A Robust Workflow for Quantitative Seismic Reservoir Characterization: Example From Noisy Land Data

Geophysical reservoir characterization provides valuable information about lithology, porosity, fluid saturation, fluid type and flow zone. The ideal seismic data for this task should have a high signal-to-noise ratio, good bandwidth and large spatial aperture. The lack of these requirements can make the results of seismic data analysis and inversion unfeasible for quantitative interpretation and reservoir characterization. This situation happens more frequently in the case of land data, which is significantly noisier and has less frequency content than marine data. We address this problem by proposing a workflow incorporating the latest advances in signal processing, elastic inversion and artificial intelligence. A meticulous prestack seismic data pre-conditioning using multispectral filtering for noise removal, inverse Q filtering and residual moveout corrections ensures the compatibility between observed and modeled AVO signatures at well locations. Then, a full-waveform prestack elastic inversion can consistently recover high-quality elastic properties in the reservoir. Finally, reservoir properties of shale volume porosity and water saturation are estimated from the inversion results using the machine learning random forest method. We apply the proposed workflow on a data set from an onshore complex tectonic-sedimentary area consisting of noisy 3-D land seismic data and a limited number of wells. Results demonstrate that our workflow successfully delineates reservoir property distributions, where a previous reservoir characterization workflow based on conventional routines had failed.