Porphyry Cu 1995
Porphyry Cu 1995
Porphyry Cu 1995
ABSTRACT
The magma metal series classification of ore deposits
developed by S. Keith of MagmaChem Exploration, Inc.,
and neural networks, a new type of pattern recognition software, have profound implications to metallogenesis and
have promising applications to mineral exploration and to
alteration and zoning studies. These techniques were evaluated by assigning 43 deposits to six subclasses of the magma
metal series classification defined on eight variation diagrams, training neural networks to classify analyses of 205
igneous and 887 mineralized samples, and testing the networks on their ability to classify new data.
Porphyry copper deposits are characteristic of six different alkalinity and ferrictferrous ratio categories in the magma
metal series classification. Such deposits occur in calcic
weakly oxidized, calc-alkalic oxidized, calc-alkalic weakly oxidized, quartz alkalic oxidized, quartz alkalic weakly oxidized,
and nepheline alkalic oxidized subclasses. The calc-alkalicoxidized subclass includes world class porphyry copper deposits
with the best grade and tonnage characteristics.
Only two of the foregoing six subclasses were studied in
detail; these were compared to four other deposit types of
different alkalinity and oxidation character. Whole rock
oxides of fresh igneous rocks were correlated with trace elements in rock chip samples from temporally and spatially
associated ore deposits of the six alkalinity and oxidation
subclasses. The K,O versus SiO, diagram best defined the
alkalinity classes of calc-alkalic and alkali-calcic; SiO,/K,O
ratios of alkali-calcic igneous rocks range between 14 and
20, and those of calc-alkalic rocks are between 20 and 30.
Iron mineralogy and the Fe,O,/FeO versus SiO, diagram
best defined oxidation subclasses of oxidized, weakly oxidized, and reduced; Fe,OpeO ratios are more than 0.8 and
magnetite and sphene arc abundant for oxidized subclasses,
between 0.5 and 1.2 with magnetite, sphene, and rare
ilmenite for weakly oxidized snbclasses, and less than 0.6
with only ilmenite for reduced subclasses.
Whole rock analyses from fresh igneous rocks were
obtained from mining districts for which trace element
geochemistry was also available. Lead-zinc-silver deposits
such as Tombstone, Tintic, and Park City are related to oxidized atkali-calcic igneous rocks. Polymetallic lead-zinc-copper-tin-silver deposits such as Santa Eulalia, Railroad,
Taylor, and Tempiute are associated with weakly oxidized
alkali-calcic igneous rocks. Tin-silver deposits of Llallagua
and Potosi are correlated with reduced alkali-calcic intrusives. Porphyry copper deposits such as Ray, Christmas,
Mineral Park, Highland Valley Copper, and Sierrita are
derived from oxidized calc-alkalic plutons. Gold-rich porphyry copper deposits such as Copper Canyon, Ajo, El Salvador, El Teniente, Hedley, and Morenci are linked to weakly
COl<l<ESI'ONI)t<N(Ii;1
01' hi.i<:\I..iNiSY AKI) FERRIClFliRROilS KAIIOS 0 1 I(iNF,Oi:S ROCKS
Table 1. Magma metal series elassilicatian showing logical subdivisions of aluminum content, alkalinity,oxidatioionstate, and selected metal specializations
(from Keith and others, 1991).
Peacock I n d e x MCAo
.i50:
o
E62:
b
b
$
a
sioz
Figure 1. Variation diagram illustrating the technique used by Peacock
(1931) to determine the alkali-lime index, often called the Peacock index.
2t
Y/ZI
- Ray i
~ l a1l.
. r.d
~ e i g h pcn.rnl
l
SiO: ,cr,tr\ Fu0'filgO uhuru kcO' i, 0.YtFc~0,)+ FuO. 311'
is \tnlngl) in,"-pow. \Ill' i\ mlodcralrl) inln-puor. \\IK ir vwhl! inm
rich.
iron rich. and SIK i\ \tn,ncl,
.
...., l l l K is it~t~durat(.l\
.. in," rich. 2r. Ycrricl
fprrnnr variation diaeram where weieht
Fe.0, divided br weight
.. oercent
.
kc() i\ plottrd ayain*t ~ c i g h prrrznt
t
SiO:. SO i, itmngl! o\i#li,ed.
\ I 0 is modrralcl) #oridi,rd. \\'O is uuukl! a,\i<li,cd. \YK i\ wr.uhl! redurnl, and SK i\ slnongl! rcdurcd. If.\-7.r rari;#liun diarrian~in ppm.
