il Principal Component Analysis for Alteration Mapping* W. P. Loughlin U.K. National Remote Sensing Centre, Farnborough, Hants, United Kingdom Apstmact: Reducing the pring the numberof image bends input for principal component analysis (PCA) ensures that certain tates wil not be mapped end ineates te thlfood tet oer le ons vocally mapped into only one of 3! component images. In arid terrain, PCA of four TM bands will avoid iron-oxade and thus more reliably mage iy Aetect hydroxyl-bearing minerals if only one input band is from the visible spectrum. PA for iron-oxide mapping wil avoid hydroxyls if only one of the SWIR bands fused. A simple be created in which anomalous concentrations of hydroxyl, rincpal component color compost nage ea then mops irom, and none mt played ‘rightly in red-green-blue (RGB) color space. This composite allows qualitative inferences on alteration type and intensity to be made which can be widely applied. INTRODUCTION Discs aimucanoss orvetonseer nose (Loughin, 90) a the UK. National Remote Sensing Cenve (MISC), a new and simple methodology was developed for alteration Tapping using ATM and "i imagery of the Cret Basin region ofthe western United Stats, The technique i fst, bust and Felabl,reqlres no atmospheric oradiometiccomrecton, and Can be carried outwith standard Principal Component Anlyis (FCA) software The technique requires only a mentary an derstanding ofthe spectral properties of minerals and vegeta tion and it rele onthe aii of the principal component tranform fo map inceasng aubetes of dita variance no suc. cessive components, No detalled knowledge of the spetal [Popstn of specicurgt requ The matodony has cen informally named the "Crosta technique” afer the re. searcher whose intl studies (Crosta and MeM. Moore, 1983) inspired the NRSC development ‘he NISC study was guided by detailed field information on numerous and varedssneral prospects and deposts coved bythe ata imagery, and it drew onthe colective experience of exploration geologits who had been using the imagery a8 Prospecting aid. The prime requirement was to develop taly Effective processing techniques which coud be implemented by Brologse using inexpensive image processing systems Tages prodced By FA were Judged the mast ecarate in delineating alteration tones and the most relable in that they ‘enol prone to spurous or false” anomalies. A modified fore Pa, which uses sets of four selected mage band, was then developed forthe Ata imagery and tested on Landa Ti agery. The technique was next applied to archived ATM and Ta image data acquired over some fecent western US. gold dis: coveties before or soon after their discovery It wao found that many of these could have been located using the motiied FA method, and this i demonstrated for the Gal Bar and Gold Tee Extension orebodies in Bureia County, Nevada Subsequent Was ofthe methodclogy on Tht images fiom other pars of the westemn US, southern Span, the eastern Medi feranea, the Nidle East, and South America have demon- stated that thas wide application inard and seman tran, Frincipal component color composite images ("Cross com posts”) produced by the Crosta technique sre easy to unde FRc he High Thematic Conference on Geol Remote Sein Dey Coty 3 apg ‘oun wn Calg Constants (LA, Denjgonely County Fermanagh, Nowthers land. Fhorocsaramrmc ENGINEERING & ReMoTs SENSING, Vol. 57, No. 9, September 1991, pp. 1165-1169. stand and can be interpreted in a qualitative manner. Photogeological inferences on color relationships, and their im- plications for mapping alteration type and intensity, can be ex- {rapolated to any arid or semi-arid region. ‘THE CROSTA TECHNIQUE ‘The principal components transformation is a multivariate statistical technigue which selects uncorrelated near combi. futons (lgenvestor loadings) of vals in ech away that tach successively extracted linear combination, or principal component (Pc), has a smaller variance (Singh and Harton, 196). The statistical variance in mullspettral images i related to the spectral response of various surficial