Академический Документы
Профессиональный Документы
Культура Документы
net/publication/339292133
CITATIONS READS
0 377
5 authors, including:
Some of the authors of this publication are also working on these related projects:
Echohydrology, Soil Moisture Dynamics and Soil Water Repellency View project
All content following this page was uploaded by Sina Saneiyan on 15 February 2020.
3D Geoelectrical Problems data from both electrical resistivity tomography (ERT) and
induced polarization (IP) surveys, and has recently been
With ResiPy, an Open Source updated with 3D inversion capability. The software is based
around the mature inversion codes R2, cR2, R3t, and cR3t. In
Graphical User Interface this paper we focus only on the 3D aspect. Processing of 3D
for Geoelectrical Data ERT data with ResIPy is demonstrated and two case studies
are showcased. The first case study is a 3D time lapse study of
Processing an active landslide with complex geology and topography, the
second is a high resolution 3D survey of a river terrace deposit.
Jimmy Boyd In both cases, ResIPy allows relatively rapid data analysis and
Lancaster Environment Centre, Lancaster University, reduction of complexity normally associated with 3D survey
Lancaster, United Kingdom data inversion.
j.boyd1@lancaster.ac.uk
Bio
Introduction
Guillaume Blanchy
Electrical resistivity tomography (ERT) and induced
Lancaster Environment Centre, Lancaster University, polarization (IP) surveys are non-intrusive, relatively quick
Lancaster, United Kingdom and straightforward to perform in the field. Additionally,
g.blanchy@lancaster.ac.uk these methods are sensitive to several subsurface properties
Bio and states, providing an advantage over conventional point
Sina Saneiyan sampling methods often employed in environmental and
geotechnical studies. The sensitivity of electrical methods to
Department of Earth and Environmental Sciences,
subsurface geology and hydrological parameters make them
Rutgers, The State University of New Jersey, powerful tools for environmental studies. Furthermore, although
Newark, New Jersey, USA the majority of studies are concerned with 2D, the ability to
s.saneiyan@rutgers.edu collect and model 3D geoelectrical data provides more spatial
Bio sensitivity, increasing the value that geophysics can bring to
Paul McLachlan subsurface characterisation (e.g. Yang and Lagmanson, 2006).
3D geoelectrical methods have a range of applications, such as
Lancaster Environment Centre, Lancaster University,
mapping bedrock interfaces (Chambers et al., 2012), observing
Lancaster, United Kingdom subsurface fluid flow (Doetsch et al., 2012) and monitoring
p.mclachlan@lancaster.ac.uk landslide moisture dynamics (Uhlemann et al., 2017).
Bio
Interpretation of geoelectrical data acquired from ERT
Andrew Binley and IP surveys requires data inversion to convert the raw
Lancaster Environment Centre, Lancaster University, measurements into a distribution of electrical properties (i.e.
Lancaster, United Kingdom electrical resistivity for ERT surveys). The inversion process
a.binley@lancaster.ac.uk attempts to produce a geoelectrical model which is consistent
Bio with the observed data. Given the non-linear nature of the
inversion, several iterations of the geoelectrical model may be
Abstract required in order to obtain a final model. Once convergence
is reached, the resulting model can be interpreted (Binley
The applications of geoelectrical methods are becoming and Kemna, 2005). Several non-commercial inversion codes
increasingly widespread for near surface investigations in many exist for this purpose, but often require programming ability
disciplines, due to their sensitivity to a wide variety of geological or specialist knowledge of geophysics to be used effectively.
and hydrogeological properties. For field applications, Furthermore, they can lack other pre-processing steps such as
collection of geoelectrical data is relatively straightforward, mesh generation or data conditioning. ResIPy aims to address
however, data processing can be challenging. Mature modelling those issues by providing an intuitive GUI that guides the user
codes exist, but they often require advanced knowledge of through the different steps needed for inverting the data. We
geophysics or programming to be used effectively. This paper have previously presented ResIPy (Blanchy et al., 2019 under
complements recent presentations of ResIPy, an open source, review, Saneiyan et al., 2019) for 2D inversion, here we focus
and user-friendly, alternative. ResIPy consists of a graphical on recent updates to the ResIPy software which facilitates 3D
user interface (GUI) which utilises a modern tabbed design static and time-lapse ERT inversion.
