Automated well log interpretation for seismic Inversion
Internal Report
Christopher Dyt, Irina Emelyanova, Michael Clennell
EP177125
August 2017
EXPLORATION GEOSCIENCE, DATA ANALYTICS
CSIRO Energy
Citation
Dyt C, Emelyanova I and Clennell MB (2017) Automated well log interpretation for seismic inversion. The CORE Data Analytics Project: Milestone report. CSIRO Energy, Australia Citation
Copyright
© Commonwealth Scientific and Industrial Research Organisation 20XX. To the extent permitted
by law, all rights are reserved and no part of this publication covered by copyright may be
reproduced or copied in any form or by any means except with the written permission of CSIRO.
Important disclaimer
CSIRO advises that the information contained in this publication comprises general statements
based on scientific research. The reader is advised and needs to be aware that such information
may be incomplete or unable to be used in any specific situation. No reliance or actions must
therefore be made on that information without seeking prior expert professional, scientific and
technical advice. To the extent permitted by law, CSIRO (including its employees and consultants)
excludes all liability to any person for any consequences, including but not limited to all losses,
damages, costs, expenses and any other compensation, arising directly or indirectly from using this
publication (in part or in whole) and any information or material contained in it.
CSIRO is committed to providing web accessible content wherever possible. If you are having
difficulties with accessing this document please contact [email protected].
Internal Report | i
ii | Internal Report
Contents
Executive summary .................................................................................................................. v
Part I Classifying Rock Types 1
1 Introduction ................................................................................................................. 2
2 Classifying rock types .................................................................................................... 3
2.1 Methodology ................................................................................................... 3
2.2 Study area location........................................................................................... 4
2.3 Expert classification rules ................................................................................. 4
2.4 Well data ......................................................................................................... 5
2.5 Rules for log classification................................................................................. 5
2.6 Velocity and impedance profiles ....................................................................... 6
3 Results 7
3.1 Lauda 1 ............................................................................................................ 7
3.2 Martell 1 ........................................................................................................ 12
3.3 Iago 1............................................................................................................. 14
3.4 Achilles 1 ....................................................................................................... 16
4 Discussion .................................................................................................................. 18
5 Conclusion.................................................................................................................. 19
6 References ................................................................................................................. 20
Internal Report | iii
Figures
Figure 1: Location of the 4 exploration wells found used in the survey; Iago 1, Martell 1, Lauda 1
and Achilles 1 ........................................................................................................................... 4
Figure 2. The impedance versus depth for rocks classified as shale in Lauda 1 ............................ 7
Figure 3. The impedance versus depth for rocks classified as marls in Lauda 1 ........................... 8
Figure 4. The impedance versus depth for rocks classified as medium to fine grain siliciclastic in
Lauda 1 .................................................................................................................................... 8
Figure 5. The impedance versus depth for rocks classified as medium to coarse grain siliciclastic
in Lauda 1 ................................................................................................................................ 9
Figure 6. The impedance versus depth for rocks classified as sands in Lauda 1 ........................... 9
Figure 7.The impedance versus depth for rocks classified as gas filled sands in Lauda 1 ............ 10
Figure 8. The impedance versus depth for rocks classified as radiolarite in Lauda 1 .................. 10
Figure 9. The impedance versus depth for rocks classified as pyritic material in Lauda 1 ........... 11
Figure 10. The impedance versus depth for rocks classified as limestone in Lauda 1................. 11
Figure 11. The velocity profile for all material from Lauda 1. the grey data points are the actual
velocities from the well log. The orange data points are the values from the various calculated
velocity curves for the rock type at that depth. ....................................................................... 12
