European Geologist Journal 45
The hunt for phi structures
by A. Lambert1
1 BSc Liège University 1973 EurGeol 38
Phi (Φ) structures are geological entities that combine ring shaped fault segments with crossing shear zones. Their closeness to mineral occurrences is recognised but many geologists do not consider them useful for prediction of deposits and challenge their detection. This paper reviews the geology, genesis and economic worth of the structures, focusing on gold mining in Europe. In addition, 203 large mines or mining districts worldwide are investigated for the presence of phi structures. Methods used to identify them include Computer Vision and Machine Learning algorithms, Hough and Radon transforms, and global gravity gradient maps. Results show phi patterns in the vicinity of mining in 61% of the cases. A companion web archive provides learning images, routines and detailed examples.
Phi (Φ) structures are tectonic sets combining ring-shaped fault segments with crossing shear zones; they cover surfaces between a few to several thousands of square kilometres. Their proximity to mineral occurrences is frequent and intriguing: many among the Φ patterns cover or are tangent to deposits of iron, gold, copper, lead-zinc, diamonds, rare earth, lithium or uranium.
The concept of the phi structure is a catchall. It includes identified meteorite impacts, alkaline ring complexes and intrusive stocks calderas as well as discrete features without any surface clues to their origin. The distinction between magmatic or impact genesis often remains blurred: large impacts are frequently followed by intrusions. When the surface scar is eroded, the remaining phi signatures are quite similar in both the intrusive and impact cases: it is often impossible to distinguish them without deep drilling or a seismic survey.
Figure 1: Discovery of Olympic Dam by O’Driscoll, regional view from Bourne and Twydale 2004, modified. Red dot: Olympic Dam (South Australia).
Many geologists have tested the predictive capacity of the phi structures. Outstanding among them is O’Driscoll (1980, 1985) who gave credit to the phi concept for contributing to the Olympic Dam mine discovery. O’Driscoll coined the “hydrothermal highway” metaphor (Figure 1). Witschard (1984), Eggers (1979, 1981), Saul (1978), Kutina (1995), Robinson (2005) and Bubner (2000) present western world examples. Florenskii (2006), Gavrilov (2012) and Piloyan et al. (2016) give abundant references to eastern European publications about these structures and their relations to ore deposits (note that citations given in grey are for references listed in the companion web archive). Watchorn (2017) maintains a website outlining giant ring structures in Western Australia and discusses their possible ties to gold and nickel occurrences. All of the studies converge in showing that the phi structures are favourable sites for hosting mineral occurrences: they combine a probable heat gradient with a dense fracturing pattern.
In this study, 203 large mining sites or districts were searched to test the presence of phi structures. A few, mainly from the gold exploration scene, are described here to illustrate the concept.
The main challenge was to design a way to locate the structures in a reproducible, probabilistic way so as to deflect identification critics: the human eye is prone to find patterns where there are none…and to ignore obvious ones. Detection methods are presented for the ring and for the cross-cutting components.
Digital terrain models (DTM) provide the best support for locating phi structures; the most efficient algorithms found are Hough transform and the Convolutional Neural Networks (ConvNets) image recognition.
2.1.1. Computer Vision/ Hough Transform
The Kolwezi ring structure in the south of the Democratic Republic of the Congo was chosen as first test example. The area is the largest cluster of strata-bound copper occurrences in the world (Broughton et al., 2010). The 40 km wide half-ring is easy to spot on the digital elevation models; (in this case ALOS – Jaxa 2018 – Figure 2).
The Circular Hough Transform method is expected to find triplets of (x, y, R) that belong to circles in a given image with a high probability. The algorithm has been implemented on Octave software (Octave 2018); it was applied to river network images generated from digital elevation data with SAGA Gis (SAGA 2015) (Figure 3). The Hough transform plane reflects a probability distribution (Figure 4) and shows that a properly shaped circular feature is very unlikely to emerge at random.
Figure 2: Illuminated DTM of Kolwezi area. The diameter of the phi structure spans over 40 km. The red star is the location of Kolwezi town, Democratic Republic of the Congo.
Figure 3: Drainage network calculated from above DEM, same frame.
The Hough transform plane with the peak response from the twin curved segments is shown in Figure 4; the circle segment is confirmed by more than six standard deviations.
Figure 4: Transform plane with peak showing the centre of segments highlighted in Fig. 3 (Octave routine is given in the companion web archive).
