Sie sind auf Seite 1von 4

Lower Exploration Risk Using a maximum likelihood clustering and AVO analysis

Randy L. Nickerson* and Scotty Tuttle, North Central Oil Corporation


Summary
AVO is one of the primary tools used for offshore exploration with varying success. Attributes such as: intercept, gradient, fluid
factor, and combinations of these have been used for AVO interpretation. The fluid factor volume has been used successfully in
both class 2 and class 3 gas sands. There are potential pitfalls in using the fluid factor without tight geological calibration. In this
case study, multiple AVO attributes were clustered using an maximum likelihood system. This has proven to be a more reliable
method for saving time and quantifying risk.
Introduction
The Upper and Middle Miocene section in Southeastern Louisiana has been successfully explored using 3D seismic and AVO
analysis. In 1998 a dry hole was drilled using the fluid factor AVO volume (Figure1). After reviewing the results of this well, it
was determined that a better calibration between existing gas fields and dry holes within the survey was required. The study area
involved a 185 square mile survey in the state waters of Louisiana. This survey was reprocessed prior to the drilling of the initial
well for full AVO analysis. The following attribute volumes were generated for use in this study: original RAP migration, new
RAP migration, near angle stack (0 to 15 degrees), middle angle stack (15 to 30 degrees), far angle stack (30 to 45 degrees), fluid
factor (derived from cross plotting intercept and gradient, and fluid factor (derived from cross plotting near and far angle stack).

Figure 1. Fluid Factor Volume showing horizon of interest


drilled in 1998.

Table 1 Amplitude Thresholds for each of the AVO


attributes. *Data loaded 16 bit versus 8 bit.

The calibration study looked at all producing fields and dry holes that penetrated the zones of interest within the 3D survey. The
results of this study were a series of amplitude thresholds for each AVO attribute that statistically indicated the presence of
hydrocarbons. A sampling of prospects and fields within this survey was used as input into an algorithm that groups or clusters
horizon amplitudes from different AVO attributes into modes based on similar characteristics. This process should group gas
fields into unique modes that are different for tight sands, wet sands, and shales. In this paper we will describe the initial
calibration of wells to various AVO attributes, and expand this calibration using this clustering method.
Many articles have been written about the use of AVO with regard to the Miocene section in the Gulf of Mexico. Hilterman et. al.
maintain that pore pressure is the main attribute that determines impedance contrasts between sands and shales within the
Miocene section. There are three major pore pressure zones within the study area that effect the AVO characteristics.
Normal Pressure ( 0 to 7,500 depth)
In the shallow part of the study area the pressure profile is hydrostatic (8 - 9lb/gal). In this zone the shales generally have larger
impedance values than the adjacent sands. A gas sand in this section should have a typical bright spot or Class 3 AVO. A quick
check of bright spot anomalies is done by comparing the near to the far angle stacks. If it is a true Class 3 AVO, a large amplitude

SEG 2000 Expanded Abstracts

Lower Risk using Maximum Likelihood Clustering and AVO Analysis


associated with both attribute volumes will be observed. This section typically has large reflection amplitudes within the entire
section, sometimes making it difficult to separate bright spots associated with gas and normal reflections from thick blocky wet
sands.
Transitional Pressure ( 7,500 to 11,500)
In the intermediate depths within the study area the pore pressure starts to exceeds hydrostatic (9 to 14 lb/gal.) pressure. In this
transitional-pressure zone the wet sands and shales have very small impedance contrasts, with the sands having a larger
impedance than the shales as the pore pressure profile increases. The gas sands in this depth range can be either a Class 2 or Class
3 AVO anomaly. This is the principal zone of interest in this study. Gas fields in transitional pressure in the study area usually
have small to moderate amplitudes on near angle stacks, which become very large on far angle stacks. The fluid factor method
can work well in this zone, but success depends on the recognition of large differences in Poissons ratio which can be caused by
either a tight sand or a hydrocarbon bearing sand. The initial well was drilled in 1998 on the fluid factor amplitude anomaly
shown in Figure 1. This well encountered a very tight sand (average less than 15% porosity), which had the lowest Poissons
ratio of any zone in the well. A typical gas field is shown with both a fluid factor and RAP horizon amplitude extraction (Figures
2 & 3). Both show amplitude anomalies, with the fluid factor map being more pronounced. The seismic data over this field has a
typical class 2 AVO anomaly with very weak near angle amplitudes and very large far angle amplitudes. There are some fault
shadow problems where the two pressure trapping faults converge in the southwest portion of the field. While the field has
structural/faulting trapping components, the major trapping mechanism is a stratigraphic pinchout on the north and east flank of
the field.