materials such as rocks, soils, and vegetation, and it i also influenced by the Statsfial dimensionality ofthe image data, When multispectral og channel re ened ates and ajo ee formation it fellows that the ordering of the ping ‘ents is influenced both by the spatial abundaace of he varus Suicial materials and by the image statistics, We can take ad- vantage ofthe inftence of scene statistic, which can be bath ‘measured and adjusted, in order to “force” the transformation to give us information on the spatial distribution and relative abundance of particular surficial materials, thats using CA os an image interogation technique, (Grosta and MeN Moore (198) describe a methodology called Feature Oriented Principal Components Selection (FPC3). FPCS {s based on the examination of PCA eigenvector loadings to decide which ofthe principal component images will concen trate information directly related to the theoretical spectral signatures of specific targets. An important aspect of this ap- roach is that it predicts whether the target surface type is ghlighted by dark or bright pixels in the relevant principal component image. The methodology developed at the NASC felies specifically on the selective input of only four image bands for FA. “To introdice the technique, a series of principal component transformations on both sbe and four bands ae presented as tables of statistics and principal component transformation fac- tors. Selected PC imaget from some ofthese transformations are reproduced in the figures to suppor the arguments in the text. ‘Tue Roaenrs MounTAN AREA, EUREKA Coun, NEVADA The Landsat 1 image reproduced in Fgue 1s 52 by 52a sateen (Eom pl 2) coveing the southern pt of tte Roberts Mountains, Eureka County, Nevada, The image Was acquired on 1 Apa 1985, not long fer the discovery of the Gali Bar deposit and before the dscovery or disturbance 0099-1812 91/5708-1163808.0000 (©1991 American Society for Photogrammetry sand Remote Sensing1164 Fr. 1, Roberts Mountains area, Nevada. a subscene, Pct albedo image. of the group of small drebodies known collectively as the Gold Bar Extension deposits (Figure 2). The orebodies are bulk-mine- able and are presently being exploited by Atlas Precious Metals Inc. There are also a number of gold prospects within the area. ‘The gold mineralization is epithermal in origin, of probably early Tertiary age, and is hosted in Palaeozoic carbonate rocks. Min- eralization is controlled by NW-SE trending structures within a similarly trending major mineral belt known as the Battle ‘Mountain - Eureka trend, itself parallel to the famous Carlin, ‘Trend. The area is semi-arid and vegetation type and amount is influenced by elevation, aspect, and availability of soil mois- ture. Sagebrush and sparse grasses are ubiquitous around both the Goldbar and Goldbar Extension deposits. Principal Component Analysis of Six TM Bands Table 1 is a listing of the image statistics, eigenvalues (which sive an indication of decreasing variance in successive principal components), and eigenvector loadings (linear combinations of ‘weighted input images in the principal components) for a prin- pal components transformation, using the covariance matrix, ‘on all six reflective bands on the Roberts Mountains subscene. The transformation has been carried out on raw data. Table 2 {s a listing for the transformation as applied to stretched data where each band has been given a Gaussian-type stretch to give all bands approximately equal mean and standard deviation values prior to the transformation, In both transformations the first principal component (PCr) is composed of a positive weighting from all bands. As indicated by the eigenvalues, Pci accounts for 89.636 percent of the total ‘variance forthe raw data PCA and 85.903 percent of the variance for the stretched data. Overall scene brightness, or albedo, is responsible for the strong correlation between multispectral im- age channels, and PCA has effectively mapped this into PC! of both transformations. The remaining components should there- fore