2.1.2 Machine learning/ Neural Networks
Convolutional Neural Networks (ConvNets) have been developed during the last decade. ConvNets derive their name from the “convolution” operator which scans an input image, extracts features and give them a rating through a combination of weights and biases learned by stacked layers of digital neurones. Convolution preserves the spatial relationship between pixels by learning image features using small squares of input data (Walkarn, 2016). The number of earth and planetary sciences applications is growing exponentially (e.g. Palafox et al., 2017; Kaggle website, 2017).
The most popular ConvNets are fed with several million images and trained for weeks on supercomputers. Some of the features picked up by them are useful to most computer vision problems and can be transposed for targets other than those initially planned. Smaller vision studies typically re-use the trained lower parts of larger nets to rebuild a new and efficient network for their own purposes, with a much lower computation cost (Chollet, 2016, 2017). VGG16 and Xception nets were tested here, the program used is Keras (2017), a Python wrapper applied over Theano (2017) or Tensorflow (2017); both are math libraries designed to handle large tensors. The network was trained on 3,640 images and validated on a different set of 1,160, similar to those in Figure 5 (224×224 or 299×299 pixels). The prediction of whether or not a phi structure is on the image has an accuracy of about 91% on VGG16 net with 4% false negative and 5% false positives, while 93% accuracy is reached on Xception.
Some layers of the networks respond particularly well to phi characteristics; for instance layers 15 and 16 of VGG16 react to curved edges while layers 25 and 27 respond to circular “cloudy” features, as shown in Figure 6. Those four layers can be combined into a specific filter which can be applied over wider images applying a method described by Perone (2016) and Blier (2016). The identified PHIs can be tested on smaller images for a final diagnostic (Figure 5). Routines and methods are given and discussed in the companion web archive. The grey shadings in Figure 5 represent illuminated DTM scenes containing phi structures.
Figure 5: Extract of images library (all here are positives, i.e. they contain one or several PHI structures; the plain, negative, images are available in the archive).
Figure 6: Test images over Richat structure in Mauritania: left: DTM; centre: combined layers 16,25,27 of VGG ConvNet; right: heatmap ( the lighter colours are warmer).
2.2 Cross-cutting lineaments
Shear zones cut across and may truncate the ring structures. A range of methods are available for their detection, starting with the plain visual study. To thwart personal interpretations, a Radon transform algorithm may be used to outline the most significant bundles of lineaments. Radon (Hoilund, 2007) is an integral transform which takes a function f defined on the image to a function Rf defined on the two-dimensional space of lines in that same image. The value at a particular line is equal to the line integral of the function over that line. Said plainly, with a given orientation theta, all pixels in the image are looked across “rays” perpendicular to that section and the aligned pixel values are summed up (Figure 7).
Figure 7: Clockwise: 1- original drainage map with test line drawn, 2 – radon peak in 3D corresponding to the black line, 3 – the three best alignments , 4 – caption to show radon projection mode (Octave routine is given in the companion web archive).
Shear zones can be detected from geophysical data, when available. Gravimetric or magnetic gradient maps, like TXX or TYY, are very sensitive and best for detecting crossing shear zones at a high angle. Even low resolution surveys (Figure 8) like global gravimetric syntheses may give reasonable indications for large crustal shears.
Figure 8: Tasiast Gold Mine (Mauritania), TXX gravity gradient map (ring segment in red, mine location shown with the red symbol, white circles: gold anomalies, in blue are possible other rings. Range: -50 to +50 Eotvös).
3. Case Histories, Geological Context
Three origins are possible for phi structures: meteorite impact is exogenous while the other two are endogenous: magmatism and fluids buildup. Phi structures can be transposed into overlaying sediments or a volcanic blanket, deposited later, through isostatic corrections. The geothermal cells have practical engineering and environmental applications.
3.1. Impact origin
At Vredefort RSA, one of the largest impact sites studied worldwide, the initial shock was dated at 2023+-4 Ma. After that, “anatectic granites” raised and metamorphism developed, giving dates at 2017 +- 5 Ma (Reimold, 2015), interacting with previous gold occurrences.
Apart from Vredefort and a few other examples listed by Grieve (1994, 2005) and Reimold (2005), it is extremely difficult to prove a meteoritic impact origin for an old and deeply eroded feature. They are always largely eroded, shatter cones are washed out and fresh pseudotachylites remain difficult to spot and diagnose. Siljan impact is presented below and other examples are discussed in the companion web archive.
