Seismic attribute-based AUTHORS characterization of ...

AUTHORS

Ivan Dimitri Marroquın � Earth andPlanetary Sciences, McGill University, 3450University, Montreal, Quebec, Canada H3A 2A7

Ivan Dimitri Marroquın is a Ph.D. student ofgeophysics at McGill University, Canada. Hereceived his B.Sc. combined degree in physicsand geology from the Universite de Montrealin 1994 and his M.Sc. degree in geophysicsfrom Ecole Polytechnique in 1998. His currentresearch includes characterization of conglom-erate and coalbed reservoirs by 3-D seismic-based means.

Bruce S. Hart � Earth and PlanetarySciences, McGill University, 3450 University,Montreal, Quebec, Canada H3A 2A7;[email protected]

Bruce Hart holds a Ph.D from the Universityof Western Ontario. He held positions with theGeological Survey of Canada, Penn State, andthe New Mexico Bureau of Mines and MineralResources prior to joining McGill Universityin 2000. His research interests focus on theintegration of 3-D seismic data with other datatypes for reservoir characterization programs.He has been an associate editor of the AAPGBulletin since 2000.

ACKNOWLEDGEMENTS

Principal funding for this project was providedby a Department of Energy–funded investiga-tion of optimizing infill drilling in naturallyfractured tight-gas reservoirs (contract numberDE-FC26-98FT40486; Larry Teufel, principal in-vestigator). Portions of this project were fundedby a Natural Science and Engineering ResearchCouncil award (238411-01) to Hart. We thankWilliams Energy (in particular, Ralph Hawks) andTom Engler (New Mexico Tech) for supplyingdata and insights into Rosa field. Software usedin this study was graciously provided by Land-mark Graphics Corporation (GeoGraphix) andHampson-Russell. We thank both companiesfor their ongoing support of our research andformer AAPG editor John Lorenz and two anon-ymous reviewers for their helpful commentsthat helped improve the focus of this paper.

Seismic attribute-basedcharacterization of coalbedmethane reservoirs: An examplefrom the Fruitland Formation,San Juan basin, New MexicoIvan Dimitri Marroquın and Bruce S. Hart

ABSTRACT

The Fruitland Formation of the San Juan basin is the largest pro-

ducer of coalbed methane in the world. Production patterns vary

from one well to another throughout the basin, reflecting factors

such as coal thickness and fracture and cleat density. In this study,

we integrated conventional P-wave three-dimensional (3-D) seis-

mic and well data to investigate geological controls on production

from a thick, continuous coal seam in the lower part of the Fruitland

Formation. Our objective was to show the potential of using 3-D

seismic data to predict coal thickness, as well as the distribution and

orientation of subtle structures that may be associated with en-

hanced permeability zones. To do this, we first derived a seismic

attribute-based model that predicts coal thickness. We then used

curvature attributes derived from seismic horizons to detect subtle

structural features that might be associated with zones of enhanced

permeability. Production data show that the best producing wells

are associated with seismically definable structural features and thick

coal. Although other factors (e.g., completion practices and coal type)

affect coalbed methane production, our results suggest that conven-

tional 3-D seismic data, integrated with wire-line logs and produc-

tion data, are useful for characterizing coalbed methane reservoirs.

INTRODUCTION

The San Juan basin is the world’s largest producer of coalbed meth-

ane. Most coalbed methane in this basin is produced from coal

seams of the Upper Cretaceous Fruitland Formation. Production

began in 1951 with the Ignacio Blanco-Fruitland gas field at Ignacio,

AAPG Bulletin, v. 88, no. 11 (November 2004), pp. 1603–1621 1603

Copyright #2004. The American Association of Petroleum Geologists. All rights reserved.

Manuscript received August 25, 2003; provisional acceptance December 31, 2003; revised manuscriptreceived May 3, 2004; final acceptance May 19, 2004.

DOI:10.1306/05190403088

Colorado. Today, Fruitland coalbed methane wells have

a cumulative production of 9 tcf (�0.25 � 1012 m3),

an annual production of 0.9 tcf (�0.025 � 1012 m3),

and contain in-situ gas resources estimated at 55 tcf

(�1.6 � 1012 m3; Dugan, 2002).

The production patterns in the Fruitland vary on

basinwide and local scales, reflecting differences in geo-

logic settings and engineering factors (e.g., completion).

Geologic factors affecting the capacity of coal seams to

store and produce coalbed methane include coal com-

position and rank, gas composition, water production,

coal thickness, and degree of fracturing (e.g., Kaiser

and Ayers, 1994). An understanding of the controls

acting on these seams can be used to guide exploration

and development activities. Despite the increasing im-

portance of coal seams as sources of methane gas, little

seismic data have been acquired to aid in characterizing

coalbed methane reservoirs (Shuck et al., 1996). Shuck

et al. (1996) and Ramos and Davis (1997) used seismic

data to help characterize a coalbed methane reservoir

in the Fruitland Formation at Cedar Hill field in the

San Juan basin (Figure 1). These studies sought to de-

termine fracture location, direction, and density. Shuck

et al. used multicomponent seismic data, whereas Ramos

and Davis used both amplitude variation with offset

(AVO) and modeling analysis. Selected aspects of these

studies are summarized in the Discussion.

This paper aims to show how two of the primary

geologic controls on coalbed methane production, coal

thickness and subtle structures that may be associated

with enhanced permeability zones, may be predicted

from conventional (P-wave) three-dimensional (3-D)

seismic data. We use 3-D seismic attributes to generate

a model that predicts the thickness of the coal seam and

then use curvature analysis to delineate subtle struc-

tural features. We note that the best production trends

are associated with thick coal and seismically detectable

structural lineaments. This methodology should have

application in other coalbed methane plays. Issues that

arose during the course of our attribute analyses shed

insights on how to address certain technical aspects of

these studies.

GEOLOGICAL FRAMEWORK

The San Juan basin lies in the east-central part of the

Colorado Plateau in northwestern New Mexico and

southwestern Colorado. This basin is roughly circular

and covers an area of about 6700 mi2 (17,353 km2;

Laubach and Tremain, 1994). During the Late Cre-

taceous, shoreline progradation (to the northeast) led

to deposition of the northwest-southeast–striking

Pictured Cliffs Sandstone (Ayers et al., 1994). Inter-

mittent transgressive-regressive shifts of the shoreline

resulted in the interfingering of the upper tongues of

the Pictured Cliffs Sandstone and continental facies of

the overlying Fruitland Formation. The Fruitland For-

mation is the primary coal-bearing formation in the San

Juan basin.