Figure 2. AVO fluid factor amplitude map of type field.

Figure 3. RAP migrations amplitude map of type field.

Hard Geopressure (>11,500)


The hard geopressure interval is below the transitional pressure zone (>14lb/gal). The wet sands in this section have larger
impedance values than the surrounding shales. When gas is present the sands still have larger impedance values than the adjacent
shales. Gas sands in this high-pressure zone will have Class 2 (polarity reversals) or Class 1 AVO anomalies. This section was
not calibrated by this study.
Initial Calibration
The initial calibration study incorporated all the producing and dry holes within the 3D seismic survey. Within the survey there
are 167 wells drilled deeper than 7,500. All of these wells were studied to determine which attribute or combination of attributes
best isolated the known gas pay from the rest of the geologic section. The amplitudes used in this study were the total amplitude
inflection for each horizon (maximum peak amplitude minus maximum negative amplitude). Table 1 shows the amplitude
thresholds determined from this calibration for known gas fields. The largest percentage of gas fields within the 3D seismic
survey was identified using single attribute amplitude thresholds for the far angle and fluid factor volume attributes. The highest
predicted drilling success rate is accomplished by combining all of the attributes. To best evaluate a potential drilling location, it
was determined that a combination of far angle and fluid factor amplitude threshold is required.

SEG 2000 Expanded Abstracts

Lower Risk using Maximum Likelihood Clustering and AVO Analysis


Using this methodology, it was found that 90% of the gas produced (>110BCFG) within the survey had amplitude anomalies that
met this criterion. Moreover, if drilling had been restricted to only wells associated with an anomaly that met or exceeded these
amplitude threshold, the chance of finding hydrocarbons would have been greater than 75% for wells drilled deeper than 7500.
Clustering
The clustering was performed using the GAMLS (Geologic Analysis via Maximum Likelihood System) software. This software
can be used for clustering and prediction of any kind of digital data. Samples are grouped into clusters (modes) using fuzzy
probability; each sample is assigned to each mode with a fractional probability. Any number of modes and any number of
variables can be used. The model utilized in clustering is that the probability density distribution of each mode of each variable is
normal (Gaussian). During iteration, the sum of the density distributions for each mode (for each variable) approaches the total
measured density distribution. The modeling is done in N-dimensional space where N is the number of variables used. This
method produces a maximum likelihood solution within which the relationships among the variables are different for each mode.
In this study the following seven attributes were used as variables: near angle, middle angle, far angle, initial RAP migration, new
RAP migration, fluid factor, and fluid factor produced by near/far angle crossplot.

Figure 4. Near/Far crossplot showing distribution of the


modes in the cluster.

Figure 5. Type field showing gas production associated


with modes 8,9, and 10.

This software is normally used to cluster digital well data, but for this study extracted horizon data were converted into LAS
format to be used as input. Horizons amplitude extractions from all seven of the AVO attributes were input into the program for
each of the studied gas fields and prospects, and all of the data were clustered in the same computer run. Ten modes were
selected for clustering so that a reasonable diversity of geological facies would be defined. Each mode should relate to a similar
geological interface, with gas sands separating out from other geological features. Figure 4 shows a crossplot in 2 dimensional
space showing the relationship of near angle stack amplitudes versus far angle stack amplitudes and the derived cluster modes.
However, since the program was run in 7-dimensional space to derive these modes, it is difficult to describe the results using 2
dimensional space. Though the modes appear to overlap in 2-dimensional space, they are discrete bodies of data in 7-dimensional
space.