Figure 12: The velocity profile for all material from Lauda 1. the grey data points are the actual
velocities from the well log. The orange data points are the values from the various calculated
velocity curves for the rock type at that depth. ....................................................................... 12
Figure 13. The impedance versus depth for rocks classified as shale in Martell 1 ...................... 13
Figure 14. The impedance versus depth for rocks classified as sand in Martell 1....................... 13
Figure 15. The impedance versus depth for rocks classified as coal in Martell 1........................ 14
Figure 16. The velocity profile for all material from Martell 1. The grey data points are the
actual velocities from the well log. The orange data points are the values from the various
calculated velocity curves for the rock type at that depth. ....................................................... 14
Figure 17. The impedance versus depth for rocks classified as shale in Iago 1 .......................... 15
Figure 18. The impedance versus depth for rocks classified as sand in Iago 1 ........................... 15
Figure 19. The impedance versus depth for rocks classified as marl in Iago 1 ........................... 16
Figure 20. The velocity profile for all material from Iago 1. the grey data points are the actual
velocities from the well log. The orange data points are the values from the various calculated
velocity curves for the rock type at that depth. ....................................................................... 16
Figure 21. The velocity profile for all material from Achilles 1. the grey data points are the actual
velocities from the well log. The orange data points are the values from the various calculated
velocity curves for the rock type at that depth. ....................................................................... 17
iv | Internal Report
Internal Report | v
Executive summary
To accurately perform seismic inversions, accurate velocity and impedance profiles need to be defined for the various rock types present in the subsurface, both reservoir lithology (e.g. clean sandstone, porous limestone) and the overburden rocks (shales, marls). In order to generate these
profiles rapidly and in an objective and repeatable manner we attempt to automatically determine lithology from well logs before analysing the velocity and density trends for each of the identified categories, so that they are “inversion ready” facies descriptions, for a number of commonly occurring rock types.
To classify the rock sequence into type’s relevant to seismic inversion we first captured an expert’s methodology of interpreting facies from the well logs, and convert these into mathematical algorithms.
These algorithms were implemented in a single computer code, which was applied to four oil and gas exploration wells drilled in the Northern Carnarvon Basin offshore Western Australia during
the period 2000 to 2009. The classifications derived from these wells generally seem to match the geologist’s interpretation, and in each case, take less than a second to run, as opposed to many hours of interpretation when performing the task manually.
Importantly, velocity and impedance profiles for each rock type are produced that could potentially be fed into seismic inversion software.
There are however some situations, either due to sparse data for that rock type, or extremely variable/noisy data, the produced depth-impedance curves of poor quality. It may also suggest that
there may be strong variation within the individual rock type groupings produced (e.g. marls may vary widely in there rock properties and hence there log response) , suggesting that they may be further separated by method such as clustering techniques.
Internal Report | 1
Part I Classifying Rock Types
Using wireline logs to determine facies
2 | Internal Report
1 Introduction
Seismic inversion is the process of turning seismic reflections into quantitative rock properties,
and is a relatively standard technique in exploration and development of oil and gas fields.
To conduct a seismic inversion, you need both seismic data and a wavelet statistically calculated
from the data. Importantly, the quality of the wavelet estimation is vital to the outcome of the
seismic inversion. The wavelet estimation can be greatly enhanced by a good understanding of
sonic and density curves of the material in the well.
Different rock types will have different velocity and impedance (velocity * density) characteristics
due to their grain’s size, mineralogy, degree of cementation etc. Similarly, the impedance of a rock
type can vary due to its burial history and degree of compaction.
Wireline well logs are often used to attempt to determine the velocity/impedance profiles of the
rocks in a well, so they can then be applied to the local field being explored. This can be a labour-
intensive practice requiring expert knowledge of a geophysicist to firstly interpret the log data to
determine the rock types and then put together the velocity curve
This work is an attempt to automate this process.
Internal Report | 3
2 Classifying rock types
2.1 Methodology
The aim of this work is to create an automated system to determine the impedance vs depth
profile of the different rock types (lithology) in a well and then use multiple wells to determine
how the different impedance profiles vary regionally.
