We present a methodology for estimating uncertainties and mapping probabili
ties of occurrence of different lithofacies and pore fluids from seismic am
plitudes, and apply it to a North Sea turbidite system. The methodology com
bines well log facies analysis, statistical rock physics, and prestack seis
mic inversion. The probability maps can be used as input data in exploratio
n risk assessment and as constraints in reservoir modeling and performance
forecasting.
First, we define seismic-scale sedimentary units which we refer to as seism
ic lithofacies. These facies represent populations of data (clusters) that
have characteristic geologic and seismic properties. In the North Sea field
presented in this paper, we find that unconsolidated thick-bedded clean sa
nds with water, plane laminated thick-bedded sands with oil, and pure shale
s have very similar acoustic impedance distributions. However, the V-p/V-s
ratio helps resolve these ambiguities.
We establish a statistically representative training database by identifyin
g seismic lithofacies from thin sections, cores, and well log data for a ty
pe well. This procedure is guided by diagnostic rock physics modeling. Base
d on the training data, we perform multivariate classification of data from
other wells in the area. From the classification results, we can create cu
mulative distribution functions of seismic properties for each facies. Pore
fluid variations are accounted for by applying the Biot-Gassmann theory.
Next, we conduct amplitude-variation-with-offset (AVO) analysis to predict
seismic lithofacies from seismic data. We assess uncertainties in AVO respo
nses related to the inherent natural variability of each seismic lithofacie
s using a Monte Carlo technique. Based on the Monte Carlo simulation, we ge
nerate bivariate probability density functions (pdfs) of zero-offset reflec
tivity [R(0)] versus AVO gradient (G) for different facies combinations. By
combining R(0) and G values estimated from 2-D and 3-D seismic data with t
he bivariate pdfs estimated from well logs, we use both discriminant analys
is and Bayesian classification to predict lithofacies and pore fluids from
seismic amplitudes. The final results are spatial maps of the most likely f
acies and pore fluids, and their occurrence probabilities. These maps show
that the studied turbidite system is a point-sourced submarine fan in which
thick-bedded clean sands are present in the feeder-channel and in the lobe
channels, interbedded sands and shales in marginal areas of the system, an
d shales outside the margins of the turbidite fan. Oil is most likely prese
nt in the central lobe channel and in parts of the feeder channel.