You are on page 1of 26

Depth Error

Accuracy
of
Depth
Conversion

PGS Reservoir (U.S.) Inc. Doe Contract #DE-AC-22-94-PC 91008

Error Analysis Depth Error

Well Misties
Depth conversion with analytic functions or with these
regionally calibrated seismic velocities will not tie individual
wells exactly.

These misties are a measure of the uncertainty in the


velocity model and depth map.

In some projects, e.g. exploration mapping, the map may be


perfectly adequate with these misties. In other projects, e.g.
field unitisation, a map is required that ties all the well
control.

13.1
Error Analysis Depth Error

Mistie Analysis
Errors in the velocity model or preliminary depth map should
be analysed to ensure that there are no regional trends
remaining. We want a model that only contains random errors.

If our mistie analysis contains any regional trends then our


velocity model is not a good representation of the velocity
variations in the earth.

If our errors are not random then we need to try a different


approach to building the velocity model.

Error Analysis Depth Error

Map Misties

A map of the misties may


clearly indicate that there is
a trend remaining in the
data.

In this example the positive


misties are on the crest of
the structure and the
negative misties on the
flanks.

From a Paradigm Geophysical brochure

13.2
Error Analysis Depth Error

Histogram of Misties
Histogram of size of
errors. A normal
distribution with a
small mean indicates
the errors are probably
random.

Any other distribution


indicates that there is
a non random element
in the velocity model
i.e. it could be
improved.

Error Analysis Depth Error

Mistie Cross-plots
Error
To find out why the errors
150
are not random it may be 100

necessary to cross plot 50


0
both velocity errors and -504000 5000 6000 7000 8000 9000 10000 11000
Mistie ft

-100
depth errors against -150
velocity, depth, TWT, -200
-250
spatial parameters -300

(northing, easting, -350


Depth ft
distance from geological
feature) etc.

13.3
Error Analysis Depth Error

Mistie Semivariogram

Derive a semivariogram
The gradient of the sill
for the misties. Any

Semivariance of Variable
indicates a regional trend.
gradient in the sill, or no
sill at all, is indicative of
a trend in the misties.

Range

Distance

Error Analysis Depth Error

Non-Random Error
If the errors are not random we may:

Revise the velocity model or choose another method of


depth conversion so that the errors are small and random.

Model the errors with respect to an independent parameter


so that the errors are small and random.

Krige the errors. This will separate a regional trend (which


may or may not be meaningful) and a random element.

Model the errors with a trend surface to obtain regional and


random elements.
13.4
Error Analysis Depth Error

Quantifying Accuracy
We frequently wish to quantify the depth conversion
accuracy.

Kriging produces a map of standard deviation of depth


error.

We can also obtain a map of standard deviation of depth


error by error analysis.

Error Analysis Depth Error

Error from Kriging


Kriging minimizes the
variance of the estimation
error. Since the variance
is calculated at each grid
point it is an easy matter
to output the standard
deviation of the depth
estimation error. The
depth error is always a
minimum around the
wells.
Example after Laurtent Moinard, Application of Kriging to the Mapping of a Reef from Wireline Logs and Seismic Data : a
Case History, in Geostatistical Case Studies, G. Matheron and M. Armstrong (editors) 1987, D. Reidel Publishing Co.

13.5
Error Analysis Depth Error

Quantitative Error Analysis


Consider the formula from average velocity:-
z = VA x t
From our seismic interpretation we can estimate the standard
deviation of our travel times based on well ties or our
uncertainty in the pick.
If we can estimate the standard deviation in the average
velocity then there are some simple rules to follow so that we
can find the standard deviation of our depth estimation error.