The Laramide orogeny began during the Late Cre-

taceous and continued into the Eocene. Various as-

pects of this tectonic event have been discussed by

Hamilton (1987) and Lorenz and Cooper (2003). The

San Juan basin is structurally delimited by the Hogback

monocline to the northwest, Archuleta anticlinorium

to the northeast, Nacimiento uplift to the east, Defi-

ance uplift to the west, and the Zuni uplift to the south.

Coalification of the Fruitland began during the Lar-

amide orogeny, and this tectonic activity may have in-

fluenced the development of cleats (the term applied to

fractures in coals), minor folds, and faults in this unit.

Based on the variability of cleat strike in Fruitland coals,

Tremian et al. (1994) noted the existence of two prin-

cipal face-cleat domains, northwest-southeast strike

in the north of the basin (domain 2), and northeast-

southwest strike in the south (domain 1), separated by

an east-trending transition zone (Figure 1). They stated

that permeability in the face-cleat direction should

be greater than in the butt-cleat direction because of

greater connectedness in the former direction. Laubach

and Tremain (1994) suggested that fracture swarms,

which are associated with increasing fracture and cleat

connectivity and intensity, should be considered as ex-

ploration targets in coalbed methane reservoirs.

DATABASE

Our database consisted of a time-migrated 3-D seismic

data volume and wire-line logs from 59 wells (Figure 1)

within and around the survey area (27 of which were

in the seismic area). Records of gas production from

36 wells in the seismic area were used to analyze spatial

trends in coalbed methane production. Unfortunately,

we did not have wire-line logs for all of these wells. The

3-D seismic grid covers an area of about 17 mi2 (45 km2)

with a bin size of 150 � 150 ft (45 � 45 m). The seismic

data were acquired with a northeast-southwest ori-

entation and then processed to have crosslines oriented

1604 Seismic Attribute-Based Characterization of Coalbed Methane Reservoirs

north-south and inlines running east-west. Seismic

data quality is variable but is uniformly poor around

the margins of our data set.

Various combinations of gamma-ray, caliper, den-

sity, and resistivity logs were available for wells; sonic

logs were not available for any of the wells in our study

area. As such, we generated pseudosonic logs from den-

sity logs using the Gardner equation (Gardner et al.,

1974). This equation provided unreasonably low ve-

locities values for coal because coals fall off the main

trend used to derive the equation. Accordingly, we

assumed a coal velocity of 8695 ft/s (2650 m/s) based

on handbooks published by geophysical logging com-

panies. Sonic logs were used for modeling purposes

Figure 1. Location map showing cleat strike domains of the Fruitland Formation coal seams in the San Juan basin and majorstructural elements bounding the basin (modified from Tremain et al., 1994). The study area is in a boundary region between aregion (domain 1) dominated by northeast-southwest–striking face cleats and a region (domain 2) dominated by northwest-southeast–striking face cleats. The inset at the upper right shows the 3-D seismic area with well locations. Line AA0 is the location ofthe log cross section shown in Figure 2.

Marroquın and Hart 1605

and to make well ties. No checkshot surveys or other

sources of velocity information were available. Image

logs, oriented core, or other independent means of de-

tecting fracture orientation were not available to us.

RESULTS

Geologic Analysis

We identified coal seams and established the vertical

and lateral extents of the Pictured Cliffs Sandstone and

Fruitland Formation using gamma-ray and density logs.

Figure 2 shows a sample log cross section through the

Fruitland Formation in the seismic area. The top of

the Pictured Cliffs Sandstone was identified by high

gamma-ray and low-density log values. This hot shale

signature is thought to be a marine flooding surface.

Upper Pictured Cliffs tongues have a blocky well-log

response and are considered to be stratigraphically lo-

cated between the Picture Cliffs Sandstone and lower-

most Fruitland coal seam (e.g., Ayers et al., 1994).

These strata were found to have an average thickness of

135 ft (41 m). Coal is identified on logs by a combi-

nation of low-density and low gamma-ray values. Fol-

lowing Ayers et al. (1994), we identified coal using

density and gamma-ray cutoff values 2.0 g/cm3 and

75j API, respectively. Two coal-bearing intervals were

delineated: a thick, continuous coal seam in the lower

part of the Fruitland Formation (with an average thick-

ness of 26 ft [8 m]) and an upper succession of thin,

laterally less continuous coal seams interbedded with

clastic strata. The thick accumulation of the lower Fruit-

land coal seam is believed to have been deposited

during periods of shoreline stillstands (cf. Ayers et al.,

1994) and produces most of the coalbed methane in

the study area. The overlying succession of thin, dis-

continuous coal seams is interpreted to have been

deposited in unstable flood-plain settings related to the

renewed progradation of the shoreline. An irregular

low thickness value (2.89 ft [0.88 m]) was observed for

the lower continuous seam in one well at the extreme

west margin of the seismic area. We were not able to

determine the cause of this anomalous value (e.g., fault-

ing); the well is located in an area of poor seismic data

quality, and so it was disregarded from further analyses.

We hand drew the isopach map of the thickness of

the lower coal seam in and around the 3-D seismic

survey area (Figure 3). Our intent was only to generate

a map that shows the principal trends in the data. We

note that the coal is thicker in the southeastern part of

the seismic area and shows a northwest-southeast trend

that is approximately parallel to the inferred Pictured

Cliffs shoreline orientation. The coal seam generally

thins toward the north part of the seismic area.

Seismic Character of Coal Seams and Related Rocks

We generated seismic models by convolving a Butter-

worth zero-phase wavelet of bandwidth 5–15–55–

65 Hz (picked to approximate frequency characteristics

Figure 2. Log cross section through the Fruitland Formation. The logs, from left to right, are the gamma ray (GR) and bulk density(RHOB). Note the presence of the thick coal seam overlain by a succession of thin coal seams and the general thickening trend of thisseam to the southeast (right). See Figure 1 for location.


of our 3-D seismic data; see below) with bulk density

and our pseudosonic logs from wells in the seismic area.

An example of a seismic model is shown in Figure 4.

