Seismic Inversion
Spectrum uses the Hampson–Russell STRATA software which is an interactive 2D and 3D modelling and inversion program. The software transforms seismic traces to velocity or acoustic impedance traces. There are options for wavelet extraction, well log editing and interpretation as well as five methods of seismic inversion.
Pre-stack STRATA uses angle gathers or angle stacks to produce acoustic impedance, shear impedance and density. Lambda, mu, rho and Simultaneous Inversion are especially useful for analysing data with AVO anomalies.
Band-limited InversionThis is the classical recursive inversion procedure which involves three steps:
The first step is to build a low frequency velocity model using sonic logs or RMS velocities (or a combination). This is done by picking major reflectors on the seismic section and providing velocity information at control points, with the velocities between these points being interpolated.
The second step is to filter the velocity model to differentiate between ‘low’ and ‘middle’ frequency components. Seismic data generally do not have frequencies below 10 Hz.
The third step is to invert the seismic traces to produce the middle frequency range using the formula below:
Seismic amplitudes are scaled down to match true reflectivity by computing a scalar from, for example, the RMS amplitude. The low and middle frequency ranges are then combined.
Model Based (Blocky) InversionA velocity model is built as for the band-limited inversion, only the pseudo-logs produced at each common-mid-point or bin location are used to produce synthetic seismograms which are then compared with the real seismic data.
The error between the synthetic seismogram and corresponding real seismic is minimised using a conjugate-gradient algorithm which, in effect, means that the pseudo-logs are modified at each bin location.
The impedance derived for any one layer is dependent upon the impedance derived for the layer immediately above it. This can lead to large cumulative errors in the impedance, often called the ‘low frequency trend error’ and this effect is related to the non-uniqueness of inverting seismic data. There may be many combinations of reflection coefficients that minimise the error in a similar way.
Constraints can be used to limit the number of possible results. ‘Hard’ constraints are so called because they limit the variation allowed in derived impedance from the starting impedance value, e.g. +-25%. Stochastic inversion, which is model based inversion using ’soft’ constraints, sees the synthetic and real seismic traces as two possibly conflicting pieces of information and applies some user-specified weighting to the two.
Sparse Spike InversionSTRATA provides two sparse spike inversion algorithms.
Linear ProgrammingThis algorithm creates a sparse reflectivity that produces the best match between the derived synthetic and the seismic trace, subject to the constraint that the number of spikes should be a minimum.
Maximum LikelihoodThis algorithm uses the model to perturb a reflectivity series estimated from the seismic data. Estimates are made about the wavelet and the sparse spike reflectivity.
The maximum number of spikes expected is provided by the user. Using the ‘single-most-likely-addition’ technique, each step in the iterative process attempts to find the next optimum spike to add to the reflectivity series.
Neural NetworkThis algorithm applies a Probabilistic Neural Network to the seismic trace to produce the impedance trace. Before this operation can be used, at least one network must be trained at a well location. Best results are obtained with multiple wells.
Coloured InversionThis algorithm approximates an unconstrained sparse-spike inversion by deriving an inversion operator that matches the amplitude spectrum of the seismic to that of the acoustic impedance.
Spectrum have experience of 3D Inversion studies from the Tertiary sand plays and Chalk of the North Sea as well as various surveys from the Middle East.
