A reservoir description incorporating the precision and scale of all available data (geological, petrophysical, seismic, tracer and well test data) is critical for reservoir modeling. The heterogeneous variables sampled by these data are composed of several superimposed random functions, each characterized by a unique spatial and temporal scale or support. For example, core, well log, well test and seismic data have supports ranging from inches to thousands of feet.
Current geostatistical methods map lithofacies, porosity, and permeability on a network of grid nodes called the geologic modeling cells. Pseudo point properties that assimilate information from all available data are modeled onto model cells using one of several available conditional simulation techniques. Some methods attempt to combine data with varying support and data with multiple scale support through simple correlations. For example, one approach to incorporate geophysical data is to use a direct transform of the seismic signal to rock properties through a linear regression or crossplot. Reservoir models built using such linear correlations tend to be case specific with little generality.
This paper presents a method for identifying the impact of multi-scale data (data that measure average property over multiple flow units) on reservoir modeling. It examines the information about the reservoir system each data type carries. For example, what fraction of core scale variability is captured by well log data. We also present a consistent method for integrating multi-scale data. Through a series of numerical simulations, we show the impact of heterogeneity of reservoir properties on the fluid flow performance.
Reservoir modeling is a systematic approach to estimate constitutive properties based on a few experimental observations. In other words, it is an algorithm that produces multi-dimensional distribution of rock properties forming a prototype of the actual reservoir. The main goal is to impose the same boundary conditions as in the reservoir and obtain actual responses. The benefits are that once an appropriate model is formed it can be examined for a variety of boundary conditions.