From this example, we note that the top of the Fruit-

land Formation corresponds to a zero crossing, although

for ease of picking, we interpreted the Fruitland horizon

as a moderate-amplitude continuous peak in the 3-D

seismic data. A moderate-amplitude trough observed

in the seismic model, between the Fruitland and the

top of the thick coal seam, represents the succession

of thin coal seams. We interpreted this horizon in the

3-D seismic data as a continuous and moderate- to

weak-amplitude trough. This reflection is caused by

the interference effects from the individual reflections

Figure 3. Hand-drawn isopachmap of the lower Fruitland coalseam with 5-ft (1.5-m) contourintervals showing general thick-ness trends. Symbols show datacontrol points. The coal seamis thickest in the southeasternpart of the study area.

Figure 4. Seismic model showing thepredicted response of the Fruitland coalseams and related strata. Line generatedusing wells shown in Figure 2. Seismichorizons: FRLD = top of Fruitland Formation,TopTC = top thick coal seam, BottomTC =bottom of thick coal seam, MainPC = topof the Pictured Cliffs Sandstone.


associated with the succession of thin coal seams.

Interference results from short-path multiples, such

that their overlapping waveforms superpose and

outweigh the amplitude of the initial primary reflec-

tion (Gochioco, 1991, 1992). The top of the thick coal

seam, on both seismic model and 3-D seismic data, is a

robust, continuous, high-amplitude trough. The bottom

of the thick coal seam, on the seismic model, corre-

sponds to an intermediate position on the waveform

that is close to the reflection peak, and therefore, we

interpreted this horizon on the 3-D seismic data as the

underlying robust, high-amplitude peak. Moreover,

the bottom of this coal defines the top of the underlying

upper Pictured Cliffs tongues. The top of the Pictured

Cliffs Sandstone, in the seismic model, corresponds to a

trough of weaker amplitude. The corresponding hori-

zon on the 3-D seismic data is a fairly low-amplitude

trough of poor quality. The interpreted horizons in the

3-D seismic data are shown in Figure 5 on an arbitrary

northeast-southwest seismic line trough the survey area.

A statistical wavelet was extracted from the 3-D

seismic data to determine the frequency content of the

data in a time window that covers the horizons of

interest. We observed a predominant peak frequency

( fp) of 40 Hz. By assuming an interval velocity (V ) of

8695 ft/s (2650 m/s ) for coal seams (see above), we

calculate the dominant wavelength (l) to be

l ¼ V=fp ¼ 8695 ft=s = 40 Hz � 217 ft ð66 mÞ ð1Þ

yielding a tuning thickness (l/4) of 54 ft (16 m). This

value is at least twice the average thickness of the lower

Fruitland coal seam in our study area. Therefore, we

conclude that this seam is a seismic thin bed, for which

amplitude variations have been used to estimate the

true coalbed thickness (Gochioco, 1991; Brown, 1999).

SEISMIC ATTRIBUTE ANALYSIS

Seismic attribute studies are useful for integrating

wire-line log and 3-D seismic data to make predictions

of physical properties (e.g., porosity and thickness)

throughout a 3-D seismic survey area (e.g., Schultz

et al., 1994; Hampson et al., 2001). We used a window-

based approach (Chen and Sidney, 1997) to derive a

statistical relationship between coal thickness and seis-

mic attributes (Taner et al., 1979; Brown, 1999; Chen

and Sidney, 1997). The time window for seismic at-

tribute extraction was defined by the picks for the

top and base of the thick coal seam. The thickness of

this seam in each well was determined from the bulk

density log. The multiattribute analysis was based on

27 geophysical well logs in the 3-D seismic survey and

35 amplitude, frequency, and phase attributes. This

list was further increased by applying the following

nonlinear transforms to each attribute: natural log, ex-

ponential, square, inverse, and square root.

Figure 5. Arbitrary seismic transectshowing the interpreted seismic horizons.Transect corresponds to seismic modelshown in Figure 4. Seismic horizons:FRLD = top of Fruitland Formation,TopTC = top thick coal seam, BottomTC =bottom of thick coal seam, MainPC =top of Pictured Cliffs Sandstone.


The choice of attributes to use in our regression

expression was made using statistical methods de-

scribed by Hampson et al. (2001). The attributes were

first analyzed to determine which single attribute cor-

related the most strongly with coal thickness (both

values defined at well locations). This attribute was then

paired with all the remaining attributes to determine

which combination of attributes (e.g., two, three, or

more attributes) best predicted coal thickness. As more

attributes are added into the regression model, the pre-

diction error will continuously decrease and conse-

quently improve the fit to the data. This could lead to

overfitting of the data, however, so cross-validation

was used to determine the optimum number of attri-

butes to keep (Hampson et al., 2001). This method con-

sists of dividing the data into two subsets (e.g., the

training data and validation data). Although the train-

ing data set is used to train the model, the validation

data set is used to estimate the validation error. The

validation error is a measure of the quality of the fit to

the validation data set. Therefore, the point at which

adding new attributes starts to increase the validation

error determines the optimum number of attributes to

use in the analysis (Hampson et al., 2001).

Figure 6 shows the total validation error and pre-

diction error against the number of attributes used in

the stepwise linear regression. We note that the valida-

tion error curve begins to increase after the third attri-

bute, defining this as the optimum number of attributes

to be used. The analysis yielded the following linear

relationship:

where

X1 is maximum absolute amplitude; X2 is integrated

trace; and X3 is total energy;

with a correlation coefficient of R2 = 0.87 and average

error of 3.2%. Figure 7 shows the crossplot of the pre-

dicted thickness vs. the actual thickness values.

The attributes composing the regression expres-

sion are defined as follows:

� The maximum absolute amplitude is the maximum

value in the analysis window interval. Theoretically,

below tuning thickness (e.g., the range of measured

coalbed thicknesses in the present study) amplitude

is directly proportional to bed thickness because of

the effects of destructive interference from reflec-

tions at the top and base of the bed. This trend is

observed in our data (Figure 8a). This attribute is

the single best predictor of bed thickness, with a

correlation coefficient of 0.84.� The integrated trace is essentially a band-limited

(recursive) inversion, with low impedance being

represented by negative numbers. The value we used

is the average value of the integrated trace in the

analysis window. For a thinning wedge of constant

acoustic impedance, the inversion result will best

estimate the impedance of thick beds, but will not be

able to reproduce the impedance of thinner beds,

such as the coal bed studied in this paper. Thick coals

in our data are associated with strong negative values

of integrated trace, and thin coals have weaker

negative values (Figure 8b). As such, there is a strong

negative correlation between integrated trace and

thickness. Strangely, equation 2 suggests an inverse,

Figure 6. Validation errorand prediction error plottedagainst the number of attri-butes used in the stepwiselinear regression. The validationerror indicates that a combina-tion of three attributes bestmodels predicted coal thickness(cf. Hampson et al., 2001).

