We carried out a multicomponent electromagnetic (EM) survey of the Scarborough gas field in 950 m water on the northwest Australian shelf. Magnetotelluric (MT) data, along with transmitter inline horizontal electric (Ey), vertical electric (Ez), and horizontal magnetic (Bx) field controlled-source electromagnetic (CSEM) data were collected. The Scarborough reservoir is a challenging EM target because it lies between a resistive overlying siltstone and a resistive basement. We carried out 2D inversions of various data combinations to determine how well they recover the expected geology. In particular, we examined the value of the vertical electric CSEM fields. Individual inversions of the Ey and Bx components generate almost identical models, suggesting that these two data sets do not carry independent data, although model studies suggest that this may not be the case in shallower water. Both models smear the siltstone, reservoir, and basement resistors together. The Ez-only inversion includes a resistor with a clear lateral extent at reservoir depths that is separated from basement, but when combined with other CSEM components, Ez provides only marginal improvements in resolution. Not surprisingly, an MT-only inversion is blind to the thin reservoir resistor but combined with CSEM data produces a clear separation of the reservoir from the basement. The combination of Ey, MT, and Ez also separates the siltstone horizon from the reservoir. The sensitivity of MT to horizontal conductivity makes it a powerful complement to the standard Ey CSEM data. The Ez CSEM component adds some value, but perhaps not commensurate with the logistical costs of data collection. The horizontal magnetic CSEM field appears to add little value at these water depths, but if simultaneous MT data are being collected, this component will be available at little cost.

The Scarborough gas field is situated in a water depth of approximately 950 m, approximately 300 km offshore on the Exmouth Plateau, part of the northwest shelf of Australia (Figure 1). It is approximately 20 km across and estimated to hold 9 Tcf of gas. In May/June 2009, we collected controlled-source electromagnetic (CSEM) and magnetotelluric (MT) data over Scarborough as part of our efforts to develop electromagnetic (EM) methods for hydrocarbon exploration. The data acquisition, processing, and data uncertainty are described by Myer et al. (2012). The MT data interpretation is presented by Myer et al. (2013), and the CSEM data interpretation is presented by Myer et al. (2015). The Myer et al. (2015) study involved 2D inversions of the radial horizontal electric field component of the CSEM data, inverted jointly with the MT data. The horizontal electric field component is the one most commonly collected and interpreted in CSEM studies, academic and commercial, and it is well-known that the radial (inline) component has much better sensitivity to thin subhorizontal resistors than the azimuthal (crossline or broadside) component (e.g., Cheesman et al., 1987; Constable and Weiss, 2006). However, as part of our Scarborough study, we collected vertical electric field CSEM data, as well as CSEM data on the horizontal magnetometers used for the MT study. Here, we investigate whether these additional CSEM components contribute to the resolution of the reservoir in CSEM inversions.

The Scarborough gas field has proved to be a challenging CSEM target because the reservoir, approximately 1000 m below the seafloor and having a resistivity-thickness product of approximately 900  Ωm2, is overlain by the Gearle, a resistive siltstone at a depth of 700–800 m having a resistivity-thickness product of approximately 300  Ωm2. The reservoir sits approximately 1000 m above a resistive “basement,” which also limits our ability to resolve the target horizon. A simpler target might have made a better case study, but this is the only academically available data set with all three components of CSEM data available, and the ability to separate the three resistive layers provides a good test of how multicomponent inversions might assist interpretation.

The MT method is a well-established exploration technology that is used broadly throughout academia and industry. Natural variations in earth’s magnetic field provide the source of energy, from which a transfer function between horizontal magnetic and electric fields is estimated as a function of frequency. The transfer function contains information about earth conductivity, and on land bandwidth it can extend from kilohertz frequencies (and even higher if man-made radio sources are used) to periods of 100,000 s, providing information on subsurface electrical conductivity from the meter scale near the surface to upper mantle depths.

On the ocean floor, the high conductivity of the overlying seawater robs the source field of high frequencies, imposing a short period limit of 1–10 s on the MT method (dependent on the water depth and seafloor conductivity). Nevertheless, ocean-bottom MT measurements have proved to be useful in delineating geologic structure on scales extending from exploration targets to plate boundary tectonics. It is also important to note that skin depth (the characteristic length scale of inductive attenuation) alone does not determine the resolution of EM methods — near-surface conductivity contrasts can produce galvanic effects in the data that reveal their existence.

