REMOTE SENSI NG OF ENVI RONMENT 21:119-128 (1987) 119

Combining Panchromatic and Multispeetral

Imagery from Dual Resolution Satellite Instruments
JOHN C. PRICE
USDA-ARS Remote Sensing Research Laboratory, ASRI, Beltsville, Maryland 2070,5.
A procedure is developed for combining high spatial resolution panchromatic data with lower resolution multispectral
data in order to produce high spatial resolution digital data in multispectral | onn. Data simulating the French SPOT
satellite were processed to resemble high altitude aerial photography, but image artifacts can hamper photointerpreta-
tive methods.
Introduction (USDA, 1982). At present photointerpre-
tation of aerial photography is being con-
Since t he launch of Landsat 4 in 1982 sidered for assessing land use and land
t he scientific communi t y has been pro- condition. Application of automated tech-
vi ded wi t h high quality digital images of niques to digital satellite data may be an
t he earth' s surface. In the mid and late effective tool at some future time,
1980s dat a of 10 m spatial resolution will dependi ng on the quality and cost of
be acqui red from the Satellite Pour l' Ob- satellite data, and the availability of pro-
servation de la Terre (SPOT), and future cedures for extracting the desired assess-
Landsat satellites will add a 15 m pan- ment of large areas.
chromat i c band to the current 30 m In this paper a technique is developed
Themat i c Mapper channels. Such data for combi ni ng dual resolution digital data,
provide opportunities for detailed ob- as from SPOT and the planned Thematic
servation of processes at the earth' s Mappers, to produce high spatial resolu-
surface, and for mapping of urban areas tion multispectral data. The goal of this
and description of land use. As pointed work is essentially equivalent to that of
out in a recent paper by Cliche et al. Cliche et al. (1985), except that the anal-
(1985), t he SPOT data may be processed ysis is based on statistical properties in
by image enhancement algorithms to yield t he data, rat her than an ad hoc approach
visual product s resembling color aerial whi ch seeks an optimum image display.
photography. In particular, the analytic approach may
Although many applications of such serve as a basis for design of future earth
dat a will involve mapping of urban and observing systems, and the resultant
suburban areas, there are possible uses products, representing a reconstruction of
requiring bot h high spatial resolution and high spatial resolution multispectral data
large area coverage. For example, each 5 from a sampled data set, are suitable for
years the Soil Conservation Service of the furt her numerical studies, such as com-
U. S. Depar t ment of Agriculture carries put er classification. The basis for the
out an i nvent ory of the condition of non- t echni que depends on the substantial re-
federal rural lands in the United States dundancy existing in most high resolution
00344257/87/$0. 00
120 I.C. PRICE
mt dt i spect r al satellite data. This re- I n this paper "i nformat i on" is used in
dundancy implies t hat properly chosen t he original sense of Shannon (1948), i.e.,
dat a subsets, as from SPOT, may be as a statistical represent at i on of t he vail-
processed to yield dat a simulating t hat ability in a dat a set. As has been poi nt ed
from a less sophi st i cat ed but more power- out (Ter Harr, 1961; Price, 1984b), t he
ful (and expensive) observing i nst rument , t er m " ent r opy" is often used in t he en-
Despi t e t he attractiveness of this ap- gi neeri ng literature as a synonym for t he
pr oach some sacrifices are expect ed when wor d " i nf or mat i on, " despite t he possibil-
t he dat a acqui red by a remot e i nst rument ity of confusi on wi t h t he t hermodynami c
r epr esent a sampl e rat her t han a corn- defi ni t i on of ent ropy. In this paper infor-
pl et e coverage high resolution multispec- mat i on refers to t he variability of a set of
tral dat a set. Variability associated wi t h a observat i ons havi ng a probability distil-
number of unknown causes cannot be but i on p/,
fully removed, so t hat resultant dat a
pr oduct s have a residual noise compo- DN,,~
nent . The const ruct i on of hi gh spatial H = - Y'~ p i l o g 2 p i ,
resolution dat a sets from combi ned pan- ~= 0
chr omat i c and multispectral dat a is dis-
cussed in t he next section. I n t he third wher e t he Pi represent t he probability of
sect i on t he limitations of t he pl anned a gi ven measur ement value, often called
t echnol ogy for image analysis are il- digital number or DN. For example, a
l ust rat ed t hr ough an example drawn from series of observations correspondi ng to
r andom number s in t he range of 0-255,
Themat i c Mapper data. In some cases
wi t h each value having a probability of
i mage artifacts may be present whi ch are
1/ 256, has an information value of
di st ract i ng at best and whi ch may con-
fuse phot oi nt erpret at i on. 255
H = - Y~ ( 1/ 256) I og2( 1/ 256) = 8 bits.