Thickness ¼ 3:922 þ 2:005e 8X1 þ 3:183e9 ð1=X2Þþ6:812e18 ð1=X3Þ ð2Þ


not a negative, correlation between integrated trace

and coal thickness.� The total energy is the sum of all squared trace am-

plitudes in the window interval. Like the maximum

absolute amplitude, there is a strong positive cor-

relation between this attribute and coal thickness

(Figure 8c). As was the case for the integrated trace

attribute, the inverse relationship between total en-

ergy and bed thickness implied by equation 2 seems

counterintuitive.

We suggest that equation 2 integrates the three

attributes in a way that captures subtle variations in

waveform shape. Examination of that equation sug-

gests that the inverse transform attributes (integrated

trace and total energy) only have a significant influence

on the result when the linear attribute (maximum ab-

solute amplitude) is small. They fine-tune the correla-

tion at low-coal thickness values. This is consistent with

the only slight improvement in correlation coefficient

that was obtained using all three attributes (0.87) vs.

the maximum absolute amplitude (0.84). The details of

this interaction are investigated below through forward

modeling.

We obtained some negative values, because our

model (equation 2) is statistical in nature, and these

values were set to zero in our map. These values, as well

as some exceptionally high thickness values close to the

margins of the seismic area, are thought to be caused by

poor input data quality, although it is also possible that

the model performs poorly when extrapolating beyond

the range of the input data. We thus tested the per-

formance of the model by searching for hidden extra-

polation points (Montgomery and Peck, 1992). This

procedure consists of defining the smallest possible

ellipsoid enclosing the original data points. If a given

data point lies outside of the boundary region, the pre-

diction involves an extrapolation point. For this test, we

used the diagonal elements hii of the hat matrix H =

X(X0X)1X0, where the matrix X is made up of the

attribute values at well locations (Montgomery and

Peck, 1992). The largest value of hii (e.g., hmax) defines

the external boundary of the ellipsoid. Any point that

satisfies

h00 ¼ x0 ðX0XÞ1 x

h00 � hmax ð3Þ

is considered to be an interpolation point, and x is a

vector of attribute values at any location in the seis-

mic area. The largest value of hii, 0.934, is much larger

than the second next larger value, 0.249, and therefore

could be an outlier (Montgomery and Peck, 1992).

However, this observation corresponds to the lowest

measured coal thickness (e.g., 11.85 ft [3.6 m]), and we

demonstrate the importance of this point to the re-

gression analysis in the paragraph below. We therefore

decided to use this value as hmax. Table 1 shows per-

formance test results for six points across the seismic

area (locations shown in Figure 9). We note that the

three first h00 values exceed hmax, and consequently

are extrapolation points. These results confirm our pre-

vious observations concerning the poor data quality on

the margins of the seismic area. The rest of the h00

values lie in the boundary and therefore represent in-

terpolation points.

Composite amplitude is commonly used to esti-

mate bed thickness below tuning (e.g., Gochiocco, 1991;

Brown, 1999), although this attribute was not selected

for our analysis using stepwise linear regression anal-

ysis. Composite amplitude is the sum of the absolute

amplitudes of reflections identified at the top and base

of the reservoir (Brown, 1999). We measured the Spear-

man’s rank correlation coefficient (r s) between various

amplitude attributes and coal thickness to determine

how composite amplitude ranked in comparison to

other attributes we had generated. The Spearman’s

rank correlation coefficient is a measure of the mono-

tonic association between two variables (Lehmann and

D’Abrera, 1975). Unlike the correlation coefficient,

Figure 7. Crossplot of the predicted thickness vs. the actualthickness values. Note the tight fit and the presence of a singlelow thickness value.


Figure 8. Crossplots of the maximum absolute amplitude (a, d), integrated trace (b, e), and total energy (c, f ) against coal seamthickness or wedge model thickness. Attributes extracted from the seismic data are in the left column, and attributes extracted fromthe wedge model are in the right column. Note the similarity in trends between model results and data.


Spearman’s rank correlation coefficient works on ranked

data and measures both linear and nonlinear relation-

ships between variables. Table 2 shows that maximum

absolute amplitude and total energy attributes have a

high-ranking correlation with coal thickness (e.g., r s =

0.74 and 0.72, respectively). However, the integrated

trace has a moderate correlation of 0.62. We also note

that composite amplitude has a strong correlation with

coal thickness (0.68). According to Hampson et al.

(2001), the stepwise linear regression methodology

will not select attributes that provide essentially the

same information to the predicted model. As such, the

composite amplitude attribute must provide the same

information provided by the maximum absolute am-

plitude and/or total energy, both of which scored higher

in Table 2.

We undertook the modeling of various bed thick-

ness and lithology profiles (i.e., do the beds above and

below the coal have the same acoustic impedance?) in

an attempt to determine under what circumstances

composite amplitude might provide a better result than

maximum absolute amplitude. Our results indicated

that different combinations of bed thickness and lithol-

ogy profiles determine which of the two attributes cor-

relates best to bed thickness. However, the results are

complex, in ways that we do not yet fully understand.

We generated a simple wedge model to examine

the physical basis of the observed empirical relation-

ships between attributes and coal thickness (Figure 10).

This model was generated using the Butterworth zero-

phase wavelet employed in our original seismic mod-

els (Figure 5). From the wedge model results, we ex-

tracted the same attributes identified in equation 2,

Table 1. Hidden Extrapolation Test Results Showing the

Performance of the Regression Model to Predict Coal Thickness

Data Point h 00 Category

1 132.780 hidden extrapolation



4 0.254 interpolation



Figure 9. Attribute-basedthickness map of the lowerFruitland coal seam. The thickestcoal is predicted for the south-eastern region of the survey area,in accordance with the welldata (Figures 2, 3). Northwest-southeast–striking thicknesstrends, delineated by yellow,red, and green colors, are evi-dent in the southeast part ofthe survey area. The numberedpoints (1–6) show the locationof data points used to test theperformance of the regressionmodel to predict coal thickness(Table 2). Points 1–3 are identifiedas extrapolation points causedby bad data quality around thesurvey margin. Our porosity pre-diction is not expected to be validin these areas. Points 4–6 areidentified as interpolation pointsthat are within the expected rangeof attribute variation, so the po-rosity values are considered tobe valid in these areas.


then crossplotted them against coal thickness. The

results (Figure 8d, f) show the same trends as those

observed in the data (Figure 8a, c), leading us to con-

clude that they are truly responding to variations in coal

seam thickness (i.e., the observed correlations are not

spurious; cf. Kalkomey, 1997). On a 3-D crossplot of

attributes against coal thickness (Figure 11a), the low-

est coal thickness point appears to be an outlier that

may be associated with bad data. However, an identical

plot of the model results (Figure 11b) shows a trend

that neatly reproduces the trend observed in the data.