The idea of joint inversion of MT and resistivity data dates back to Vozoff and Jupp (1975), and it was introduced into the original 1D Occam inversion algorithm of Constable et al. (1987). Because the resistivity method might be considered a low-frequency EM method, this could be viewed as the beginning of joint MT and CSEM inversion. Key (2016) naturally includes the capability for joint inversion with MT in his 2D CSEM inversion code, which Myer et al. (2015) used in all their joint CSEM and MT inversions of the Scarborough CSEM data. Constable et al. (2015) note that the inclusion of CSEM data in a MT inversion not only changed the structure at the shallower depths, at which one would expect the CSEM data to have an influence, but it also changed the structure at depths beyond the sensitivity of the CSEM data. As mentioned above, MT data are sensitive to near-surface conductivity structure, so changes in the near-surface structure driven by the CSEM data may require compensation by changing the deeper structure to fit the MT data.

Because of the desirability of collecting MT data simultaneously with CSEM data, most CSEM receivers are equipped with induction-coil magnetometers, which will record CSEM signals along with the electric field sensors. Although several modeling studies have considered the effects of horizontal magnetic field data (e.g., Um and Alumbaugh, 2007; Orange et al., 2009; Løseth et al., 2010), in practice, one rarely sees the magnetic field used in CSEM studies. The main reason is that the motional noise is greater on the magnetic field sensors than on the electric field sensors because any rotation can couple magnetometers to Earth’s magnetic field, which is 10 million times larger than the CSEM amplitudes being measured. Note that this issue is not a function of magnetometer sensitivity. A fairly standard geophysical induction coil sensor has noise of approximately 0.1  pT/Hz at 1 Hz, which translates to a CSEM noise floor of 1019  T/(Am) for a 60 s stack frame and the transmitter moment used in the Scarborough survey. However, a sensor rotation of only a few nanoradians in a 40  μT Earth field will result in a (noise) signal of the same magnitude as the sensor noise. One nanoradian is a rotation of 1 mm at the end of a 1000 km baseline.

Figure 2 shows the cumulative distribution of observations as a function of the source-receiver range for all of the inline data on Scarborough line 2 with amplitude errors of less than 30%. We see that 90% of the electric field data extend to source-receiver ranges of 8000 m, but 90% of the magnetic field data are limited to source-receiver ranges of 6400 m. The electric field data extend to a maximum range of 14.3 km, but the magnetic field data extend only to 11.1 km.

Of course, this limitation on maximum range would be of limited significance if the magnetic field data had superior sensitivity to the targets of interest, in this case hydrocarbon reservoirs. However, in deep water, the magnetic field has almost identical sensitivity to the horizontal electric field, as evident in the 1D inversions of Key (2009) and the 2D inversions shown in this paper, so there appears to be no advantage in using it. This may not be the case in shallow water, as we show below.

The vertical electric field is not normally collected during CSEM surveys using a towed horizontal electric field transmitter, probably because it is logistically inconvenient, but it has been considered in many model studies, and it has been the subject of field tests by Scripps Institution of Oceanography (SIO), described for the first time in this section. More recently, vertical electric field CSEM measurements have become a key element of PetroMarker’s vertical transmitter/vertical receiver surveying system (Helwig et al., 2019).

Early model studies

When CSEM modeling algorithms were first developed, it was natural to include calculations of the vertical electric fields generated from a horizontal electric dipole, but as Chave and Cox (1982) note, “the vertical field amplitudes are smaller by at least an order of a magnitude” compared with the horizontal fields. This led to the emphasis on horizontal field measurements as the marine CSEM method developed. That the inline vertical electric field Ez might carry additional information to the horizontal fields was first hinted at by Chave et al. (1991), who note that “it is clear … that the vertical electric field is sensitive to the conductivity of the half-space at smaller ranges” than the horizontal field and that “the relative changes in amplitude of the vertical and horizontal electric fields are comparable for a given change in medium conductivity.” However, making measurements of the horizontal fields was challenging enough at that time, and the horizontal electric dipole-dipole method continued to be the only one used in practice during the early days of marine CSEM.

