’Imagerie Spatiale)
correlators: COSI-Corr, Medicis and MicMac. Medicis is developed by CNES (Centre National d’Etudes Spati-
ales, France). It can be used for sensor calibration, digital elevation
model (DEM) and digital surface model (DSM) computation, image
2.2.1. COSI-Corr (Co-Registration of Optically Sensed Images and overlay and computing deformation between images. Its applica-
Correlation) tion stretches to every field where measuring the similarity
COSI-Corr is a software module developed using IDL (Interactive between pixels is needed, such as remote sensing, geology and
Data Language) and integrated with ENVI, implemented at the medicine.
Caltech Tectonics Observatory (USA). In Ayoub et al. (2009), Medicis offers both a spatial and a frequency correlator. We
COSI-Corr is described as providing tools to accurately orthorectify, chose the frequency domain correlator because it has shorter com-
co-register, and correlate optical remotely sensed images (aerial putation time and gives similar results. The computation method is
and pushbroom satellite images) with the ultimate objective of the following: first, the pixelwise disparities in lines and columns

Fig. 5. The pair of ortho-images used for the study of the 2001 Kokoxili earthquake. The pre-event image is from SPOT1 (29 September 1989, incidence angle: L4.60 ) and the
post-event image is from SPOT4 (30 November 2002, incidence angle: L4.50 ). The area where the results are presented (Fig. 6) is marked in red. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx 5

Fig. 6. East-west component of the co-seismic offset field from the 2001 Kokoxili earthquake, using the pair of SPOT ortho-images from Fig. 5. The small correlation windows
are 8  8 px (for COSI-Corr) and 9  9 px (for Medicis and MicMac), and the large correlation windows are 64  64 px (for COSI-Corr) and 65  65 px (for Medicis and
MicMac). Perpendicular profiles were stacked over a width of approximately 6:5 km and a length of 6:5 km using the weighted median method. For the small correlation
window, COSI-Corr underestimates the amplitude of the co-seismic offset; Medicis’ result is a bit noisier than MicMac’s, but the amplitude of the offset is the same for both,
and the same than with the large correlation window. For the large correlation window, all the three correlators retrieve the same amplitude of the offset ( 4:5 m).

are estimated using the peak correlation method and then the sub- 2.2.3. MicMac (Multi Images Correspondances par Méthodes
pixel displacement is computed. A ‘‘maximum exploration area’’ Automatiques de Corrélation)
(or ‘‘search space’’) is given around an estimated disparity. An ‘‘ini- MicMac1 is free open-source software, distributed under the CeC-
tial resemblance threshold’’ defines the lower bound of the ILL-B license, implemented at IGN (Institut National de l’Information
correlation coefficient, at pixel level; the sub-pixel disparity is Géographique et Forestière, France). It computes multi-image
computed only for couples (line, column) with the correlation sub-pixel correlation in the spatial domain.
score above this threshold. Sub-pixel displacements can be MicMac’s fields of application include three-dimensional
measured by applying fractional shifts to the sliding window modeling in archeology, architecture for cultural heritage2 and
(Inglada et al., 2007), done by image interpolation. The parameter geology and surveying for environmental applications (Guérin
‘‘precision of localization’’ manages the fractional shifts and it rep- et al., 2012; Lisein et al., 2013).
resents the required sub-pixel precision – the accepted values are Among the three correlators presented in this paper, MicMac is
between 0 (extremely good precision) and 0.05 (poor). Dichotomy the only one using regularization. This technique can provide very
is used for an iterative precise search by resampling using the good results when using small correlation windows, which is
apodized sinc function. A correlation solution is validated only if well-suited to our purpose of obtaining results with a high spatial
its correlation score is above the ‘‘final resemblance threshold’’.
Global shifts between the images are computed by averaging
locally measured shifts obtained on all retained sampling points 1
http://logiciels.ign.fr/?Micmac.
(Bicheron et al., 2011). 2
http://www.tapenade.gamsau.archi.fr.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
6 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