This observation leads us to conclude that this was not a

bad data point, therefore justifying its inclusion in the

extrapolation point analyses described previously. Our

previous efforts with seismic modeling to validate

attribute studies (e.g., Hart and Balch, 2000; Hart and

Chen, in press; Tebo and Hart, in press) have likewise

provided important information.

Our attribute-based map of coal seam thickness

(Figure 9) suggests that coal is generally thicker in the

southeast portion of the seismic area, and thickness

patterns show a northwest-southeast trend. This belt of

thicker coal coincides with the local depocenter observed

from the log-based isopach map (Figure 3). We believe

that the thickness trends are geologically plausible be-

cause they are consistent with depositional models and

regional mapping of northwest-southeast–striking

belts of thicker coal associated with paleoshoreline po-

sitions (e.g., Ayers et al., 1994). Furthermore, the area

of thicker coal in the southeast part of the study area

overlies a similar relative thickening at the level of the

underlying Dakota Formation (Hart and Chen, in press),

suggesting that this may have been a localized area of

enhanced subsidence during the Upper Cretaceous.

Table 2. Spearman’s Rank Correlation Coefficient between

Seismic Amplitude Attributes and Coal Thickness

Amplitude Attribute

Spearman’s Rank

Correlation Coefficient

Maximum absolute amplitude 0.74

Total energy 0.72

Variance in amplitude 0.72

Kurtosis amplitude 0.70

Average energy 0.69

Root mean square amplitude 0.69

Composite amplitude 0.68

Average reflection strength 0.66

Average absolute amplitude 0.63

Total absolute amplitude 0.62

Integrated trace 0.62

Trace length 0.58

Amplitude envelope 0.58

Total amplitude 0.52

Reflection strength slope 0.49

Integrate absolute amplitude 0.48

Mean amplitude 0.47

Amplitude 0.41

Skew amplitude 0.23

Figure 10. Wedgemodel and correspondingsynthetic traces. Attributesextracted from this modelwere used to investigatethe relation between attri-butes composing the re-gression model and coalthickness (Figure 8d, f; 11b).


Curvature Attribute Analysis

Important opening-mode fractures contribute to

permeability pathways for gas in coalbed methane

reservoirs. Strata that have been faulted or folded may

also influence gas production (Pashin, 1998). Kaiser

and Ayers (1994) pointed out that minor folds and

faults in the San Juan basin enhance fracture and cleat

permeability in the Fruitland.

Horizon attributes, such as various types of cur-

vature, have been successfully used to detect subtle

faults in reservoirs (e.g., Steen et al., 1998, Roberts,

2001). Hart et al. (2002) showed that horizon at-

tributes might also be used to detect highly productive

fracture swarms in tight-gas reservoirs. Accordingly,

we sought to investigate whether curvature maps could

be used to predict the location and orientation of subtle

structural features too small to be detected on vertical

seismic transects, time slices, or horizon slices. We de-

rived 11 curvature attributes (Roberts, 2001) and com-

puted these attributes from the top thick coal seismic

horizon using aperture values of one, three, five, and

seven bins. The aperture value defines the size of a 3� 3

grid of points used in the computation of curvature

attributes (Stewart and Wynn, 2000; Roberts, 2001).

With an aperture of one bin, our maps were excessively

noisy. As larger aperture values were used, the resulting

curvature horizon shrank inward from the margins

of the survey and curvature lineaments identified only

regional trends. We thus defined the best results in

terms of delineation and distribution of curvature lin-

eaments. We found that the maximum-, strike-, and

dip-curvature attributes defined below (all derived using

an aperture of three bins) provided the most geolog-

ically reasonable results. A depth-converted structure

map, generated by integrating log picks and seismic

horizons (methodology described by Hart, 2000), and

shaded relief maps of the top thick coal seismic hori-

zon helped us to identify and interpret other struc-

tural features.

Figure 12 shows the depth-converted structure

map of the top of the thick coal (TopTC seismic pick).

The map shows that the study area is crossed by a

northwest-southeast–striking structural low that ap-

proximately coincides with the trend of the thick coal

inferred from both log-based and attribute-based thick-

ness maps (Figures 4, 9, respectively). Shaded relief

maps of this surface (Figure 13a, b) show subtle struc-

tural features that are not readily apparent on the

depth-converted structure map, including a variety of

Figure 11. Three-dimen-sional crossplots of coalthickness vs. the maximumabsolute amplitude anda combination of inversetransform attributes (i.e.,integrated trace and totalenergy). (a) Points showingattributes from seismic dataand well-based coal thick-ness. (b) Line showing con-tinuous variation in attri-butes and thickness alongthe wedge model (Figure 10).Note the similarity in trendsbetween the two results. Asingle low-coal thicknessdata point appears to be anoutlier, but its presence ispredicted by the model re-sults. The incorporation ofthe integrated trace andtotal energy attributes intoour regression analysis(equation 3) helps themodel to fit this point.


northwest-southeast– and northeast-southwest–trending

structures. The poor data quality along the western mar-

gin of the survey is obvious from this display.

Maximum curvature shows the locations where

the horizon forms anticline, syncline, or flat-dipping

surfaces as positive, negative, and zero curvature val-

ues, respectively (Roberts, 2001). Lineaments of high

positive maximum curvature values are well defined

across the seismic area (Figure 13c) and generally cor-

respond to features mapped from the shaded relief

maps (Figure 13a, b). Strike curvature describes the

shape of the surface into areas of valley shapes (nega-

tive curvature values) and ridge shapes (positive cur-

vature values) (Roberts, 2001). Curvature lineaments

on the strike curvature map (Figure 13d) display are less

pronounced. However, these lineaments mostly corre-

spond to structures delineated on maximum curvature

display (Figure 13c). Dip curvature shows relief var-

iations in the surface and is a measure of the rate of

change of dip in the maximum dip direction, i.e., at

right angles to the strike curvature (Roberts, 2001).