SEG 2000 Expanded Abstracts

Lower Risk using Maximum Likelihood Clustering and AVO Analysis


Each of the clusters are shown as an ellipse around the center of the density distribution of each mode (Figure 4; the ellipse are
drawn at 2 standard deviations from the mean of each mode). Figure 5 is a map showing the same clustering. Notice that modes
8,9, and 10 (dark gray, light gray, and dark blue). are shown to be present within the gas field. Table 2 shows the mode threshold
values of each attribute, compares them to the initial values picked manually.. There is a very close correlation (within one
standard deviation) between the cluster amplitudes and the initial amplitudes picked manually. Comparing near to far in each of
the modes, it appears the 8th and 10th modes are class 2 anomalies, with the far angle amplitude being significantly larger than the
near angle. The 9th mode correlates with a class 3 AVO anomaly, where both the near and far are large amplitudes, with the near
angle amplitude being slightly larger than the far angle amplitudes. The fact that the near angle amplitudes are slightly stronger in
the 9th mode (which we associate with known gas) than the far angle amplitudes can be explained by processing artifacts, NMO
tuning, and other acquisition factors. It is important to note that the middle angle amplitudes are much higher on average than
those of either the nears and fars, this is a function of how the data was loaded, with more data being clipped for the middle
angle than others. Comparing known production with these clusters, mode 10 has a strong correlation with production, usually in
excess of 35 of gross pay, whereas cluster 8 is associated with wells having between 10 and 35 of gas pay. Mode 9 does
correlate with gas production, but it fits more into a class 3 AVO anomaly, because the near angle amplitudes would be
considered bright spots.

Table 2. Comparing thresholds from initial study to the top


three clusters that correlate with known gas fields.

Figure 6. Original Horizon drilled as a dry hole in 1998.

Fluid Factor versus Clustering Results


Large fluid factor (I/Grad) amplitudes (Table 2) that were clustered into mode 9 correlates closer to a class 3 than class 2 AVO
anomalies. It appears that class 3 gas sands in transitional pressure produce large fluid factor amplitude anomalies, making it a
viable exploration tool. Whereas, the fluid factor from the near/far angle crossplot is much larger for modes 8 and 10 than for
mode 9. The fluid factor derived from the near/far angle stack crossplot produced a larger amplitude for purer class 2 AVO
anomalies (modes 8 & 10). Mode 9 samples had a very small amplitude for this type of fluid factor volume. This fluid factor is
looking for very large difference between near and far angle stack amplitudes, which makes it a very good tool for class 2 AVO
sands. By comparing figures 2 and 5 it is apparent that clustering produced a much better result in determining the extents of the
gas field. The fault shadow problem areas that the fluid factor had trouble with, were handled much better and were clustered into
modes associated with gas in these zones. Figure 6 shows the results of the objective horizon of the dry hole. This horizon
clustered predominately into mode 7, which describes a class 1 AVO anomaly, or high impedance sand, and which is consistent
with the well data.
Conclusions
Calibrated AVO has been successfully used in the Miocene of Louisiana to find hydrocarbons. While the fluid factor volumes
can be very useful in locating potential gas fields, clustering using a maximum likelihood system has proven to be more reliable.
The clustering method uses multiple attributes (versus only 2 attributes for fluid factor crossplotting) which results in more robust
results. Further, the traditional crossplot of intercept and gradient involves somewhat dependent attributes. Clustering limited
angle stack amplitudes involves independent attributes, which gives more accurate results. Clustering doesnt have to be limited
to AVO attributes; it can be expanded to include: instantaneous phase, instantaneous frequency, or any other post stack derived
attribute. The results of this new clustering technique lead to the discovery of two new gas fields.

SEG 2000 Expanded Abstracts

Das könnte Ihnen auch gefallen