By substituting rock skeleton modulus expressions into Gassmann approximate fluid equation, we obtain a seismic porosity inversion equation. However, conventional rock skeleton models and their expressions are quite different from each other, resuling in different seismic porosity inversion equations, potentially leading to difficulties in correctly applying them and evaluating their results. In response to this, a uniform relation with two adjusting parameters suitable for all rock skeleton models is established from an analysis and comparison of various conventional rock skeleton models and their expressions including the Eshelby-Walsh, Pride, Geertsma, Nur, Keys-Xu, and Krief models. By giving the two adjusting parameters specific values, different rock skeleton models with specific physical characteristics can be generated. This allows us to select the most appropriate rock skeleton model based on geological and geophysical conditions, and to develop more wise seismic porosity inversion. As an example of using this method for hydrocarbon prediction and fluid identification, we apply this improved porosity inversion, associated with rock physical data and well log data, to the ZJ basin. Research shows that the existence of an abundant hydrocarbon reservoir is dependent on a moderate porosity range, which means we can use the results of seismic porosity inversion to identify oil reservoirs and dry or water-saturated reservoirs. The seismic inversion results are closely correspond to well log porosity curves in the ZJ area, indicating that the uniform relations and inversion methods proposed in this paper are reliable and effective.
Deconvolution denoising in the f-x domain has some defects when facing situations like complicated geology structure, coherent noise of steep dip angles, and uneven spatial sampling. To solve these problems, a new filtering method is proposed, which uses the generalized S transform which has good time-frequency concentration criterion to transform seismic data from the time-space to time-frequency-space domain （t-f-x）. Then in the t-f-x domain apply Empirical Mode Decomposition （EMD） on each frequency slice and clear the Intrinsic Mode Functions （IMFs） that noise dominates to suppress coherent and random noise. The model study shows that the high frequency component in the first IMF represents mainly noise, so clearing the first IMF can suppress noise. The EMD filtering method in the t-f-x domain after generalized S transform is equivalent to self-adaptive f-k filtering that depends on position, frequency, and truncation characteristics of high wave numbers. This filtering method takes local data time-frequency characteristic into consideration and is easy to perform. Compared with AR predictive filtering, the component that this method filters is highly localized and contains relatively fewer low wave numbers and the filter result does not show over-smoothing effects. Real data processing proves that the EMD filtering method in the t-f-x domain after generalized S transform can effectively suppress random and coherent noise of steep dips.