The adoption of marine CSEM methods by industry in the earliest years of this century meant that suddenly many more people were studying the method, often with the goal of obtaining some advantage over existing technology. The vertical electric field was again examined. With access to 2D and 3D modeling codes, it became apparent that Ez may be preferentially sensitive to the edge of structures (Constable and Weiss, 2006). In addition, concern about the reduction in sensitivity to hydrocarbon targets caused by energy propagating through the air in shallow water (the “air wave”) made the measurement of Ez of interest (Andréis and MacGregor, 2008) because, at least in theory, Ez is insensitive to the horizontal air wave signal. Figure 3, drawn from a model study by Orange et al. (2009), illustrates both of these effects and shows the 0.1 Hz inline horizontal field (Ey) and Ez for a 2D model of a target reservoir, 100  Ωm and 100 m thick buried 1000 m in a background of resistivity 1  Ωm. The response has been normalized by the response of the background without the reservoir. In deep water (3000 m), the Ey and Ez anomaly responses are of similar magnitude, and the air wave response, which still carries information about the target, is expressed at much larger ranges than the primary response. However, in shallower water (1500 m), the Ey air wave response merges with the primary response and reduces its magnitude as well as extending its width. The Ez response is largely unchanged (indeed, it is slightly larger and sharper), with no air wave component.

With sensitivity tuned to the vertical resistivity, it is likely that measurement of Ez would help resolve anisotropy (see, e.g., Figure 12 of Constable, 2010). Another advantage comes from the fact that there is little or no MT signal in the vertical electric component, and so for low-frequency CSEM transmissions (approximately 0.1 Hz), it is sometimes possible that Ez signal-to-noise ratios are better than for Ey (L. Srnka, personal communication, 2006). Indeed, early in the development of marine MT methods for exploration, model studies of Ez (which could be called an “electric field tipper”) showed that there was not a large MT signal and that it added little value in determining the structure (M. Hoversten, personal communication, 1998).

Instrument development

These considerations led ExxonMobil to fund SIO to develop a vertical electric field sensor in 2003 following preliminary research of its own (Summerfield et al., 2005). There are two obvious approaches for collecting vertical electric field data: (1) floating an electrode above the receiver instrument on a cable and (2) attaching electrodes to a rigid pole mounted vertically on the instrument.

The advantage of a cable array is that a relatively large separation between the electrodes can be achieved — in a study of internal waves and tides, Bindoff et al. (1986) use a separation of 160 m in this way. However, large separations are only effective at reducing noise in electric field measurements if the source of the noise is voltage generated in the electrodes and amplifiers: Noise created by environmental electric fields will scale with the antenna length along with CSEM signals. Deployment of such a long antenna requires paying out the array behind a moving vessel, and for CSEM inversion large antennas will add to the computational burden because electric fields need to be integrated along their length. However, shorter antennas of the order the same length as the horizontal electric field sensors (10  m) may be practical, and they were tested by SIO.

The use of a rigid pole simplifies deployment but limits the electrode spacing to a few meters or less. To maximize electrode separation, in initial tests the lower electrode was mounted near the seafloor at the base of the instrument frame. This proved to be a mistake because data analysis showed that the insulating instrument frame was deflecting electric currents associated with the larger horizontal CSEM signals into the vertical direction, contaminating the Ez measurement. This means that the lower electrode for an Ez measurement needs to be mounted above the instrument frame.

During a 2003 marine CSEM experiment over the Gemini prospect in the Gulf of Mexico (described by Constable et al., 2015), we deployed five instruments with a 10 m cable floated above the instrument and eight instruments using polycarbonate poles with electrode spacings of 1.6 m. The cable/float assemblies exhibited much larger, and much more variable, noise across the entire spectrum than the data collected on the poles, even when the longer dipole is factored into the electric field calculation. However, the pole instruments did show narrow resonances between 0.5 and 2 Hz, presumably associated with strumming. The conclusion is that the main source of noise in the vertical electric field measurements is motion of the antenna in Earth’s magnetic field for both types of sensors, but noise is far worse on the cable array. The company PetroMarker started to investigate a marine time-domain EM method using vertical transmitters as well as receivers in 2005. They, too, discovered that rigid dipoles had lower noise than the floated cable antenna (Holten et al., 2009) and published seafloor noise spectra broadly consistent with the spectra for rigid poles shown in Figure 4 (Håland et al., 2012).

The observation of strumming led to a refinement of the polycarbonate pole, replacing it with a pole made from spun aramid fiber, which is light (a consideration because the instrument has to be floated to the surface under its own buoyancy) and very stiff. These were the sensors used during the 2009 Scarborough survey. Figure 4 shows representative spectra of the various sensor types collected during the Gemini and Scarborough surveys. The noise is more consistent for the aramid poles than for the polycarbonate poles, and the resonance has narrowed to a more predictable frequency at approximately 1 Hz, although the amplitude of the resonance is still variable: One of the instruments has a peak of two orders of magnitude in power, and one instrument has no peak at all. For comparison, the spectra of the horizontal electric fields from the Scarborough experiment are also shown.