The dip curvature display (Figure 13e) shows curvature

lineaments that are bigger and more continuous in

comparison to lineaments mapped on other displays.

According to Roberts (2001), this attribute will tend to

exaggerate local relief along the surface.

The lineaments mapped from these displays

(Figure 13a, e) indicate the presence of low-amplitude

structural features. A seismic transect that shows the

expression of typical curvature-defined features is shown

in Figure 14. Lineament azimuths were analyzed to de-

termine if a consistent trend existed between mapped

lineaments and cleat strike from the San Juan basin

(Figures 1, 15a). The analysis was performed on binary

images (e.g., black lines on white background) with

processing image software (methodology described by

Hart et al., 2002). The software extracted the orien-

tation of each lineament, and the lineations so defined

were plotted as rose diagrams. Mapped lineaments from

shaded relief maps (Figure 13a, b) show major trends that

strike toward 45 and 145j (Figure 15c, d, respectively).

These lineaments are subparallel to major face-cleat

strikes identified in outcrop (Tremain et al., 1994) and

probably reflect real structural features. In contrast,

lineaments observed on maximum-, strike-, and dip-

curvature maps (Figure 13c, e) show multimodal pat-

terns (Figure 15e, g, respectively). To identify clusters

in lineament trends, we combined all mapped linea-

ments into a single rose diagram (Figure 15h). Although

this composite rose diagram shows considerable scatter,

major trends centered at 45 and 145j are concordant

with face-cleat strikes as measured in outcrop.

Figure 12. Depth-convertedstructure map (elevation abovesea level in feet) of the top ofthe thick coal seam. Note thenorthwest-southeast structurallow access cutting the seismicarea and contours, indicatingpoor data quality around thewestern and northern marginsof the survey area.



Acquisition artifacts could conceivably influence

the orientation of structures derived from our horizon

attribute analysis. The orientation of acquisition arti-

facts was inferred from amplitude striping on the TopTC

amplitude horizon. Amplitude lineaments show a strong

preferred trend of 45j (Figure 15b), the acquisition

orientation. Although we identified some lineaments

that strike in this direction, other northeast-southwest

trends are different and may be differentiated from

acquisition artifacts because of their orientation (cf.

Hart et al., 2002). Analysis of outcrop and core cleat

orientation (Tremain et al., 1994) (Figure 1) and pro-

duction trends (Kaiser and Ayers, 1994) from the

Fruitland suggest that both northwest-southeast– and

northeast-southwest–striking structural trends could

be present in our study area. We sought to further re-

duce the problem of discerning real structure features

from noise by focusing on lineaments that were iden-

tified via several approaches. Lineaments identified

using only one approach were considered to be possibly

the result of map generation (e.g., gridding) or pro-

cessing artifacts (Hesthammer and Fossen, 1997).

We created a map combining our coal thickness

prediction with curvature analysis results and a bubble

plot of cumulative gas production normalized by the

number of years of production to evaluate stratigraphic

and structural controls on coalbed methane production

(Figure 16). Stratigraphic factors controlled the devel-

opment of the lower Fruitland seam and, indirectly, the

distribution of coalbed methane resources. As previ-

ously discussed, the predicted coal thickness trends are

geologically realistic. The combination of stratigraph-

ic and structural data suggests that the relative de-

crease in coal thickness at the north-central part of

the seismic area is probably caused by paleotopo-

graphic variations on which peat swamps developed;

Figure 13. Horizon attributes derived from seismic horizon corresponding to the top of the thick coal, with their interpretedlineaments. (a) Shaded relief map with illumination from the northwest (subtle structural features show up as reddish lineaments).(b) Shaded relief map with illumination from the northeast (subtle structural features show up as reddish lineaments). (c) Maximumcurvature map. Positive curvature (concave down) lineaments are shown in red and yellow, and negative curvature (concave up)lineaments are shown in dark blue and purple. (d) Strike curvature map. Positive curvature (concave down) lineaments are shown in redand yellow, and negative curvature (concave up) lineaments are shown in dark blue and purple. (e) Dip curvature map. Positivecurvature (concave down) lineaments are shown in yellow and orange, and negative curvature (concave up) lineaments are in greenand blue. Also shown is the location of seismic transect BB0 shown in Figure 14.

Figure 14. Seismic transect BB0 throughFruitland Formation at Rosa field show-ing the expression in cross section ofcurvature-defined lineaments (arrows)at the thick coal level that might be re-lated to minor structural features. Linelocation shown in Figure 13a–e.


areas of topographic highs may have produced thin

coal accumulations. Despite the lack of well control in

this area, the depth-converted structure map (Figure 12)

shows a subtle structural high in this area that sup-

ports this interpretation.

Structural features mapped through horizon at-

tribute analysis also appear to influence coalbed meth-

ane production. The best producing wells (e.g., wells

producing gas exceeding 80 mmcf/yr [2.3 � 106 m3/yr])

are concentrated in areas of thicker coal and in the

vicinity of structural lineaments, especially northwest-

southeast–striking features. Another area of above-

average gas production is near the west-central margin

of the 3-D survey area, but the poor quality of seismic

data in the margins makes the interpretation uncer-

tain. Four wells that are found in a patch of thicker

coal and in the proximity of lineaments (outlined by

yellow dashed line in Figure 16) have lower coalbed

methane production. These wells have the longest

records of production in Rosa field (e.g., 1988–2001),

so engineering factors, including improvements in com-

pletion technologies with time, may be at least partially

responsible for the lack of correlation to thickness and

structural lineaments. As a general rule, most wells with

lower normalized cumulative production (e.g., gas pro-

duction lower than 30 mmcf/yr [0.8 � 106 m3/yr]) are

drilled in areas of lower coal accumulation that are away

from structural lineaments. We conclude that the in-

tegration of attribute-based coal thickness, structural

trends, and production data yields a consistent inter-

pretation that explains much of the variability in coal-

bed methane production.

Figure 15. Rose diagrams derived frommapped lineaments: (a) principal cleatstrike measured in outcrop by Tremainet al. (1994) (Figure 1); (b) amplitudehorizon, showing strong northeast-south-west trend related to acquisition foot-print; (c) shaded relief display with illu-mination source position northwest (Fig-ure 13a); (d) shaded relief display withillumination source position northeast(Figure 13b); (e) maximum curvaturedisplay (Figure 13c); (f) strike curvaturedisplay (Figure 13d); (g) dip curvaturedisplay (Figure 13e); and (h) the combi-nation of all lineaments derived fromhorizon attributes. Two dominant trendsare apparent in (h–a) northwest-south-east set (apparently corresponding todomain 2 orientations) and a bimodalnortheast-southwest trend. The latterprobably contains trends associated withacquisition artifacts (see b) and realstructural trends. Hart et al. (2002)describe a methodology for distinguish-ing acquisition artifacts from real struc-tural trends that have similar orientations.


