Reservoir monitoring with controlled source electromagnetics - a case study from a producing oil field in NW Germany

Injection of fluids (e.g. brines, CO₂, steam) is commonly used in enhanced oil recovery (EOR) techniques to push crude oil in place towards the production wells. To optimize EOR procedures, it is essential to know the spatial propagation of injected fluids in the subsurface. Electromagnetic monitoring methods are particularly useful to decipher the spatio-temporal distribution of typically resistive oil versus typically conductive fluids.

We present an overview of soft- and hardware developments, modelling results, and time-lapse field data obtained over five years in an oilfield in NW Germany. CSEM modelling studies showed that conventional surface-based measurements alone do not provide sufficient resolution to changes within a thin (<15 m) reservoir structure located at ~1200 m depth. Combination with sources and/or receivers with vertical components increase sensitivity to such reservoirs very significantly. Based on these findings, a novel horizontal-vertical dipole source using the steel casing of a 1.3 km deep abandoned oil-well was successfully used for current injection in three time-lapse CSEM surveys (2014-2016) across the oilfield. We developed a novel numerical framework to compute the effect of metal casings on CSEM data and included it into our existing modelling and inversion (imaging) software. We also developed a receiver chain to measure the vertical electric field in a shallow observation borehole. Repeatability of the measured data – an essential prerequisite for any monitoring application – was excellent between the repeat surveys despite of high noise levels in an active oil field.

We also show results of a new numerical framework for 4D (time-lapse) CSEM inversion which allows direct imaging of changes within the 3D electrical conductivity structure of a reservoir. A cascaded inversion scheme in combination with a-priori information (conductivity constraints) and weighting of subdomains of the modelling space shows promising results in solving this mathematically ill-posed problem.