~
~~
I8
!i
1I"
D!
b
p
6"
4:
Porphyry copper deposits are common in six different alkalinity and femclferrous ratio subclasses in the magma metal series classification as indicated by the solid squares on tahle 1.
Porphyry copper deposits occur in calcic weakly oxidized, calcalkalic oxidized, calc-alkalic weakly oxidized, quartz alkalic
oxidized, quartz alkalic weakly oxidized, and nepheline alkalic
oxidized subclasses.
The calcic weakly oxidized subclass generally produces
copper-gold stockwork porphyry deposits in or near low-potassium granodioritic epizonal plutons. Examples include Panguna
in Bougainville, Yandera, Ertzberg, and Plesyumi.
The calc-alkalic weakly oxidized subclass produces several
types of porphyry copper deposits generally associated with
mesothesmal mediurn-potassium granodioritic plutons which
contain either biotite and hornblende or only hiotite. Examples
of moderate-sulfur porphyry copper-gold-molybdenum deposits
in or near granodioritic plutons include Ajo, El Salvador, Frieda
River, and Morenci. High-sulfur gold-skarn deposits include
Hedley and McCoy. High-sulfur gold-lead-zinc-silver fringes of
gold-copper-zinc porphyry and or skarn deposits are exemplified by Copper Canyon, Nevada. Hypothermal tungsten
skm-copper-molybdenum deposits at the contact of dioritegranodiorite plutons include Elk Mountain and Mill City.,
The calc-alkalic oxidized subclass contains world-class
porphyry copper deposits with the best grade and tonnage
characteristics. These mesothermal high-sulfur porphyry deposits are primarily copper-silver-molybdenum deposits. The
central zones of these districts contain porphyry copper-dncmolybdenum deposits. These central zones are surrounded by
zones of copper-zinc mines, which are in turn surrounded by
zones of lead-zinc-silver mines. Furthest from the central zones
are zones of silver-manganese mines or prospects. Examples
from Arizona of mesothermal high-sulfur porphyry coppersilver-molybdenum deposits are Bagdad, Christmas, MiamiInspiration, Ray, Mineral Park, San Manuel, Siemta-Esperanza,
Silver Bell, and Twin Buttes-Mission. Bethlehem and Valley
Copper in British Columbia are Canadian examples. Other
examples are Tyrone, Yerington, Escondida, Sar Cheshmah,
Cananea, and Chuquicarnata.
The quai-ir alkalic wcakly oxidized suhclass iiicliides porpliyry copper deposits and or skarn deposits with gold, high
sulfur, and sometimes platinum group elements. Examples of
quartz alkalic weakly oxidized porphyry copper deposits include Bingham, Copper Flat-Hillsboro, Bajo de la Alumbrera,
Ok Tedi, Ely, and Ingerbelle.
The quanz alkalic oxidized subclass includes the porphyry
copper deposits of Bisbee, Equity Silver in British Columbia,
and Cerrillos in northern New Mexico.
The nepheline alkalic weakly oxidized to oxidized subclass
includes epigenetic mesothermal hase metal veins and stockworks with very minor gold associated with nepheline-monzonite to syenite plutons. Examples include Carihou-Bell and
Galore Creek in British Columbia.
g w.
gw-
.:
60-
@ m-
3 loP+
w-
Q m-
LU("I0
,MIX r m*n r ~ r z u r ~ l
*.
'* J
.*',u)*
>m--
o,
to-
I
01
rn
50
55
3~
NU
- Ray
BJ
1
'0
5'4
Wl.+
Incl. olt.r.0
potassic
alteration
fresh or other
40
4s
50
** @
55
60
alteration
65
XI
75
CO
4e
K-CO
- LIW
011plots
mucl? scatter oil alkaliiiity plots that coi~zlatioilsbetween miileralization and whole I-ockgeochemistry of associated plutons are
obsci11-ed.
Statistical Analysis
Standard statistical analysis was used to detect general
characteristics of the data, to increase the efficiency of the
neural networks, and to classify the samples. Standard petrologic diagrams, magma metal series plots, statistical histograms,
and other diagrams were used to characterize the data (Wilt,
1993). Commercial software such as IGPET, Rockstat, and
SPSS was used to produce comparison diagrams and to categorize subclasses. Discriminant analysis was used to classify
analyses of both igneous and mineralized samples into the six
magma metal series subclasses of the study.