DISCUSSION

Several studies (e.g., Schultz et al.,1994; Hirsche et al.,

1997; Kalkomey, 1997; Pennington, 1997; Hart, 1999,

Hart and Chen, in press; Hampson et al., 2001) have

examined the methodology and concepts to be used

when conducting a seismic attribute study (i.e., our

attribute-based prediction of coal thickness), and a full

review is beyond the scope of this paper. In this study,

we used seismic modeling to identify seismic horizons

because we lacked velocity logs with which to make

well ties. The seismic models allowed us to identify and

carefully map the horizons of interest. The empirical

relationship we derived between attributes and phys-

ical properties had a high degree of statistical signif-

icance, and we successfully identified hidden extrap-

olation points in areas of poor seismic data quality.

The coal thickness map was geologically reasonable,

and the results, when integrated with curvature anal-

ysis, helped to explain patterns in coalbed methane pro-

duction in the study area.

Our horizon curvature attribute analyses identified

two dominant structural trends, one striking northwest-

southeast and the other northeast-southwest. These ori-

entations are consistent with basin-scale mapping of

face-cleat orientations in outcrop (Tremain et al., 1994)

and production trends from the Fruitland (Kaiser and

Ayers, 1994). Multicomponent 3-D seismic data (in-

cluding a multicomponent 3-D vertical seismic profile)

and AVO analysis from the Cedar Hill field (approx-

imately 50 km [30 mi] west of our study area) also

indicated the presence of northwest-southeast and

northeast-southwest structural trends (Shuck et al.,

1996; Ramos and Davis, 1997). The highest coalbed

methane production from that area is associated with

northwest-southeast–trending structures or from areas

of enhanced fracture density where different structures

intersect. These results are in general agreement with

the results presented in this paper; however, we em-

phasize that our study was conducted with conven-

tional P-wave 3-D seismic data instead of the more

expensive approaches used at Cedar Hill field.

Our simple seismic modeling efforts neglected

elastic attenuation effects in coal seams but were able to

reproduce complex attribute behaviors (e.g., Figure 11).

Hughes and Kennett (1983) used seismic models to

Figure 16. Bubble plot ofcumulative production normal-ized by number of years ofproduction and curvature lin-eaments superimposed overpredicted thickness map. Lin-eaments mapped from shadedrelief lineaments are in violet,maximum-curvature lineamentsare in black, strike-curvaturelineaments are in light blue,and dip-curvature lineamentsare in yellow. Except for thewells circled by the yellowdashed line, the best produc-tion seems to be associatedwith wells located where thecoal seam is thickest (red andyellow) and are associatedwith structural lineaments.


study attenuation in coal and concluded that loss of

seismic energy is small because coal seams are thin in

comparison to the predominant wavelength. We con-

clude that simple acoustic models are useful for un-

derstanding coal-bearing rocks, at least for coals that

are seismic thin beds.

CONCLUSIONS

The controls on coalbed methane production are many

and include a variety of geologic, hydrodynamic, and

engineering factors. In this paper, we have shown the

potential application of using 3-D seismic and curva-

ture attributes to predict coalbed thickness and the

location of enhanced permeability zones in a coal seam

from the lower part of the Fruitland Formation in our

study area. We draw the following general conclusions:

1. Seismic attribute studies may be used to predict coal

thickness. In our case, a combination of three at-

tributes (maximum absolute amplitude, integrated

trace, and total energy) yielded a multivariate linear

regression expression that has a 0.87 correlation co-

efficient with the input data. The map generated

from this analysis has geologically reasonable coal

thickness trends. Coal thickness variations that strike

northwest-southeast are related to the paleoshore-

line orientation. An area of thicker coal accumula-

tion in the southeast part of our study area appears to

be related to a local depocenter that was present

throughout at least some of the Late Cretaceous.

2. Horizon attributes derived from seismic mapping of

the top of the coal, including shaded relief and var-

ious measures of curvature, identified geologically

reasonable structural features in the study area. We

needed to vary the aperture of our curvature calcu-

lations to generate these maps. Structures that strike

northwest-southeast and northeast-southwest cor-

respond to the orientation of face cleats measured

in outcrop.

3. Integration of our attribute-derived thickness map,

our map of seismically defined subtle structures, and

coalbed methane production data yield a picture that

is consistent with known or inferred geological con-

trols on coalbed methane production. The best pro-

duction is associated with thick coal deposits that are

close to seismically defined structures.

4. Seismic modeling is a useful technique for validating

seismic attribute studies. Seismic modeling helped

us to confirm that the attributes selected by the step-

wise linear regression methodology were indeed

physically related to coal thickness, thus minimiz-

ing the possibility of basing our results on spurious

correlations. This same seismic modeling helped us

to identify that an apparent outlying data point

should be kept for the analysis and was not caused

by bad data.

5. Our results were derived from mediocre-quality

conventional (P-wave) 3-D seismic data. Similar re-

sults should be obtainable for coalbed methane res-

ervoirs elsewhere.

REFERENCES CITED

Ayers, W. B. Jr., W. A. Ambrose, and J. S. Yeh, 1994, Coalbed meth-ane in the Fruitland Formation, San Juan basin—Depositionaland structural controls on occurrence and resources, in W. B.Ayers, Jr., and W. R. Kaiser, ed., Coalbed methane in the UpperCretaceous Fruitland Formation, San Juan basin, New Mexicoand Colorado: Bureau of Economic Geology Bulletin, v. 146,p. 13–40.

Brown, A. R., 1999, Interpretation of three-dimensional seismicdata, 5th ed.: AAPG Memoir 42, 514 p.

Chen, Q., and S. Sidney, 1997, Seismic attribute technology forreservoir forecasting and monitoring: The Leading Edge, v. 16,p. 445–456.

Dugan, T., 2002, A history of Fruitland Formation coalbed methanedevelopment, in the San Juan basin, New Mexico and Colorado:AAPG Rocky Mountain Section Annual Meeting, 2002 Tech-nical Program, p. 23.

Gardner, G. H. F., L. W. Gardner, and A. R. Gregory, 1974, For-mation velocity and density—The diagnostic basics for strat-igraphic traps: Geophysics, v. 36, p. 770–780.

