Some Psychological Aspects of the Visual Perception of Cartographic Information

The widespread development of geographic information technologies for geological applications, alongside their obvious benefits, has in a number of documented cases led to very serious and costly errors and misconceptions arising from uncritical use of the software provided. Such errors occur most frequently when actual geological data (geophysical, geochemical, mineralogical, etc.) are treated as abstract numerical fields or images detached from their subject matter, and still more often when several types of data are used together.

Within the emerging scientific discipline of geological psychology, particular attention is devoted, among other topics, to how specialists in the geological sector form appropriate image-based conceptual models of the geological phenomena under study. Errors of perception and representation receive specific consideration (4).

Below, the psychological characteristics of visual perception and analysis of simple and multifactor cartographic information are used to examine some mechanisms by which specialists develop illusions, false stereotypes and misconceptions.

1. Illusions

Let us conduct a small psychological experiment. Pages 366–370 of Tables of Mathematical Statistics (3) contain a table of uniformly distributed random numbers from 0 to 99. By definition, these numbers are entirely unrelated, and any coincidence is accidental. Let us open the table on page 367 and copy the values in the first twenty rows and twenty columns. This gives a table of 20 by 20 cells (Fig. 1.а.). Let us retain 5% of all the numbers, selecting the lowest values (0 to 5), and 5% with the highest values (94 to 99). Their positions in the table are shown in Fig. 1.б. Examination of the figures shows that the highlighted cells within the square do not form any clear, recognizable shapes, confirming the random nature of these numbers. Nevertheless, some vague “patterns” can be discerned in the distribution of the marked cells.

Let us reduce the number of points to 2%; readily recognizable patterns emerge. If each circle in these figures is taken to represent a geological body, such as a kimberlite pipe, then kimberlite fields are clearly discernible, with pipe clusters identifiable within them (circles connected by lines). As a first approximation, these point patterns resemble the Mirny and Alakit–Markha kimberlite fields of western Yakutia.

This example is sufficiently illustrative. When factual data are severely limited, prior knowledge of their nature is absent, and a sufficient stock of elementary spatial templates is available in the mind, it is almost always possible, consciously or more often unconsciously, to devise a simple “explanation” or identify “patterns” that have no relation to reality.

Let us continue the experiment using the same array of random numbers. Dividing the array into pairs of vertical columns gives a random set of number pairs. Let one number represent the X coordinate of a point and the other its Y coordinate. This yields 200 points, with coordinates ranging from 0 to 99 on each axis. In this case, the points occupy 2% of the plot area (Fig. 2.а). The figure illustrates the arrangement of points with random coordinates in the plotting plane. It is easy to see that this arrangement is not uniform and that the points form certain “recognizable” clusters. Some clusters are linear, others arcuate. The adjacent figure (Fig. 2.б.) shows an example of combining point clusters into shapes (interpretation). Someone may wish to repeat the operation in their own way and obtain a different result. However, some shapes in the interpreted figure, or parts of them, can be expected to recur; that is, part of the figure will be reproduced.

The existence of such reproducibility in these examples indicates a regularity. However, this regularity lies not in the arrangement of the points in the figure (which is random by definition), but in the characteristics of human perception.

Human perception of the external world is known (2) to exhibit structural organization (the tendency to combine what is perceived into coherent, relatively simple structures), recognition (assigning a perceived object to a familiar category), completeness (sensory and mental completion of a set of perceived elementary objects into a whole image), and object-based perception (representing the world as coherent images rather than isolated sensations). In our case, structural organization, recognition, completeness and object-based perception manifested themselves in our “recognition” of the simplest “structures” in these figures—straight-line and arc segments—which we unconsciously “completed” into straight lines, arcs and, in places, “ring structures”.

These definitions fully explain the reproducibility of the results and allow everything drawn in the above figures to be classified as illusions: phenomena of perception, imagination and memory that exist only in the human mind and do not correspond to any real phenomenon or object (2). In the cases considered, we were dealing with optical, or more precisely, optical-geometrical illusions.

Let us make the situation more complex by using random numbers to define lineaments (coordinates and azimuth), fixing only their total number (N) and length (L). Depending on the parameters specified, some general mechanisms of perception and interpretation of the resulting figures can then be observed. At low line densities (N=100–300), approximately equidimensional structures are “recognized” most clearly (Fig. 3). With sufficiently large numbers of lines (N=1000–3000), “complex” linear structures, and even entire assemblages and combinations of them, are readily “recognized” (Fig. 4).

Interestingly, examining the results makes it possible to devise specific quantitative algorithms for “interpreting” linear and arcuate structures that formally identify, combine and map linear and arc-shaped elements in the figures. However, the way the illustrated lineaments were generated establishes that no spatial relationship exists between them. Combining them into linear, arcuate and concentric structures produces the same optical-geometrical illusions, whose identification is given a spurious formal basis during algorithm and software development.