Fig. 7. The pair of ortho-images used for the study of the September 2005 mega-dyke episode in the Afar region. Both the pre-event image (19 December 2004, incidence
angle: L0.484217 ) and the post-event one (13 January 2006, incidence angle: R0.22152 ) are SPOT4 images. The blue box indicates the area where the results are presented
in Fig. 8. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

resolution in near-fault area. MicMac uses multi-directional dy- A great advantage of MicMac is that it is open-source, which
namic programming when regularizing, with an energy minimiza- means that users may adapt it to their specific problems and also
tion approach based on the ‘‘data attachment term’’ (representing be involved improving the software.
the data consistency) and the ‘‘regularization term’’ (expressing the
a priori assumption of surface regularity) (Pierrot-Deseilligny and
Paparoditis, 2006). A multi-resolution approach consists in starting 2.3. Preliminary tests
the computation at a coarse resolution and improving the resolu-
tion at each matching level. For MicMac, multi-resolution approach A series of preliminary tests were performed on satellite
is mainly used in three-dimensional modeling where the combina- images. First tests were carried out on images containing a syn-
torial uncertainty is very high. For two-dimensional correlation, as thetic sub-pixel deformation field, in order to test the capability
the initial uncertainty is low, the multi-resolution approach is not of the correlators to retrieve this kind of deformation and evaluate
used, we work at full resolution at each matching level. The dispar- their performances in the ideal case, where the deformation is per-
ity solution is computed in an iterative process, the solution being fectly known. First, we used images without diachronism, where
successively refined at each matching level. Within the ‘‘correla- the slave image was created by applying synthetic sub-pixel defor-
tion window’’, the ‘‘search space’’ (or ‘‘exploration area’’) is cen- mation on the master image (an example of images containing
tered on the initial pixel position in the epipolar geometry from synthetic sub-pixel deformation is presented in Fig. 2 and the cor-
the previous correlation level. Within the search space, the corre- relation results obtained with the three correlators on these
lator computes a disparity for every position defined by the ‘‘dis- images are in Figs. 3 and 4).
cretization step’’ (which controls the sub-pixel precision). An These tests allowed us to evaluate the three correlators’ param-
apodized sinc kernel (Thévenaz et al., 2000), obtained by multipli- eters and do first comparisons between the correlators’ results and
cation of the ‘‘ideal kernel’’, the sinc kernel, by a differentiable and the theoretical displacement field. Second, tests were performed
bounded support function, the Hann function, is used for resam- on two diachronic images, in order to test the influence of diachro-
pling. The disparity with the best similarity/correlation coefficient nism on the correlation results. In the diachronous case, we used
is the correlation solution. two satellite ortho-images from the same place acquired several
We adapted MicMac for dealing with the correlation problems months apart, with no a priori deformation. The synthetic sub-pixel
that may occur when measuring two-dimensional ground deformation (for a fault offset) was generated with Coulomb3 (a
displacements. Among its adjustments, it is important to point MATLAB add-on used by the geoscience community for seismotec-
out its method of dealing with great spatial inhomogeneities in tonic simulations). This deformation was applied to the most recent
the resemblance between images with long time baselines (e.g. image. Using synthetic diachronic images gives the opportunity of
combination of rocky areas which change very little and snowy working on realistic scenario with the advantage of control over
areas which completely decorrelate between the two acquisitions). the amplitude and localization of the deformation. We found that
In order to prevent the corruption of measurements in well-corre- the precision of the correlators on synthetic images is around
lated areas by poor-correlated ones, MicMac provides the option of 1=100 px on non-diachronic synthetic images and 1=10 px on dia-
using non-linear costs (Pierrot-Deseilligny, 2013) associated with chronic ones.
the normalized cross-correlation coefficient (Fig. 1). A correlation Tests on the size of the correlation window have shown that
threshold, C min , is taken into account, the correlation below this MicMac, unlike the frequency-domain correlators, allows the use
threshold having no influence. Another parameter, c, controls the of a small correlation window without degrading the results, as
influence of correlation scores: the higher the value of c, the higher it uses a regularization algorithm. For the frequency-domain
the influence of correlation scores close to 1. When a priori knowl- correlators, the correlation noise increases significantly with
edge of the fault slip direction is available, MicMac offers the decreasing size of the correlation window (Binet and Bollinger,
possibility of using non-isotropic regularization, where the fault 2005). According to Leprince et al., 2007b and Ayoub et al., 2009,
slip direction is taken as a privileged direction for regularization
(Pierrot-Deseilligny, 2013). 3
http://www.coulombstress.org.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx 7

Fig. 8. East-west component of the offset field from the 2005 Afar dyke intrusion, using the pair of SPOT ortho-images from Fig. 7. The sizes of the correlation window used
are 32  32 px for COSI-Corr and 9  9 px for Medicis and MicMac. Profiles are stacked over a width of 2:6 km and a length of 10:3 km using the weighted median method. All
the three correlators detect two displacements corresponding to extension of the rift of  7 m on the east-west axis.