A fairing could be added to the pole to reduce the resonance, but outside the resonance we are achieving the amplifier-limited noise floor of the instrument when the shorter dipole length is considered. Because the resonance is narrow, avoiding CSEM transmission at that frequency is an effective strategy. Figure 5 shows a photograph of a CSEM receiver equipped with a vertical sensor pole.

In the rest of this paper, we present modeling and inversion studies using the 2D adaptive finite-element forward and inversion package, MARE2DEM, described by Key (2016). This publicly available code can simultaneously invert MT and frequency-domain CSEM data using the regularized inversion algorithm of Constable et al. (1987). Inversion is carried out on a triangular or quadrilateral parameter mesh of conductivities that are isotropic or can have up to triaxial anisotropy in the principal coordinate directions. Forward calculations are carried out on a triangular finite-element mesh in a dual-grid approach, and adaptive refinement ensures a specified forward model accuracy, usually set to 1%. The code is optimized for performance on a multicore computer cluster. In the framework of this modeling package, the inline electric field is designated as Ey, the vertical electric field as Ez, and the crossline magnetic field inline with the transmitter as Bx.

We start with a synthetic forward and inverse model study based on the model shown in Figure 6, a simplified version of the Scarborough gas field geology in which a resistive horizon (the Gearle) overlies a gas reservoir at 2 km below sea level, which in turn is above a resistive formation (the Mungaroo) at 3 km depth.

We generated synthetic forward model data at frequencies of 0.25, 0.75, 1.75, and 3.25 Hz for seafloor receivers spaced at 1 km intervals and horizontal electric field transmitters spaced at 250 m. The parameters are similar to those for the data collected over Scarborough. The Ey, Ez, and Bx field components were degraded with 2% Gaussian noise applied proportionately to the amplitudes and phases of the field components, but inversions were carried out using log(amplitude) and phase, which Wheelock et al. (2015) show to be faster and more stable than inverting the linear amplitude. Figure 6 shows the result of inverting Ey, Ey + Bx, and Ey + Ez in a water depth of 1000 m.

The addition of the Bx or Ez data does little to improve the resolution of the inversion. In both cases, there is a barely perceptible sharpening of the reservoir and Gearle, and the addition of Bx slightly sharpens the resolution of the reservoir edge and creates a separation of the Gearle at the edge of the reservoir. Within the limitations of the simplified model, the conclusion is that we are likely to see only modest improvement by including Ez and Bx in our inversions for Scarborough-like targets and high-quality Ey data. However, heterogeneity in the background sediments, along with anisotropy, might very well modify this conclusion.

Before moving to real data, we note that for a water depth of 300 m, there is a larger improvement obtained by adding the magnetic field component (Figure 7). The magnetic field component interacts with the atmosphere resistor differently than the horizontal electric field does. This can be seen in Figure 5 of Orange et al. (2009), which shows that in deep water, the Bx anomaly over a 2D reservoir model looks very similar to the Ey anomaly, but as the water shallows, the Bx response of the reservoir gets much larger than the Ey response (this was noted by Orange et al. [2009] without further comment).

About three quarters of the receiver instruments in the Scarborough survey were equipped with vertical electric field sensors. Initial studies tended to confirm the conclusion that the inclusion of Ez in inversions did little to improve resolution, so our previous publications (Myer et al., 2012, 2015) concentrated on the Ey data, which are, after all, representative of most of the CSEM data collected. All of the receivers were equipped with induction coil sensors to collect MT data and, as noted above, Myer et al. (2015) include MT in the Ey inversions. However, the calibration of the magnetic sensors, although adequate for MT data, was initially not comparable to the 1% uncertainties associated with the Ey data (see Myer et al., 2012). This was because the external magnetic sensors were calibrated separately from the amplifiers housed in the data logger pressure cases, and we did not match sensors and amplifiers during deployments. For this study, we modeled the nonlinear interactions between the amplifiers and sensors to generate accurate calibrations and allow us to include the Bx data alongside the Ey data.