account for decreasing variance caused by differences be- tween spectral regions and between individual bands. The statistical dimensionality of data (related to sensor gain and offset as well as spectral differences) and, in particular, the ‘magnitude of the standard deviations (SD) has greatest influ fence over the actual weighting of original bands mapped into PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1991 3km Fia. 2. Sketch map. Goldbar (1), Goldbar extension (2), and prospects (@) and (4), successive components In the raw data transformation of Table 1 the large sb for Ta is responsible for that band’s dominance in Pci, and the low SD ofthe vegetation band (rt) is largely, but aot entirely, responsible for i's very lowe contaibution to 16.58 pect of whe DN men 72.8 er ow weighting compared othe 65.07 percent of Tus where the mean Ilsa incetecqurn=, toe WO FCL “chet eps coms be identical. The difference fs not readily apparent, however, and the substitution of the raw data PC1 in Fi 1 (which is Fer from stretched data) would make litle diference to the visual appearance of Figure 1. The transformation to some ex tent sell-compensates fortis effect in higher numbered pain: cipal components tothe extent that the weighting of individual bbands with low SDs wil be significantly greater in at least one ofthe successive Pea. In Table 1 Tat (With the lowest SD) has 83.92.27 percent loading in FCS, Ta has one of 91.43 percent in C6, and TMD (with the third lowest SD of tie se) has 2 6635 percent loading in PC. Compare these loadings with those from the Pca of stretched data in Table 2, where the ss are identical for all ands and the eigenvector magnitudes for Tv, TM, and TMB never achieve the levels forthe raw data FCA. Singh and Harrison (1985) recommend using “standardized” Pca fo over- come this effect. For empircal information extraction a6 pre- sented in this study, standardization of image bands is Unnecessary and PCA’ on raw, unstretched data is elective in nearly al cases. Examination ofthe magnitude and sign (Le, postive ong. ative) of eigenvector loadings gives an indication of which spet- teal properties of vegetation, rocks, and sols are responsible for the statistical variance mapped into each FC, and this the basis of the Crosta and McM. Moore (1989) approach. Eigenvector Toadings for PCa of Table 1 indicate that it descrbes the difer- ence between the visible channels (Ii, 2, and 2) and the ik channels (Ta, 5, and 7). Materials which Nave the highest e- flecance inthe visible spectral region will appear on Pe2 asthe brightest pixels (postive eigenvector loadings) and those with highest 1 reflectance will appear as darkest pixels (negativePRINCIPAL COMPONENT ANALYSIS FOR ALTERATION MAPPING nes TABLE 1. Prncieat CouponenT ANALYSIS ON So TM Banos, RosenTs MmNs. Suoscene. RAW DATA input Bonds TM Tw TNS Twa TMS, ™ Band Means 106.6 5050 8.60 7220 1332 74.90 SD of Bands 37 29 33 29 48 39 Eigen. Eigenvector Matix values() rer +3725 +2354 +4047 “16.58, +6507 +372 (69.635) PC +6339 ta +3729 =1300 “5823 279 (5280) PCs 2! tosa7 “09s, +9227 “asia =H (3510) Po = 48.05, 406.82 +3699 $30.20 = 466 +5805, (0590) PCS $40.69 “iss 66.35 +0928 =19.00 +5781 (0305) PCS 23.89 +9148, 3233 =06.65 +0307 =0187 (oor, (Cigenvector loadings expressed as percentages ofengial bands) Taste 2._Pancira. Courone ANALYSts oN So TM Banos, ROBERTS Mrns. Suescene. STRETCHED DATA Input Bands Ta 72 ™3 Ta TNs To, Band Means wa 170 wa 1270 wa 72 SD of Bands 39 53 53 3 59 53 Bigenvector Matric vals() Per +4218, +2292 +432 “$31.57 441.68 +4239 (85.903) PQ “248 “7 =2157 +9183, +0807 Be (939) PS +4364 +3322 +1671 +2047 “65.15 527 (3a, PC “321 +0875 +1798 +1042 09 +68 (0519) PCS +6223 “3173 “5663 toa 15.99 +4055 (0346) Pos +1885 a5 +6214 $052 =05.51 =03.94 (0200) igenvector loadings). It follows that a simple negation of FC2 dict that “limonitic”iron-oxides (including jarosite and goeth- wl inde ph refectance by bt pale Eigenvector loadings for Pc3 (of Table 1) indicate that dominated by vegetation, highly reflective in Ta an