234 Geophysics in Geothermal Exploration The prediction power of impedances, at fine scale and at seismic scale, represents their ability to isolate well-defined clusters corresponding to lithologies or the ability to derive trends. This is validated using cross-plots, before and after upscaling, to ensure that the cluster organization is conserved. In Figure 7.9, the pale color points in background represent the data at a well scale, while the darker points represented the data at seismic scale, sampled at the seismic rate. In practice, the trends should be valid in both scales to be applied on inversion results. Figure 7.9 Petro-elastic model with and without upscaling, showing the possibility to separate sands from shales (left) and to derive a porosity trend in sands only (right). In the section concerning Machine Learning techniques, the propagation to 2D or 3D inversion results will be discussed using a classifier (discrete lithology) or a regression (continuous properties) based on this well data, called “training samples”. 7.3.2 Seismic attributes related to faults and fractures Avoiding fastidious work for the geophysicists, the computation of seismic attributes for fracture detection is efficient but presents various challenges: • To separate geological discontinuities from random noise. Model-based seismic inversion helps reduce the noise content. On synthetic seismic data, parameters for attribute computations can therefore be better tuned to better unveil meaningful fracture response. • To separate the fault and fracture response from other major structural features, such as highly tilted blocks. • To sort or separate the regional from the local features. Several types exist (Chopra and Marfurt, 2007): • Geometrical attributes, such as dip and curvature. • Correlation-based (coherency) attributes, including a steering to tilt the computational window. • Attributes linked to energy, such as envelopes, RMS or spectral decomposition.
RkJQdWJsaXNoZXIy NjA3NzQ=