Seismi is mic c inve inv ersio rs ion n
Seismic inversion course, UPPA, 2008
Cours ou rse e Outli ut line ne Ge Genera neral Gene eneral rall aspects aspects of seismic i nversion
Concept and purpose
Petrophysical basis
Data requirements
Overview Overview of methods and basic inversion inversion process process
Poststack inversion – practical practical workflow
Data QC
Wavelet extraction
A priori model building building Inversion parameterization parameterization and QC of results
Prestack inversion – pra practical cticall workf low practica workflow
Data QC
Wavelet extraction
A priori model building building Inversion parameterization parameterization and QC of results
A d v an Ad anc c ed i n v er ers s i o n t ech ec hn i q u es
Geostatistical inversion
Joint PP/PS inversion
4D inversion
Cours ou rse e Outli ut line ne Ge Genera neral Gene eneral rall aspects aspects of seismic i nversion
Concept and purpose
Petrophysical basis
Data requirements
Overview Overview of methods and basic inversion inversion process process
Poststack inversion – practical practical workflow
Data QC
Wavelet extraction
A priori model building building Inversion parameterization parameterization and QC of results
Prestack inversion – pra practical cticall workf low practica workflow
Data QC
Wavelet extraction
A priori model building building Inversion parameterization parameterization and QC of results
A d v an Ad anc c ed i n v er ers s i o n t ech ec hn i q u es
Geostatistical inversion
Joint PP/PS inversion
4D inversion
Various rio us inve i nversi rsion on me m ethods tho ds Sparse Spik Sparse Spike e algori thm t o find fi nd an impedance impedance model ma tching ing tth he pike: e:: algorithm algorit algo rithm hm to f ind mode mod el match matchi ng the e seismic bandwith , but w with smalle st number n umber on zero zero reflectiviti es (limite (limit ith the smallest smallest numb er of nnon reflectivit ies (limit ed number of homogene hom ogeneous ous layers) homo geneous l ayers) la Model -base based: d: us e s an al low lo w -freque frequency ncy impeda impedance nce model whic h is i s pe perturb ed uses es an initi ini tial initial p erturbe rturbed Layer based: Layer based: simi ssimilar imilar lar to model m odel based, based, except except that t hat the model is la l a yere yered d and inverted i nverted sim ilar both for thick thickne nesses layers rs and pedance pe dances s nesses of laye and im i mpeda mpedances nces Stochasti c: multi mu lti ple realiz re alizations ations mode els tha th a produce produc e the seismic seismic ltiple rea aliza lizati tions ons of i mpedance mod that att can can r e eproduce amplit mplitudes udes withi n a given error error These four can be post -stack (inversion These (inversion of a full s stack) severa tack) or pre pr e-stack (inversio n of several severall s u b -stacks)
Poststa osts tack ck inversio inversion n 35
Inversio nversion n of full ful l stack or near near stack st ack data
30
25
% y t i 20 s o15 r o P
Resul sult: t: acou cousti stic c impeda im pedance nce model Often related related to por osit osity y
10
5
0 6000
Quasi zero offset
7000
8000
9000
10000
11000
12000
13000
14000
Ac A c o u s t i c Imp Im p edan ed anc ce
15000
16000
Poststack inversion workflow Data QC Seismic to well tie and wavelet extraction A-priori model building Inversion parameterization Interpretation of the inversion results
Required input data Well logs: Vp, Vs, Rhob, Phie, caliper, GR, …. Check-shots, VSP or T=f(depth) Seismic data: full stack or near stack Stacking or migration velocities Smoothed seismic interpretation (horizons, faults) Notions of depositional mode within units
Quality control of all input data is paramount!
Poststack inversion Data QC Wavelet extraction A priori model building Inversion parameterization and QC of results
Seismic data: processing issues Proper amplitude preserved processing with correct zero phasing is necessary
Preservation of relative amplitude and phase information Post-stack FK filter or strong spectral whitening can harm relative amplitude information
Sufficient amplitude-preserving multiple attenuation
Inversion assumes that the seismic consists of primary reflectio ns
Accurate NMO corrections necessary Footprint removal required Trace dependent operators o perators have to be avoided Strong lateral amplitude variations can be a problem
Difficulties with local attenuation, e.g. gas, hydrates, etc.