the pos. {tive loading of Tin this FC also indicates that strongly veg. fated pixels will be bright inthis PC image. The percentage of variance mapped into this “vegetation” Peis only 3.514 percent which is obviously not a measure of vegetation abundance in the Roberts Mountains azea where vctwally al peels wil con fain some vegetation and a significant proportion wil be well vegetated, By imposing the same SD and very similar means on the input bands prior fo PCA, as show in Table 2, vegetation {s mapped into Pc2 (1.83 percent positive eigenvector load from Tt, and negative loadings from the visible channels an The, which closely matches the theoretical spectral response of vegetation) and the corresponding eigenvalue for Po2fadicton that this is responsible for a more “reasonable” 9.592 percent of scene variance. ‘An instructive result of the equalization of band means and SDs in Table 2i the effect on te eigenvector loadings to ct. For most bands the eigenvector loadings are very similar in magnitude. The low eigenvector loading from Tuts distinctly anomalous, and this an be explained by the significant popu Inton of high ONS in TaHt which are solely due fo vegetnton, This vegetation variance is then mapped to the next principal component, Pc. It isa general rule that, when means and $03 are equalized, the lowest contributor to PC willinvarably dom. Jnate Pca and thereby account forthe second highest variance. This isnot always vegetation, especially in extremely aid ter- rain, Having mapped albedo to PC1 and visible to m ditferences, oF vegetation, fo PCs 2 and 3, the remaining three PCs can be expected to contain information due tothe varying spectalre- sponse of iron-onies (absorption invisible bande] end 2 and higher reflectance in TMS) and fydrox-bearing minerals (ab- Sorption in Ta7, higher reflectance in TNA). By Looking for mod erate of large eigenvectors loadings for both Tylana TM, in CS where these loadings ae also opposite in sign, vre can pre- A ite) will be distinguished by bright pixels in Pe of Table 1, by dink pels inet Tale, ond by ack pd in Peso TS 2. Hematite (characterized by greater reflectance in TM than in ‘Tia) should be mapped as dark pixels in PCs of Table 1 and bright pixels in PCs ofthe stretched data transformation (Table 5 reining mtd Se abe ad a Seog ae en ee eae ee ea ee ae re eee aa ig ul ot wd ate oe a at ee eae eS eae Se gerne a rae aera bassin fo tebe ese Dae Se ee ipa myer ico tnt bea cen en ee earn eee orca eee ee ere Sane ae rte ae eee ens ee a Be ee eee ee pene pene ee ee ae ee oh a ‘The decision process for both of the above six-band PCAs has cae ee Se eee eee Se ee iene ree el epee tk Tee ae te a ore nee re ee eee ES rea ee eee re eee eee ae1166 Fra. @. Negated Pos from sichand Fa, Hydroxy and iron-oxido bright PCA on Four Bands - Hydroxyl Mapping Table 3 describes the principal components transformation on unstretched Ta bands 14,5, and 7 of the Roberts Mountains subscene. TM bands 2 and 3 have been deliberately omitted to avoid mapping iron oxides, and it should be noted that T¥2or TiB could substitute for Ts in this transformation. Following the reasoning process described above, we can predict that Ct Js the “albedo” image, PC2 describes the contrast between the SIR and the visible region, FCS is brightest for vegetation, and FCt highlights hydroxy-bearing minerals as dark pixels. This “hydroxy?” (H image is therefore negated in Figure 4 to show anomalous concentrations of Has brightest zones. One fortui- tous property of this H image is the fact that it has a negative contribution from vegetation in TM (.e, when negated to show Has bright pixels) which compensates forthe similarity in re- sponse between certain vegetation classes (e.g, those with high leaf water content) and hydroxyls inthe sw. This is a problem when only two or three TM bands are input for PCA, and itis a ‘major drawback in ratio or band subtraction techniques. The simple methodology for hydroxyl mapping by FCA on mM bands, 1,4, 5, and 7 isto examine the eigenvertor loadings for