?I
6d
6c
MACr
Figure 6. Comparison of K,O versus SiO, variation diagrams for metaluminous alkali-ealcic (left side) versus calc-alkalic (right side) and for
oxidized (top), weakly oxidized (center), and reduced (bottom) classes. C is
MCAo
6e
MCAow
6f
MCAr
7c
MACr
7e
MCAow
7f
MCAr
8a
8c
MAC0
MACr
Ed
8f
MCAO
MCAr
9~
MACr
9e
MCAOW
9f
MCAr
Calc-akalic oxidized
:d
Mineralogy
abund. sulfides (py, cpy, gal, sphal);
some enargite;
abund. Pb minerals, sulfides,
Pb sulfosalts, oxides, sulfates,
Pb arsenates, phosphates,
Pb vanadates, molybdates;
large variety of Ag minerals;
Ag sulfides, sulfosalts, halides, Ag
tellurides, esp. argentite;
Zn carbonates, oxides, sulfates;
small amts. Cu minerals;
As, Sh, Sn not mentioned;
Mn oxides common;
native Ag, Au reported
abund. sulfides (py, cpy, gal, sphal);
smaller variety of minerals than MACo;
native Ag, Au infrequent;
fewer Cu minerals; Cu carbonates reported
fewer Pb or Zn minerals; but cemssite,
hindheimite, duftite are common;
few Ag minerals; argentite, rare sulfosalts
and halides are present; no Ag tellurides;
Mn oxides, some W, Sb, Fe minerals;
no Mo, Bi, As, Sn, Co, Cd, Ni, TI, Hg or
Te min
conunon sulfides (py, cpy, gal, sphal);
no mention of native Ag, Au;
no Cu, Zn minerals, except tetrahedrite,
enargite;
anglesite, cerussite reported, not other Pb;
lead snlfosalts nunierous;
huehnerite, wolframite minor;
bismuthinite, Sb minerals present;
Sn, esp. cassiterite, stannite are common;
no Mo, As, Mn, Co, Cd Ni, TI, Hg, Te
minerals;
Cu/(l'WZn)
7 to 200,000;
Table 2 continued.
(:()KR~~SPC)NI>iNCtCE
OF ALK/\lLINU'Y 1Ni) FERRICll'lRROllS Iil<TIOS OF I C N B O l j S ROCKS
piled for each rnine and grouped to detennine metal charactcristics of each subclass (Wilt, 1993, appendicesj. General pautems
are summarized in table 2.
Si1ver:gold ratios of more than 40:l are a conspicuous characteristic of production from metalumirlous alkali-calcic ore deposits. High base-metal contents, high manganese contents, and
high Lead-zinc ratios are also typical. Alkali-calcic suites are characterized by significant production of silver, lead, zinc, tin, boron,
molybdenum, and beryllium; they also have significant byproduct
indium, fluorine, lithium, manganese, and tungsten with trace
amounts of gold, copper, gallium, iron, uranium, thorjum, phosphorus, selenium, mercury, antimony, bromine, arsenic, barium,
and magnesium (Keith and others, 1991). Silver production is at
least 100 times gold production, and lead and zinc production is
ten times copper and molybdenum or tungste~tproduction. Alkalicalcic ore deposits lack molybdenum, cobalt, nickel, thallium, and
mercury minerals. Sulfosalts of most metals are present or abundant; tellurides, pa~ticularlysilver tellurides, are present but rare
in oxidized alkali-calcic districts.
Alkali-calcic oxidized districts used in this study include
Tombstone, Tintic. Park City, and Castle, Montana. These
include vein, skarn, and replacement deposits in which zinc production is greater than lead and silver and manganese are commonly produced as byproducts. Lesser amounts of boron,
bismuth, ba-ium, antimony, mercury, gold, beryllium, tellurium
and fluorine are also present in these districts.
Alkali-calcic weakly oxidized deposits include the tin-hearing orebodies at Santa Eulalia, Mexico, and Candelaria, Nevada.
This type of deposit includes epithermal veins, skams, and replaceme~ltsfrom which lead, zinc, and silver are produced and
that have significant associated occurrences of tin, bismuth,
manganese, boron, mercury, and arsenic. Districts also include
epithermal hotspring deposits that have greater silver production
than gold; that include some lead, zinc, and beryllium production; and that have a significant presence of arsenic, antimony,
mercury, bismuth, scandium, and fluo~ine.Alkali-calcic weakly
oxidized mining districts used in this study include Taylor,
Railroad, Kinsley, Tempiute, Cherry Creek, Linka, and Swales
Mountain, Nevada. Alkali-calcic weakly oxidized deposits generally have 100 times less volume than, but similar grades to,
alkali-calcic oxidized deposits. Weakly oxidized deposits have
higher silver-gold ratios and higher proportions of base to precious metals and of copper and rinc to lead than do deposits of
the oxidized subclass.