Est i mat i on of 10-m Multispectral Data / = 0
I n several r ecent paper s (Price, As di s cus s ed by Pri ce (1984a),
1984a, b) t he substantial r edundancy of Themat i c Mapper dat a from agricultural
Landsat and SPOT simulation dat a has areas generally are rat her restricted in
been illustrated by comput at i on of t he range, and t end to group about a mean
i nformat i on cont ent of actual data, as value, yi el di ng an information cont ent of
compar ed wi t h t he information whi ch 4 - 5 bits per pixel per spectral channel
coul d, in principle, be acquired by t he i nst ead of t he theoretically possible 8 bits.
r emot e sensor. It appears t hat in a typical Correl at i on among spectral channels fur-
agricultural scene t he information con- t her reduces t he information cont ent of
t ent represent s about 40% of t he dat a mul t i spect ral data. The choice of spectral
l engt h of st andard format digital data. channel s for t he SPOT i nst rument results
Thus, in principle, such dat a could be in a very hi gh correlation among channel
encoded and compressed by approxi- 1 (0. 50-0. 59 /Lm), channel 2 (0. 61-0. 68
mat el y 60% wi t hout reduci ng t he utility /zm), and channel 4, t he panchromat i c
of t he data. channel ( 0. 51- 0. 73/ zm) . Thus t he count
COMBINING DUAL RESOLUTION SATELLITE IMAGERY 121
values from each pixel in one of these smaller. In fact, the 10 m SPOT data can
channels may be used to predict the count be encoded in the spacecraft in an
value of this pixel in the others, to within 8 - 5 - 5 - 8 format for four successive pixels,
a small probable error. A joint probability wi t h the 5-bit data words representing
funct i on may be comput ed from which it t he difference bet ween the actual count
is possible to estimate the reduction of value and t he interpolated value derived
information content associated with the from the end-point 8-bit words. Obvi-
r edundant spectral channels (Malila, ously, the 5-bit words cannot, in general,
1985). It is this redundancy which per- retain the flfll accuracy of the 8-bit val-
mits the estimation or reconstruction of ues, but a lookup table with expanded
high spatial resolution multispectral data. ranges for the large (and improbable) dif-
Because the scale of 10 and 20 m reso- ference values is used to retain the major
lution dat a is much smaller than the typi- part of the information of the full 8-bit
cal scale of spatial variability for most data.
l and areas t here is a pronounced tend- As illustrated in Fig. 1, the dual resolu-
ency for spatial redundancy in the image tion image dat a from SPOT and the fu-
data, i.e., each pixel tends to be similar to t ure Themat i c Mappers combines the
t he surrounding ones. This suggests the spectral redundancy due to the overlap of
use of a suitable transformation, e.g., a t he spectral intervals in the visible, with
del t a or difference transformation, in t he spatial redundancy of the panchro-
whi ch each pixel value except the first is mat i c channel, as four 10-m pixels lie
repl aced by its difference with its neigh- wi t hi n a single multispectral pixel. Be-
bor in t he dat a stream. This has the effect cause the 10-m data are very highly corre-
of replacing larger numbers representing lated with channels 1 and 2, but much
t he sateflite observed radiances, with the less correlated with channel 3, the two
differences, whi ch are generally much cases will be discussed separately. The
Si 2
8i I
!
!
- ~ [ s , ' , s , ~
I
I '
f s , ' , s l ,
l O m S 2
Pi1 Pi 2
P I3 P I 4
FIGURE I. The considerable spatial and spectral redundancy of the 10 and 20 m data perudts the
construction of high resolution mtdtispectral data.
122 J.C. PRICE
met hodol ogy to be devel oped resembles If b , , 4 is zero, t hen this result is t he same
t he pr ocedur e used in t he t r eat ment of as t hat used in t he SPOT simulation pro-
t he SPOT simulation dat a (SPOT Image gr am ( SPOT Image Corp., 1983, p. 25),
Corp. , 1983). i.e.,
Combination of channels / 4 j / [ 4 4 i l
1 and 2, and panchromatic data xi~'j = X ~' x 0. 25 ~ x .