bands'S and 7, in both of FCS and PC4, to find which has a moderate or strong loading from both bands, with opposite signs. Negate only those Pcs where the TW7 loading is positive, so that the anomalous pixels are brightest in all cases. This is the “Crosta hydroxyl image” (H). The PCA should be caried ‘out on unstretched data. In those rare instances where hydroxy] Is mapped into both Fes and Pc, TMs and TAP can be sretched before FCA. An alternative strategy, where hydroxyls are ma Into both Fes and Pc, isto ence tat both show hydrant bright pixels (by negation, as required), followed by a “pai wise” PCA of PC3 and PCs, having Bist stretched PCS and PCA to give each similar means and S0s- One of the to PCS from this transformation will have two positive eigenvector loadings, and that will be the H image. PCA on Four Bands - Iron-Oxide Mapping Table d describes the principal components transformation on unstretched TM bands 1, 3,4, and 5 of the Roberts Mountain subscene. TAT could be substituted for TMS in this analysis with PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1991 TrOUE3._Prncieat Couponen AuaLysis Fon HyoRaXY. MAPPING inpattands STNG EindMens 10eé 7222 7A SbotBnds 372948 gen, igen Mite velo) rer tang ison 47489 +4938 “9S PQ = tas? +0150 asl “esas ( 3am PO = “aaa 5336 oat —s568 (438) Pet ane +3089 “ase Ml (08) Fra. 4. "Hydroxy!" image (H), TABLE 4. PRINCIPAL COMPONENT ANALYSIS FOR IRON-OxiDE MAPPING inputBands TMI TMS TMG TMS Band Means 1066 68.6 72.2 1332 SDofBands 9 373829 Eigen Eigenvector Matrix values) Pei 44290 F465 419.78 +7497 (68521) PC $0078 44135-2087 “5827 (6942) PCS $1419 —1161 +9482 -2595 (3.963) PCH “5.16 +750 +1352 —1763 (0578) litle effect on the result; the omission of one swIR band is to deliberately avoid hydroxyl mapping. The PCs can be inter- preted as albedo in FCI, TR versus visible in PC2, vegetation in PC3, and iron-oxide as bright pixels in PCA. This iron-oxide (F) mage is reproduced in Figure 5. ‘The rules for iron-oxide mapping are similar to those for hy- <éroxyl mapping. The magnitude of eigenvector loadings for TM ‘and TMB in either PCS or Pct should be moderate or strong, and ‘opposite in sign. To enhance iron-oxides as bright pixels, those ‘PC images with negative loadings from TMG should be negated. ‘Where the result is equivocal in that iron-oxide is mapped t© both ofPC3 and FC4, the input bands can be pre-stretched befor? transformation or a pairwise PCA can be run on PC3and FCé, in ‘a manner similar to that described for derivation of the hydrox!PRINCIPAL COMPONENT ANALYSIS FOR ALTERATION MAPPING ‘image. Regardless of the number of steps taken to produce the image, it should be saved as the “Crosta iron-oxide image” (F). Hematite can be mapped by the substitution of Twa for TMi in the above transformation. Tt has been found that hematite ‘mages are invariably noisier than either the F or H images, and. it must be presumed that this is due to a greater proportion of systematic noise in T62. CROSTA COMPOSITES AND ALTERATION MAPPING The monochrome hydro and irn-xide images produced by Aon four bands (och ar thowe in gure) ae) ee Ctsy to interpret in that snomalousconcenteona ese pire trl eategey re represented by te brightest pals on cash iage. There is no'necd to com he Sgervecr seen ate the images tave ben etate,toundestond tnd ntrpet these mages which would be te tase for Pe images ect se band pana component tensors The rosa images have another advantage i that hey can teladded together oprsce an imagen HP nag) on vihich pals ith anomatous concentestons ofboth teow tnd ironies are he bight pas The hydro age of Figure # ad the ro-onideimage at Rgure§ have en nope inthis way to crete the image seprotuced in Figure 6s shouldbe tompared with Raat he socband Peas Hgure The addon ha been achieved here by a parvice eh weg the H and F images a the fo input ars One ofthe Cs fom this tansonm wil have Fostive pangs ras bok, inputs and this athe 14 inage, Care should be hen suing ARG asfomaton to equalize sats of the oper soe ‘uch that the eigenvecerlondng