Anda di halaman 1dari 8

MAKALAH RINGKASAN JURNAL BAHASA INGGRIS

DISUSUN OLEH :

DAMAR KUMARA SAKHI


F1D322013
Kelas A

DOSEN PENGAMPU :

Juventa, S.T., M.T.

MATA KULIAH BAHASA INGGRIS

PRODI TEKNIK GEOFISIKA

FAKULTAS SAINS DAN TEKNOLOGI

UNIVERSITAS JAMBI

JAMBI

2023
KATA PENGANTAR

Puji syukur kami panjatkan kepada Tuhan Yang Maha Esa, atas rahmat dan karunia-
Nya kami dapat menyelesaikan tugas penulisan makalah tepat waktu. Adapun judul dari
tugas ini adalah “Resume Jurnal”. Tugas ini dibuat dengan tujuan memenuhi tugas dari
Bapak Juventa,S.T., M.T. dalam Mata Kuliah Geofisika Umum.
Saya menyadari bahwa dalam penulisan tugas ini tidak terlepas dari bantuan banyak
pihak, sehingga makalah ini dapat terselesaikan. Saya juga menyadari bahwa tugas ini
masih memerlukan penyempurnaan, terutama pada bagian isi.
Oleh karena itu, saya menerima segala bentuk kritik dan saran yang membangun dari
berbagai pihak. Apabila terdapat banyak kesalahan pada makalah ini, saya memohon
maaf. Demikian yang dapat saya sampaikan. Semoga tugas ini dapat bermanfaat bagi
perkembangan dunia pendidikan.

Jambi, 10 Mei 2023

Damar Kumara Sakhi

1
JOURNAL SUMMARY
Writer : Mark Pilkington dan Pejman Shamsipour
Years : 2014
Title : Noise Reduction Procedures for Gravity-Gradiometer Data
Journal Type : Geophysic Journal
Pages : 69-78

ABSTRACT

Noise suppression of airborne gravity-gradiometer data is a crucial part of the data


processing stream. We considered two approaches to removing noise kriging and
directional filtering. Kriging is an estimation procedure for the interpo- lation of spatial
data. The estimator is calculated from the data variogram, which characterizes the noise
level and cor- relation length of the measurements. Directional filtering uses a user-
defined operator that is oriented to preferentially smooth the data along the strike, but it
leaves short-wave- length components in the cross-strike direction for definition of the
trend edges. Both methods were applied to a recently collected offshore gravity gradient
survey.

1. INTRODUCTION

Ongoing improvements in gravity gradiometry technology have led to its


establishment as a powerful technique in mining and hy- drocarbon exploration. In
addition to providing high-resolution gravity data, the availability of the gravity gradient
tensor offers, in comparison with the vertical gravity component, a much richer sig- nal
with increased options for processing and interpretation meth- ods. Any such
methodologies must address the possibility of noise in the data and include appropriate
steps to mitigate the unwanted effects of a noisy signal. This is the focus of this study, the
suppres- sion of noise in gravity-gradiometer data. The sources and approaches to
treatment of noise in airborne gravity-gradiometer AGG systems are numerous and have
been discussed extensively in the literature Hammond and Murphy, 2003 Boggs and
Dransfield, 2004 Barnes and Lumley, 2011 Di- Francesco, 2013 Dransfield and
Christensen, 2013. Certain noise sources are related to the AGG instruments, e. g. ,
measurement error and noise induced by aircraft turbulence. Other sources arise from the
subsequent processing of data, as in terrain correction errors and artifacts introduced from
leveling and gridding operations. Standard processing of AGG data is designed to
suppress the unwanted noise and emphasize the geologic signals present. With continuing
im- provements in survey design and operation and data processing methods, predicted
noise levels are expected to decrease further in the future DiFrancesco, 2013.
Nonetheless, currently produced AGG data often still contain high enough levels of noise
to interfere with qualitative and quantitative interpretations. Our approach here is to
explore noise-reduction methods and determine their appli- cability and effectiveness in
removing unwanted noise in AGG data. A major strength of gravity-gradiometer surveys
is the high- resolution content of data that can be achieved, particularly when compared
with airborne gravity surveys. Hence, any noise-removal approach should preserve this
key asset. But the most common method to combat noise is to smooth or low-pass filter

2
the data, which effectively degrades resolution. To avoid smoothing, Lyrio et al. 2004
use a wavelet-based approach, making the assumption that the signal and noise can be
separated in the wavelet domain, once a suitable threshold scale is found. Denoised data
are deter- mined by transforming back from the wavelet domain after the selected noise
component contributions are set to zero. Their application is applied only to line data, but
it should be applicable to gridded data.