Fr om statistical analysis of t he SPOT L j = 1 .1
dat a one finds t hat dat a from channels 1
and 2, and t he panchromat i c channel, to However, b is generally nonzero.
be number ed 4, are all highly correlated. The validity of this procedure was
t est ed usi ng SPOT simulation dat a from
Thus
Washi ngt on, DC. Since 10-m four-chan-
X~ " = a , , , , X ' ~ + b . . . . m :~ n = 1, 2, 4 nel dat a wer e not available, t he verifica-
t i on was carried out by averaging t he
wher e t he superscri pt represents t he t hr ee channel s of 20-m dat a and t he single
channel number , t he subscript i t he pixel channel 10-m dat a to 40 and 20 m, re-
number , i.e., on an image line, and a , , , , spectively. Application of t he above pro-
and b,,,, are det er mi ned by a best-fit cedur e to t he averaged dat a pr oduced
pr ocedur e to a sample of t he data. This 20 m values for channels 1 and 2, whi ch
rel at i onshi p is sufficient to estimate t he wer e compar ed with t he original t rue val-
10-m mul t i spect ral values from t he chan- ues at 20-m resolution. Residual errors
wer e of t he order of 2 counts as com-
nel 4 dat a wi t h a hi gh degree of accuracy.
The residual error may be r educed by par ed to t he st andard deviation of about
usi ng t he known value of t he 20-m pixels 20 cot mt s in t he original data, i.e., t he
for channel s 1 and 2 , as follows: Let t he 20-m channel 4 dat a described 99% of t he
vari ance of 40-m data in channels 1 and
smaller (10 m) pixels be identified by x i , ,
j = 1-4. The n using t he channel 4 values 2. This suggests that t he panchromat i c
to est i mat e 1 and 2, dat a can provi de an excellent estimate of
t he 10 m values in channels 1 and 2.
X I ' j ~ - a m 4 x 4 j + b , , 4 + 8"'ij, m = 1, 2
Combination of
Channel 3 and panchromatic data
wher e 8ij is t he error in t he est i mat ed
value x i' ~. However, t he average of t he The pr ocedur e described in t he previ-
four subpixel values must equal t he mea- pus subsect i on is unsatisfactory for t he
sured value X~" of t he 20-m pixel. Thus channel 3 dat a because t he visible and
one may comput e a correction factor f = near i nfrared spectral channels are not
X m 4 ,, h
/[0. 25F~j=lx~, ] sue t hat t he correct linearly related. In t he SPOT Simulation
sum is obt ai ned for t he four 10-m pixels, Program cubi c convolution interpolation
i.e., was used to create 10-m pixels from t he
20-m data. This met hod discards entirely
x ;'j = X m ( a . , 4 x 4 + bm 4 ) / t he i nformat i on provi ded by t he panchro-
mat i c channel of data. In addition, it does
[ 0 . 2 5 ~ _ _ ( a m 4 x 4 + b m 4 ) ] . not satisfy t he integral pr oper t y t hat t he
1 four 10-m pixels must average to t he 20-m
COMBINING DUAL RESOLUTION SATELLITE IMAGERY 123
val ue. A s c a l e d one - di me ns i ona l e xa mpl e wh e r e ~3( x4) is t h e me a n val ue of x 3 f or
is a p p r o p r i a t e ; if a me a s u r e me n t S i is all p i c t u r e e l e me n t s i n t h e i ma ge wh i c h
o b t a i n e d at x = 0, r e p r e s e n t i n g t h e i nt e r - h a v e t h e g i v e n val ue x 4. Th e s e 256 val ues
val ( - 1 / 2 , 1 / 2 ) , t h e n i n t e r p o l a t e d f or x 3 ( I ) f or I = 0 - 2 5 5 ar e o b t a i n e d whi l e
va l ue s S~1 at - 1 / 4 , r e p r e s e n t i n g t h e t h e a , , n a n d br an val ues ar e b e i n g de -
i nt e r va l ( - 1 / 2 , 0) a n d S~2, at + 1 / 4 , r i ve d f or c h a n n e l s 1 a n d 2. De r i va t i on of
r e p r e s e n t i n g t h e i nt er val (0, + 1 / 2 ) , mu s t 10- m d a t a f or c h a n n e l 3 is t hus a s i mpl e
a v e r a g e t o t h e or i gi nal val ue S i. How- l o o k u p of t h e p r e d i c t e d val ue, gi ven f r om
ever , f r o m c u b i c c onvol ut i on t h e val ues t h e k n o wn va l ue of x 4. Th e r es ul t i ng f our
ar e gi ve n b y S~1= 0. 070Si _ 2 +0. 227Si _ 1 n e s t e d pi xel s ar e t h e n c o r r e c t e d as i n B
+ 0.867S~ - 0. 023S~+ 1 a n d Si2 = - f or t h e f a c t t h a t t h e i ni t i al e s t i ma t e of t he
0 . 0 2 3 S i _ 1 +0 . 8 6 7 S i +0 . 2 2 7 S ~ 1 - 0 . 0 7 0 S i 2 , s u m doe s n o t gener al l y agr ee wi t h t h e
wi t h t h e a v e r a g e equal t o - 0.035(S~ 2 + me a s u r e d val ue.