for a frequency-domain correlator such as COSI-Corr, 32  32 px is 3. Correlation results on real-case applications

considered a small window size. Therefore, testing it with what we
consider to be a small correlation window (8  8 px) is an extreme In this paper we present the correlation results using the three
case for this correlator, and less reliable results are to be expected. correlators (MicMac, COSI-Corr and Medicis) on two co-seismic
Thanks to MicMac’s parametrization flexibility, we were able to cases (instantaneous ground displacements due to the earthquake)
test a pyramidal approach by incrementally decreasing the correla- and a dyke intrusion case (a rather slow event, which took place
tion window size. The results obtained in areas which correlate over 20 days). For each case, we use two ortho-images bracketing
well are comparable to those obtained using the regularization the event. Real cases are of course more challenging, especially if
method, but they rapidly degrade in difficult areas, with a greatly there is a large interval of time between the two acquisition dates
increased computation time. of the images which implies important changes, and therefore an

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
8 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

Fig. 9. The pair of ortho-images used for the study of the 2008 Yutian earthquake. Both the pre-event image (25 August 2002, incidence angle: L1.733520 ) and the post-
event one (26 June 2008, incidence angle: L2.206474 ) are SPOT5 images. The fault is in an area covered almost entirely by the snow; surface ruptures are faintly visible across
the glaciers in the after-event image (inset). The green box indicates the area where the results are presented in Figs. 10 and 11. (For interpretation of the references to color in
this figure legend, the reader is referred to the web version of this article.)

increased risk of decorrelation. In all cases, the ground displace- On the section of the fault covered by our images, the horizontal
ments were calculated from sub-pixel correlation of ortho-rectified displacement is considered to be fairly constant with an average
SPOT panchromatic images. The topographic signal was removed displacement of  3:5 m and a maximum of 4.5–5 m (Xu et al.,
from the optical images using the 3-arc-second SRTM digital eleva- 2006).
tion model. The correlation was performed with the three correlators using
In order to quantify the displacement, we developed FDSC small and large correlation windows. The use of small correlation
(Fault Displacement Slip-Curve), a free, open-source program dis- windows is important in order to preserve the signal.
tributed under the CECILL-B license and available in the MicMac For all three correlators, no significant discontinuity is detect-
package.4 FDSC computes profiles perpendicular to the fault and able in the north-south component of the co-seismic offset field.
then stacks them using the weighted median method, with the cor- The deformation is detected only in the east-west component of
relation scores as weights. the correlation.
The correlation parameters for COSI-Corr, Medicis and MicMac In the case of large correlation windows, the correlation was
valid for all three study cases are summarized in Table 1. The calculated on a 64  64 px window (for COSI-Corr) and a
parameters are specific to each correlator. For MicMac, all the 65  65 px window (for Medicis) with a step of 16 px yielding a
parameter values, including the correlation window size ground pixel size of 160 m. For MicMac, the correlation window
(9  9 px), were the same in all cases. For the two other correlators, size was 65  65 px. The sizes of the small correlation windows
it was necessary to adapt the correlation window size according to were 8  8 px (for COSI-Corr) and 9  9 px (for Medicis) with a step
the noise level of the images. We tested many sizes of correlation of 4 px (yielding a ground pixel size of 40 m), and 9  9 px for
windows. The choice of the window size is rather qualitative; for MicMac. The MicMac results are in full resolution; therefore they
every correlator we present results from the smallest correlation are downsampled in order to have a pixel size comparable to the
window size which gave the clearest results. other two correlators.
For the small correlation window, COSI-Corr’s result is very
noisy and the amplitude of the co-seismic offset is underestimated
(Fig. 6). MicMac gives good results in terms of noise and amplitude
3.1. The 2001 Kokoxili earthquake of the deformation retrieved from the stacks of profiles perpendic-
ular to the fault. The results of Medicis are a bit noisy, but the
The Kokoxili earthquake (moment magnitude M w  7:8, 14 amplitude of the offset corresponds to the amplitude retrieved
November 2001) ruptured the Kunlun fault on the northern Tibe- by MicMac.
tan Plateau over a total distance of 450 km. It is one of the largest For the large correlation window, all three correlators retrieve
continental strike-slip earthquakes ever recorded (Klinger et al., the same amplitude of deformation ( 4:5 m, consistent to the
2005). In the case of a strike-slip fault, the fault surface is nearly displacements measured by Xu et al. (2006), Klinger et al. (2005),
vertical and the displacement is mainly horizontal. In Klinger Klinger et al. (2006)). The results of Medicis and MicMac are more
et al., 2005; Klinger et al., 2006 the displacements on the Kunlun blurry than when using small correlation windows.
fault after the 2001 earthquake were estimated to  4 m on Therefore, for the other studies, we present the COSI-Corr’s re-
average, with a maximum of 10 m. sults for correlation windows with a minimum size of 32  32 px
We study the co-seismic displacements on a small section of (Leprince et al., 2007b). For Medicis we present the results either
this fault using the sub-pixel correlation on a SPOT1 pre-event im- for the small correlation window (when possible, if the correlation
age and a SPOT4 post-event image, with 10 m pixel size (Fig. 5). It noise permits it) or the large one (a window size equivalent to the
is important to note the long temporal baseline (more than one used for COSI-Corr). All results for MicMac are for the small
13 years) between the two images. correlation window (9  9 px). For the sake of consistency, all the
correlators’ results presented were donwsampled by a factor equal
4
http://logiciels.ign.fr/?Telechargement,20. to the step used for the COSI-Corr correlation.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx 9