In order to do this, the lithologies must first be identified. The aim is to break the rocks into broad
general classifications, not to exactly describe the lithology at each depth. This will allow each
lithology to have enough data points to create a usable impedance curve. To this end a geologist
was asked to interpret the well log date for a test well, determine what well logs were most
important and derive a set of rules to determine the rock type. Because some rock properties vary
with depth due to compaction the rules were requested to be as broadly applicable throughout
the well as possible.
The rules derived from the first well were then developed into a computer program and applied to
the initial well to test accuracy and then to subsequent wells to explore robustness in the
technique.
4 | Internal Report
2.2 Study area location
Figure 1: Location of the 4 exploration wells found used in the survey; Iago 1, Martell 1, Lauda 1 and Achilles 1
Four wells were used in this study; Lauda 1, Iago 1, Achilles 1 and Martell 1. They were chosen as
they are relatively recent wells, with the oldest being Iago 1 drilled in December 2000, and thus
have well logs taken with quite modern tools. They also all have very comprehensive well logs that
are publically available. All wells reside within the Carnarvon Basin, North West Shelf, Western
Australia and are within 160 km of each other as shown in Figure 1. It was decided to use just one
basin at this initial stage as the wells are likely to have similar rock types and properties, whilst
showing subtle variations between them.
2.3 Expert classification rules
Lauda 1 was chosen as the test well to determine a set of rules. It was chosen due to its
comprehensive suite of 42 logs, long interval of logging, and decent quality readings. Initially a well
log expert analysed the logs and determined broad interpretation of the rocks, whilst capturing
why he chose those classifications. This then allowed him to derive a general set of initial rules for
determining five criteria for rock types; sand, shale, intermediate grain, radiolarite, and marl. Later
after reviewing the results against actual well logs, a further four lithologies were added to
include; limestone, coal, pyritic material, and sand containing gas (gas sand). Also, it was decided
Internal Report | 5
to break the intermediate grain sized material into coarse-medium and fine-medium as the
previous category was thought to be too broad.
Importantly it was possible to derive criteria for the different lithologies without needing to utilise
hard to define subjective terms such as “conforming” and “wavy” for example. Terms such as
these are often observable to a trained eye, but can be close to impossible to determine
mathematically from a noisy well log reading.
2.4 Well data
The amount of variables measured collected from a different wireline tool in a well can vary quite
significantly from more than fifty to just a few basic well logs. In designing our algorithms, we
chose to use those that are most commonly recorded across most wells, so as to give our solution
the most future flexibility. To this end we utilise, Photoe lectron factor (PEF), Gamma Ray(GR),
neutron porosity(NPhi), grain density (RhoB), slowness (DTCO) (from which you can calculate the
compression wave velocity (Vp)) and shear wave velocity(Vs) . The grain density and PEF are
utilised to calculate the density porosity (Dphi), which in turn is used to calculate the difference
between the two porosities (ND).
2.5 Rules for log classification
The classifications were determined as follows:
Limestone: PEF > 4.0 be, GR < 45 APS, DTCO < 80us/ft.
Coal: RhoB <1.7, DTCO >115 us/ft, NPhi > 0.6.
Pyritic Material: ND > 0.2, RhoB > 2.71, PEF > 3.2
Gas Sand: ND < -0.1
The remaining material after these distinctive lithologies were identified was broken up as:
Shale: ND > 0.15.
Radiolarite: 0.15 > ND > 0.05, RhoB < 2.4, NPhi > 0.25.
Fine medium: 0.15 > ND > 0.10.
Coarse medium: 0.10 > ND > 0.05.
Sand: ND < 0.05.
The classification of Marl was more problematic. Marl is a calcium carbonate or lime rich
mudstone, so it’s log characteristics are very similar to the medium and fine -grained material. The
addition of the calcite in Marls tends to make it harder and subsequently faster to transmit sound
than its comparable mudstone. However as the velocity in the rocks is also compaction and hence
depth related, separating the cemented Marl from its un-cemented Mudstone is difficult.