Error Analysis Depth Error

Velocity Error
A simple Velocity Error Distribution
approach is to 35
use the RMS
30
velocity error or
Mean = - 5 ft
the standard 25
SD = 73 ft/sec
deviation from 20
our mistie error
15
analysis.
10

0
-150 -100 -50 0 50 100 150 200
Velocity Error ft/sec

13.6
Error Analysis Depth Error

Velocity Function Error


Average Velocity One Way Tim e in secs
0 0.2 0.4 0.6 0.8 1
Alternatively when we determine 0

our analytical functions we can 1000


also find the standard deviations
of the parameters. 2000

[If we use a ‘trendline’ approach 3000

(based on normal equations) the

Depth in feet
4000
estimates of the standard deviations of
the parameters are the square roots of 5000
the diagonal of the covariance matrix.
6000
Experience tells us that this approach
underestimates the error in velocity 7000
z = (6839.2±1.1)t2 + (3782.5±0.8)t
models.]
8000

9000

Error Analysis Depth Error

Velocity Function Error


If we use an optimisation Average Velocity
by Optimization
approach the estimate of the Time secs
Depth ft

standard deviations is from the 0


0 0.2 0.4 0.6 0.8 1

tangents to a form of the


1000
equivalence diagram (solution
trough) in parameter space. 2000

3000
The results returned by these
4000
two approaches are quite
different. 5000

6000
The choice depends on whether
7000 z = (3825.6±96)t
you are buying or selling! + (6780.0±133)t2
8000

9000

13.7
Error Analysis Depth Error
Parameter Error Analysis
The equivalence diagram Delta Chi^2

shows the normalized SD 6640

values with respect to the 6650


6660
optimum solution. The 6670
contour interval is 1. It is the 6680

tangents at one standard 6690


6700
deviation that are used to

UN
6710 K

IT
estimate the SD of errors in

CO
SD
6720

NT
OU
the parameters. 6730

R
6740

Experience tells us that this 6750

approach gives realistic estimates 6760


6770
of errors and is therefore the
6780
preferred method.
3925

3915

3905

3895

3885

3875

3865

3855

3845

3835

3825
OPTIMUM
Vo SOLUTION
SD

Depth Error
Parameter Error Analysis
When we considered the formulae for instantaneous velocities
we had two equations for second and subsequent layers.
∆z = (V0 /K + ztop)(eK∆ t - 1) when the velocity model is referenced
to the surface, or datum, and ∆z = Vtop (eK∆ t - 1) /K when the
velocity model is referenced to the top of the layer.

Referencing the velocity model to the top of the layer reduces


the uncertainty in V0 whilst using velocity logs rather than
checkshot / VSP time-depth data reduces the uncertainty in k.

13.8
Depth Error
Parameter Error Analysis

For this example the SD reduces from ~1000 m/s in V0 to ~350 m/s
so the choice of velocity model can have a big impact on the
uncertainty or error in the final depth structure map.

Depth Error
Parameter Error Analysis

For this example the SD reduces from ~0.25 /s in k to ~0.155 /s so


the use of velocity log data rather than VSP data can have a big
impact on the uncertainty or error in the final depth structure map.

13.9
Depth Error
Parameter Error Analysis

By using digital log data rather than VSP data and referencing the
velocity model to the top of the layer rather than the surface
reference datum we can greatly reduce the uncertainty in the
velocity model parameters.

Error Analysis Depth Error


Time Error

We can either determine our time pick errors by comparing


them to travel times from the well velocity surveys, i.e.
checkshot/VSP travel times and then statistically determining
the S.D. of the travel time errors, just as found the S.D. of the
velocities by comparison of estimated/modelled velocities to
observed well velocities.

In practice we often will not have sufficient well velocity


surveys for this approach to be meaningful.

13.10
Error Analysis Depth Error
Time Error
In the absence of lots of well velocity surveys we can estimate
the expected error (and therefore S. D.) in our time picks.

If we are close to well control and there is a good match


between our synthetic seismograms and the seismic data then
we are probably comfortable that the tie is correct to within half
a cycle or better. Depending on the dominant frequency of the
seismic event this is probably of the order of 10msec.

If our error is <10msec then our S.D. is <3msec

If we are less certain about our time pick then the S.D. will be
greater.

Error Analysis Depth Error


Velocity Grid Errors

SD = 877 ft/s SD = 704 ft/s

The SD of the error in the smoothed grid is 729 with respect to


the unfiltered grid and 568 with respect to the filtered grid.
13.11
Error Analysis Depth Error

Combining the Errors


There are some rules that have to be followed to combine the
different standard deviation (S.D.) estimates. The following
holds for independent variables.