Table 1 shows the distribution of the data and the ranges inverted (source-receiver distance for CSEM, period for MT). We used CSEM frequencies of 0.25, 0.75, 1.75, and 3.25 Hz for Ey, Bx, and Ez, with error floors as given in Table 1. The CSEM data were processed using the algorithm of Myer et al. (2011), which produces a statistical stacking error. The larger of the error floor or stacking error is taken, to which a range and frequency-dependent navigation error estimate described by Myer et al. (2012) is added geometrically. Finally, a minimum 2% error was enforced and data with errors larger than 30% were excluded. CSEM data were limited to ranges of 1 km or greater to avoid the larger systematic navigation errors associated with small source-receiver offsets. The Ey data were obtained by rotating the horizontal electric fields into the direction of the transmitter tow line (113.9° east of north) using receiver orientations derived from recording compasses. Myer et al. (2012) estimate a standard error of 3.3° for the compass directions, which contributes a standard error of 0.4% to Ey. The compasses also record the pitch and roll of the instruments, but because the noise level is quite different in the vertical and horizontal electric fields, and because the seafloor may be sloping, the data were not rotated into exactly horizontal and vertical components. Instead, the dip of the field components is recorded in the data files and the inversion computation is rotated into measurement coordinates. The average departure from vertical/horizontal is 2.1° with a standard deviation of 2.4°, which is probably comparable to the tilt measurement accuracy.

The transmitter for his survey output 300 A zero-to-peak on a 250 m neutrally buoyant antenna, modulated using waveform-D (Myer et al., 2011) with a fundamental frequency of 0.25 Hz. This produces a source dipole moment of 66,000 Am at the third harmonic of 0.75 Hz. The transmitter antenna had a dip of approximately 5°, which was, again, included in the modeling.

Figure 8 shows maps abstracted from the raw CSEM data at one frequency (0.75 Hz) and a source-receiver range of 3000±200  m. To produce the maps, all data with a signal-to-noise ratio of more than 10 and having transmitter azimuths ϕ within 60° of radial (inline, or ϕ=0°) were rotated to the radial and azimuthal directions, and the radial component was divided by cosϕ to recover the radial mode. These data were then associated with source-receiver midpoints and fit to a smooth map using the method described by Constable et al. (2018). The maps are only plotted for points within 5 km of data, and the number of points in the map and rms misfit to the smooth surface are noted on the maps.

All data components show elevated amplitudes, and smaller phase shifts, over the seismically determined gas-water contact. The magnitude of the signal from the reservoir appears to be similar for all of the components, with the phase being generally smoother than the amplitude, which would be consistent with near-surface variations in resistivity producing galvanic effects in the amplitude data but leaving the phase less affected. Such an effect would be particularly pronounced in Ez, in which the amplitude will be affected by the resistivity contrast between the sediment immediately beneath the receiver and the seawater, but the phase would not be altered. The appearance of the Ez amplitude and phase maps and the lower misfits for the phase data supports this concept.

Various combinations of the line 2 Scarborough data set (see Figure 8 for the location) were inverted using a quadrilateral inversion mesh conforming to bathymetry, with cells 250 m wide by 65–90 m tall to a depth of approximately 150 m below the seafloor, increasing to 1000 m wide to a depth of 2500 m. Below this depth, the width and height are doubled. The horizontal roughness was penalized three times more than the vertical roughness, but the large aspect ratio for the inversion mesh was used to increase the effective smoothing in the horizontal direction in addition to this. The model parameters allowed anisotropy with equal horizontal resistivities and differing vertical resistivity (transverse isotropy in the vertical direction). The penalty between the horizontal and vertical resistivities was the same as the vertical roughness penalty (i.e., 1).

All of the inversions shown below converged from a 1  Ωm starting half-space to a smooth model with an rms misfit of 1.5, with two exceptions as noted. Figure 9 shows the vertical resistivity resulting from inversion of the individual data components. The Ey and Bx inversions are very similar, as expected. The reservoir, situated at a depth of 1.9–2.0 km and extending between 18 and 0 km in horizontal distance, shows as a slight increase in resistivity, but it is merged with the electrical basement at a 2.5 km depth. The basement resistivity increases smoothly with depth from 2.5 km, until our color scale saturates at 3.5 km and 10  Ωm. What appears as basement in the resistivity inversions is actually the top of the Mungaroo Formation, which, as noted by Myer et al. (2015), is more resistive than expected.

The Ez inversion could be run to a lower misfit of rms 1.0, and it produces a clear resistor at reservoir depths that is well-separated from the basement, although the basement is not well-resolved by this component. The edges of the reservoir appear to be well-defined, in fairly good agreement with the seismically determined edge to the west, but several kilometers inside the seismic estimate to the east. It is perhaps not surprising that there is little sensitivity to the basement horizon, at a depth of 2.5–3.5 km, when the data only extend to source-receiver ranges of approximately 5 km (Figure 2). That there is some expression of basement beneath the reservoir could be a consequence of the increased resistivities at reservoir depths increasing the signal strength at larger range. Examination of the data suggests that this is the case; over the reservoir, the Ez data extend to ranges of 5.5 km or more, whereas off the reservoir, the ranges only extend to approximately 4.5–5 km.