S~2) +0. 102( S~_ 1 + Si 1 ) +0 . 8 6 7 S ~. We Thi s t e c h n i q u e was a ppl i e d t o aver -
n o t e i n p a s s i n g t h a t " c u b i c c o n v o l u t i o n " a g e d 10- a n d 20- m d a t a as de s c r i be d i n
is n o l o n g e r u n i q u e l y de f i ne d, as t h e coef - t h e p r e v i o u s s ubs ect i on. Thi s p r o c e d u r e is
f i c i e nt s u s e d f or i nt e r pol a t i on of La n d s a t onl y mo d e r a t e l y successf ul , wi t h t h e pr e -
MSS d a t a we r e mo d i f i e d at s ome p o i n t i n d i c t e d val ues f or c h a n n e l 3 a c c o u n t i n g f or
t i me d u r i n g a ppl i c a t i on t o Th e ma t i c a b o u t 75% of t h e va r i a nc e of t he or i gi nal
Ma p p e r d a t a ( Fi s chel , 1984) Th e s e da t a . Sever al a l t e r na t i ve a l gor i t hms we r e
we i g h t i n g coef f i ci ent s ar e gi ve n b y t e s t e d , wi t h ne a r l y i dent i cal concl us i ons ;
i . e. , t h e r e doe s n o t s e e m t o exi st a gener al
p r o c e d u r e wh i c h can der i ve hi ghl y a c c u-
w 1 = a d ( 1 - d ) 2, r a t e val ues f or 10-m pi xel s of c h a n n e l 3.
w 2 = ( a + 2 ) d 3 - ( a + 3 ) d 2 + 1, T h e r e a s on is r eadi l y es t abl i s hed f r om
i n s p e c t i o n of t h e t wo- di me ns i ona l hi s t o-
w 3 = - ( a + 2 ) d ~ + ( 2 a + 3 ) d Z - a d , g r a m of c h a n n e l 3 d a t a vs. c h a n n e l 4
w 4 = a d 2 ( 1 - d ) , da t a , as i l l us t r a t e d i n Fi g. 2.
Ev i d e n t l y l o w val ues of r ef l ect i vi t y i n
c h a n n e l 4 ma y be as s oci at ed e i t he r wi t h
wi t h a = - 1 f or MSS d a t a a n d ol der l o w val ues of r ef l ect i vi t y i n c h a n n e l 3, as
T h e ma t i c Ma p p e r dat a, a n d a = - 1 / 2 f r o m s h a d o ws or wat er , or wi t h hi gh val -
f or n e we r T h e ma t i c Ma p p e r dat a. He r e d ue s of r e f l e c t i vi t y i n c h a n n e l 3 d u e t o
is t h e d i s t a n c e ( 0, 1) f r om p o i n t x 2 a n d x 3 v e g e t a t i o n . L o w c or r e l a t i on b e t we e n
i n t h e ser i es of f our poi nt s , x 1, x2, x 3, x4. s pe c t r a l c h a n n e l s is f avor abl e f or s pect r al
A mo r e ge ne r a l r e l a t i ons hi p t h a n t h e d i f f e r e n t i a t i o n a n d cl assi fi cat i on, b u t u n -
l i ne a r f i t of t h e pr e vi ous s ubs e c t i on b e - f a vor a bl e f or i n f e r e n c e of hi gh spat i al res-
t we e n t h e 10-m d a t a a n d c or e gi s t e r e d o l u t i o n mu l t i s p e e t r a l val ues f r om dual
20- m d a t a is o b t a i n e d b y c o mp u t i n g t h e s pat i al r e s ol ut i on dat a. T h e c h a n n e l 3 a m-
e x p e c t e d ( me a n ) val ue f or c h a n n e l 3 f or b i g u i t y is e s pe c i a l l y s e ve r e i n t h e
e a c h g i v e n va l ue i n c h a n n e l 4, a f t e r t h e Wa s h i n g t o n , DC s cene wh i c h was us e d
10- m d a t a h a v e b e e n a v e r a g e d t o 20 m. f or t hi s s t udy, as ma n y ci t y pi xel s c ont a i n
T h u s b o t h s h a d o ws of bui l di ngs a n d t rees. Th e
p r o b l e m is e x p e c t e d t o be less s ever e f or
x ~ i ( x 4 ) = ~ 3 ( x 4 ) , a gr i c ul t ur a l scenes, b u t t he a c c ur a c y of
124 J. C. PRICE
VEGETATION
SHADOWS
|
COUNTS
255
255
0 CFt AI~NEL $4
FIGURE 2. Two-dimensional histogram of SPOT channels 4 and 3,
illustrating that a value in channel 4 does not specify adequately the
value in channel 3, especially for low values in channel 4.