If C = fn(A,B) such that the S.D. of A is a and that of B is b then


the S.D. of C, c = √[(δf/δA*a)2 + (δf/δB*b)2]

Rules we can use are: -


1. C = A*B or A/B then c = AB√(a2/A2 + b2/B2)
2. C = A ± B then c = √(a2 + b2)

Error Analysis Depth Error

Example
z = VA x t
VA = 6652 ft/s with a standard deviation of 64 ft/s and t = 0.775
sec with a standard deviation of 0.0035 sec
Then z = 5155 ft with a standard deviation given by
5155.√[(64 / 6652)2 +(0.0035 / 0.775)2] = 55 ft

Unlike the depth error map obtained by kriging the depth errors
in this approach do not decrease towards the well ties.

13.12
Error Analysis Depth Error

Cross Validation
Depth conversion by analytic functions is probably the most
accurate form of depth conversion when there is sufficient
well control to determine the model parameters.

A technique called ‘Cross Validation’ is used to answer the


question “will I get a better depth conversion if I have more
well data?”

Error Analysis Depth Error

Cross Validation
The whole depth conversion is
repeated a number of times
using different numbers of
wells. The wells may be
selected on a historical or
ERROR

statistically random basis. The


overall depth conversion error
is measured each time against
NUMBER OF WELLS
the total well database. The
curve of error v. the number of
wells levels out if there are
sufficient wells.

13.13
Error Analysis Depth Error

Summary
By analysing the errors in our depth conversion we can see
whether or not our velocity model is a reasonable one.

A good macrovelocity model will have small random errors ,


with a mean close to zero, a near normal distribution and no
residual trend.

We can quantify the accuracy of our velocity model and


depth map.

In field areas where we have lots of well control available


we can also determine whether or not our velocity model
will improve with the addition of more well control.

Error Analysis Depth Error

13.14
Exercise 13.1 Error Analysis

Time (s)
-1000

0 0.2 0.4 0.6 0.8 1


0

1000

Depth (ft) 2000

3000

4000

5000

6000

7000
y = 1658.21x 2 + 6193.65x - 98.69

After the well in the Depth Conversion exercises (chapter 12) is drilled we find:-
z = P + Q.t + R.t2.
P = -98.69 and p = 0.59, Q = 6193.65 and q = 2.60, R = 1658.21 and r = 2.56
Using the two rules for combining standard deviation estimates find the standard
deviation of the depth z.
Remember that t is one way time (0.85 sec). If you are uncertain that you have picked
the correct event then the estimate for the standard deviation in t will increase. The
error in T may become 60 msec, so that the error in t is 30 msec and the standard
deviation in t is σt = 10 msec.
(The error bar is usually taken as 3 x Standard Deviation and the symbol σ is usually
used for standard deviation.)
Hint.
Find the standard deviation in the product Q.t, then find the standard deviation in the
product R.t2 and finally find the standard deviation in the sum P + (Q.t) + (R.t2)
How does the above result compare with your earlier estimate of the error in your
depth prognosis?

13.15
Depth Error

Stochastic
Depth Conversion

Stochastic Depth Error

Stochastic Depth Conversion


The uncertainty of depth conversion can be estimated through
stochastic modelling. The method is presented as an extension
of geostatistical (kriging) mapping.

Structure-in-time and velocity data are input to the process.

Output are a depth map and maps of probability relating to the


structural closure, e.g. the 10%, 50% and 90% probability of the
structure being above the spill point.

13.16
Stochastic Depth Error

Structure in Time
Note the elongated high and the large structural range.

Stochastic Depth Error

Velocity Distribution
A histogram of the stacking velocity distribution shows that it is
far from normal and unsuitable for use in kriging.

13.17
Stochastic Depth Error

Velocity Variogram of VS
The sills of the variogram are not flat and the data cannot be
modelled

N-S

E-W

Stochastic Depth Error

Stacking Velocity
Stacking velocity cross-plotted against time shows virtually no
correlation.

13.18
Stochastic Depth Error

Stacking Velocity
After preparation (dip and bias corrections? Normalisation +
smoothing?) the velocities show a good correlation with travel-
time.

Stochastic Depth Error

Velocity Distribution
A histogram of the velocity distribution after preparation shows
that it is near normal and suitable for use in kriging.

13.19
Stochastic Depth Error

Velocity Variogram
The variogram sill is now flat and it can be modelled.

From a Scott-Pickford brochure

This map of variance is an


alternate form of 2D
semivariogram.