2. KRINGING

Kriging is an estimation procedure commonly used in geostatis- tics for the


interpolation of spatial data. The method is based on the theory of regionalized variables
Matheron, 1963 and provides the best linear unbiased estimator for a given problem. The
basis of the method is to find an estimator Z0 of some field from a set of neigh- boring
field values Z xi i 1 n . The field Z xi is assumed to be stationary with a constant average
value and known covariance Ch. The estimator is constructed as follows a 8000 E 10 Z X
Zx 1 i i 5 i1 0 6000 4000 2000 5 10 15 20 25 30 35 where the unknown weights i are to
be determined. The estimator Z0 is unbiased, so the sum of the weights i equals one. The
weights are a function of the distance between Z0 and the surrounding Z xi , with the
closer points exerting more influence on the estimated value than distant ones.
Importantly, kriging incorporates the correlation between data points into the estimator
Z0 through the variogram h, given by k h 1 X Zx Zx 2 2 0 b E 40 2k i1 ih 8000 6000
4000 2000 0 0 2000 4000 6000 8000 Easting m 25 20 15 10 5 0 5 10 15 20 25 30
35.
where k is the number of pairs of points that are separated by a distance h. If the field
Z is a stationary process, then h is inde- pendent of location and a single variogram can
be used for process- ing the entire data set. As h increases to a value known as the range,
h levels off to a constant value 2. The variogram h is re- lated to the covariance Ch by h 2
Ch 3 where 2 is the process variance. The variogram may differ accord- ing to the
direction considered, so that each h can be subdivided by orientation and anisotropic
effects easily included. When modeling typically isotropic variograms, we need to define
three parameters the range, sill, and nugget effect. The range is the distance after which
no more correlation exists between two obser- vations. The sill is the upper limit value of
the variogram at a dis- Figure 1. Two examples of noise in AGG data. a Data Tzz from
Blatchford Lake, Northwest Territories, collected at 250-m line spacing and 100-m mean
flight height. The contour interval is 5 E. b Data Tzz from McFaulds Lake, Ontario,
collected at 250-m line spacing and 100-m mean flight height. The contour in- terval is 5
E. tance equal to the range. If it exists, the sill represents the variance of the field. The
nugget effect represents the combined effects of short- scale variation of the field,
positioning errors, measurement inaccur- acy, and any variability of the field occurring at
small enough scales to be unresolved by the sampling interval. Therefore, it can be
considered as the total residual random noise remaining in the data after initial
processing. Once the variogram is calculated, the estimator Z0 is found by minimizing
the error variance Z0 Z0 2, leading to the ordinary kriging system The synthetic data
example in Figure 2 demonstrates the effective- ness of ordinary kriging in removing
uncorrelated noise. Figure 2a and 2b shows two examples of synthetic Tzz data corrupted
by un- correlated Gaussian noise with variances of 1. 5 and 3. 0 E2, respec- tively. Little
coherent Tzz signal is apparent.
DIRECTIONAL FILTERING
3
where is a Lagrange multiplier and ij xi xj and 0j x0 xj . Subscript 0 refers to the
location of the interpo- lated point. The system in equation 4 is solved at each data grid
location. To ensure stability of the linear system in equation 4, the calculated variogram
equation 2 is replaced by a best-fitting ana- lytic function such as an exponential. Once
the weights i are found from equation 4, they are used in equation 1 to determine the
kriged values Z at each grid point. There is a close relationship between geostatistics and
linear fil- tering from a mathematical point of view. They both have roots in the
mathematical developments by Kolmogorov and Weiner Chils and Delfiner, 1999. For a
variable or field comprising two components with different correlation properties,
factorial krig- ing is a geostatistical technique that can separate out each compo- nent. In
this study, we suppose that we have one field or variable, plus noise. This reduces to the
particular case of ordinary kriging. Let Zi indicate the samples and Zi the underlying
random variable. The model is then Zi Zi i 5 Removing or suppressing noise through
simple low-pass filtering is at odds with the goal of preserving gravity signals in the data
caused by geologic sources of interest. The latter contributes, within the limits imposed
by the flight height and sampling rate, to the observed power spectrum at all wavelengths
in the tensor compo- nents. Importantly, the shorter wavelength portion of the spectrum
provides the definition or sharpness of signal required to define de- tail in the observed
data. Removing too much of the short wave- length content results in loss of resolution at
the smallest scales and poorer definition of anomaly shape, continuity, and amplitude.
Fortunately, the type of anomaly that is most useful for geologic mapping purposes is 2D,
even though we are usually dealing with map-based 3D data sets. Quasi-linear 2D
features in the data may correspond to litho- logical contacts, but even if this is not the
case, such trends can help in distinguishing structural regimes, and deformation styles and
trends. Truly 3D features, e. g. , caused by sources with horizontal dimensions smaller
than the flight height, are not of much interest if geologic mapping is the primary aim of
the survey. Because a 2D feature varies more smoothly along the strike compared with
perpendicular to the strike, low-pass filtering to remove noise where i are random errors
defined only at the sample points. We start with the simplest sce- nario in which the
errors are nonsystematic so E i 0, uncorrelated with the data E Z x 0 and uncorrelated
with themselves E ij 0. The aim is to estimate a noise-free value of Z0 from noisy
samples Zi. In signal processing, this task is known as fil- tering. However, in
geostatistics, this is ad- dressed with factorial kriging. For the case of uncorrelated errors,
at any point other than a sample point, the kriging mean square error is EZ Z02 EZ Z02
E2. 6 Because the second term on the right hand side is zero by definition, the error
variance for the Zi is the same as that defined for the case of ordi- nary kriging, and
equation 4 can be used to do the filtering. So for uncorrelated noise, the krig- ing
estimator defined in equation 1 provides an unbiased estimate of a noise-free Zi in equa-
tion 5. For correlated noise, factorial kriging does a b 60 40 20 c d 60 40 20 20 40 60 80
Easting m E 4 2 0 2 4 20 40 60 80 Easting m not reduce to ordinary kriging but must
be modi- fied to incorporate the spatial correlation of the noise present. Figure 2. a
Synthetic

