Summary
In this research we use the fuzzy c-means (FCM) clustering technique to add petrophysical information from borehole data to model-based seismic impedance inversion. Model based inversion is a common seismic impedance inversion algorithm because it integrates low frequency data from boreholes and is robust. However, beyond the borehole the solutions are as non-unique as many general geophysical inversion problems and output depends greatly on the initial model. Our approach incorporates prior information from well log or core measurement to build a more realizable earth model by using FCM clustering on the petrophysical measurements. This approach tends to produce earth models with less parameter variation and is well suited for crystalline, or hard rock, inversion where there are only a few distinctive rocks units, but considerable structural complexity. Using synthetic examples we show that our method can effectively recover the true model despite structural complexity. The application to real data from the Kevitsa Ni-Cu-PGE (platinum group elements) deposit in northern Finland shows that our inversion results are consistent with well log data and produces impedance models that are more interpretable than the seismic image alone.
Introduction
Seismic reflection methods tend image well the subsurface boundary, but cannot deduce the seismic impedance of media. This limitation may be resolved by utilizing a seismic inversion process such as model based inversion with well control from acoustic logging, a common seismic inversion approach. However, the result of seismic inversion is ambiguous and depends on accurate well data due to the fact that it is band-limited. In order to deal with the ill-posed problem, prior information can be added to constrain the solution (Meju, 1994). In conventional inversion, a smoothness condition is often used to choose a model, but such smooth models are often unrealistic geologically. Therefore, further information such as petrophysics (Sun and Li, 2011) is needed to build a more reliable model. Another critical problem of seismic inversion is that it can depend greatly on an initial model; Cooke and Schneider (1983) stated that the initial guess should be within the "region of convergence", which is difficult to define.