ae apprninatey eal it the output re. Am aleve to sigiy alte snd Images and resale the eulant iage te 2 ey evs Coton Composires The Crosta composite image is created by stretching the H, H+E, and F images so that the brightest pixels in each are favorably enhanced, the means are less than 128, and the dar- est portion of each distribution is clipped (undersaturated) 10 4 certain extent. These three images can then be combined in usr Fa. 6. “Hydroxy plus iron-oxide" Image (H + F). Compare wit Figure a various ways to suit the personal preferences of individual pho- togeotogsts Biferent combinations of Crosta images have been empiri cally assessed, andi appears thatthe combination Hy +P, F in redgreen bine (A) fe most suited fo photogeaogiel infer. pretation, While ican be argued that a universal’ standards esizable in any case, there may be some technical mest inthe Hy HLF, F (Res) combination. The human eye is more adept Bt diferentiating worm reddish colors than col blu color. The ssc cond (glen mead he Hage §S often truly “anomalout” for mineral exploration purposes, specially where it can be seen thatthe stoma ae oh out, trop and residual soils. Vetsly all rocks and sis are afested ty lroncosie staining to some degree and the F images are merely a measur of on-staining intensity thoughout the ares covered by each scene. INTERPRETATION Routine Interpretation By combining the H, H-+F, and F images as red, green, and blue, respectively, a dark (bluish) color composite image is cre- ated ‘on which alteration zones are unusually bright and can be interpreted as follows: White pixels within alteration zones are potentially the most prospective in that they are both iron stained and argillized, bright reddish to orange zones are more argi- lized than iron-stained, and bright cyan to bluish zones are ‘more iron-stained than argillized. ition of Intense Alteration ‘An important alteration type is often encountered where in tensely atgillized or slicified rocks are also heavily iron-stained, and the “mixed” pixels in these zones have higher reflectance in Tag than in Tu firon-oxide here reversing and overprinting the usual TMs>TM7 spectral contrast for argilic materials). These zones may be encountered at deeper erosion levels in mesoth- ‘ermal systems and as the dense siliceous and pyriized caprocks Gjasperoids) often associated with epithermal mineralization. ‘These mixed pixels may also lack typical iron-oxide signatures in the visible spectrum due to the overprint by highly reflectiveUE in 1168 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1991, argilc minerals such as kaolinite. In other words, the presence of iron-axide destroys the “argilic” response in WIR, and, con- versely, the presence of argillic minerals tends to destroy most ofthe fron-oxide response inthe visible spectrum. This impor- tant alteration type can often be recognized on Crosta hydroxyl images by very dark or black pixels closely assocated with bright pixels representing hydroxyl alteration, and on Crosta compos- Ites (H, H+F, Fin Rt) by black or very deep blue colors aso in assocation with obvious alteration colors. Not all zones with higher reflectance in Tat” than in TMs ate altered in this way, as this (Tu?> TMs) spectral contrast can also be due to disturbed ‘ground, recent sol erosion, or reduced vegetal cover (Loughlin, 1990) The recognition ofa contextual association with genuine argilization is therefore extremely important "The dark core of the concentrically zoned alteration system around Long Peak in the Battle Mountain Range (Lander County, Nevada), illustrated by the hydroxyl image of Figure 7 (ATM, 512 by 512 pixel, 75m resolution, is an example ofthis type of mixed pixel, intensely altered zone. This is & potential por- Phyry type system Figure 8 (TM) is an example from the Chilean Andes where inueoey alle rok form a dark “doughnt between inet and outer zones of argilzation and iromstaining, in the north west of the image. In this case iron-ovide response from the Visible spectrum is not completely destroyed