Wall et al. (2005, 2014) propose the “TAG: thermal aureole gold” model. Although not designed for phi structures, the model outlines the centres and ring peripheries as target areas. Part of Wall’s discussions are on the Muruntau deposit, which is on a phi structure. It should be observed that the true Muruntau ring structure is eight times larger than the one shown in Wall’s block model (Figure 9) and deposits are not in the same structural position.
Figure 9: Muruntau shows a typical phi pattern. Left: view from SE, the radius of Muruntau structure is about 40 km. Right: TAG model from Wall (2014). Symbols are for mine sites.
The rosary of gold deposits along Muruntau is a good example of how the crossing shear zone influences the mineralising process. Copper/gold porphyries are presented below in the European test section; kimberlites are other examples of magma related structures.
3.3 Gases: structures without magmatic or volcanic indications
Most of the phi occurrences reviewed do not give any indication of a significant intrusive body nearby or of a significant volcanic activity suggesting calderas. This is particularly observed for gold deposits in West Africa. These Φ structures might originate as a ground swell under a pocket of gases, fluids or dry steams which later escaped, leaving scant traces of their passage. Gilat (2005, 2012) indicates that tremendous gas activities are generally overlooked.
3.4 Diachronic effects
Interaction with ore deposits can be less than obvious. While phi structures may directly remobilise and enrich earlier deposits, isostatic replays can drive the hydrothermal systems upwards, improving the economic value of later stage occurrences. Some examples:
- the Vredefort structure was impacted before the Karoo deposition but ring faults have replayed and penetrated the Karoo cover;
- In the Central African Republic, Bakouma U mine is in Neoproterozoic sediments but ring fractures nearby cut across nappes of various ages;
- In Mali, near Sadiola, some ring fractures stop at the escarpment of the Infracambrian sandstones while several structures nearby are developed in the sandstones, also on the Senegal side,
- the phi structure west of Kolwezi is covered by the NW thrust sheet; however, later ring cracks in the neighbourhood penetrate the overlaying nappe.
Figure 10: Sub-level caving models, applicable to the ring features (from Rusin et al. 2007)
Phi structure collapse generates three associated fracture types: steep cracks along the ring structure, décollement on top and en echelon faults within the perimeter. The collapse environment could explain why, in some copper, uranium or gold provinces, steep rich shoots coexist with strata bound lenses or with abundant dolerite sills (e.g. Rusin et al., 2007; Figure 10).
3.5 Geothermal and Environmental aspects
Phi structures provide valuable aquifers; their permeability frame must be properly understood when conducting impact studies, and all the more so when exploiting water next to mineral deposits or mines that the structures may host.
A quick search over the Internet with the three words “geothermal-ring-structure” will return several hundred relevant references. Most relate to tertiary or recent volcanic calderas of relatively small size, but a number of them are linked to older and larger structures, including known impacts. For instance, in Sweden, three old impacts were drilled and found to have geothermal potential: Siljan, Bjorko and Dellen. Unfortunately the old fracture network became cured with calcite, quartz or albite. Bjorko, at the SW outskirts of Stockholm, is plugged by calcite and could be exploited after fracking, but more economic sources of warm water are available at the moment (Henkel et al., 2005). The Siljan impact structure is about 50 km wide, the largest in Europe, with two lead zinc deposits at its SE edge and showing seeps of oil.
4. European Test Set
203 large mines or districts worldwide were tested with the above methods for their possible associations with phi structures. Of these, 61% were found touching or within phi structures (see list and images in the companion archive). Findings related to European gold mines and districts will be discussed here.
The largest gold deposit in Europe is Rosia Montana (Gabriel) in Romania. This is a copper-gold porphyry that gives a small but distinct phi signature (Figure 11). The nearby advanced project, Rovina, can be spotted on Google Earth (Figure 12).
Figure 11: Rosia Montana (diam. 12 km), N is up. Source: ALOS
Figure 12: Rovina ( diameter 8 km), N is up Source: Google Earth
It is relatively straightforward to identify a number of similar porphyry signatures around the Carpathian arc: in Romania, south of the Carpathian Range, there are for instance the Rinink and Horezu rings. Other provinces with similar signatures are Biely Vrch and Kremnica in Bohemian Slovakia, Kirovorhad in Crimea, Kisladag (Gumuskol) and Ovacik in Turkey, Pribram in Tchequia and Transkarpathia in Ukraine, as well as those in West Turkey. One fascinating case is the giant oval structure covering Slovakia (Figure 13, major gold occurrences in red).