4
data with noise nugget effect 1. 5E2, range 40 units. b As panel a but with a nugget
effect of 3. 0E2. c Kriged data from panel a. d Kriged data from panel b. can be tuned to
take this into account. Short-wavelength features along the strike are preferentially
removed while being preserved in the cross-strike direction. In this fashion, short-
wavelength com- ponents that provide the detailed definition of the 2D features are
unaffected, whereas the short-wavelength noise that overprints and disrupts along strike
trends is suppressed. This is the essence of the contact lineament processing approach of
Brewster 2013. This kind of varying directional low-pass filter is used in image process-
ing for the detection of object edges from noisy data and the robust detection of lines in
computer vision applications e. g. , Freeman and Adelson, 1991 Geusebroek et al. , 2003.
In seismic data processing, structure-oriented filtering or smooth- ing Fehmers and
Hocker, 2003 Hale, 2009 uses similar principles on seismic images to enhance reflection
continuity while preserving discontinuities such as faults. A simpler form of
directional filtering is the basis of microleveling of aeromagnetic data, where short-
wavelength variations perpendicular to the flight-line direction are suppressed Minty,
1991. In this case, only a single filter di- rection is needed for the whole data set.
Enhancing the continuity of linear features in aeromagnetic data is also the goal of several
trend reinforcement algorithms that use approaches that differ from directional filtering,
e. g. , Keating 1997 and Smith and OConnell 2005. The successful application of a
directional filter requires the ori- entation of the filter to be known at each grid point.
Several options exist to calculate this orientation, for example, by computing a spa- tial
average in several directions and then choosing some measure to define the optimum
direction. This could be done using the spatial average that is closest to the original grid
value or using the maxi- mum spatial average Lakshmanan, 2004. For gradiometer data
applications, we have the advantage of knowing that the major trends in the data are
caused by the dom- inant geologic strike of the causative bodies. The strike can be de-
termined from the complete gradient tensor or from each tensor component separately.
Two methods exist for finding the strike of a 2D body from the gradient tensor. The first
is based on finding a rotation angle such that the sum of squares of the first row of the
rotated gradient tensor reaches a minimum Pedersen and Rasmus- sen, 1990. This strike
angle value is given by solutions to Figure 3. Response of a Gaussian smoothing filter
with N30E, x 5 and 2 10 km. The x-axis is eastwest, and the y-axis is northsouth. Units
have been arbitrarily set to kilo- meters. The maximum value is unity before normalizing.
tan 2 2TxyTxx Tyy TxzTyz 7 Txx2 Tyy2 Txz2 Tyz2 The second approach is based on
first calculating the eigenvalues and eigenvectors of the gradient tensor matrix. Beiki and
Pedersen 2010 note that the eigenvector associated with the smallest eigen- value gives
the strike direction of a simple line source. For quasi-2D bodies, which are thicker and
dike-like, the same eigenvector can be used to estimate the strike direction. Both of these
approaches as- sume that the tensor components are produced by a 2D or quasi-2D
body. To verify this assumption, the dimensionality invariant I given by Pedersen and
Rasmussen 1990 can be used. It varies from zero for a 2D body to unity for a 3D source,
so using values of I 0. 3 is a practical definition of two dimensionality for real data Beiki
and Pedersen, 2010. Similarly, values of I can be plotted for the whole data set, and a
threshold can be chosen. The final method, which is adopted in this work, does not
require the complete gradient tensor values simultaneously and is not reliant on
assumptions regarding the dimensionality of the source bodies. It determines a strike
direction for each tensor