t he procedure depends on the nature of section. Considerable detail is evident,
t he scene, in general, suggesting t hat the increased spatial reso-
Although derivation of the fitting lution (10 m) of SPOT data will be of
paramet ers for the spatial interpolation great value for studies of urban land use
appears to require an additional step of and for photointerpretation generally.
comput at i on, such statistical properties Fi gure 4 shows that even very small fea-
may be readily obtained during calibra- tures may be inferred from the 10-m data
tion and production processing of the under opt i mum conditions.
satellite dat a before distribution to data Unfortunately, even the high spatial
users (e.g., NASA, 1978). resolution dat a are susceptible to sam-
pling limitations due to within pixel aver-
Artifacts and Spatial Interpolation aging. This averaging can produce
peculiar effects when evaluated by photo-
Wi t h t he advent of higher spatial reso- interpretation. From canals on Mars to
lution digital data it is possible to resolve linears as mapped by geologists, line seg-
not only evi dence of man' s influence on ment s produce a strong response from
t he earth' s surface, such as agricultural t he human visual system. Caution must
fields, dams, street and road patterns, be exercised when high contrast linear
etc., but also considerable spatial detail in features are present in image data. The
these features. Figure 3 illustrates com- following example and its quantitative ex-
bi ned panchromat i c and multispectral planation illustrate the potential problems
i magery of t he nation' s capitol in the associated wi t h narrow linear features as
District of Columbia, as derived from the well as a possible solution for image
met hodol ogy described in the previous restoration. However, there appears to be
COMBINING DUAL RESOLUTION SATELLITE IMAGERY 125
FIGURE 3.' Washington, DC: the Capitol building is at the center of the 2.5 km square area.
no general procedure for eliminating recorded by the Landsat 4 Thematic
artifacts except oversampling, i.e., imag- Mapper on 12 December 1982 (Scene I D
ing each point more than once. Of course, 40149-17444). The peculiar ladderlike ap-
oversampling provides very little ad- pearance results from the scale of the
di t i onal information about the land display, wher e magnification permits dis-
surface, serving mostly to smooth out the crimination of individual detector re-
image defects which occasionally result sponses as t hey cross over the road
from the 1-1 sampling of instruments surfaces. Figure 6 illustrates an idealized
such as SPOT and the Landsat Thematic nadi r vi ew of a narrow dark linear feature
Mappers. in Fig. 6(a), the averaging effect of the
Fi gure 5 illustrates the appearance of large pixel size and the resultant detector
dark, narrow strips, i.e., parallel roads " responses for three adjacent detectors in
separat ed by several hundred meters, as Fig. 6(b), and an analytic solution in Fig.
126 J . C. PRI CE
FI GURE 4. Unde r opt i mum condi t i ons feat ures as small
as aut omobi l es ma y be recogni zed in t he SPOT 10 m
dat a. Car s ar e not resol ved in t he ma t c hi ng aerial pho-
t ogr aphy, but t hei r pl acement wi t h respect to l ane
mar ker s is suf f i ci ent for positive identification.
COMBI NI NG DUAL RESOLUTI ON SATELLITE IMAGERY 127
FI GURE 5. Themat i c Mapper image of a di vi ded high way 15 km west of Blythe, CA. The st ri pi ng represents
responses of individual detectors.
Orr 3 I S C A N - " /
I - l L ~'~'~R 0 A D
I-
N A D I R V I E W
S-
/
n.. I -.~j,~.~ 2 f l /
Lkll-- ( ~ I
I-- :3 ~ ~ A'~-.~ I . . . .