At present, approximately ten thematic fields are known (1) in which remotely sensed photographic information is used for practical purposes: agriculture, climatology, land use, forestry, monitoring of water bodies and water resources, emergency monitoring, geological investigations, and others. For each application, the optimal technologies for image acquisition, image processing and thematic interpretation are selected according to its purpose. Each interpretation method is an algorithm for converting digital models of photographic images of an area into purpose-specific logical-semantic models appropriate to the subject.

Geological interpretation of remotely sensed data is itself multipurpose: geological in the strict sense (using the terminology and logic of a geological map), geomorphological (the equivalent for a geomorphological map), mapping of Quaternary deposits, mapping of exploration conditions, and so on. Consequently, interpretation in each case can only be performed according to a defined thematic legend, using factual thematic information previously acquired for the area and the theory of the relevant method or discipline.

The concepts of “ring structure” and “lineament” are not geological terms; accordingly, image interpretation using these models is not geological interpretation. If individual components of such structures are identified in photographs, each must undergo individual geological interpretation, without grouping them, based on the actual geological evidence previously acquired for the specific area.

Without presenting the supporting arguments within the scope of this journal article, we list the principal indicators of illusions in geological research systems that may help assess the reliability of predictive exploration decisions:

1. A conscious or unconscious prior bias in selecting facts from the full diversity available, based on elementary ideas or impressions;

2. Geological agnosticism or disregard of geology (in the full sense of systemic, genetic and historical understanding) when working with primary evidence, comparing data from different methods, and performing subsequent interpolations and extrapolations;

3. Lack of formalization of the solution, which is presented solely at an intuitive level without attempts at formalization;

4. Absence of mathematical proofs, even where they are entirely applicable.

It must be emphasized that these indicators do not directly and unambiguously identify an element of a geological research system as illusory. Nevertheless, we believe the list may help identify elements that should be examined first for possible illusory character.

2. False stereotypes

A FALSE STEREOTYPE is generally understood as the stereotyped, uniform execution of a usually multicomponent procedure in which some components do not correspond to the nature of the phenomena under study. Primary false stereotypes arise when such a technology is first developed without specifically examining whether each element corresponds to the nature of the phenomenon. Secondary false stereotypes arise when a technology is inappropriately transferred from previously effective solutions.

The practical use of primary false stereotypes can be illustrated by the following example. Figure 5 presents an example of “computer-processed images” with “emphasis on ring-shaped elements of structures controlling diamond-bearing districts”, obtained by a research group attempting to identify elements of a predictive exploration model of a kimberlite field in space photographs. The authors do not disclose the processing algorithm. However, knowing the algorithms available in principle, one may suppose that various combinations of image spectra were tried until a satisfactory result was obtained as “elements of ring structures”. The paper contains no references to attempts at geological interpretation of each combination, or to the use of geological premises in the algorithm to trace structures of known geological origin. Significantly, the authors regard the workflow that produces a combination of image spectral components in which the “diamond-bearing district” is marked by the indicated “ring structure” as a workflow for predicting “districts” from remote sensing data—in other words, as a particular stereotype.

Let us supplement our psychological experiment with another example, formulating the problem as follows. Suppose a deposit is known to occur at the centre of an area. How many times must we “throw” a set of one hundred points with random coordinates to obtain recognizable “elements of ring structures” around the deposit?

Figure 6 presents random-number modelling results addressing this question. The large circle marks the fixed position of the “deposit”; the coordinates of the other points are random numbers. Panel “а” shows one complete realization, and panel “б” a possible interpretation. The small insets on the right show the central portions of individual modelling results, numbered according to the sequence of models. Of 300 individual models, a satisfactory result—the presence of “elements of ring structures”—was obtained, on average, in every tenth model.

This experiment shows that even random, entirely unrelated and meaningless spatial elements, when repeatedly generated, quite often form elementary recognizable combinations around or near a fixed, predefined point.

Consequently, the stereotype described above, in which diamond-prospective areas are identified from “ring structures” in space photographs, is a primary false stereotype. It is primary because the entire procedure constructs a predictive exploration model using a particular reference example.

To illustrate secondary false stereotypes, we shall again use models based on random numbers.

A number of publications on formal analysis of geological, geophysical and geochemical fields contain examples of calculating the density of their elements per unit area. A square cell is most commonly used as the areal unit for such calculations. Subsequent conclusions are usually based on contour maps of the density of the relevant elements.

Figure 7а shows a grid whose nodes are assigned the same random numbers from the set described above. Using the Chart Wizard in Microsoft Excel, we drew a contour map dividing the entire “site” into areas with values below 50 and above 50. The result readily shows that the contours clearly emphasize two orientations of “structures” on the map: northeast and northwest. If desired, interpretation of the map could include drawing the axes of positive and negative anomalies, identifying breaks in these axes, and so forth, thereby producing a “tectonic map” of the site.

However, the random nature of the numbers at the grid nodes compels us to seek the cause not in spatial patterns, which are absent by definition, but in the procedure provided by the Chart Wizard for converting values on a square grid into a map. We thus find that the map’s apparent “structure” is caused, on the one hand, by the square grid of input values and, on the other, by the scheme used to interpolate these values along the sides of each square cell. Figure 7б supports this conclusion: the grid cells are arranged in a staggered pattern, and the “structures” extend predominantly in roughly east–west and north–south directions.