3.2. The 2005 dyke intrusion in Afar equivalent moment magnitude (M eq ) of  7:0 is computed based
on inversion of geodetic data and the geometry of the dyke (Gran-
Between 2005 and 2010, a rifting episode with 14 dyke intru- din et al., 2009). The September 2005 event is considered to be the
sions occurred in the Manda-Hararo rift in Afar (Ethiopia). The first largest rifting episode ever observed on land (Yirgu et al., 2006).
and largest dyke intruded on 26 September 2005 and ruptured the In order to measure the horizontal displacements induced by
entire 60-kilometer-long rift segment (Yirgu et al., 2006; Ayele this intrusion, correlation was performed on two 8 m-resolution
et al., 2007; Grandin et al., 2009), opening the rift with a maximum SPOT4 ortho-images separated by 13 months (Fig. 7). The correla-
extension of about 6 m (Ayele et al., 2007; Grandin et al., 2009; tion window size selected for COSI-Corr was 32  32 px with a
Barisin et al., 2009). Such rifting events include tectonic and mag- 16 px step, yielding a ground pixel size of 128 m. For Medicis
matic processes which lead to the separation of two divergent tec- and MicMac, the correlation window used was 9  9 px and their
tonic plates. In the case of a dyke intrusion, most of the results were downsampled by a factor equal to the COSI-Corr step.
deformation is aseismic (or ‘‘silent’’), so the seismic energy (quan- The east-west results (Fig. 8) show that the three correlators detect
tified by the moment magnitude scale M w ) is less relevant, but an two displacements corresponding to extension of the rift by a total

Fig. 10. East-west component of the co-seismic offset field from the 2008 Yutian earthquake, using the pair of SPOT ortho-images from Fig. 9. The sizes of the correlation
windows used are 32  32 px for COSI-Corr, 9  9 px for Medicis and 9  9 px for MicMac. Profiles are stacked over a width of 1:6 km and a length of 1:6 km using the
weighted median method. COSI-Corr (with a large correlation window) and MicMac (with a small correlation window) retrieve both a co-seismic offset of  2:7 m; Medicis’
results for a small correlation window are not satisfactory, the amplitude of the offset is not well retrieved.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
10 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

of  7 m. All three correlators detect the same amplitude of dis- extension and the movement on the fault includes both vertical
placement, which exceeds the  6 m maximum extension value and horizontal displacement. The co-seismic horizontal displace-
from (Grandin et al., 2009; Barisin et al., 2009; Ayele et al., ment is measured using correlation of two 5 m-resolution SPOT5
2007). This discrepancy seems to be caused by a problem in the ortho-images (Fig. 9). In addition to a time interval of nearly 6
ortho-images, independent of the correlation, most likely due to years between the two images, snow covers large percentage of
inaccuracies in the determination of the satellite attitude (Grandin the images (including the fault area) which makes the correlation
et al., 2009) and distortions induced by charge-coupled device extremely difficult to compute. The surface deformation associated
(CCD) misalignments (Barisin et al., 2009). with this earthquake is primarily in the east-west direction (Elliott
et al., 2010), so it is retrieved by the east-west correlation map
3.3. The 2008 Yutian earthquake (Fig. 10 and Fig. 11). The maximum horizontal offset measured in
the field by Xu et al. (2013) is  3:6 m.
The Yutian earthquake (M w  7:1, 21 March 2008) on the In this case, for COSI-Corr the best results were obtained with a
Tibetan Plateau is one of the largest recent continental normal fault correlation window of 32  32 px with a 8 px step, yielding a
earthquakes (Elliott et al., 2010). Normal faults develop in areas of ground pixel size of 40 m. For Medicis, as the results obtained