5
component independ- ently, based on the two orthogonal horizontal gradients of the
com- ponent at each grid point. Due to noise, using horizontal gradients directly can lead
to highly variable strike directions unsuitable for further processing. Therefore, to reduce
noise effects on strike es- timates, the component data are first low-pass filtered. Second,
rather than using calculated horizontal gradients from finite differences, the strike
direction is found from fitting a plane, in a least-squares sense, to a windowed portion of
a given component. A 5 5 window is used for the planar surface fit. If the fitted plane has
the form ax bx c 0, then the strike direction is given by tan1 cb , where is measured
clockwise from the north. Using a finite window coupled with some low-pass filtering
produces more continuous strike estimates that are suitable for use in the sub- sequent
directional filtering. The estimated strike direction at each grid point defines the ori-
entation of the directional filter. We use a Gaussian filter for this purpose, which is
defined at a point x, y by g x y expx cos y sin 22 y cos x sin 22 8 where the origin is at 0,
0 and is the angle between the filter and the y-axis. Variables x and y determine the width
of the filter and hence the degree of smoothing. If is defined also to be the strike of the
component anomaly, then y is always kept greater than x be- cause this determines the
direction of maximum smoothing. Filter values using equation 8 are always positive and
are normalized to sum to unity Figure 3. At each grid point, the strike is determined and
used to calculate the appropriate filter coefficients, which are then applied to each
component. Although the same strike values will often occur at several points throughout
a given component grid, recalculating filter coefficients using equation 8 at every grid
point is not computationally demanding, so no attempt was made to optimize the filtering
using lookup tables, etc. Because each com- ponent grid is processed separately, the
resulting strike grids will be different. This reflects the difference between the responses
produced by a given source for each tensor component.

ST. GEORGES BAY SURVEY

A gravity-gradiometer survey was conducted by Bell Geospace over St. Georges Bay,
Newfoundland and Labrador, to assist with mapping regional basin structures in a mostly
offshore region Du- mont and Jones, 2013. Flight lines were spaced 500 m apart with a
northeastsouthwest orientation whereas tie-lines were spaced at 5000-m intervals with a
northwestsoutheast orientation. The sur- vey was flown at a nominal altitude of
100 m on a gentle drape surface. The 90-m SRTM elevation model was used for
computing the drape surface. The data were leveled and then denoised using the 48 30
FFT approach Sanchez et al. , 2005. Based on correlation tests, a density of 2200 kgm3
was chosen for the terrain correction. The gradiometer survey covers most of the
Carboniferous Bay St. George subbasin, which is bounded to the south by the onshore
northeast-trending Long Range Fault and constitutes part of the larger Maritimes basin
Shillerif and Williams, 1979. Carbon- iferous and Upper Devonian sedimentary rocks
that constitute the subbasin fill are exposed onshore to the north, east, and south,
overlying Precambrian basement.

6
3. DISCUSSION

We have discussed two main options for noise suppression in AGG data directional
filtering and kriging. Incidentally, simple low-pass filter- ing was also considered.
Kriging and directional filtering have different foundations, theory, and implementations.
Kriging is computationally intensive but is almost completely data driven the variogram
model is chosen based on the calculated variogram. The lack of user input might be
viewed as an advantage, increasing the objectivity of the method, but if the results are
unsatisfactory, there are limited options for improvement. Direc- tional filtering, on the
other hand, requires several user-defined parameters, so processing can be based on
experience and any avail- able geologic information, e. g. , structural strikes from
mapping. It also has the advantage of incorporating strike information to pro- duce
increased coherency along these orientations. Kriging can also 60 40 20 0 20 40 60 0 km
4 raw lp4 df include directional information by using a spatially varying aniso- tropic
variogram, with a longer range in the strike direction, but directional filtering will achieve
the same goal much faster. A practical approach for noise reduction is to use kriging and
take the results to guide the directional filtering. Kriging gives an esti- mate of how much
noise is present by calculating the nugget effect and using this in the interpolation of the
noisy data. Because the variogram calculation also specifies the correlation character of
the data, this is reflected in the level of smoothing of the kriged a 80 60 40 20 0 20 40 20
48.

4. CONCLUSION

Ordinary kriging and directional filtering are shown to be viable options for noise
suppression of gravity-gradiometer data. Kriging is based on the statistical character of
the data and produces esti- mates with a commensurate level of smoothness and noise
reduc- tion. For directional filtering, the degree of smoothing is user defined but has the
advantage of being sensitive to strike informa- tion derived from the data. For the St.
George’s Bay survey, both approaches successfully reduce noise levels so that coherent,
geo- logically meaningful anomaly patterns are revealed. In this particu- lar case, the
directional filtering was carried out independently of the kriging, but in retrospect,
kriging of the data appears useful in guiding the choice of parameters used for filtering.

Anda mungkin juga menyukai