forall ofthe dark zone on the hydroxyl image. A narrow and discontinuous ring Structure where (visible region) iron-oxide response is lacking, within the broader doughnut, coincides precisely with a very resistant ridge which is red on the TM7,4,1 (RGB) composite and Black on the Crosta composite. in context, this ridge can be interpreted as pyrtized jasperoid. ‘These intensely altered zones can be aesthetically enhanced by substituting a Image which shows high SW reflectance pixels (for instance, a negated version of PCs in Table 2) as the fed component of an image with H and F images as the green and blue components. This complication is no strictly necessary in reconnaissance alteration mapping, provided that Fic. 7. aTa “Hydroxy” image (H). Long Peak, Battle Mountain Range, ‘Nevada, the significance of dark zones in association with recognizable alteration is appreciated during photogeological interpretation. CONCLUSIONS AND RECOMMENDATIONS The methodology described above has been tried and tested. fon TM and ATM images from various parts of Nevada and Or- egon, and on TM images from southern Spain, the eastern Med- iterranean, the Middle Eest, and the Andes. The processed images are very reliable for rapid alteration mapping, for finding the ‘most prospective outcrop within large alteration zones, and for drawing attention to subte, often unsuspected, alteration zones. For more specific applications, such as detecting extensions cof known mineralization, the analyses are much more sensitive ‘when the area for the determination of image statistics is care- fully selected to avoid areas of well exposed alteration, playa clays, or dense woodland. (Oren APPLICATIONS ‘The Crosta technique relies on the principal component trans- form to map information common to all bands—overall scene brightness or albedo—into the first principal component and specific spectral contrasts into successive components. The technique should also be used for processing Geoscan or GERIS image data to determine subtle spectral differences between closely spaced channels such as those in both sWik and TIR regions on the Geoscan scanner. In the case of multispectral ‘Tk, the frst principal component will account for heat effects and successive components should enhance subtle differences in emissivity. SoFTWARE REQUIREMENTS Standard principal components analysis software has been designed for operations such as information compression and decorrelation stretching, not for information extraction as de- scribed above. The computation of Crosta images will be made snuich faster and simpler if a number of interactive options are built into the software. The software options should include the ‘opportunity to view the image statistics and eigenvector matrix before any transformations are performed, the opportunity togr FRINCIPAL COMPONENT ANALYSIS FOR ALTERATION MAPPING 16 abort the process at that stage, and the further option of se- REFERENCES lecting only one principal component for computation and au- tomatic negation of the selected component, f required Crosta, AP, and J. MeM Moore, 1969. Enhancement of Lands The- = . os matic Mapper Imagery for Residual Soil Mapping in SW Minais Gerais State, Brat A Prospecting Case History in Greenstone Bet Terrain. Precedings ofthe 70s (ERIM) Thomatc Conference, Renote ACKNOWLEDGMENTS Sensing for Exploration Geniogy. Calgary, 2-6 08, pp. HSB? Lough, W. P., 1990, Geological exploration inthe western United This work would not have been possible without the contin. "Sates by usd ofasbome Saones apes IMM Cope: Ree uued support and encouragement of Mike Hammond and his Sensing, en OpestionalTecmley forte Ming end Peru staff at the United Kingdom NRSC. Hunting Technical Services dart. Landon, 28-3 Oct pp. 223-241 supplied the U.S. aT and Chilean TM image data Kathy BOw- singh, A.,and A. Harrison, 1985, Standardized Principal Components. den of Rio Tinto Zine supplied the Nevada TM imagery Ini. Remote Sesng, Vol 6, No.6 pp. 883896 Forthcoming Articles Eugene M. Barnaba, Warren R. Philipson, Arlynn