Fig. 11. East-west component of the co-seismic offset field from the 2008 Yutian earthquake, using the pair of SPOT ortho-images from Fig. 9. The sizes of the correlation
windows used are 32  32 px for COSI-Corr, 33  33 px for Medicis and 9  9 px for MicMac. Profiles are stacked over a width of 1:6 km and a length of 1:6 km using the
weighted median method. All three correlators retrieve a co-seismic offset of  2:7 m.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx 11

using a small correlation window (9  9 px, Fig. 10) are not satis- constructive comments. We are grateful to two anonymous
factory, the results are then presented for a correlation window reviewers that helped to improve this manuscript. This is IPGP
size comparable to the one used for COSI-Corr (Fig. 11). For this contribution number 3501.
very difficult case, MicMac proves to be the only correlator capable
of using a small correlation window and it also offers a clearer
References
image of the fault offset. In Fig. 11, the amplitudes of displacement
retrieved by MicMac (correlation window size 9  9 px), COSI-Corr Ayele, A., Jacques, E., Kassim, M., Kidane, T., Omar, A., Tait, S., Nercessian, A., de
(32  32 px) and Medicis (33  33 px) are  2:7 m. This value is in Chabalier, J.-B., King, G., 2007. The volcano–seismic crisis in Afar, Ethiopia,
agreement with Xu et al. (2013). starting September 2005. Earth Planetary Sci. Lett. 255, 177–187. http://
dx.doi.org/10.1016/j.epsl.2006.12.014.
Ayoub, F., Leprince, S., Keene, L., 2009. User’s Guide to COSI-CORR, Co-registration of
Optically Sensed Images and Correlation. <http://tectonics.caltech.edu/
4. Discussion/conclusion slip_history/spot_coseis>.
Barisin, I., Leprince, S., Parsons, B., Wright, T., 2009. Surface displacements in the
Numerous tests have been carried out to determine the best September 2005 Afar rifting event from satellite image matching: asymmetric
uplift and faulting. Geophys. Res. Lett. 36 (L07301). http://dx.doi.org/10.1029/
parameters for the three correlators and the results selected were 2008GL036431.
the best ones for each correlator. Unlike MicMac and Medicis, Bicheron, P., Amberg, V., Bourg, L., Petit, D., Huc, M., Miras, B., Brockmann, C.,
COSI-Corr does not have many parameters for the user to choose, Delwart, S., Ranéra, F., Hagolle, O., Leroy, M., Arino, O., 2011. Geolocation
assessment of MERIS GlobCover orthorectified products. IEEE Trans. Geosci.
which can be seen as an advantage (it is very easy to use), but also
Remote Sens. 49 (8), 2972–2982. http://dx.doi.org/10.1109/
as a drawback as the user does not have a lot of options especially TGRS.2011.2122337.
for difficult correlation cases. Medicis is fairly easy to use despite Binet, R., Bollinger, L., 2005. Horizontal coseismic deformation of the 2003 Bam
the high number of parameters and no graphical user interface. (Iran) earthquake measured from SPOT-5 THR satellite imagery. Geophys. Res.
Lett. 32 (L02307). http://dx.doi.org/10.1029/2004GL021897.
MicMac also has a large number of parameters but a simple Casson, B., Delacourt, C., Allemand, P., 2005. Contribution of multi-temporal remote
command line interface, dedicated to measuring ground displace- sensing images to characterize landslide slip surface – application to the La
ments, is available (Pierrot-Deseilligny, 2013). Clapière landslide (France). Nat. Hazards Earth Syst. Sci. (5), 425–437.
Delacourt, C., Allemand, P., Casson, B., Vadon, H., 2004. Velocity field of the ‘‘La
Our suite of examples allowed us to evaluate the performance Clapière’’ landslide measured by the correlation of aerial and QuickBird satellite
of each correlator. For the Kokoxili earthquake, the primary images. Geophys. Res. Lett. 31 (L15619). http://dx.doi.org/10.1029/
challenges were the long temporal baseline between the two 2004GL020193.
Dominguez, S., Avouac, J.-P., Michel, R., 2003. Horizontal coseismic deformation of
SPOT1 and SPOT4 images used, the highly mountainous snowy the 1999 Chi–Chi earthquake measured from SPOT satellite images:
area of the fault, and the use of multi-sensor images. It allowed implications for the seismic cycle along the western foothills of central
us to study how each correlator managed to find the fault signal Taiwan. J. Geophys. Res. 108 (B2). http://dx.doi.org/10.1029/2001JB000951.
Elliott, J.R., Walters, R.J., England, P.C., Jackson, J.A., Li, Z., Parsons, B., 2010. Extension
in a difficult case, testing small and large correlation windows. on the Tibetan plateau: recent normal faulting measured by InSAR and body
Our main interest is to have very detailed results in order to finely wave seismology. Geophys. J. Int. (183), 503–535. http://dx.doi.org/10.1111/
evaluate the ground deformation, therefore using small correlation j.1365-246X.2010.04754.x.
Grandin, R., Socquet, A., Binet, R., Klinger, Y., Jacques, E., de Chabalier, J.-B., King,
windows is essential if the correlation noise allows it. The Afar
