Seismically regularized controlled-source electromagnetic inversion

Abstract

Marine controlled-source electromagnetic (CSEM) data can be highly sensitive to the presence of resistive hydrocarbon bearing layers in the subsurface. Yet, due to the relatively poor depth resolution of CSEM data and the smoothness constraints imposed by electromagnetic (EM) inversion methods, the resulting resistivity models are often highly smoothed-out, typically underestimating the reservoir resistivity and overestimating its thickness. Conversely, seismic full-waveform inversion (FWI) can accurately recover the depths of seismic velocity changes, yet, is relatively insensitive the presence of hydrocarbons. In spite of their low depth resolution, CSEM data have been shown to be highly sensitive to the resistivity-thickness product of buried resistive layers, suggesting that if the thickness of a target layer can be constrained a priori, very accurate resistivity estimates may be obtained. We developed a method for leveraging the high depth resolution of FWI into a standard CSEM inversion algorithm so that the resulting resistivity models have depth constraints imposed by the seismic structure and consequently may obtain more accurate resistivity estimates. The seismically regularized CSEM inversion that we propose is conceptually similar to minimum-gradient support (MGS) regularization, but it uses regularization weights based on gradients in the seismic velocity model rather than the self-reinforcing model resistivity gradients used in the typical MGS scheme. A suite of synthetic model tests showed how this approach compares with standard smooth and MGS inversions for a range of rock types and hence, levels of correlation between the seismic and resistivity structures, showing that a significantly improved resistivity model can be obtained when the velocity and resistivity profiles are correlated in depth. We also found that this regularization weighting method can be extended to use depth constraints from geophysical data other than seismic velocity models. Tests on a real data example from the Pluto gas field demonstrated how the regularization weights can also be set using a nearby well log, resulting in a more compact estimate of the reservoir resistivity than possible with a standard smooth inversion.

Two‐dimensional magnetotelluric inversion using reflection seismic data as constraints and application in the COSC project

“…To overcome this problem, inversion with seismic regularization has been considered. Brown et al [] proposed a weighted regularization operator based on the gradient of a seismic velocity model v : where β is a stabilization parameter. Note that in model regions with spatially constant seismic velocity (i.e., ), simplifies to the gradient or first‐order differences of the resistivity model .…”

“…Note that in model regions with spatially constant seismic velocity (i.e., ), simplifies to the gradient or first‐order differences of the resistivity model . Brown et al's [] regularization operator is a modification of Portniaguine and Zhdanov's [] minimum gradient support inversion that replaces the gradient of the resistivity model in the denominator with the gradient of the seismic velocity model. Since full 2‐D or 3‐D seismic waveform inversion is still impractical in reflection seismics, Brown et al's [] algorithm and examples are for 1‐D models.…”

“…One such inversion scheme is Brown et al . 's [] 1‐D weighted regularization which is based on the gradient in seismic velocity and, hence, requires sequential inversion of full seismic waveform and EM data. Zhou et al's [] inversion processes a guiding seismic image and determines its discontinuities and coherent regions to assign small and large smoothness weights, respectively, across the associated cell boundaries.…”

Inversion of controlled-source electromagnetic reflection responses

“…This search is done by minimizing in some way the difference between the measured and the forward modeled data based on the latest guess of the subsurface conductivity distribution. Because this is a major topic of CSEM research, a wide variety of papers have been published on inversion (Christensen and Dodds, 2007;Plessix and Mulder, 2008;Key, 2009;Brown et al, 2012;Ray et al, 2013;Sasaki, 2013;Wiik et al, 2013;Grayver et al, 2014).…”

Probing the solution space of an EM inversion problem with a genetic algorithm

“…There are many ways to do this inversion and, correspondingly, many publications about the topic: 1D (e.g., Christensen and Dodds, 2007;Key, 2009) or 3D (e.g., Grayver et al, 2014), constrained by seismic data (e.g., Brown et al, 2012), combined with magnetotelluric (MT) data (e.g., Sasaki, 2013;Wiik et al, 2013) or using CSEM data only (e.g., Ray et al, 2013) just to name a few.…”