Prediction and Mapping of Karst Features in Mishrif Carbonate Reservoir Through Neural Net Process - Zubair Field

Abstract

Carbonate reservoirs are often characterized by karst features occurrence, usually related to a significant permeability enhancement in presence of low porosity and low permeability matrix type sediments. The distribution of such karst features is generally highly heterogeneous and difficult to predict, making the reservoir management challenging. In Zubair Field (Iraq), there are numerous evidences of karst events within the Upper interval of Mishrif Formation. The production behavior of Upper Mishrif is therefore very heterogeneous, moving from wells with relatively low flow capacity, as expected from petrophysical interpretation, to wells with a very high flow capacity, hence related to karst enhanced permeability. The integration of petrophysical interpretation, well test and multi-rate production logging allowed to preliminary highlight the improved permeability intervals associated to karst. In addition, accurate image log analysis on the same wells investigated a possible relationship between vug densities and production data, to be extended also to wells lacking the latter data. This process allowed to define a karst flag in more than 60 wells. Then, correlations between karst features and different seismic and geological attributes were identified. The most meaningful parameters were used as input data for a Neural Net Process, leading to the definition of a probability 3D Volume of karst occurrence. The final outcomes of the workflow are karst probability maps, used as a driver for the definition of new wells targets and associated trajectories. The recent drilled wells, with optimized paths according to these prediction-maps, have demonstrated the reliability of this approach intercepting the desired karst intervals. This study represents a valuable opportunity in terms of understanding of the reservoir behavior and impact on the ongoing intensive drilling campaign and related field performance.