taking the user through each step of the geoelectrical data ResIPy (resistivity and induced polarisation with Python) is
processing workflow, including data filtering, mesh generation, an open source software consisting of a standalone GUI
inversion and visualisation. ResIPy is capable of processing (Blanchy et al., 2018), which handles all processing steps
Page 86 Vol 24, 4 2019
Workflow
The GUI has a tabbed design (Figure 2), each tab is associated
with a different processing step in an ERT/IP processing
workflow: data importing, data conditioning, mesh generation,
inversion settings, inversion and post processing. This design
Figure 3. Example of a reciprocal error model computed within the ResIPy GUI. In blue
allows the user to change the inversion parameters, data filters are the mean measurement of each reciprocal pair versus the reciprocal error. The
or mesh in a nonlinear fashion before running an inversion. Here reciprocal error model (red line) is fitted to binned reciprocal errors (orange circles).
we focus on a 3D processing workflow within the GUI. Data condition and reciprocal error modelling
If the imported dataset contains normal and reciprocal
Data importing measurements, whereby the measurements are recorded in
Raw electrical measurements can be directly imported into a normal and reverse quadrupole configuration respectively,
ResIPy (Figure 2) and certain formats such as the output from a ResIPy will automatically pair these measurements and compute
Syscal device (IRIS Instruments, 2019) are natively supported. a reciprocal error for each paired measurement. This error can
The inversion code R3t expects transfer resistance measured by be used to filter the data but also to fit an error model using the
a quadrupole of electrodes, i.e., the potential difference measured logarithmically binned errors (Figure 3). This error model is then
by a pair of potential electrodes divided by the current injected used to generate a weighting for each measurement used in the
at a different pair of current electrodes. Custom data formats inversion whether it has a reciprocal measurement or not. An
in any, human readable, file can be imported using the ‘custom option is also given in the GUI to remove the quadrupoles without
parser tab’. For 3D surveys the transfer resistance is displayed reciprocal measurements. We advise capturing both normal and
against measurement number (Figure 2) as, unlike conventional reciprocal measurements in the field and fitting an error model as
2D surveys, it is not possible to compute a pseudo-section. Note this often improves the quality of the final inverse model.
that the inversions codes, R2, cR2, R3t and cR3t, were designed
to work with arbitrary geometry, e.g. for cross-borehole, surface Meshing
to borehole surveys. Also note that any quadrupole and electrode Only unstructured tetrahedral meshes are supported for 3D
geometry configuration can be used within ResIPy; pole dipole inversion with ResIPy, which are generated by Gmsh (Geuzaine
configurations can also be imported, in this case ResIPy inserts and Remacle, 2009), although R3t and cR3t also permit the
a remote electrode in the modelling mesh. use of structured triangular prism meshes. A tetrahedral mesh
The next part of the workflow is to inform the software where consists of nodes and elements, each element is defined by 4
the electrodes are positioned in 3D space given a set of XYZ nodes, and additionally, each electrode must occupy a single
coordinates. Electrode coordinates can be entered manually node. ResIPy constructs fine elements around the electrodes,
or imported from a .csv file. In environments with complex and coarsens the mesh with distance from the electrodes.
topography the user can also import extra topography points, This is to allow for improved modelling of the potential field
which will be considered during the meshing process. in parts of the mesh where high gradients exist, whilst not
creating an excessively fine mesh in other regions in order to
Page 88 Vol 24, 4 2019
Inversion
Prior to inversion there is a tab with inversion settings, which
allows for adjustments to advanced inversion parameters. For
each parameter, a descriptive help is provided in a side panel.