6 | Internal Report
Therefore, we divide the shales and medium grained material into a fast and slow grouping by
determining a depth related cut of velocity. Assigning the fast rocks to the category ‘Marl’.
To determine the cut off velocity, we calculate use an exponential fit of the compression wave
velocity (Vp) versus depth profile for all of the Marls, and another Vp versus depth profile for all
the sands. We then average these two profiles. This becomes the cut-off velocity between
cemented and un-cemented Marls. Sands typically have much high sound velocities than shales
under the same compression.
This is a somewhat crude cut off, as most siliciclastic sedimentary rocks have a degree of
cementation and therefore “marliness”, however a general classification was required to separate
the categories.
The above algorithm was coded as a Fortran90 program which runs on a basic laptop. The
software “ShaleType”, takes less than a second to classify a typical well log.
2.6 Velocity and impedance profiles
The shear wave velocity versus depth was calculated for each facies type using an exponential
least squares fit. http://mathworld.wolfram.com/LeastSquaresFittingExponential.html .
In general, this produced reasonable curves, however, when the rock type was poorly represented
or was clustered in just one part of the well, its ability to extrapolate was poor as will be seen in
the results section. This is also true of linear and quadratic extrapolations, meaning that an
extrapolation technique should be used cautiously outside of the depths where the rocks are
actually found.
Finally, an artificial velocity curve was created for the entire well segment analysed. This was done
by assigning the velocity profile for the interpreted rock type at that depth, reproducing quite well
the measured velocities of the well log.
Internal Report | 7
3 Results
Four wells were analysed with the software, to explore its robustness. All the wells are from the
Canarvon Basin, North West Shelf, Australia. The wells are Lauda-1, Archillies-1, Iago-1 and
Martell-1. All had reasonable log coverage which was freely available. For the sake of brevity, only
the complete results from Lauda 1 will be shown, with only interesting features of the other wells
being presented
3.1 Lauda 1
Location: Lat/Long: -20.818645/114.722635.
Figure 2. The impedance versus depth for rocks classified as shale in Lauda 1
All the impedance curves for Lauda 1 show trend lines that would be ex pected for most wells, with
the impedance for each rock type increasing with depth in a reasonable curve (Figure 2 to Figure
9). The results for the sand profile shown in Figure 6 are somewhat fortunate, with only one small
packet of sand occurring high in the well providing enough influence to keep the curve reasonable.
It should be noted that even with the rocks grouped into their respective lithologies, there is still a
relatively large range of data values at each depth for each classification, and hence quite large R2
values. The large amount of noise is to be expected due to the natural variance in the rock as we
attempt to classify them into only a few categories.
y = 2.492E+03e3.641E-04x
R² = 9.396E-01
4000
5000
6000
7000
8000
9000
1800 2000 2200 2400 2600 2800 3000 3200
Imp
eda
nce
Depth
Lauda 1 Shale
8 | Internal Report
Figure 3. The impedance versus depth for rocks classified as marls in Lauda 1
Figure 4. The impedance versus depth for rocks classified as medium to fine grain siliciclastic in Lauda 1
y = 2.668E+03e3.816E-04x
R² = 7.212E-01
4000
5000
6000
7000
8000
9000
10000
11000
12000
13000
1800 2000 2200 2400 2600 2800 3000 3200 3400
Imp
eda
nce
Depth
Lauda 1 Marls
y = 2.313E+03e3.973E-04x
R² = 9.517E-01
4000
4500
5000
5500
6000
6500
7000
7500
8000
1800 2000 2200 2400 2600 2800 3000
Imp
eda
nce
Depth
Lauda 1 Medium - Fine Grain
Internal Report | 9
Figure 5. The impedance versus depth for rocks classified as medium to coarse grain siliciclastic in Lauda 1
Figure 6. The impedance versus depth for rocks classified as sands in Lauda 1
Figure 7 shows the only gas sands for the entire well log. In this case the resulting curve is
unusable as there is simply not enough spread of data to perform a meaningful extrapolation.
Likewise, for limestone shown in Figure 10, we see that that it only occurs low down in the well, so
extrapolating the date much beyond the depths covered could be misleading.
y = 2.445E+03e3.697E-04x
R² = 8.851E-01
4000
4500
5000
5500
6000
6500
1800 1900 2000 2100 2200 2300 2400 2500 2600 2700
Imp
eda
nce
Depth
Lauda 1 Medium-Coarse Grain
y = 4.283E+03e2.725E-04x
R² = 2.354E-01
4000
5000
6000
7000
8000
9000
10000
11000
12000
1800 2000 2200 2400 2600 2800 3000 3200 3400
Imp
eda
nce
Depth
Lauda 1 Sands
10 | Internal Report
Figure 7.The impedance versus depth for rocks classified as gas filled sands in Lauda 1
Figure 8. The impedance versus depth for rocks classified as radiolarite in Lauda 1
y = 4.256E+102e-6.982E-02x
R² = 8.332E-01
1800 2000 2200 2400 2600 2800 3000 3200 3400
Imp
eda
nce
Depth
Lauda 1 Gas Sands
y = 3.608E+03e2.909E-04x
R² = 6.372E-01
4000
5000
6000
7000
8000
9000
10000
1800 2000 2200 2400 2600 2800 3000 3200 3400
Imp
eda
nce
Depth
Lauda 1 Radiolarite
Internal Report | 11
Figure 9. The impedance versus depth for rocks classified as pyritic material in Lauda 1
Figure 10. The impedance versus depth for rocks classified as limestone in Lauda 1
Figure 11 shows 3 different curves for the shear wave velocity. The grey curve is that measured by
the logging tool. The orange curve is created by using the calculated velocity profile for the rock
type identified at that depth. It does a reasonable effort at matching the actual curve given the
high range of variability in the data, indicating that our results are reasonable. Figure 12 is
identical to Figure 11, however it shows just the lower section of the well to highlight detail.
y = 2.780E+03e4.188E-04x
R² = 6.855E-01
4000
5000
6000
7000
8000
9000
10000
11000
12000
1800 2000 2200 2400 2600 2800 3000 3200
Imp
eda
nce
Depth
Lauda 1 Pyritic Material
y = 4.765E+03e2.389E-04x
R² = 6.421E-01
4000
5000
6000
7000
8000
9000
10000
11000
1800 2000 2200 2400 2600 2800 3000 3200 3400
Imp
eda
nce
Depth
Lauda 1 Limestone
12 | Internal Report
Figure 11. The velocity profile for all material from Lauda 1. the grey data points are the actual velocities from the
well log. The orange data points are the values from the various calculated velocity curves for the rock type at that
depth.
Figure 12: The velocity profile for all material from Lauda 1. the grey data points are the actual velocities from the well log. The orange data points are the values from the various calculated velocity curves for the rock type at that
depth.
3.2 Martell 1
Location: Lat/Long: -19.395022/114.310692
2000
2500
3000
3500
4000
4500
5000
1800 2000 2200 2400 2600 2800 3000 3200 3400
Vel
oci
ty
Depth
Lauda 1 Velocity Measured and Calculated
2000
2500
3000
3500
4000
4500
5000
2800 2850 2900 2950 3000 3050 3100 3150 3200 3250
Vel
oci
ty
Depth
Lauda 1 Velocity Measured and Calculated
Internal Report | 13
Figure 13. The impedance versus depth for rocks classified as shale in Martell 1
Figure 14. The impedance versus depth for rocks classified as sand in Martell 1
The calculated output for Martell 1 appears to be quite successful, although caution should be
exercised in interpreting the marls, as the sand profile (Figure 14) is used to distinguish between
marls and shales. As can be seen, there are few sands and they only occur low down in the well .