Predictably, the MT-only inversion is blind to any thin resistors at the depth of the Gearle and reservoir, but it does see an increase in resistivity at the depth of the Mungaroo, although the increase in resistivity starts a little deeper than for the Ey and Bx inversions. The resistors near the surface are probably artifacts that have been included to improve the fit to the TM mode resistivities by generating galvanic (static) effects. We could include a static shift as a free parameter in the inversion, but true static shifts are rare in the marine environment (Key and Constable, 2011), and the inclusion of CSEM data in the joint inversions shown below removes these artifacts.

Figure 10 shows the vertical resistivity resulting from inversions adding the Ez and Bx components to the Ey CSEM data. The addition of Bx CSEM data to the Ey data does not change the character of the inverted resistivity at reservoir depths, but it does increase the resistivity of basement, bringing the 10  Ωm threshold from approximately 3.5 to 3.0 km depths. The reservoir shows a slight increase in resistivity in the inversions that include the Ez component, but it is even more badly merged with the electrical basement. All inversions produce a slight sharpening of the Gearle Formation off the reservoir to the east.

The addition of MT data to the various CSEM data combinations is shown in Figure 11. Adding MT to the Ey CSEM data produces a perceptible improvement over any of the combinations of CSEM data without MT. The increase in resistivity at the basement depth is much sharper, and there is a dip in the basement toward the east. There is an increase in resistivity at the reservoir location, and we now see a clear separation between the reservoir resistor and basement. For the Ey + MT inversion, the reservoir and Mungaroo depths are a little shallower than depths derived from seismic data and well logs. Note that this inversion is basically the same as the one presented in Figure 7a of Myer et al. (2015), but we have used different data selection criteria, which included removing approximately 200 data with weighted residuals more than six standard errors. Perhaps more importantly, we forced a larger degree of horizontal smoothing by building the inversion mesh with 1 km wide rectangular parameters at reservoir depths.

Adding Ez data to the Ey + MT inversion continues to improve the model. The peak resistivity of the reservoir is now higher, and the depths of the reservoir and top Mungaroo now agree with the seismic data and well logs. We are even seeing a separation between the Gearle, at a depth of 1.5 km, and the reservoir, as well as a sharpening of the Gearle resistivities to the east, which, like the Mungaroo, are deepening toward the east. However, we have lost some of the lateral continuity of the subreservoir sediments.

The addition of the Bx CSEM data to the Ey + MT inversion pushes the reservoir and top Mungaroo slightly deeper than the Ey + MT inversion, but not quite to the expected depths, and it sharpens up the Gearle a little in the east. It does not greatly improve the resistivity image of the reservoir. Inverting all four data sets together produces an image that is similar to the Ey + MT inversion but with the reservoir at the expected depth. For this inversion, we converged to a smooth model at rms 1.7, and so we are perhaps finally seeing some level of incompatibility between the different data sets. This may be caused by subtle differences in the different data sets’ abilities to incorporate small amounts of non-2D structure.

It is reasonable to ask why the addition of MT data makes such a profound improvement, given that MT has poor resolution in resistors, and the shortest period, 16.5 s, has a skin depth of approximately 2 km in the sediments. The answer is that the MT data have sensitivity to the conductive sediments beneath the reservoir, so much so that the MT only inversion underestimates the resistivity of the top Mungaroo. In the joint inversion, this prevents the Ey inversion from smearing the reservoir resistivity with basement resistivity, forcing a conductor between the two. However, the Ey CSEM data are still sensing the resistivity of the Gearle/reservoir and basement, which it must place above and below the conductive layer, in the reservoir and in a shallower, sharper, basement layer. The MT data now sense a second conductor at depth, suggesting that the deeper parts of the Mungaroo are less resistive.

All of the inversions above show vertical resistivity from anisotropic inversions. Figure 12 shows anisotropy ratios for the simpler combinations of data. As expected, the MT-only inversion does not require anisotropy to fit the data because electric and magnetic fields are predominantly in the horizontal directions. The Ey only inversion includes a small amount of anisotropy at reservoir depths and below. The addition of MT, Bx, or Ez data produces much more anisotropy. Even though the addition of the Ez and Bx data in Ey inversions does not change the resistivities greatly, it does produce an increase in anisotropy at reservoir depths, which could be a clue for the presence of the reservoir. The inclusion of MT data generates more anisotropy, but mostly above the reservoir depth, perhaps as a result of the Gearle. All produce some anisotropy in the Mungaroo Formation, which consists of successive sequences of siltstone, sandstone, and coal, which one might expect to have some anisotropy.