For the 3D inversions, ResIPy calls R3t or cR3t (if IP data
Figure 4: Mesh tab of ResIPy GUI after creating a tetrahedral mesh. Note that only a is available), which use an “Occam’s” type solution (Binley,
specific portion of the mesh around the electrodes location is actually displayed in 2015). The solution is based on regularised objective function
the GUI. The full mesh is much bigger in order to accurately model the current flows
and can be fully viewed in Paraview. The ‘material’ colour scale corresponds to the
combined with weighted least squares, which aims to yield a
discretisation of the mesh, in this case the inversion will treat the entire mesh as one smooth distribution of electrical resistivities that fits the data
zone as only one material has been defined.
(Binley, 2015). During inversion, the output of the R3t/cR3t
maintain practical levels of computer memory and computation executable is output to the GUI. Note that 3D inversions can
demands. The overall mesh is extended beyond the extent of take some time depending on the size of the problem.
the survey, which is necessary for infinite half space problem.
The boundaries of the mesh need to be sufficiently far away Modelling error
from electrode nodes as to not interfere with electrical current In addition to the reciprocal errors described earlier, a forward
flow modelling during inversion. ResIPy provides two primary modelling error can also be estimated. The option can be
options for 3D mesh construction, first is the “characteristic accessed through the inversion settings tab. This error accounts
length” of elements on electrode nodes, generally this length for errors due to discretisation of the problem in order to solve
should be at least half the electrode spacing. The second the governing equations in forward modelling. To estimate
option is the “growth factor”, which describes how much this error, ResIPy creates a flat mesh with one resistivity value
mesh elements will expand with distance from the electrode and the elevations of electrodes are normalised to 0 m. A
nodes. These two options can be changed in order to make flat topography for model error calculations is used to allow
the mesh finer or coarser, the finer the mesh the more RAM comparison of a forward model with an analytical solution.
and computing time needed. For 3D inversions R3t and cR3t is run in forward mode using
To construct a mesh, ResIPy calls Gmsh to make a flat 3D the same quadrupole configurations given in the field survey.
half-space given the electrode positions, topography is then Deviations in measured apparent resistivities modelled for a
super imposed on to the mesh nodes through 2D interpolation. homogenous subsurface with a resistivity of 100 Ohm.m give
If supplementary topography points are provided, they will be the modelling error. The total error is then computed as
used in addition to the electrodes' elevation in the interpolation. (1)
By default a bilinear interpolation scheme is used for mesh
where is the total error, is the reciprocal (measurement)
nodes. Through the API other interpolation options are available,
error and is the modelling error. The total error is used to
such as nearest neighbour look up which we find works well in
weight the inversion and in our experience its inclusion can
the case of high resolution terrestrial LiDAR (Light Detection
improve the quality of final inverted image.
and Ranging) scans. Additionally, ResIPy offers the option
to import a tetrahedral mesh created externally, and has the
ability to import Tetgen meshes (Hang, 2015), visual took kit
(.vtk) files and Gmsh (.msh) files for advanced cases (e.g.,
Applications
tank experiments or subterranean studies). When importing Successful uses for 3D inversions are well established in the
an external mesh, the node associated with each electrode is literature, many focusing on the assessment of hydrological
matched based on its closeness. conditions given the sensitivity of electrical resistivity to fluid
For mesh visualisation ResIPy natively displays a 3D mesh saturation and salinity. In the following sections we present
(Figure 4) using the Matplotlib Python package (Hunter, 2007), two recent case studies demonstrating the 3D capabilities of
however, this has limited functionality for advanced viewing ResIPy.
angles and interpretation. We suggest displaying 3D models
in Paraview, which is an open source software designed for Hollin Hill
viewing 3D volumetric data (Henderson, 2007). R2, cR2, R3t, Hollin Hill (0° 57’ 34.24”, 54° 6’ 40.67”) is an active landslide
cR3t, and ResIPy natively handle .vtk files which can be directly (Figure 5) studied by The British Geological Survey. In order
read into Paraview. From the GUI .vtk files can be exported to access the geological and hydrological conditions of the
Page 89 Vol 24, 4 2019
from the middle to lower Jurassic Period, the stratigraphy in slope, as well as surface impressions due to the flow lobes.
ascending order is: the Redcar Mudstone, Staithes Sandstone, Ignoring these topographic features would likely result in
Whitby Mudstone and Dogger formations (Merritt et al., 2013). inversion artefacts.