G.C.P., Lasserre, C., Tait, S., Tapponnier, P., Delorme, A., Pinzuti, P., 2009.
dyke intrusion was a favorable study case in terms of imaging con- September 2005 Manda Hararo–Dabbahu rifting event, Afar (Ethiopia):
ditions: it occurred in a desertic area with no significant seasonal constraints provided by geodetic data. J. Geophys. Res. 114 (B08404). http://
variations, and the time interval between the two SPOT4 images dx.doi.org/10.1029/2008JB005843.
Guérin, C., Binet, R., Pierrot-Deseilligny, M., 2012. Détection des changements
was relatively short. The Yutian earthquake was an especially dif- d’élévation d’une scène par imagerie satellite stéréoscopique. In:
ficult case: the relatively long temporal baseline of almost 6 years Reconnaissance des Formes et Intelligence Artificielle (RFIA), lyon, France.
between the two SPOT5 images and the snow made it extremely Hermas, E., Leprince, S., El-Magd, I.A., 2012. Retrieving sand dune movements using
sub-pixel correlation of multi-temporal optical remote sensing imagery,
difficult if not impossible to retrieve the fault signal due to the high northwest Sinai Peninsula, Egypt. Remote Sens. Environ. 121, 51–60. http://
risk of decorrelation. Due to all these unfavorable conditions, this dx.doi.org/10.1016/j.rse.2012.01.002.
was an extreme test of the correlators’ efficiency. Hollingsworth, J., Leprince, S., Ayoub, F., Avouac, J.-P., 2012. Deformation during the
1975–1984 Krafla rifting crisis, NE Iceland, measured from historical optical
As expected for a frequency-domain correlator, COSI-Corr gave imagery. J. Geophys. Res. 117 (B11407), http://dx.doi.org/10.1029/
very good results when using large correlation windows 2012JB009140.
(minimum 32  32 px, as specified in Leprince et al. (2007b) and Inglada, J., Muron, V., Pichard, D., Feuvrier, T., 2007. Analysis of artifacts in subpixel
remote sensing image registration. IEEE Trans. Geosci. Remote Sens. 45 (1),
Ayoub et al. (2009)), but not satisfactory results when using 254–264. http://dx.doi.org/10.1109/TGRS.2006.882262.
smaller correlation windows. However, despite using the Klinger, Y., Xu, X., Tapponnier, P., der Woerd, J.V., Lasserre, C., King, G., 2005. High-
frequency domain, Medicis retrieved accurate deformation ampli- resolution satellite imagery mapping of the surface rupture and slip distribution
of the Mw  7:8, 14 November 2001 Kokoxili Earthquake, Kunlun Fault,
tudes for the first and second study cases using a small correlation
Northern Tibet, China. Bull. Seismol. Soc. Am. 95 (5), 1970–1987. http://
window. For the most difficult (Yutian) study case, MicMac was the dx.doi.org/10.1785/0120040233.
only correlator capable of giving good results at small correlation Klinger, Y., Michel, R., King, G., 2006. Evidence for an earthquake barrier model from
window and it offered a better picture of the fault signal. Thanks Mw  7:8 Kokoxili (Tibet) earthquake slip-distribution. Earth Planet. Sci. Lett.
(242), 354–364. http://dx.doi.org/10.1016/j.epsl.2005.12.003.
to its features (e.g. regularization, non-linear correlation cost), Leprince, S., Ayoub, F., Klinger, Y., Avouac, J.-P., 2007a. Co-registration of optically
MicMac appears to be less vulnerable to noise and to be able to sensed images and correlation (COSI-Corr): an operational methodology for
deal with large spatial inhomogeneities due to long time intervals, ground deformation measurements, in: IEEE International Geoscience and
Remote Sensing Symposium (IGARSS), Barcelona, Spain.
snow and seasonal changes. MicMac proves to be a quality free Leprince, S., Barbot, S., Ayoub, F., Avouac, J.-P., 2007b. Automatic and precise
open-source alternative for image correlation for ground deforma- orthorectification, coregistration, and subpixel correlation of satellite images,
tion measurements. application to ground deformation measurements. IEEE Trans. Geosci. Remote
Sens. 45 (6), 1529–1558. http://dx.doi.org/10.1109/TGRS.2006.888937.
Lisein, J., Pierrot-Deseilligny, M., Bonnet, S., Lejeune, P., 2013. A photogrammetric
workflow for the creation of a forest canopy height model from small
Acknowledgements
unmanned aerial system imagery. Forests 4 (4), 922–944. http://dx.doi.org/
10.3390/f4040922.
The authors would like to thank CNES for funding this project Massonnet, D., Feigl, K.L., 1998. Radar interferometry and its application to changes
through the TOSCA program and for kindly lending us Medicis. in the Earth’s surface. Rev. Geophys. 36 (4), 441–500.
Michel, R., Avouac, J.-P., 2002. Deformation due to the 17 August 1999 Izmit, Turkey,
We thank Stéphane May for his valuable help with Medicis, and earthquake measured from SPOT images. J. Geophys. Res. 107. http://dx.doi.org/
Raphaël Grandin, Stéphanie Dumont and Belle Philibosian for their 10.1029/2000JB000102.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002
12 A.-M. Rosu et al. / ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

