Reservoir Characterization Integrating Seismic Attributes and Petrophysical Properties: Case Study from North Ustyurt Jurassic Reservoir

OBJECTIVE Highly heterogeneous m ulti-stacked channel system of the Jurassic sequence has been modeled within one of the fields in the North Ustyurt Basin. The main objective of the work was to predict the distribution of different rock types in highly heterogeneous depositional setting, and to forecast undrained sweet spots to increase oil production. METHODS PROCEDURES, PROCESS The complex reservoirs assemblage has been characterized with the use of seismic, log and core data, and implemented rock typing methodology, based on machine learning technique (ML). 3D seismic data combined with well logs have been utilized for mapping fluvial reservoirs away from the well control, engaging co-rendering of seismic attributes (Envelope, RMS, Coherence), eXchroma and spectral decomposition. Interpreted geobodies were further utilized for rock typing based on classification with supervised training. The model was trained on core-derived estimates from routine and special core laboratory measurements of 7 wells to forecast log data. The outcomes were further applied to generate a 3D static model to predict facies. RESULTS, OBSERVATIONS, CONCLUSIONS Analysis of well data with 3D seismic in the studied area allowed to delineate extensive multi-stacked channel systems within the Jurassic sequence, The channelized systems identified on the log and core data appeared to be highly corelative with the results of 3D seismic imaging. In total, twelve superimposed stacked channel systems of various morphologies ranging from relatively straight to meandering were mapped by co-rendering of seismic attributes. Extracted geobodies were further utilized as main guidance for rock-typing procedure based on ML technique. Petrophysical rock types were estimated on core samples using Winland plot employing porosity vs. permeability and MICP-derived pore throat radius (PTR) cutoffs. The following three major PTR rock types were identified: <0.8 µm, 0.8-2.4 µm, > 2.4µm, corresponding to permeability ranges: <1 mD, 1-10 mD, > 10 mD, accordingly. This outcome was further utilized in training ML on core data to implement the forecast on log data. Among various tested Dataiku classification algorithms Random Forest showed the highest confidence prediction on the confusion matrix and was selected for PTR rock typing. Implemented rock type classification showed general consistency with observed well performance and fluid behavior, enabling geologically consistent history-matching to get forecasts for future reservoir management activities, NOVELTY The main output of this work was delineation of rock typing of the reservoir rocks within the channel systems using ML technique. Our prediction model fits well with production data, demonstrating advantages of applied ML technique in classification and characterization of heterogeneous channelized systems. This Is important in terms of evaluation of the impact on dynamic behavior in the main oil-bearing reservoir parts of the case study.