SUMMARYModelling of multi-scale fracture and fault networks is highly uncertain, relying typically on a range of parameters (size, shape, flow properties, connectivity, spacing, orientation) with individually large or sometimes unknown ranges which are very difficult to quantify. Analog data can be used, but resulting models are often unsuitable for well placement or detailed production monitoring in fields under development.Borehole image data are critical in the static (geological) description of the fracture network, but in the absence of dynamic information (e.g. mud losses while drilling, PLT, production, long-term well test) the relative contribution of fractures and fracture networks is speculative. This study highlights the challenges of building a hierarchical description of fractures and connected conduits in a sour gas reservoir which has not started production. In addition, constraining dynamic behaviour is hampered by extreme operational limitations (only short duration well tests and production logging is not feasible).This case study describes a workflow for characterising fault / fracture networks in the pronounced mechanically layered Arab Formation (Late Jurassic) reservoirs of onshore UAE.

A high quality broadband impedance solution by inverting seismic amplitudes is one of the most common industrial standards to integrate geophysical inputs into geological models during field development studies and planning. The use of accurately quantified petrophysical volumes derived from seismic data during geological model building can help improve the understanding of the reservoir and in maximizing the value of seismic data. During this case study for two sour gas carbonate reservoirs- reservoir units C & D, in an onshore Abu Dhabi field, the acoustic impedance volume from seismic inversion and the seismic data itself were further converted into a porosity volume by a probabilistic neural network (PNN) approach. This non-linear transform approach establishes the link between seismically-derived impedance and porosity through an optimized training correlation and error method approach at each well location. A combination of the following factors pertaining to training dataset, established a successful neural network based porosity prediction workflow for the reservoirs: Amplitude preservation of seismic data (target-reservoir specific re-processing)Many calibration wells with good quality calculated log porosityHigh quality inverted seismic impedance. A high correlation of predicted porosity to measured well porosity was achieved at all the well locations. In addition, reservoir unit C porosity grades from a very high but thinner porosity at the top to medium-high thicker porosity at the base, which was well resolved within the available seismic bandwidth and the PNN method of porosity prediction matched this variation of the log porosity. High cross-validation scores achieved during PNN training for porosity prediction, in effect provided the confidence in prediction away from the wells. Furthermore, a close match of predicted and measured porosity at blind test wells drilled in the field extended the confidence in the porosity prediction results. The porosity prediction successfully delineated regionally known non-porous lineaments and faults (encountered while drilling wells) on mean porosity maps of the two reservoirs. The main objective of reservoir-focused reprocessing was to minimize the amplitude damaging effects of the near-surface, mainly due to presence of large sand dunes, and adequately prepare the data for seismic inversion. Successfully estimating seismically derived porosity was an important tool for well planning, geosteering long reach horizontals and also a key input for reservoir characterization to improve our understanding and distribution of reservoir properties in our geologic model.

No abstract

No abstract