Coal (Figure 15) was identified (having been absent from Lauda 1) near the base of the well,
although only a few layers were observed leading to a poor approximated curve.
The calculated velocity profile again does a reasonable job of tracking the measured data, and
picks out sharp changes in lithology (which would show up as seismic reflections).
700
1200
1700
2200
2700
7000 7500 8000 8500 9000 9500 10000 10500 11000
Imp
eda
nce
Depth
Martell 1 Shales
700
1200
1700
2200
2700
7000 7500 8000 8500 9000 9500 10000
Imp
eda
nce
Depth
Martell 1 Sands
14 | Internal Report
Figure 15. The impedance versus depth for rocks classified as coal in Martell 1
Figure 16. The velocity profile for all material from Martell 1. The grey data points are the actual velocities from the well log. The orange data points are the values from the various calculated velocity curves for the rock type at that
depth.
3.3 Iago 1
Location: Lat/Long: -19.930484/115.33546
The results for Iago 1 were less successful when identifying shales and marls. This highlights the
risk in using the sand profile when splitting the two categories. In this case the impedance curve
for the sand profile (Figure 18) is actually decreasing with depth due to sparse data.This in turn
700
1200
1700
2200
2700
7000 7500 8000 8500 9000 9500 10000 10500 11000
Imp
eda
nce
Martell 1 Coals
550
750
950
1150
1350
1550
1750
1950
7500 8000 8500 9000 9500 10000 10500 11000
Vel
oci
ty
Depth
Martell 1 velocity v depth. Measured and calculated
Internal Report | 15
lead to a possible over classification of marls (Figure 19) in the top half of the well and an under
identification of shales (Figure 17). These in turn resulted in a poor fit of the shear wave velocities
in the top of the well
Figure 17. The impedance versus depth for rocks classified as shale in Iago 1
Figure 18. The impedance versus depth for rocks classified as sand in Iago 1
y = 2.437E+03e3.235E-04x
R² = 2.920E-01
4000
4500
5000
5500
6000
6500
7000
7500
8000
8500
2000 2200 2400 2600 2800 3000 3200 3400
Imp
eda
nce
Depth
Iago 1 Shales
y = 1.311E+04e-1.196E-04x
R² = 4.255E-01
4000
5000
6000
7000
8000
9000
10000
11000
12000
13000
2000 2200 2400 2600 2800 3000 3200 3400 3600
Imp
eda
nce
Depth
Iago 1 Sands
16 | Internal Report
Figure 19. The impedance versus depth for rocks classified as marl in Iago 1
Figure 20. The velocity profile for all material from Iago 1. the grey data points are the actual velocities from the
well log. The orange data points are the values from the various calculated velocity curves for the rock type at that
depth.
3.4 Achilles 1
Location: Lat/Long: -20.215221/114.537134
Achilles 1 like Lauda 1 produced excellent curves for nearly all lithologies due to a good spread of
data throughout the log. The only anomaly was limestone, with the resulting impedance curve
decreasing marginally with depth. The resulting artificial velocity profiles seem to do a good job of
representing the actual shear wave velocities.
y = 4.554E+03e1.983E-04x
R² = 1.864E-01
4000
6000
8000
10000
12000
14000
16000
2000 2200 2400 2600 2800 3000 3200 3400 3600
Imp
eda
nce
Depth
Iago 1 Marls
2000
2500
3000
3500
4000
4500
5000
5500
2000 2200 2400 2600 2800 3000 3200 3400 3600
Vel
oci
ty
Depth
Iago 1 velocity v depth. Measured and calculated
Internal Report | 17
Figure 21. The velocity profile for all material from Achilles 1. the grey data points are the actual velocities from the well log. The orange data points are the values from the various calculated velocity curves for the rock type at that
depth.