Alkali-calcic reduced igneous rocks are associated with tinsilver deposits similar to the metaluminous tin-silver deposits of
Bolivia. These include epithermal vein, porphyry, skam, and
replacement deposits with tin and silver production and lesser
amounts of gold, copper, indium, gallium, arsenic, boron,
yttrium, fluorine, and phosphorous. Examples of alkali-calcic
reduced mining districts include Llallagua, Omro, and Potosi,
Bolivia. Production from alkali-calcic reduced districts indicates
that they are very large volume silver-rich tin deposits with
no byproduct gold or base metals. The mining districts of
Llallagua, Omro, and Potosi, Bolivia, each had production of
several hundred million kilograms of tin, several tens of millions of kilograms of silver, and no lead, zinc, copper, or gold.
Metalun~inouscalc-dkalic districts are characterized by
significant PI-oduction of copper, gold, tungsten, and magne-
<jy
IIMACo
OMCAo
m
a MACow 0MCAow
I
I
MACr
I
I
MCAr
rameters of elements of mineralized samples (Wilt, 1993). Because most whole rock oxides varied with silica, their means did
not reveal differences among subclasses, but ratios such as SiO,/
&O and FelO,/FeO clearly differentiated the six subclasses.
High variances from the means of each subclass were found
for most elements in mineralized rock chip samples, yet many
means differed enough to allow the subclasses to be distinguished.
For example, alkali-calcic suhclasses had 10 to 100 times as much
silver as did calc-alkalic subclasses. Most elements did not have a
normal distribution; asymmetrical distribution resulted partly from
a wide range of data being grouped into 21 equal classes so that
most samples fell in the bin closest to zero. The asymmetry also
occurred because many samples were not analyzed for all 40 elements and therefore had zero concentrations for some elements.
In spite of the diversity in the data, discriminant analysis
was able to properly classify whole rock data for 88 percent of
the samples and trace element data for 76 percent of the
samples. When mineralized samples were first classified as alkali-calcic or calc-alkalic and then subdivided into oxidation
subclasses, discriminant analysis correctly classified 96 percent
of alkali-calcic samples and 83 percent of calc-alkalic samples
(Wilt, 1993). Despite high variances in the data, the six magma
metal series classes were clearly distinguished.
Rb Y
Ni
Co Mo
Cr Cd Sr Ag W Sn
!
I
MAC0
MACow
I
MACr
nMCAo
UMCAow
Ir MCAr
Table 3. Traceelement cl~aracteristicsofsix classes studied (means of samples in parts per ntillion).
diice ore deposits, so standard economic evaluations iiie also rcqoised. However, the assigiiment of prospects to magma iirctal series subclasses can indicate iT further evaluation is warranted.
The magma metal series classification is an inexpensive preliminary step that can eliminate prospects that do not fall into the
targeted type of deposit. For example, even though some alkalicalcic oxidized silver deposits have high gold and copper concentrations, they would not make good exploration targets for
porphyry copper deposits or Carlin-type gold deposits. If only
igneous rocks x e present without obvious veins or mineraliration, potential mineralization of a particular ore deposit type
could be predicted by the magma metal series subclass.
Applications of Neural Networks
The success of neural networks in classifying geochemical
data in this study indicates they can he useful in mineral exploration and in deciphering geological systems. Trace element
data from exploration projects of unknown ore types can he
tested on the networks trained in this study to determine which
types of deposits may be found. Once a network is trained on
data from either fresh igneous rocks or mineralized trace elements from mining districts of known magma metal series subclasses, the network can he tested on similar types of data from
districts whose subclasses are unknown. It is not necessary to
obtain whole rock data on fresh igneous rocks related to mineralization in order to correctly assign a mining district to a
magma metal series subclass; using neural networks, trace element data from mineralized rock chips can he used to classify
the district equally well.
Neural networks can be applied to geologic pattems at many
different scales. This study demonstrated patterns in the broadest
range of data that could be used to classify types of ore deposits
and no attempt was made to distinguish between proximal and
distal samples. Within each deposit type, more detailed representations of the areal distribution of the data may allow a neural network to find zoning patterns withn trace element data, alteration,
fracture density, ore grade or type, or whole rock data. If there are
numerically definable differences in the data, neural networks can
be trained to detect those differences. The challenge is to define
training patterns that have predictive value in exploring for ore deposits. The magma metal series classification has this predictive
power because it is empirically based on the correlation between
ore deposit types and natural geological processes.
ACKNOWLEDGMENTS
My mother, Jeannette L. Rasmussen, is gratefully acknowledged for financially and emotionally supporting my doctoral