The Whitby Mudstone is actively failing, with progressive
deformation observed in flow lobes of mudstone material For data processing we use ERT data captured in September
(Merritt et al., 2013, Uhlemann et al., 2017). 2017 by the geoelectrical monitoring array, which consists of 5
survey lines, 9.5m apart, each with 32 electrodes at a spacing
As the field site is an active landslide the topography is also of 4.75 m, using a dipole-dipole configuration. Measurements
complex and requires periodic UAV (unmanned aerial vehicle) without reciprocal pairs are removed, and a power law
or LiDAR surveys to capture changes in topography and reciprocal error model is used for weighting the measurements.
slope movements (e.g. Peppa et al., 2019), suggesting that Additionally, a forward modelling error is computed. After filtering
topography and electrode movements should be included out measurements with over 5% reciprocal error (and unpaired
when constructing the mesh. Here we use updated electrode measurements) 1718 of 2021(85%) measurements remained.
coordinates for our 3D ERT study, and topography of the mesh The results are comparable to previous 3D inversions published
is informed by a recent terrestrial LiDAR survey from August in the literature (Chambers et al., 2008, Merritt et al., 2013,
2018. The mesh topography retains a rotational back scarp Uhlemann et al., 2017), however, this model also includes the
towards the top of the slope (Figure 6a), this is important for back scarp feature (Figure 6b) which was not present during
accurate electrical current flow modelling in that area of the previous studies as it developed in early 2016.
The ERT volume (Figure 6a), captured in September 2017, shows
the low resistivity lobes over higher resistivity material. This is
interpreted to be the lobes of mudstone material which have
flowed onto the underlying sandstone (Merritt et al., 2013).
Towards the top of the slope an area of comparatively low
resistivity can be observed, which is interpreted to correspond
to an outcrop of in situ mudstone. The apparent increase in
resistivity compared to a baseline image (Figure 6b) is likely due
to relative drying of the hillslope compared to a baseline inversion
in March 2010 reported by Uhlemann et al. (2017). In this case 3D
ERT monitoring of landslide can be linked to spatial geological
variations, but time-lapse comparison allows to see the dynamics
of the moisture content of the material temporally, which is of
the upmost interest for assessing the likely stability of a slope.
Willington
River terrace deposits represent economical, hydrological and
archaeological significance, but their structures can be complex
in 3D space (Chambers et al., 2012 and references therein).
Chambers et al. (2012) explored a river terrace deposit with ERT
for the purposes of accessing 3D bedrock detection methods
(Hsu et al., 2010) in these environments, and demonstrated the
value of 3D geophysical surveys in site characterisation. The
original 3D ERT data collection and subsequent processing was
Figure 5. Satellite image of the Hollin Hill landslide with locations of electrodes performed by Chambers et al. (2012). The location of the field
shown. The grid of electrodes has been distorted due to slope movements.
Coordinates are in British National Grid. site is in valley of the Great Ouse, near the village of Willington,
Page 90 Vol 24, 4 2019
Conclusion
We have briefly discussed the benefits of 3D ERT surveys and
Figure 7. Insert map of the 3D geophysical survey undertaken near Willington,
reproduced from Chambers et al. (2012) (with permission). Coordinates in British inversions: they yield spatial information that 2D surveys lack.