2500
3000
3500
4000
4500
5000
5500
6000
10500 11000 11500 12000 12500 13000 13500 14000 14500 15000
Vel
oci
ty
Depth
Achilles 1 velocity v depth. Measured and calculated
18 | Internal Report
4 Discussion
Automated well log interpretation is by no means a new idea. Most commercial well log software
provides some form of automated log analysis from small start-up companies to massive service
companies such as Schlumburger. A list of the software will not be listed here, however a quick
search of “well log interpretation” will reveal many different products, all attempting to do the
same thing. Nearly all of them use much more advanced algorithms than that presented here, and
provide a much more refined interpretation.
We are deliberately attempting to keep our classifications extremely broad, and the number of
logs we utilise as small as possible. Our end target is to provide general impedance profiles for
different lithologies across multiple wells in a field extremely quickly. The end target is being able
to produce an accurate seismic inversion. Our current model shows promise, however the meth od
whilst working reasonably well for 3 wells, did not work for all 4. Sparseness of data for different
lithologies (sands in particular) causes a great deal of uncertainty. The accurate determination of
the degree of cementation in the siliciclastic rocks is the major problem. The use of clustering
techniques is being explored to break these rocks into meaningful groups, with very encouraging
initial work being realised when applied to the full range of lithologies (Emelyanova et al 2017).
Statistical techniques such as BFAST (Verbesselt et al 2010, 2010, 2013) were trialled. These
algorithms which have been designed to determine trends in data sets, however they could not
deal with the complexity and noise of well logs, nor the rapidly varying data. They also become
prohibitively time consuming when run at a resolution required to capture the detail required.
Internal Report | 19
5 Conclusion
Whilst the software needs further improvement, the initial results are quite encouraging. The
largest problem is in interpreting the distinction between the marls and the shales/medium grain
sediments. Utilising the sand profile as a constraint only works when we have sufficient sand along
the length of the logged segment to construct a decent profile. Several other methods have been
attempted, (for example simply splitting the shales along their mean velocity depth curve), but all
have had their problems similar to the current approach.
The next stage will be to use more advanced methods such as ensemble clustering to separate
these sediments into groups as in reality the degree of calcification in a rock varies greatly and it
may be wrong to simply assign these elements just as shale or marl.
20 | Internal Report
6 References
Emelyanova, Irina & Pervukhina, M & Clennell, Michael & Dyt, Chris. (2017). Unsupervised
identification of electrofacies employing machine learning. 79th EAGE Conference and Exhibition
2017 10.3997/2214-4609.201701655.
Verbesselt, J., Hyndman, R., Newnham, G., & Culvenor, D. (2010). Detecting trend and seasonal
changes in satellite image time series. Remote Sensing of Environment, 114, 106-115. DOI:
10.1016/j.rse.2009.08.014.
Verbesselt, J., Hyndman, R., Zeileis, A., & Culvenor, D. (2010). Phenological change detection while
accounting for abrupt and gradual trends in satellite image time series. Remote Sensing of
Environment, 114, 2970-2980. DOI: 10.1016/j.rse.2010.08.003.
Verbesselt, J., Zeileis, A., & Herold, M. (2013). Near real -time disturbance detection using satellite
image time series, Remote Sensing of Environment. DOI: 10.1016/j.rse .2012.02.022.
22 | Internal Report
CONTACT US
t 1300 363 400
+61 3 9545 2176 e [email protected]
w www.csiro.au
AT CSIRO, WE DO THE EXTRAORDINARY EVERY DAY
We innovate for tomorrow and help improve today – for our customers, all
Australians and the world.
Our innovations contribute billions of
dollars to the Australian economy
every year. As the largest patent holder
in the nation, our vast wealth of intellectual property has led to more
than 150 spin-off companies.
With more than 5,000 experts and a burning desire to get things done, we are
Australia’s catalyst for innovation.
CSIRO. WE IMAGINE. WE COLLABORATE. WE INNOVATE.
FOR FURTHER INFORMATION
Energy
Chris Dyt t +61 8 6436 8785
w www.csiro.au/en/Research/EF