Robust low frequency model building
seismic data are band-limited in nature and lack the low-frequency component. As the low-frequency component holds the basic information on geological structure, the lack of low-frequency information degrades the quantitative prediction based on seismic inversion. It is therefore essential to build an accurate low-frequency model to have confidence in seismic inversion and in turn on the quantitative predictions made therefrom. The low-frequency model is constructed such that the different subsurface interval impedance values are constrained by the horizons interpreted on the seismic data. This leads to more meaningful inverted impedance data. There are different methods adopted for low-frequency trend generation. The most common one is to use single well’s trend and extrapolate it over 3D seismic data with the help of available horizons. By following this approach, surprises are expected as the inverted impedance may or may not match the impedance logs at the other well locations as shown below.
Inversion analysis carried out using low-frequency trend generated with a single well. (a) the well log curve used for generating low-frequency trend, and (b) a blind well. Notice, the inverted impedance trace in red matches the impedance log trace in blue at the well (a) used for generating the trend, but does not match the impedance log trace in blue for the blind well. (Adapted from Sharma & Chopra, 2017)
Another way to build a low-frequency model is to make use of a few wells for generating the low-frequency model for inclusion in the impedance inversion. Such a technique linearly interpolates the impedance data between the wells, constrained by horizons using weights calculated on the basis of inverse distance, and similarly extrapolates away from the well control. However, when quality checks are performed on the generated low-frequency models using this technique, they often are found to exhibit artifacts in the form of artificial tongues with anomalous impedance values, appearing more like bull’s eyes. Such patterns are not geological and do not generate meaningful impedance sections or volumes. To overcome this problem at SamiGeo, we build an accurate low frequency model that honors the spatial variation of seismic velocity and accuracy of well-log data at the location of wells by following the workflow shown below.
Workflow for generating low-frequency model using multi-attribute regression. (Adapted from Ray & Chopra, 2016)
(Left) Overlay of predicted low frequency impedance trend using preferred workflow mentioned above over the equivalent filtered well-log curve. Overall a good match between predicted (red) and measured (black) curves are noticed for all the wells. (Right) Horizon slice over a 10ms window from the low frequency impedance model derived using a multi-well approach employing (a) kriging (b) multi-attribute regression analysis. Horizons in the broad zone of interest were used for constraining the impedance trend laterally and vertically. On the left slice, notice the pronounced low-frequency impedance values appearing around wells W4 and W5, and high impedance values near well W1, which will surely result in artifacts when used in impedance inversion and hence should be avoided. However, on the right slice a gradual transition of low frequency impedance from one well to another well is observed as expected. (Adapted from Sharma & Chopra, 2017)
References
Amit Kumar Ray and Chopra, S., 2016, Building more robust low-frequency models for seismic impedance inversion published in First Break, May 2016 p29-34.
Sharma, R. K. and S. Chopra, 2017, Quantitative interpretation via more robust low-frequency models, presented at Geoconvention, held at Calgary, in May.