National Grid. Additionally, 2D survey interpretation can be complicated due
to 3D effects. ResIPy has recently been updated to handle 3D
geoelectrical data, making it suitable for practitioners, researchers
and students who are interested in 3D data analysis in an intuitive
open-source GUI. Although 2D inversions traditionally require
less computing power than 3D inversions, here we demonstrate
ResIPy can be used on a personal computer to carry out 3D
inversion. The software is open source which means anyone can
interrogate the source code and adapt, and improve, it to fit their
needs; furthermore, the software can be downloaded and utilised
to its full potential without the need of a license, which can be
advantageous in academic fields. With the advanced filtering
and error modelling options available within ResIPy, low reliability
measurements can be easily removed and more realistic inverted
models produced. The software is very versatile and accepts any
electrode geometry given a set of XYZ coordinates.
In cases of complex topography, additional XYZ topography
Figure 8. 3D resistivity volume inverted inside of the ResIPy GUI, displayed in Paraview. points can also be imported into ResIPy to aid in the development
of meshes and better representation of the actual topography.
4 km to the east of Bedford, United Kingdom (0° 24’ 11.14”, We demonstrate this with the case study from the Hollin Hill
52° 7’ 58.74”). The stratigraphy of the area in ascending order landslide observatory. The inverse model achieved is comparable
consists of Oxford clay Formation, middle Jurassic, overlain by to previous studies of the field site. We also show a static high
Quaternary sand and gravels likely to have been deposited by resolution 3D survey from the Great Ouse Valley UK that illustrates
a braided river system (Green et al., 1996, Barron et al., 2010). how ResIPy can also handle relatively large datasets.
The field site is situated in an arable field.
The original data were collected as a series of orthogonal 2D
surveys, on a 93 by 93 m grid with electrodes spaced every 3
Acknowledgements
m (Figure 7), as such this is a high resolution survey with 764 We would like to thank Prof. Jonathon Chambers and Dr. Paul
electrodes in total. Measurements were made in a dipole-dipole Wilkinson at The British Geological Survey for providing the
configuration. We compiled the 2D surveys into one 3D survey file 3D datasets used to demonstrate 3D inversions with ResIPy in
consisting of 23,462 raw measurements (including reciprocals), this paper. We would also like to thank the ENVISION Doctoral
which can be directly imported into the software (ResIPy). As a Training Program and the National Environmental Research
part of data cleaning, a 5% reciprocal error threshold is applied, Council for funding parts of this work. ResIPy is an open-source
and unpaired measurements are removed. As with the Hollin software available at: https://gitlab.com/hkex/pyr2.
Page 91 Vol 24, 4 2019
Author Bios
Jimmy Boyd Sina Saneiyan
PhD Student Postdoctoral research associate
Lancaster Environment Centre, Lancaster Rutgers, The State University
University, of New Jersey,
Lancaster, United Kingdom Newark, New Jersey, USA
j.boyd1@lancaster.ac.uk s.saneiyan@rutgers.edu
Jimmy graduated from the University of Leeds with a MSc in Sina is a postdoctoral research associate at Rutgers University-
Exploration Geophysics. He is currently a PhD candidate at Newark, USA. He holds a BSc. in Mining Engineering from
Lancaster University, UK, where he is exploring monitoring the University of Tehran (Iran), MSc. in Petroleum Engineering
landslides using geophysical methods with the British Geological from Sharhrood University of Technology (Iran) and Ph.D. in
Survey. His current research is on long-term 3D monitoring of Environmental Sciences (Near Surface Geophysics) from Rutgers
landslides, and focusing on relating geoelectrical and geotechnical University-Newark (USA). Sina has led multiple mineral exploration
properties. projects in Iran, focusing on copper and iron ore explorations. In
2017 he won the Near Surface Research award from the Society
Guillaume Blanchy of Exploration Geophysicists for his PhD work.
PhD Student
Paul McLachlan
Lancaster Environment Centre, Lancaster
Teaching Associate
University,
Lancaster Environment Centre, Lancaster
Lancaster, United Kingdom
University,
g.blanchy@lancaster.ac.uk
Lancaster, United Kingdom
p.mclachlan@lancaster.ac.uk