Stochastic Depth Error

The output from kriging is very dependent on the variogram


model.

This model controls the way kriging grids the data.

In stochastic depth conversion it is all important for the multiple


realisations of the depth maps which must honour the points of
well control (the observed well depths) but can assume any
depth values between the points of well control that are
consistent with this variogram model

13.20
Stochastic Depth Error

Stochastic Modelling 1
Produces the best result but needs a lot of wells.

Stacking,
Simple or Smoothed
RMS or
Variogram Common Seismic
Average
Kriging Velocity
Velocity

Well
Time
Average Cross Plot
Map
Velocity

Probability
Multiple
Variogram Cokriging and Mean
Realisations
Depth Maps

Stochastic Depth Error

Stochastic Modelling 1
Produces the best result with few wells.

Stacking,
Simple or Smoothed
RMS or
Variogram Common Seismic
Average
Kriging Velocity
Velocity

Well
Time
Average Cross Plot
Map
Velocity

Kriging Probability
Multiple
With and Mean
Realisations
External Drift Depth Maps

13.21
Stochastic Depth Error

Stochastic Modelling 2
If the variogram from the seismic has a large nugget or sill the
results can be very unreliable. Needs lots of wells.
Stacking,
RMS or
Variogram
Average
Velocity

Well
Time
Average Cross Plot
Map
Velocity

Probability
Multiple
Variogram Cokriging and Mean
Realisations
Depth Maps

Stochastic Depth Error

Stochastic Modelling 2
If the variogram from the seismic has a large nugget or sill the
results can be very unreliable. Works with few wells.
Stacking,
RMS or
Variogram
Average
Velocity

Well
Time
Average Cross Plot
Map
Velocity

Kriging Probability
Multiple
With and Mean
Realisations
External Drift Depth Maps

13.22
Stochastic Depth Error

Stochastic Modelling
Typically 100 realisations are run. The average of all the depth
maps produced is very similar to the deterministic depth map.
A statistical analysis of the results gives the probability map.

Stochastic Depth Error

Probability Map
Iso-probability contours of the structure having independent
structural closure.

50%
10% 90%

13.23
Stochastic Depth Error

Stochastic Depth Conversion


The uncertainty of depth conversion can be quantified through
stochastic modelling.

The results are only believable (i.e. agree closely with other
methods of depth conversion, match observed oil / water
contacts, etc.) when the variograms have: -
• small nuggets
• small sill values
• are well modelled.

Depth Error

Probability Map
It is not necessary to use stochastic modelling to define the
probability map associated with our depth map.

We can compute the probability map (grid of values) directly


from: -
our deterministic depth map
our standard deviation of depth error map
the closing contour or Oil Water Contact value/grid

13.24
Depth Error

Probability Map
We can compute the probability map (grid of values) directly
from: -
our deterministic depth map (x)
our standard deviation of depth error map (σx)
the closing contour or Oil Water Contact value/grid (X)
using the spreadsheet function: - NORMDIST(X, x, σx, TRUE)

Or we use the Normal Distribution function:


φ(z) = exp[-(X-x)2/2σx2] / [σx√(2π)]
where z = (X – x) / σx
The probability is then found from the statistical tables of Φ(z)

Depth Error

Probability Map

structure Probability Φ(z)

height = x
S.D. = σx
contact / closure = X
normal
distribution
φ(z)

13.25
Depth Error

Probability Map
Spill point 9200 9200
SD of depth
conversion
error 255 123

depth probability probability


9500 0.12 0.01
9400 0.22 0.05
9300 0.35 0.21
9200 0.50 0.50
9100 0.65 0.79
9000 0.78 0.95
8900 0.88 0.99
8800 0.94 1.00 By reducing the depth conversion
8700 0.98 1.00
8600 0.99 1.00
error we increase the risk weighted
8500 1.00 1.00 reservoir volume.

Depth Error

Summary
We can analyse the errors, or probable errors, in our depth
maps in a number of ways:
Well misties
Smoothing of seismic velocities
Uncertainty of function parameters
Accuracy of picking of interpreted events

We can use these estimates to answer the question “Will I


get a more accurate depth map if I had more well control?”

We can also use these estimates to determine the probability


of a map location being above the closing contour or oil/gas-
water contact.

13.26

You might also like