U.I 0 ~ ~ ' ~ "
,,.,,
S C A N ' - "
a . W = d s i n 0
01-3 f ~ ~ t / ==; a r = a+( f - a) ( L I W) c o s O
I I
FI GURE 6. Sc he mat i c represent at i on of a single narrow line (road)
as vi ewed from above (a), t he correspondi ng out put s for adj acent
det ect ors (b), and t he sol uti on for t he relative det ect or / r oad angle, t he
wi dt h of t he road, and t he reflectivity of t he road, given i deal i zed
meas urement s (c). Not e t hat t he result is not produced by scan
di recti on; i.e., t he same image would be pr oduced by scan mot i on in
t he perpendi cul ar di recti on.
6(c), assumi ng a background reflectivity ground, etc. , strongly suggest using aver-
a, and a road reflectivity given by a r. ages from a number of appropriately
Al though solution for 0, w, and a r is shi fted i mage lines. If the configuration is
possi bl e in this idealized case, gi ven the recogni zabl e by an i mage analyst, then
cont i nuous trace from only the one det ec- magni fi cati on and interpolation may be
tor, i n practi ce the uncertainty in estimat- used to restore the linear features to
i ng t he roundi ng width ( d) , the change thei r accust omed appearance. While an
in detector output as compared to back- i mage analyst may readily i denti fy this
128 J.C. PRICE
particular t ype of image artifact, the References
situation in geometrically complex urban
areas, given differing sizes and shapes of Cliche, G., Bonn, F., and Teillet, P. (1985),
buildings and roads, etc., is not favorable. Integration of the SPOT panchromatic
Simple interpolation of the image does channel into its multispectral mode for
not improve the display of features such image sharpness enhancement, Photo-
as t hat in figure 5. gramm. Eng. Remote Sens. 3:311-316.
Fischel, D. (1984), Validation of the Thematic
Mapper radiometric and geometric correla-
Concl usi on and Outlook tion algorithms, IEEE Trans. Geosci. Re-
mote Sens. GE-22:237-242.
I n this paper the techniques for pro- Malila, W. A. (1985), Comparison of the in-
duci ng equivalent high resolution multi- formation contents of Landsat TM and MSS
spectral dat a have been discussed in the Data, Photogramm. Eng. Remote Sens.
9:1449-1457.
context of t he information content of the
i mage data. To the extent that such NASA (1978), Heat Capacity Mapping Mis-
sion User's Guide (J. Price, Ed.), Goddard
met hods are feasible there exist a number
of promising developments: Space Flight Center, Greenbelt, MD, 120
pp., revised 1980.
1. The design of bet t er sensors, with
t he choice of spectral intervals, spa- Price, J. C. (1984a), Comparison of the infor-
mation content of data from the Landsat-4
tial resolution, and precision selected Thematic Mapper and the Multispectral
on t he basis of the information con- Scanner, IEEE Trans. Geosci. Remote Sens.
t ent of t he data vs. the cost of in- GE-22:272-281.
st rument at i on and data handling. Price, J. C. (1984b), Statistical analysis of
2. The use of coding techniques, both SPOT simulation data and comparison with
in t he spacecraft and in ground Landsat-4 Thematic Mapper data, In SPOT
processing, to reduce data volumes Simulation Applications Handbook, Am.
and simplify the task of image anal- Soc. of Photogrammetry, Falls Church, VA.
ysis and interpretation, pp. 157-164.
3. The devel opment of specialized in- Shannon, C. E. (1948), A mathematical the-
terpolation and display techniques ory of communication, Bell Syst. Tech. I.
to bri ng out the maximum detail in 27:379-423 and 623-656.
digital data, using human guidance SPOT Image Corp. (1983), 1983 U.S. SPOT
if necessary. The photo enhance- Simulation Campaign, Auxiliary Informa-
ment techniques of Cliche et al. tion Package, 90 pp.
(1985) are in this domain. Ter Haar, D. (1954), Elements of Statistical
Clearly, t he new generation of satellite Mechanics, Reinhart, New York, p. 161.
sensors open new opporttmities for appli- U. S. Department of Agriculture (USDA)
cations to t he earth sciences. (1982), Basic statistics of the 1977 National
A Fort ran listing for interpolating lin- Resource Inventory, Soi l Conservation
ear features, e.g., roads, may be obtained Service, Iowa State University Statistical
from the author. Library Bulletin 686, 267 pp.
Received 25 February 1986; revised 19 September 1986.