W. Ingram, and Jim Pim, The Use of Aerial Photographs in County Inventories of Waste-Disposal Sites. ‘Mickel Boulianne, Louis Cloutier, and Sanjib K. Ghosh, Cerebral Biopsies Using a Photogrammetric Probe Simulator. Haluk Cetin and Donald W. Levandowski, Interactive Classification and Mapping of Multi-Dimensional Remotely Sensed Data Using ‘-Dimensional Probability Density Functions (aPDF). ‘Yue Hong Chow, Slope-Line Detection in a Vector-Based GIS. ‘Raymond L. Czaplewsti, Misclassfication Bias in Areal Estimates. Heinrich Ebner, Wolfgang Kornus, Gunter Strunz, Otto Hofmann, and Franz Muller, A Simulation Study on Point Determination Using ‘MOMS-02/D2 Imagery. Peter F. Fisher, First Experiments in Viewshed Uncertainty: The Accuracy of the Viewshed Area. Peter F. Fisher, First Experiments in Viewshed Uncertainty: Simulating Fuzzy Viewsheds. G. M. Foody, A Fuzzy Sets Approach to the Representation of Vegetation Continua from Remotely Sensed Data: An Example ‘om Lowland Heath. Steven E, Franklin and Brailey A. Wilson, A Three-Stage Classifier for Remote Sensing of Mountain Environments. (Clive S. Fraser, Photogrammetric Measurement to One Part in a Million. W, Frobin and E. Hierholzer, Video Rasterstereography: A Method for On-Line Measurement of Body Surfaces. Beng Gong and J. Douglas Dunlop, Comments on the Skidmore and Turner Supervised Nonparametric Classifier (and response). Daniel K. Gordon, Paul W. Mueller, and Matthew Heric, An Analysis of TIMS Imagery for the Identification of Manmade Objects. Christian Heipke, A Global Approach for Least-Squares Image Matching and Surface Reconstruction in Object Space. Peter E. Joria, Scan C. Ahsarn, and Michael Conner, A Comparison of the SPOT and Landsat Thematic Mapper Satellite Systems for Detecting Gypsy Moth Defoliation in Michigan. D. King, Determination and Reduction of Cover Type Brightness Variations with View Angle in Airborne Multispectral Video Imagery. Richard G. Lathrop, Jr, Landsat Thematic Mapper Monitoring of Tuzbid Inland Water Quality. J. Laorea, De-Hazing Landsat Thematic Mapping Images. Zhilin Li, Bffects of Check Points on the Reliability of DTM Accuracy Estimates Obtained from Experimental Tests. Donald L. Light, The New Camera Calibration System at the U. S. Geological Survey. ‘Hans.Gerd Mass, Digital Photogrammetery for Determination of Tracer Particle Coordinates in Turbulent Flow Research. Ronald T. Marple and Eugene S. Schweig, II, Remote Sensing of Alluvial Terrain in a Humid, Tectonically Active Setting: The New Madrid Seismic Zone. Fabio Maselli, Claudio Conese, Ljiljana Pethov, and Raffaello Resti, Inclusion of Prior Probabilities Derived from a Nonparametric Process into the Maximum-Likelihood Classifier. ‘Ram M. Narayanan, Steven E. Green, and Dennis R. Alexander, Soil Classification Using Mid-Infrared Off-Normal Active Differential Reflectance Characteristic Kurt Nooak, Rectification of Digital Imagery. Laurent Polidori, Jean Chorowice, and Richard Guillande, Description of Terrain as a Fractal Surface, and Application to Digital Elevation Model Quality Assessment. Kevin P. Price, David A. Pyke, and Lloyd Mendes, Shrub Dieback in a Semiarid Ecosystem: The Integration of Remote Sensing and Geographic Information Systems for Detecting Vegetation Change. ‘Michael Rast, Simon J. Hook, Christopher D. Eloidge, and Ronald E. Alley, An Evaluation of Techniques for the Extraction of Mineral Absorption Features from High Spectral Resolution Remote Sensing Data. Randall L, Repi, Jae K. Lee, Paul W. Mausel, David E. Escobar, and James H. Everitt, An Analysis of Selected Water Parameters in Surface Coal Mines Using Multispectral Videography. Omar H. Shemdin and H. Minh Tran, Measuring Short Surface Waves with Stereophotography. Thierry Toutin, Yoes Carbonceau, and Louiselle St-Laurent, An Integrated Method to Rectify Airborne Radar Imagery Using DEM. William. Warner and Qystein Andersen, Consequences of Enlarging Small-Format Imagery with a Color Copier. Zhuogizo Zeng and Xibo Wang, A General Solution of a Closed Form Space Resection,