Figure 13: Giant ring structure of Slovakia with major gold occurrences in red (diam. 200×300 km). Grey line = crossing shear. Source: ALOS
The other European province showing significant gold is the Central Lapland with its greenstone belts in Finland and Norway. Phi structures are more difficult to spot in northern latitudes; their signatures fade relatively quickly: moraines, lakes and glacial streaks can mask them completely (Figures 14 and 15). Training models specific to glacial environments should be studied separately (Witschard, 1984; Krøgli 2008, 2010).
Figure 14: Skjellefte ring (diam. 35 km). Source: SRTM
Figure 15: Possible ring in Karasjok Province (diam. 20 km). Source: SRTM
The data set on European targets, comprising images and ConvNets predictions, is available in the companion archive.
When the first Landsat images came out phi structures became familiar to geologists. But a barrage of criticism, focusing mainly on their identification, has since pushed them back in limbo.
In this study, ring structures were searched for with the techniques described in the vicinity of 203 large mineral deposits or districts worldwide. An association with ring features was found in 61% of the cases, which is in the range of previously published evaluations (Bubner, 2004; Bourne and Twydale, 2004). The structures that were identified are definitely not figments of the imagination: their detection was monitored effectively and reproduced using different methods.
In spite of the concept being a “catch all”, the presence of a phi structure pays off and therefore it should be looked for as a standard procedure, whatever its genesis may be. If a phi structure is detected, it can add a plus to any exploration effort or even give orientation to new research. When present, phi structures are parts of the environment and should be identified for their key role: it is tempting to extend O’Driscoll’s “hydrothermal highway” metaphor to “hydrothermal roundabouts” for the phi structures, as they focus and redirect the transit of water.
The companion web archive with all the programs, cases, images used and detailed bibliography is available at GitHub: https://github.com/lambertgeo/PHI .
This study stems from the ESA-CATEGORY1 project (ref C1P7824) “GOCE – Finding the Hydrothermal Highways in Africa” (ESA 2012, with Liège University). A large number of mining districts were scanned with the aim to relate Goce gravimetric satellite data to fertile geological structures. The main finding was that large-scale gravimetric gradient maps could be used to locate large shear zones. Attention was drawn to ring features early on but only the largest rings could be detected in the large scale gradiometric maps.
I am grateful to ESA for providing data and satellite images, to Profs. E. Pirard in Liège University, C. Braitenberg in Trieste University, D. Broughton and V. Correia who gave support and comments.
Short References (citations given in grey can be found in the companion web archive)
Bourne J.A., Twydale C.R. (eds.) 2004 Crustal structures and Mineral Deposits – E.S.T. O’Driscoll’s Contribution to Mineral Exploration. Dural, N.S.W.: Rosenberg
Bubner G.J. 2004 O’Driscoll rift-ring tectonics and mineral resources in South Australia. In: in Bourne & Twydale (eds.) , pp. 195-200.
Gilat 2012 Degassing of primordial hydrogen and helium as the major energy source for internal terrestrial processes Geoscience Frontiers 3(6): 911-921
Grieve R.A. 2005 Economic natural resource deposits at terrestrial impact structures. Geological Society, London, Special Publications 1-29.pdf, http://sp.lyellcollection.org/
Henkel H., Backstrom A., Bergman B., Stephansson O., Lindstrom M. Geothermal Energy from Impact Craters.The Bjorko study. 2005. Proceedings World Geothermal Congress Antalya Turkey
Krøgli S.O. 2010 Automatic extraction of potential impact structures from geospatial data – examples from Finnmark, Northern Norway Thesis Oslo University
O’Driscoll, E. S. T., 1985, The application of lineament tectonics in the discovery of the Olympic Dam Cu-Au-U deposit at Roxby Downs, South Australia: Global Tectonics and Metallogeny, v. 3, p. 43-57; 15 Fig.
Piloyan A. & Avagyan A. 2016 The circular structures of the republic of Armenia based on a digital elevation model. European Journal of Geography number 3, 38-70
Watchorn R. 2017 YC#1 to 3 Giant Ring Structures http://geotreks.com.au
This article has been published in European Geologist Journal 45 – Environmentally sustainable mining in Europe
Read here the full issue: