Leveraging open-access remote sensing imagery to monitor dam infrastructure: Case study of the Cadia tailings dam collapse, Australia

Abstract

Satellite remote sensing data can be used to monitor environmental processes and inform disaster risk reduction and hazard early warning. This paper describes the analysis of satellite remote sensing images to investigate the partial wall collapse of a tailings dam at the Cadia gold-copper mine in Australia that occurred on 9th March 2018. Our case study uses freely available remote sensing imagery acquired by the Copernicus Sentinel-1 (radar) and Sentinel-2 (multispectral) satellite constellations to monitor land surface changes in the Cadia mine area before and after the collapse. In this paper we discuss the benefits of utilising both radar and multispectral remote sensing imagery in a holistic approach to remote sensing, which could be used for continuous, near-real time monitoring of risk-related infrastructure such as dams without the need for in-situ measurement equipment.

We applied the Interferometric Synthetic Aperture Radar (InSAR) technique to measure surface displacements and interferometric coherence maps from a stack of Sentinel-1 radar images acquired between 2nd December 2015 and 25th June 2018 at regular 12 day intervals. The time series of surface displacements show a significant increase in the rate of movement of the dam wall in the area that eventually breached in the two months prior to the collapse. This change in movement behaviour was not observed at parts of the dam wall that remained intact. This analysis demonstrates the potential for InSAR monitoring to identify issues in advance of infrastructure failure, which could allow risk mitigation strategies to be implemented by the mine operator. We used interferometric coherence data to observe changes in the dam wall and surrounding areas before and after the collapse. A drop in coherence occurred in the breached section of dam wall, indicating the surface change caused by the collapse. Coherence for unaffected parts of the dam wall remained stable. Sentinel-2 multispectral imagery acquired between 2nd July 2017 and 24th June 2018 show the timing, extent and effects of the collapse as well as the rate of tailings movement.

Full text

1
Leveraging open-access remote sensing imagery to monitor dam
infrastructure: Case study of the Cadia tailings dam collapse, Australia
Sean Minhui Tashi Chua1, Thomas Fuhrmann1 , Matthew Garthwaite1
1Geoscience Australia
Satellite remote sensing data can be used to monitor environmental processes and inform disaster risk reduction and
hazard early warning. This paper describes the analysis of satellite remote sensing images to investigate the partial wall
collapse of a tailings dam at the Cadia gold-copper mine in Australia that occurred on 9th March 2018. Our case study
uses freely available remote sensing imagery acquired by the Copernicus Sentinel-1 (radar) and Sentinel-2 (multispectral)
satellite constellations to monitor land surface changes in the Cadia mine area before and after the collapse. In this paper
we discuss the benefits of utilising both radar and multispectral remote sensing imagery in a holistic approach to remote
sensing, which could be used for continuous, near-real time monitoring of risk-related infrastructure such as dams without
the need for in-situ measurement equipment.
We applied the Interferometric Synthetic Aperture Radar (InSAR) technique to measure surface displacements and
interferometric coherence maps from a stack of Sentinel-1 radar images acquired between 2nd December 2015 and 25th
June 2018 at regular 12 day intervals. The time series of surface displacements show a significant increase in the rate of
movement of the dam wall in the area that eventually breached in the two months prior to the collapse. This change in
movement behaviour was not observed at parts of the dam wall that remained intact. This analysis demonstrates the
potential for InSAR monitoring to identify issues in advance of infrastructure failure, which could allow risk mitigation
strategies to be implemented by the mine operator. We used interferometric coherence data to observe changes in the dam
wall and surrounding areas before and after the collapse. A drop in coherence occurred in the breached section of dam
wall, indicating the surface change caused by the collapse. Coherence for unaffected parts of the dam wall remained stable.
Sentinel-2 multispectral imagery acquired between 2nd July 2017 and 24th June 2018 show the timing, extent and effects
of the collapse as well as the rate of tailings movement.
Introduction
Tailings dams are commonly used for waste storage in the mining industry. These structures have a history of failing with
significant economic, health and environmental consequences (Azam and Li, 2010). Analysis of historical tailings dam
failures has found that the primary reasons for failure include weather, heavy rainfall, poor management, substandard
construction procedures and a failure to implement monitoring programs over the long-term (Azam and Li, 2010, Rico et
al 2008). Deformation is also a key driver behind dam failures, this process can lead to liquefaction in which the tailings
lose their structure and strength due to changes in loading (Davies et al, 2002). Traditional methods of monitoring
deformation use tools such as survey reference points combined with GPS monitoring or laser scanning (Sun et al, 2012).
These equipment have to be manually installed at each site and can be costly and inconvenient to maintain, particularly
after the closure of a facility. This study used remote sensing data and techniques to investigate the effectiveness of InSAR
and multi-spectral remote sensing in the context of analysing the failure of a tailings dam wall at the Cadia mine.
Study site
Cadia Valley Operations is a gold and copper mining facility located approximately 20 kilometres south of Orange, New
South Wales, Australia. The mine was established in 1998 and contains three mining facilities and two tailings storage
facilities. The southern wall of the northern tailings dam is up to approximately 94 meters high and 6 kilometres in length,
it separates the northern and southern tailings dams. Geodetic survey prisms were located at intervals on the downstream
face of the dam wall to monitor deformation. On 9th March 2018, a 300 metre-wide segment of the southern wall collapsed.
Tailings material flowed into the southern storage facility where it was contained. However, the impact of this collapse
event was significant because operations at the mine were ceased for a month and the tailings dam remains closed (Jefferies
et al, 2019).
Materials and Methods
Data
Sentinel-1 is a mission operated by the European Space Agency (ESA) and part of the broader Copernicus initiative that
contains other Earth Observation missions (ESA, 2019a). Sentinel-2 is a partner mission to Sentinel-1 also within the
Copernicus initiative (ESA, 2019b). The Sentinel-1 mission is made up of two identical satellites that image the Earth
using synthetic aperture radar (SAR) sensors. SAR data is collected using an active source of C-band microwaves (5.6
centimetre wavelength) at currently 12-day intervals in Australia. Here we used images acquired on descending orbital

passes, approximately North-South, every 12 days over our study area. We downloaded level-1 Single Look Complex
(SLC) image products from ESA that contain the magnitude and phase measurements of the energy backscattered from the
Earth’s surface. The magnitude gives a measure of the strength of the backscattered radar signal. The phase gives a measure
of the range (distance) between the backscattering target on the ground and the satellite sensor. Spaceborne SAR systems
use a side-looking image geometry, which enables a 2D image of the Earth’s surface to be derived in the coordinate system
of the platform. We conducted further processing to obtain georeferenced SAR images that can be used for the analysis we
describe here.
The Sentinel-2 mission is made up of two identical satellites that collect optical imaging data across 13 spectral bands in
the optical, near-infrared and shortwave-infrared range with a revisit time of approximately 5 days. This study used surface
reflectance Sentinel-2 data processed by Digital Earth Australia (Dhu et al, 2017) and available within the Open Data Cube
(ODC) architecture. The ODC is a global initiative that provides a platform for satellite data management to enable large-
scale processing and high-performance computing (Lewis et al, 2019).
Methodology
InSAR is a processing technique that makes use of two or more SAR images acquired at different times to derive relative
surface displacements from changes in the measured phase signals. When a stack of SAR data is available, images of phase
change are calculated (so-called “interferograms”) resulting in a displacement time series for each pixel in the aligned
images. The phase signal may be noisy at some pixels if the backscattering characteristics within the pixel changes through
time. This is particularly the case for vegetated areas and rural environments. Ferretti et al. (2000, 2001) introduced the
concept of “persistent scatterers” which only makes use of those pixels with consistent backscattering through time. At
these pixels, the phase signal is analysed and other nuisance terms contributing to the signal, such as atmospheric or orbital
effects, are separated from the phase related to surface displacement (e.g. Hooper et al., 2004, Hooper et al. 2007; Adam
et al., 2003, Kampes, 2005). The remaining phase signal is subsequently transformed to a metric displacement using the
wavelength of the radar sensor (5.6 cm for Sentinel-1). The output of this process is a cumulative displacement time series
from December 2nd 2015 until the collapse on March 9th after which the surface has insufficient coherence. The cumulative
displacement is for the satellite sensors line of sight for a collection of persistent scatterer pixels over the Cadia mine study
site (Figure 1). A number of pixels classified as persistent scatterers were selected within and outside of the collapsed
section of the dam wall and the time-series of displacement was analysed for trends (Figure 2). It should be noted that our
approach did not have pixel coverage at the very top of the dam wall from February, 2017 onwards. This is due to
modifications to the wall resulting in changes to the radar reflection properties and subsequently reducing the radar
coherence. Valid InSAR pixels were only selected in areas with high confidence throughout the full analysed period.
The InSAR processing described above also produces an interferometric coherence product, which can be understood as a
measure of correlation between the phase signals in the two SAR images (Ferretti et al, 2007). The interferometric
coherence gives a normalised quantification of the change in the scattering properties within a pixel between the two SAR
images. Coherence values range from 0 to 1, with 0 indicating that the interferometric phase is completely decorrelated
(i.e. noise) and 1 a complete absence of phase noise. Lower coherence values typically occur over land surfaces that change
or move between images, such as water or vegetation. Higher coherence values occur over less dynamic surfaces such as
urban infrastructure, soil, sparsely vegetated or exposed rock. Thus, coherence can be a useful tool for mapping change of
the land surface. To improve ease of use and compatibility with Sentinel-2 data, the interferometric coherence data for 12-
day SAR image pairs derived from the InSAR processing was arranged into a format suitable for time-series analysis for
December 2nd 2015 until June 24th 2018. Our process reads in the coherence images, crops them to a desired study area and
stacks them into a time-series. Following this, time-series analysis was conducted on several pixels situated along the dam
wall, both within the breach and outside of the breach area to identify trends in the coherence surrounding the dam wall
failure (Figure 3). Histograms of coherence were also produced from the time-series of coherence. True colour image
sequences of the breach event were produced using Sentinel-2 data.
The tools used to undertake the persistent scatterer displacement analysis
1
and interferometric coherence analysis
2
are
freely available.
1
https://github.com/dbekaert/StaMPS/
2
https://github.com/s-m-t-c/insarex/

3
Results
Figure 1 LOS displacement rates at Cadia Mine from time series analysis of Sentinel-1 phase measurements. Red indicates movement
towards the sensor and blue away from the sensor. Green indicates no movement. Background: overview map of mining operations as
published on http://www.cadiavalley.com.au/site/about.

Figure 2 Cumulative line of sight displacements at Cadia southern tailings dam from InSAR persistent scatterer time series analysis of
Sentinel-1 phase measurements (a). Background on (a): Google Earth imagery acquired 2018-5-10. Displacement time series are shown
for four pixels located in the breach area and two pixels outside the breach area during the whole analysed period (b) and the six months
before the breach (c). The date of the breach, 9th March 2018, is marked by a black dashed line.
a
b
(c)

5
Figure 3 (a) Time series of Sentinel-1 interferometric coherence for two pixels within the dam breach and two control pixels. The time
period captured by this analysis is from July 2017 to July 2018. The date of each observation is attributed to the later of the two SAR
images used to form the interferometric coherence. The dotted line indicates the time of dam wall collapse (2018-03-08 UTC). The
location of pixels in the time-series are shown with matching colours in the Sentinel-2 multispectral images (b) & (c).
b
c

Figure 4 Interferometric coherence images at before (a) and after (b) the dam wall failure.
b
a

7
Figure 5 Sequence of Sentinel-2 true colour images covering the time of the dam breach at Cadia mine (note: labelled times are UTC).
InSAR
Figure 1 is a map overlayed with the persistent scatterer pixels from the InSAR analysis over the Cadia site where the
colour of each point represents displacement. Deformation is visible in the open-cut mine area and along the dam wall.
Figure 2 shows a view of the persistent scatterer pixels at Cadia dam and the displacement at selected pixels from 2nd
December 2017 until 25th February 2018. These results show that the dam structure has a constant rate of displacement for
all the selected pixels until approximately late December 2017. Negative displacements indicate movement away from the
satellite in the line of sight, which could be interpreted as a deformation of the dam structure. Those pixels near the top of
the dam structure have a greater rate of displacement than those pixels in the middle of the dam structure. These
observations are in line with the expected settlement of an earthen dam structure as the dam is raised over time (Gikas and
Sakellariou, 2008, Jefferies et al, 2019). After December 2017, pixels within the breach area experience a marked increase
in the rate of displacement whilst those outside of the breach area continue at the same rate. Pixel 1 at the centre of the
breach, on top of the dam wall shows a cumulative displacement of 8 cm from 8th January 2018 until 25th February 2018,
after which the dam wall collapsed on 9th March 2019. Previously, 5 cm of displacement was accumulated at that pixel
over a two year period since December 2015. An increase in the displacement rate is also seen at pixels 2, 3 and 4 which
are within or on the border of the breach. The official report on the collapse incident (Jefferies et al, 2019) notes that the
primary factors that caused the collapse are construction and excavation works along with localised weak geology
susceptible to compression. We note that the increase in the rate of displacement occurs shortly after the excavation process
is stated to have commenced (Jefferies et al, 2019). The report also notes that in-situ monitoring equipment showed similar
trends in displacement to those found in InSAR analysis (Jefferies et al, 2019). These results are strong evidence for the
value of InSAR for monitoring the deformation of dam walls and for assessing the potential for failure.

The analysis of coherence data is shown in Figure 3 and Figure 4. Of the four pixels selected for analysis in Figure 3 two
are located within the breach (pink) and two are control points outside of the breach (green). At the pixels within the breach
a slight decrease leading up to the failure is visible (Figure 3a) followed by a stronger drop in coherence whilst the pixels
outside of the breach maintain a stable coherence through the breach period. The increase in coherence of the breach 2
pixel in Figure 3a is likely due to stabilisation of the land surface following the failure. The variation in coherence visible
in the control pixels in September and October 2017 could be related to construction work that was visible in the Sentinel-
2 data. Figure 4a shows a coherence image for before the breach and Figure 4b after, the areas of high coherence are shown
in white. The loss of coherence in the breached area is visible in Figure 4b due to the change in surface characteristics.
Multi-spectral
The true colour images extracted from Sentinel-2 data before and after the breach are shown in Figure 5. The information
from Sentinel-2 helps us understand the spatial extent and spectral characteristics of the study area before and after the
failure. The collapse first becomes visible in an image acquired on the 11th of March 2018 (Figure 5c). The breached area
appears to undergo a second phase of collapse where the tailings material extends out over a larger area; this state is
captured in the image of 14th March 2018 (Figure 5d). This indicates that the total settlement of the tailings after the initial
collapse took from 24 to 72 hours. A blue pattern is visible in the Northern Tailings storage facility after the collapse. This
is due to a crusting agent applied by plane to the tailings surface (Jefferies et al, 2019). This was done because the collapse
led to drying out of the tailings material, creating problems with wind-blown dust. The crusting agent was applied to
suppress the dust created by the drying effects of the collapse. The findings from Figure 5 demonstrate the utility of
Sentinel-2 multispectral imagery for collecting information about dam failures and the surrounding environmental factors
such as the drying event and water extent.
Conclusion
Our findings demonstrate that the radar and multispectral images provided through the Copernicus Earth Observation
missions are useful for the monitoring and analysis of tailings dam infrastructure and that these data can play a key role in
the future of resilient and safe dams. In particular, the InSAR technique performed on Sentinel-1 data was able to measure
a precursory change in the rate of displacement of the Cadia tailings dam infrastructure ahead of the 9th March 2018
collapse. This demonstrates the possible use of the InSAR technique with openly available Sentinel-1 SAR data as a remote
monitoring tool for risk-related infrastructure. This tool could be instrumental in determining the possibility of
infrastructure failure with enough warning for actions to be implemented by the mine operators. For post failure analysis,
changes in the surface can also be measured through Sentinel-2 multispectral imagery and analysis of interferometric
coherence. Our results showed that the difference between the coherence in the control and breach pixels highlights the
potential for coherence to be useful for analysing changes in the land surface at tailings dams. Combination of these radar
and optical data products enable a powerful method for analysing the location, timing and extent of tailings dam failures.
References
Adam N., Kampes BM., Eineder M., Worawattanamateekul J., Kircher M., 2003. The Development of a Scientific
Permanent Scatterer System, in: Schroeder M, Jacobsen K, Heipke C, editors, Proceedings of the Joint ISPRS/EARSeL
Workshop "High Resolution Mapping from Space 2003", Hannover, Germany, 6–8 Oct 2003, pp.1–6
Azam, S. and Li, Q., 2010. Tailings dam failures: a review of the last one hundred years. Geotechnical news, 28(4), pp.50-
54.
Davies, M., McRoberts, E., and Martin, T., 2002. Static liquefaction of tailings – fundamentals and case histories.
Dhu, T., Dunn, B., Lewis, B., Lymburner, L., Mueller, N., Telfer, E., Lewis, A., McIntyre, A., Minchin, S. and Phillips,
C., 2017. Digital earth Australia–unlocking new value from earth observation data. Big Earth Data, 1(1-2), pp.64-74.
ESA 2019a, Sentinel-1 SAR User Guide, ESA, viewed 11 June 2019, <https://sentinel.esa.int/web/sentinel/user-
guides/sentinel-1-sar>.
ESA 2019b, Sentinel-2 MSI User Guide, ESA, viewed 11 June 2019, <https://sentinel.esa.int/web/sentinel/user-
guides/sentinel-2-msi/>.
Ferretti, A., Monti-Guarnieri, A., Prati, C., Rocca, F., and Massonnet, D., 2007. InSAR Principles: Guidelines for SAR
Interferometry Processing and Interpretation, Part B: InSAR processing: a practical approach. ESA, ESA-TM-19.