Figures 13 and 14 show the breakdown of the misfit by data type, frequency, and range for the MT + Ey + Bx + Ez inversion shown in Figure 11. The MARE2DEM code includes a weighting function to ensure that misfit is distributed evenly between MT and CSEM data type (Key, 2016), but as can be seen in Figure 13, there is also equal misfit among the Ey, Bx, and Ez components, even though this was not enforced. For the MT data, we see a slightly larger misfit for the TE mode apparent resistivities, which we see is weighted toward the shorter periods (Figure 14). This suggests that the cause is near-surface 3D effects that cannot be accommodated by the 2D model in the strike direction. Although the inversion has balanced the CSEM misfit by data type, there is a small but systematic increase in misfit for the phase data and misfit generally increases with frequency, which could suggest residual timing errors in the data. The CSEM misfit peaks at short range, in which the navigation errors can predominate, and large range, in which the signal-to-noise ratio is low.

Figure 15 breaks the same model misfit down by site number, using the numbering documented by Myer et al. (2012). Figure 16 shows the distribution of the normalized residuals, compared with a normal distribution with a standard error equal to the rms misfit (1.7). The residual distribution is narrower than the normal distribution, and it is slightly long-tailed. Taken all together, the distribution of residuals is not perfect, but it is quite acceptable, showing that the data errors are generally well-estimated, the 2D assumption is a good approximation, and no one part of this large and complicated data set is overly influencing the model.

As useful and important as they are, misfit breakdown plots only provide summary statistics. It is more instructive to see actual data and model fits, but with 26,199 total data points it is impractical to show all the data. In Figure 17, we present a representative example for site 12, representing 3% of all data and that has a site misfit of rms 1.67, close to the median site misfit and overall misfit of rms 1.7. There is a slight bias in the Ey and Bx phase for the lowest (0.25 Hz) frequency; otherwise, the fits are unbiased and very good.

Figure 18 shows the misfit and model roughness evolution during the MT + Ey + Bx + Ez inversion, starting from a 1  Ωm half-space. The inversion was initially run to convergence (achieving the required misfit with no further decrease in model roughness) to a target rms of 2.0, and then it was restarted and run to convergence for a target rms of 1.7. The two-step convergence is part of a process to ensure that an unreasonably small misfit has not been requested. There is a factor of six in the misfit of the various data sets for the starting half-space model, which largely disappears for the initial rms 2.0 model except for a slight overfitting of the Ez data. By the end of the rms 1.7 convergence, the misfit is almost evenly proportioned between data sets, as shown in Figure 13. Even though the Ez-only data set can fit better than the other individual data sets (rms 1.0 compared with rms 1.5) in the joint inversions, the Ez data are fit to the same misfit as the other data sets, so we chose not to reweight the errors.

When we embarked upon the Scarborough project, we had little understanding of just how challenging the reservoir target was from an EM method perspective. Our presurvey model studies showed a strong signal from the reservoir, which indeed proved to be the case, but also showed that we would be able to differentiate the Gearle from the reservoir, which proved difficult in practice. We now understand that our presurvey modeling used an overly optimistic noise model, with 1% errors down to a data cutoff at the estimated error floor. In the real world, errors become larger near the error floor. We had little information about the resistivity of the sediments underlying the reservoir, so we did not appreciate that this, too, would challenge the inversion and interpretation of the data. However, the complexity of the target is perhaps ideal for examining how the inclusion of additional data components improves the resolving power of the CSEM method.

All CSEM data components, Ey, Bx, and Ez, carry a strong signal associated with the reservoir, amounting to 20°–30° of phase shift at 0.75 Hz transmission frequency. However, inversion of Ey only data, even when four frequencies spanning one order of magnitude are used, produces a resistor that is smeared into the Gearle and basement.

The inline horizontal magnetic field Bx appears to carry almost identical information to Ey, and inversion of Bx-only data produces almost identical results to inversion of Ey-only data. However, there are two reasons that one should not consider Bx data equivalent to Ey data. One is that because of the extreme sensitivity of magnetic sensors to movement in the earth’s magnetic field, the noise floor for Bx data is greater than for Ey data, and it reduces the maximum range, in this case from 14.3 to 11.1 km. The second is that model studies suggest that in shallower water, where the air wave is expressed, the Bx component has greater resolution than the Ey component. The combined inversion of Ey and Bx data does slightly sharpen the resolution of the Gearle off-reservoir, but it does nothing to separate the reservoir from the basement or Gearle.