Having observed, understood and explained the result, we encounter a contradiction. On the one hand, we understand that the figure is meaningless; on the other, we almost always interpolate values of a field on a square grid to produce contour maps, then interpret those maps for geological purposes. In this case, the cause of the contradiction is evident. Random numbers at grid nodes—that is, values with a zero autocorrelation radius—provide no justification in principle for moving from discrete map points to areal interpolation of those values. Yet we performed this operation. Why?

Without reflection, we transferred the method of analysing spatial distributions by constructing contour maps from everyday geological practice, where we usually encounter field parameters with a nonzero autocorrelation radius, to a situation involving random numbers.

We therefore transferred an algorithm used as a standard solution to earlier problems to a new problem without assessing whether it matched the nature of the new values. Thus, OUR STEREOTYPE PROVED FALSE.

3. Simple and complex misconceptions

To conclude our psychological experiment, let us consider the comparisons in the final figure (Fig. 8). The “contour maps” shown here are interpolated grids of random numbers, including both the earlier results and new variants. These maps are overlaid with the previously obtained distributions of “object points”.

Examining these combinations, one can readily discern certain “patterns” in the positions of “object points” within the “field contours” in virtually every case. Interestingly, when the figures are viewed, the “identified pattern” appears clearer than is often the case in actual practice.

The illusory nature of the conclusion is evident from the way the figure was generated. But suppose the “contour map” under consideration were actually a map of a physical field, such as the magnetic field. Would that change our conclusion that the visually observed “pattern” is an illusion?

No, it would not, because the locations of the “object points” are random by definition and unrelated to one another.

Now suppose that the positions of our points are determined by an algorithm incorporating a quantitative relationship that places them in a specified orientation relative to one another, with a specified spatial interval between them. Further suppose that the “field” values at the nodes of the square grid are actual data, but still have a zero autocorrelation radius.

Would these assumptions make our conclusion more objective? Again, no: the interpolation of discrete data by the standard procedure has produced an illusion in the form of a so-called “contour map”.

This final example shows that EVEN WHEN OBJECTIVE DATA MAKE A PARTIAL OR VERY SUBSTANTIAL CONTRIBUTION TO A SYSTEM OF REASONING, A SMALL CONTRIBUTION OF ILLUSORY “INFORMATION”, CONCLUSIONS OR FALSE STEREOTYPES LEADS TO ERRONEOUS CONCLUSIONS FOR THE ENTIRE SYSTEM—TO COMPLETE MISCONCEPTION, RATHER THAN TO A RELATIVE TRUTH IN WHICH SOME ELEMENTS ARE TRUE AND OTHERS FALSE.

In light of the above, it seems important to distinguish PRIMARY ILLUSIONS, which have virtually no objective basis apart from the psychological characteristics of human perception of the external world; PRIMARY FALSE STEREOTYPES; and SECONDARY ILLUSORY PHENOMENA (DATA, HYPOTHESES, METHODS, ETC.), OR MISCONCEPTIONS.

MISCONCEPTIONS ARISE WHEN PRIMARY ILLUSIONS OR PRIMARY FALSE STEREOTYPES ARE COMBINED WITH DATA OF A CLEARLY OBJECTIVE NATURE. These proposed concepts appear highly useful because they allow any data, hypotheses, methods and so forth to be examined for illusory elements and false stereotyped elements.

Conclusion

In our view, these examples convincingly show that approaching the phenomena under study without reference to their subject matter—that is,

apart from the theory of the relevant research method (geological, geophysical, mineralogical, etc.), which encompasses both the theoretical premises for describing observations and their initial processing (always specific to the method and even its particular variant), and ways of interpreting the resulting data based on models appropriate to the method;

apart from prior geological and other specialist knowledge of the study area, its geological structure and the expression of that structure in the specific fields under investigation (again through models appropriate to the method);

while treating observations solely as visual images formed through the most elementary perception of cartographic information,

we will inevitably produce image-based conceptual models of phenomena, technologies and solutions on a large scale, with primary illusions, false stereotypes and misconceptions widely present.

These considerations indicate the value of integrating specialized data analysis modules more deeply into modern GIS used in subsoil resource management, based on the theory of analysis and interpretation in geological, mineralogical, geochemical and geophysical methods. Methods for investigating spatial correlations should also become an important component of GIS.

References

Garbuk S.V., Gershenzon V.E. Space Systems for Remote Sensing of the Earth. Moscow: A i B Publishing House, 1997. 296 pp., illustrated.

Nemov R.S. Psychology. A textbook for students of higher teacher-training institutions. In 2 books. Moscow: Posveshcheniye; Vlados, 1994. 576 pp. and 496 pp.

Tables of Mathematical Statistics. Bolshev L.N., Smirnov N.V. Moscow: Nauka, Main Editorial Office for Physical and Mathematical Literature, 1983. 416 pp.
Tsyganov V.A. Reliability of Geological Exploration Systems. Moscow: Nedra, 1994. 299 pp., illustrated.