9
Ferretti A., Prati C., Rocca F., 2000, Nonlinear subsidence rate estimation using permanent scatterers in differential SAR
interferometry, IEEE Transactions on Geoscience and Remote Sensing, 38(5), pp.2202–2212
Ferretti A., Prati C., Rocca F., 2001, Permanent scatterers in SAR interferometry, IEEE Transactions on Geoscience and
Remote Sensing, 39(1), pp.8–20
Gikas, V. and Sakellariou, M., 2008. Settlement analysis of the Mornos earth dam (Greece): Evidence from numerical
modeling and geodetic monitoring. Engineering Structures, 30(11), pp.3074-3081.
Hooper A., Zebker HA., Segall P., Kampes, BM., 2004, A new method for measuring deformation on volcanoes and other
natural terrains using InSAR persistent scatterers, Geophysical Research Letters, 31(23), pp.1–5
Hooper A., Segall P., Zebker HA ., 2007, Persistent scatterer interferometric synthetic aperture radar for crustal deformation
analysis, with application to Volcán Alcedo, Galápagos, Journal of Geophysical Research: Solid Earth, 112(B7), pp.1–21
Kampes BM (2005) Displacement parameter estimation using permanent scatterer interferometry. Ph.D. Thesis, Delft
University of Technology, Delft, The Netherlands.
Lewis, A., Oliver, S., Lymburner, L., Evans, B., Wyborn, L., Mueller, N., Raevksi, G., Hooke, J., Woodcock, R., Sixsmith,
J. and Wu, W., 2017. The Australian geoscience data cube—Foundations and lessons learned. Remote Sensing of
Environment, 202, pp.276-292.
Jefferies, M., Morgenstern, N., Van Zyl, D.,Waters J., 2019. Report on NTSF Embankment Failure, Cadia Valley
Operations, Orange.
<http://www.newcrest.com.au/media/market_releases/2019/Report_on_NTSF_Embankment_Slump.pdf >
Rico, M., Benito, G., Salgueiro, A.R., Díez-Herrero, A. and Pereira, H.G., 2008. Reported tailings dam failures: a review
of the European incidents in the worldwide context. Journal of hazardous materials, 152(2), pp.846-852.
Sun, E., Zhang, X. and Li, Z., 2012. The internet of things (IOT) and cloud computing (CC) based tailings dam monitoring
and pre-alarm system in mines. Safety science, 50(4), pp.811-815.