Although Key (2009) shows that the vertical electric field Ez has poorer resolving power in 1D simulations, various model studies have shown that Ez might hold particular promise for the delineation of the edges of the structure (e.g., Constable and Weiss, 2006), and of all the individual component inversions, the Ez-only inversion does the best job of delineating the reservoir and separating it from the basement resistor. However, Ez data are somewhat difficult to collect from a logistical perspective: The signals are smaller than for Ey, and the noise is larger, and so in this study we only had Ez data out to ranges of 5.7 km, compared with 14.3 km for Ey. It is also disappointing that when combined with Ey, the inversion of the two data components provides only marginal improvement over the Ey-only inversion, and the resulting model is almost indistinguishable from the Ey + Bx inversion. The inversion of all three CSEM data components (Ey + Bx + Ez) slightly sharpens up the resolution of the Gearle over the two-component or individual inversions, and it slightly increases the resistivity at the reservoir depths, but it is questionable whether the improvements are worth the additional effort of collecting the Ez data.

Inclusion of MT data by far produces the largest improvement in the resolution of the Scarborough reservoir, clearly separating it from the basement. The MT data cannot, however, help separate the reservoir from the Gearle, probably because the frequencies are too low to have significant resolution at these depths. The addition of the Bx data to the MT + Ey inversion does almost nothing to improve the resolution, but the addition of Ez data does increase the resistivity at reservoir depths and separating the reservoir from the Gearle resistor. The Ey + MT + Ez inversion is, at least subjectively, the best combination of data to use in this situation. Inverting all of the data by including Bx provides little improvement over the Ey + MT and Ey + MT + Bx inversions, perhaps because the effects of the Ez data have become too diluted.

One way to interpret these results is in terms of data independence. MT data are very different from CSEM data in that the source is a downwardly propagating horizontal magnetic field. All of the CSEM data components are related in that they were generated from the same horizontal dipole transmitter, and they are all, albeit different, measures of the same electrical current systems in the seafloor. It seems, in relatively deep water at least, that the horizontal magnetic and electric fields are sampling the CSEM fields in almost identical ways, which is consistent with the synthetic 1D studies of Key (2009). There is some evidence that the vertical electric field behaves differently enough to be of additional use. That the vertical field behaves differently was also noted by Key (2009), but in 1D the spatial resolution of Ez is poor compared with Ey or Bx. It appears that in higher dimensions, the Ez component may provide more value.

Proponents of 3D CSEM data collection, in which crossline, or azimuthal, CSEM data are collected, might claim that azimuthal data, which have increased sensitivity to horizontal resistivity, would act in a similar fashion to the MT data. We hope to test this at some future time by carrying out 3D inversion of the entire Scarborough data set. However, we believe that there is still a role for 2D predrill appraisal CSEM data collection, as well as academic and commercial data collection along profiles to image geologic structure (see, e.g., Johansen et al., 2019).

MT data provide the most useful improvement in resolution when combined with inline CSEM horizontal electric field data for inversions of the Scarborough gas field geology, and probably also other oil field type structures. Given the modest additional acquisition and processing costs, one might argue that the joint use of CSEM and MT inversion should be routine. Model studies have shown that the vertical electric CSEM field holds much promise, and our studies show that it does indeed add significant value, but only if inverted on its own or as an addition to joint MT and CSEM inversion. The simple combination of horizontal and vertical CSEM data provides little justification for the additional effort of collecting the vertical field data. Horizontal magnetic CSEM data will be collected for no additional cost or effort if MT data are being collected, but they provided little value in the inversions of the Scarborough gas field data. However, model studies show that the horizontal magnetic field may be of additional value in shallower water.

We thank BHP Petroleum for funding the data collection, and the Scripps Seafloor Electromagnetic Methods Consortium for funding the analysis and inversion. The technicians and engineers of the Scripps Marine EM Laboratory, along with the captain, crew, and science party of the R.V. Roger Revelle all contributed to the success of the field work. Special thanks go to K. Key for his MARE2DEM code and his contributions to the earlier phases of the Scarborough project, including the data acquisition. We thank the editors of Geophysics, two anonymous reviewers, reviewer R. Mittet, as well as L. Srnka and G. Liu for their comments on the manuscript.

Data associated with this research are available and can be accessed via the following URL: https://marineemlab.ucsd.edu/Projects/Scarborough/Data.html.

Freely available online through the SEG open-access option.