Gochioco, L. M., 1991, Tuning effect and interference reflectionsfrom thin beds and coal seams: Geophysics, v. 56, p. 1288–1295.

Gochioco, L. M., 1992, Modeling studies of interference reflectionsin thin-layered media bounded by coal seams: Geophysics,v. 57, p. 1209–1216.

Hamilton, W., 1987, Plate-tectonic evolution of the western U.S.A:Episodes, v. 10, p. 271–276.

Hampson, D. P., J. S. Schuelke, S. James, and J. A. Quirein, 2001,Use of multiattribute transforms to predict log properties fromseismic data: Geophysics, v. 66, p. 220–236.

Hart, B. S., 1999, Geology plays key role in seismic attributes studies:Oil & Gas Journal, v. 97 (July 12), p. 76–80.

Hart, B. S., 2000, 3-D seismic interpretation: A primer forgeologists: SEPM Short Course Notes 48, 123 p.

Hart, B. S., 2002, Validating seismic attribute studies: Beyondstatistics: The Leading Edge, v. 21, p. 1016–1021.

Hart, B. S., and R. S. Balch, 2000, Approaches to defining reservoirphysical properties from 3-D seismic attributes with limitedwell control: An example from the Jurassic Smackover For-mation, Alabama: Geophysics, v. 65, p. 368–376.

Hart, B. S., and M. A. Chen, in press, Understanding seismicattributes through forward modeling: The Leading Edge, v. 23.

Hart, B. S., R. Pearson, and G. C. Rawling, 2002, 3-D seismichorizon-based approaches to fracture-swarm sweet spot defini-tion in tight-gas reservoirs: The Leading Edge, v. 21, p. 220–236.

Hesthammer, J., and H. Fossen, 1997, The influence of seismic noise


in structural interpretation of seismic attribute maps: First Break,v. 15, p. 209–219.

Hughes, V. J., and B. L. N. Kennett, 1983, The nature of seismicreflections from coal seams: First Break, v. 1, p. 9–18.

Hirsche, K., J. P. Hirsche, and L. Mewhort, 1997, The use andabuse of geostatics: The Leading Edge, v. 16, p. 253–258.

Kaiser, W. R., and W. B. Ayers, Jr., 1994, Coalbed methane pro-duction, Fruitland Formation, San Juan basin: Geologic andhydrologic controls, in W. B. Ayers, Jr. and W. R. Kaiser, ed.,Coalbed methane in the Upper Cretaceous Fruitland Forma-tion, San Juan basin, New Mexico and Colorado: Bureau ofEconomic Geology Bulletin, v. 146, p. 187–207.

Kalkomey, C. T., 1997, Potential risks when using seismic attributesas predictors of reservoir properties: The Leading Edge, v. 16,p. 247–251.

Laubach, S. E., and C. M. Tremian, 1994, Tectonic setting of theSan Juan basin, in W. B. Ayers, Jr. and W. R. Kaiser, eds.,Coalbed methane in the Upper Cretaceous Fruitland Forma-tion, San Juan basin, New Mexico and Colorado: Bureau ofEconomic Geology Bulletin, v. 146, p. 9–11.

Lehmann, E. L., and H. J. M. D’Abrera, 1975, Nonparametrics:Statistical methods based on ranks: San Francisco, Holden-Day, 457 p.

Lorenz, J. C., and S. P. Cooper, 2003, Tectonic setting andcharacteristics of natural fractures in Mesaverde and Dakotareservoirs of the San Juan basin, New Mexico: Geology, v. 25,p. 3–14.

Montgomery, D. C., and E. A. Peck, 1992, Introduction to linearregression analysis, 2d ed.: New York, John Wiley & Sons, Inc.,527 p.

Pashin, J. C., 1998, Stratigraphy and structure of coalbed methanereservoirs in the United States: An overview: InternationalJournal of Coal Geology, v. 35, p. 209–240.

Pennington, W. D., 1997, Seismic petrophysics: An applied science

for reservoir geophysics: The Leading Edge, v. 16, p. 241 –244.

Ramos, A. C. B., and T. L. Davis, 1997, 3-D AVO analysis andmodeling applied to fracture detection in coalbed reservoirs:Geophysics, v. 62, p. 1683–1695.

Roberts, A., 2001, Curvature attributes and their application to 3-Dinterpreted horizons: First Break, v. 19, p. 85–99.

Schultz, P. S., S. Ronen, M. Hattori, P. Mantran, and C. Corbett,1994, Seismic guided estimation of log properties: Part 3. Acontrolled study: The Leading Edge, v. 13, p. 770–776.

Shuck, E. L., T. L. Davis, and R. D. Benson, 1996, Multicomponent3-D characterization of a coalbed methane reservoir: Geophys-ics, v. 61, p. 315–330.

Steen, Ø., E. Sverdrup, and T. H. Hansson, 1998, Predicting thedistribution of small faults in a hydrocarbon reservoir by com-bining outcrop, seismic and well data, in G. Jones, Q. J. Fisher,and R. J. Knipe, eds., Faulting, fault sealing and fluid flow inhydrocarbon reservoirs: Geologic Society (London) SpecialPublication 147, p. 27–50.

Stewart, S. A., and T. J. Wynn, 2000, Mapping spatial variation inrock properties in relationship to scale-dependent structureusing spectral curvature: Geology, v. 28, p. 691–694.

Taner, M. T., F. Koehler, and R. E. Sheriff, 1979, Complex seismictrace analysis: Geophysics, v. 44, p. 1041–1063.

Tebo, J., and B. S. Hart, in press, Use of volume based 3-D seismicattribute analysis to characterize physical property distribution:A case study to delineate reservoir heterogeneity at the Apple-ton field, SW Alabama: Journal of Sedimentary Research.

Tremain, C. M., S. E. Laubach, and N. H. Whitehead, III, 1994,Fracture (cleat) patterns in Upper Cretaceous Fruitland For-mation coal seams, San Juan basin, in W. B. Ayers Jr. and W. R.Kaiser, ed., Coalbed methane in the Upper Cretaceous Fruit-land Formation, San Juan basin, New Mexico and Colorado:Bureau of Economic Geology Bulletin, v. 146, p. 87–102.


Date post:	21-Nov-2021
Category:	Documents
Upload:	others
View:	9 times
Download:	0 times

Seismic attribute-based AUTHORS characterization of ...

Documents