Pierrot-Deseilligny, M., 2013. MicMac, Apero, Pastis and Other Beverages in a Van Puymbroeck, N., Michel, R., Binet, R., Avouac, J.-P., Taboury, J., 2000. Measuring
Nutshell. <http://logiciels.ign.fr/?Micmac>. earthquakes from optical satellite images. Appl. Opt. 39 (20), 3486–3494.
Pierrot-Deseilligny, M., Paparoditis, N., 2006. A multiresolution and optimization- Xu, X., Yu, G., Klinger, Y., Tapponnier, P., Van Der Woerd, J., 2006. Reevaluation of
based image matching approach: an application to surface reconstruction surface rupture parameters and faulting segmentation of the 2001 Kunlunshan
from SPOT5-HRS stereo imagery. In: ISPRS Workshop on Topographic earthquake ðM w 7:8Þ, northern Tibetan Plateau, China. J. Geophys. Res. 111
Mapping from Space (With Special Emphasis on Small Satellites), Ankara, (B05316). http://dx.doi.org/10.1029/2004JB003488.
Turkey. Xu, X., Tan, X., Yu, G., Wu, G., Fang, W., Chen, J., Song, H., Shen, J., 2013. Normal- and
Scambos, T.A., Dutkiewicz, M.J., Wilson, J.C., Bindschadler, R.A., 1992. Application of oblique-slip of the 2008 Yutian earthquake: evidence for eastward block
image cross-correlation to the measurement of glacier velocity using satellite motion, northern Tibetan Plateau. Tectonophysics 584, 152–165. http://
image data. Remote Sens. Environ. 42 (3), 177–186. dx.doi.org/10.1016/j.tecto.2012.08.007.
Thévenaz, P., Blu, T., Unser, M., 2000. Interpolation revisited. IEEE Trans. Med. Imag. Yirgu, G., Ayele, A., Ayalew, D., 2006. Recent seismovolcanic crisis in Northern Afar,
19 (7), 739–758. Ethiopia. EOS 87 (33), 325–329.

Please cite this article in press as: Rosu, A.-M., et al. Measurement of ground displacement from optical satellite image correlation using the free open-
source software MicMac. ISPRS J. Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2014.03.002