Type of stacking: full versus near XL4420 Fulls tack + TRITA TVDEF
S/N ratio : 9.9 - Freq: 8-49Hz XL4420 Near Stack
S/N ratio : 10.2 - Freq: 6-51Hz
Amplitude balancing
Before balancing
After balancing
Poststack inversion Data QC Wavelet extraction A priori model building Inversion parameterization and QC of results
Seismic to well tie & wavelet extraction Objectives
Tying the well log (depth) to the seismic data (time) such that acoustic impedance contrasts of the log correspond to seismic markers Extracting wavelet that, convolved with the impedance log, matches the seismic
Search for best seismic / well tie location Wavelet extraction at each well to determine:
Shape, amplitude spectrum
Phase / polarity, time shift
Multiwell wavelet extraction
Optimum wavelet for multiple wells
Seismic to well tie
Search for best Correlation location
Wavelet extraction and tie
Multi-well approach: Beicip (Interwell)
W1 W2 W3 W4 W5 Amp litud e of the wavel ets
Multiwell variable Phase opti mum wavelet
Wavelet extraction & selection Extraction window must be large enough (~ 5 x wavelet length), but focused on the zone of interest Amplitude of wavelets at different wells must be similar Phase of wavelets at different wells must be within 30-40° Reject wells with poor wavelets Compute a multi-well wavelet from the best single wells
Comparison of extracted mono-well wavelets Well 1
Well 2
Well 3
Well 4
Cross-correlation seismic /synthetic
Trough Amplitudes 100%
55%
250%
190%
Reasonably stable wavelet shape but strong amplitudes discrepancies
Influence of the amplitude wavelet on the AI determination Impedance Reflectivity
wavelet Seismic trace
* =
*
Impedance Log
Quantitative influence of the amplitude wavelet on the porosity determination y0.12 t i s o r0.10 o P0.08 l l e 0.06 W d e0.04 l a c s0.02 p U 0.00
7.6%
10500
11500
12500
13500
14500
15500
16500
Acoustic Impedance (g/cc*m/s)
17500
c12 s i e m t s a i e10 m s i t d s e 8 e l y a t 6 i c s s s o r i o m4 P I m2 A o r f
10.9% 7.6% 4.6% Gain factor
0
0
Phie= - 2.448E-05*AcI + 0.407 Correlation 97.4%
0.5
1
y a r t r i b a r (
2
) c e n r e f e r e
Wel l 2 Well -1 0.7 1.0 Assumption : AI shale (11000) is kept constant only AI reservoir is mis-estimated because seismic amplitudes have had a gain applied
1.5
Well-3 2.0
Mis-scaled wavelets
Impact of the wavelet shape and spectrum: resolution Impedance after inversion, 40 iterations using extracted wavelet
Extracted
Impedance after inversion, 40 iterations using Ricker wavelet (central frequency 33HZ phase -22 deg)
__ Variabl e Phase wavelet extrac ted in InterWell (used for Jan 2003 inversion)
__ Ricker 33 Hz phase - 22 deg
RICKER
Poststack inversion Data QC Wavelet extraction A priori model building Inversion parameterization and QC of results
Low Frequency (a-priori) model building Objectives
Filling in the frequency part absent from the seismic data (0 to 10Hz)
Initial solution for some algorithms
Uses well data, horizons, seismic velocities, geology Strong impact: Absolute impedance = Relative impedance (from sesimic) + LF model
LF model
Velocities
Well Interp.
HR Seismic inversion
A priori model: why do we need low frequencies? Initi al AI model with 0 - Hz frequencies Model fil tered to 10 – 80 Hz (spikes + offset+ side-lobes)
Model filtered to 10 – Hz The off set is r elated to the DC component The side lobes are due to t he lack of 1-10Hz fr eq.
Model fil tered to 0 – 80 Hz The litt le spikes are due to the lack of high frequencies: minor imp act
Courtesy P. Mesdag (Jason)
How the full inverted spectrum is built Acoustic Impedance (g/cc x ft/s) 10000
30000
9000
20ms
D e p t h (ft) 10000
Courtesy P. Mesdag (Jason)
Low f requency AI model(0-10Hz) Seismic + Low Frequency AI Model (0-60Hz) AI from Well (0-125Hz) Trend (0-3Hz) Seis. veloci ties
A m p l I t u d e 0
Seismic Bandwidth
20
40
Frequency (Hz)
60
80
A Priori model: structural & stratigraphic part Input AI log
Input key horizons
Correlation Lines along Stratigraphy (// top, // bot tom , proportional)
A Priori model: importance of the stratigraphic mode Input AI log
AI is extrapolated along stratigraphy
Proportional Input key horizons
Eroded (// base)
Onlapping // top
A Priori model: Acoustic impedance high cut 60Hz & 15Hz AI g/cm3.m/s
A-priori model building Right number of horizons, covering the whole area to invert In case of channel belts, the lateral extent of these has to be estimated in order to avoid extrapolating sand impedance everywhere in the model Good quality hori zon picking: it will impact the inversion result Faults make it more difficult
Only included if they are necessary
The velocity used has to be carefully edited and smoothed Careful choice of the highest frequency of in itial model
Model building: horizons need to be smoothed Artifacts in the inversion results related to the initial model
Stacking velocity integration
Initial velocity field
Smoothed velocity field
Ip run #1
IP run #2
F i g
Complex model building
P impedance trend model
Channel fill
Shale trend model
Final complex a priori model
Several steps: sh ale background (from s hale only AI logs), AI vs depth trend, channel sand bod ies added.
Poststack inversion Data QC Wavelet extraction A priori model building Inversion parameterization and QC of results
QC of the inversion result at wells Seismic/well tie
LF model
Ab s. A I secti on + well AI log
Inverted AI + well AI log
Seis. Resid ual
QC of the inversion result: volume
Input seismic
Seismic Residuals
Residuals should not contain geological information
Inverted r eflectivi ties
Abs ol ut e imp edan ce
QC of a post-stack inversion result Check there are no artifacts (ringing) due to a unadequate wavelet Well-tie quality has to be kept in mind when QCing the results at wells If possible, it is useful to check the results at blind wells The residuals should only contain seismic noise and no geology Reservoirs defined by specific AI bodies need to have a geological or sedimentological sense (pre-existing geological model?) Absolute impedance always has to be interpreted together with the relative impedance
Inversion outputs Absolute Impedance
Relative Impedance
Always interpret the two volumes
QC of residuals
Inversion A
Inversion B
QC: Inversion results at a blind well
HC 80Hz filtered AI log Contractor A Inverted trace Contractor B Inverted trace
Ip
Is
Vp/Vs
QC of the impedances: various issues
Ringing
Vertical striping
Artifact
Vertical striping
Interpretation Average impedance at top reservoir
Impedance – porosity cross plot 2 carbonate reservoirs present
In case of homogeneous lithologies (carbonate, tight s ands) one petroelastic parameter can be suff icient to derive one reservoir p arameter: P impedance / Porosity
Porosity realisations from inverted impedances
Definition of pore volume probabilities with uncertainties