Blending of seismic attributes with additive primary colors (red, green and blue) is a standard visualization procedure used by interpreters to integrate the information contained in them and carry out comprehensive interpretation. Attributes plotted against RGB should be of the same units and have a similar range of values. Example triplets include plotting spectral magnitudes at three frequencies, or amplitudes at three offsets. Alternatively, attributes will have to be normalized before they are input into the primary color channels of the chosen color model. Instead of displaying and interpreting such selected or optimized seismic attributes individually, they can be displayed in a composite way using RGB corendering, depicting all the information in the data at the same time. Such RGB applications have been carried out for detection of anomalies, defining complex lithofacies, identifying discontinuities and a long list of both structural and depositional geological features.
In the December 2019 and February 2020 installments of Geophysical Corner, we described how color is perceived by the human eye, and how colors are rendered on interpretation workstation monitors. We also described the additive RGB color model which forms the working model for computer monitors, as well as the subtractive CMYK (cyan, magenta, yellow and key color for black) model and the HSV and HSL color models frequently used for covisualizing seismic attributes.
In RGB color corendering, the three colors are commonly represented as red (255, 0, 0), green (0, 255, 0) and blue (0, 0, 255). The intensity of each color varies between 0 and 255, so that variations in red, green and blue individually have this range. On corendering, these color variations blend together to produce the multitude (2563= 16,777,216) of colors we see on the display. In some cases, there is a well accepted rule assigning R, G or B to a given component, with the most common example being the display of spectral magnitude components. Here, most interpreters will map a low frequency to red, a middle frequency to green and a high frequency to blue, mimicking the order of visible light components.
Sometimes, the anomalies we wish to emphasize occur at low rather than at high attribute values. Multispectral coherence exhibits this issue well, where for concreteness, let’s assume that anomalous low coherence values (<< 1) indicate faults and high coherence values (≈1) indicate coherent reflectors. If we plot the coherence computed from the 20, 40 and 60 hertz spectral voice components against R, G and B, low coherence values at 20 hertz will remove red from the image, leaving cyan, low coherence values at 40 hertz will remove green, leaving magenta, and low coherence anomalies at 60 hertz will remove blue, leaving yellow, resulting in what is effectively a CMY display of fault anomalies. Voxels that exhibit equally low coherence values at all three frequency components will appear black, whereas those that exhibit equally high coherence values will appear as white.
Reversing Polarity to Highlight Features
Although the RGB model is routinely applied to three spectral components to estimate a tuning frequency for a high amplitude reflector, sometimes the target is the analysis of a low amplitude reflector. In this situation, we can reverse the polarity of the color bar by simply flipping the range 0 to 255 to be 255 to 0. In this low reflectivity example, reversing the color bar for all three spectral magnitude components gives a CMY display that enhances subtle features in the weaker reflector. Problems arise when the geologic feature we wish to highlight – say, a fault – gives rise to anomalously low values (for example, coherence, envelope, GLCM homogeneity) for some attributes and anomalously high values for others (dip magnitude, GLCM entropy). In such situations, we may need to reverse the polarity of one or more of the three colors (RGB) before blending. In the case of a fault, we may wish to map energy ratio coherence as computed against R, but then reverse the polarity of dip magnitude against green, and GLCM homogeneity, against B, resulting in a CMY fault image, where strong fault response on all three attributes would appear as black.
Principal components present a special challenge in that they are linear combinations of a suite of input attributes defined for PCA by eigenvectors that best represent the variation in the data. Elements of an eigenvector that are negative will reverse the polarity of the corresponding input attribute. For this reason, there is no well-defined rule to define the polarity of the principal component color bar. Figure 1 shows a comparison of stratal slices 16 milliseconds above a marker that represents the base of the Middle Jurassic Hugin sandstone in the Volve Field in the southern Norwegian North Sea. The different colors represent the sandstone facies at this level. But the interesting aspect here is that with the change of polarity of one, two or all three components, a different color combination results. Some of these displays are more revealing than others. We are not advocating for any particular display or interpretation, but simply making the point that interpreters can experiment with such displays and select the one that best illustrates or backs up the interpretation.
Although the order of the principal components represents the amount of variation in the data that they represent, there is no theoretical reason to map the PC1 against red. For independent component analysis (ICA), the data are first balanced in a manner that the order of the resulting components is meaningless; rather, it is the interpreter’s task to select the three components (out of say five) that best delineate the geologic features of interest. In figure 2 we interchange and reverse the color bar of three principal components (PC1, PC2 and PC3) resulting in six displays. No display is theoretically best; rather, as with ICA, the interpreter chooses the display that best enhances a geologic feature of interest.
In conclusion, RGB and CMY color blending works simply when corendering attributes of the same type, with most workers assigning R, G and B to low, middle, and high frequency spectral magnitude components. Attributes that delineate edges at high values such as dip magnitude or variance (1-coherence) computed from low, middle and high spectral voices are best mapped against C, M and Y. Attributes that delineate edges at low values such as coherence computed from low, middle and high spectral voices are best mapped against R, G and B, but result in a CMY image.
For other combinations of attributes and targeted geologic features, the interpreter needs to give some thought as to what feature they wish to emphasize and how they want it to appear. When corendering PCA and ICA components, both the polarity and ideal order of the components plotted against RGB is poorly defined. In these cases, the interpreter is encouraged to experiment with both polarity reversal of the constituent colors (RGB) and flipping the order of the attributes being assigned to specific colors, thereby generating six candidate displays. Choice of the best displays might provide more information, clarity and detail, leading to a more comprehensive interpretation. Surely, such an experimentation will entail more time, but the results will be well worth the time invested.