Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=black. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Blackwell Publishing and The Royal Geographical Society (with the Institute of British Geographers) are
collaborating with JSTOR to digitize, preserve and extend access to Transactions of the Institute of British
Geographers.
http://www.jstor.org
270
ABSTRACT
Regionalizedvariabletheoryhasmanypotentialapplications in physicalgeography.Theprincipalone is theestimationof
propertiesdistributedover the earth'ssurfaceto assesstheirspatialvariation.Thepaperillustratesthiswith examplesin
whichthethicknessof coverloamandthestonecontentof thesoil areestimatedandmappedoversmallregions.Thesame
procedureis used to mapthe annualrainfallover EnglandandWales,the resultsof whichdemonstratethe difficultyof
doingso withtoo few data.Thetheorycanalsobe usedto designrationalsamplingschemesformapping,andanexample
of a surveyin theWyreForestdescribesanefficienttwo-stageprocedureforthis.Applyingregionalizedvariabletheoryto
hillslopeprofiledataprovidesstatisticalsummaries of theirspatialstructure.
TheexamplesfromDartmoorshow how the
wavelengthsandamplitudesof periodicvariogramswereusedto interpretandcomparethe slopeforms.
KEYWORDS:Anisotropy,Kriging,Nested surveyandanalysis,Periodicvariogram
Trans.Inst. Br. Geogr.N.S. 14: 270-286 (1989) ISSN: 0020-2754 Printedin GreatBritain
Geostatisticsinphysical applications
geography: 271
accoLimt for the very uneven establishment of crops surface of the sand and gravel which were probably
there. Corbett and Tatler (1974) surveyed the soil caused by cryogenic perturbationsin the late Pleisto-
origiinally,and later they recorded depth of the cover cene.
loam in detail. Burgess and Webster (1980) then used The kriging equations, equations (11) to (13) in
the same measurements to produce optimal maps of PartI, were solved using the coefficients in the spheri-
the A variation in the thickness of loam using the cal model above. Estimates were made at points and
methods of regionalized variabletheory. over blocks at the intersections of a 6-67 m grid from
the original data. The nearest 25 observations were
used to compute each estimate. The results can be
1000
* displayed by threading the isarithms through the
*
.:* . . resulting fine figure fields to give 'contour' maps or
... * *perspective
. ? diagrams of the statistical surface using
800
. '*** *the programSurfaceII(Sampson, 1978). Figure2a is a
?,;~~/ * * .
:perspective diagram of such a surface produced by
600
I- */ '* * punctual kriging, and Figure 2b the corresponding
E
isarithmic map. The surface passes through the
~~~~~-;/y~~~*
400
400 observed values, and the nugget effect manifestsitself
as marked discontinuities at the sampling points.
3
200 Y(h)187.0+603.81 10.2-
(1-2)3-
o<h 101.2 These can be seen clearlyon both the perspective
Y(h).187.0+603.8 h >101.2 diagramandthe map.Suchlocalirregularitiesdo not
Y(o)- occur with block kriging. The surface of estimated
0 100 200 360 values computed from kriging over blocks of
Lag/
40 m x 40 m is displayed as a perspective diagram
FIGUI 1. S a ofomtthe thickness
RE1. Sampleandmodelvariograms tv of cover
loamaitHoleFarm,Norfolk in Figure 3a, and Figure 3b is the corresponding
isarithmicmap. The regional pattern of the variation
is now much clearer.
TIle thickness of cover loam (depth to sand and In circumstances such as these, where a farmer is
graveel in cm) was measured by augering at the inter- concerned with managing areas of land in the best
secti(ons of a 20 m squaregrid over a field of approxi- way possible, block estimates of the property are
mately 18 ha to give some 450 observations. Vario- likely to be more useful as well as producing a
gramiswere computed for the four principaldirections smoother map or surface.Furthermorethe estimation
of the grid. Since these variograms differed little the variances for block estimates are considerably less, a
variation in the thickness of loam is regarded as tenth of those for points in this analysis. Correspond-
isotreopic and can be representedby a single isotropic ing to the grid of interpolated values there is a grid of
variogram for the field. There is no evidence of any estimation variancesand these can be displayed in the
local drift in the variation. A spherical model, equa- same way as the estimates. Figure 4a is a perspective
tion ((19) part I, provided the best fit to the estimated diagram showing the point estimation variances
semi--variances,as follows for Hole Farm,and Figure4b is the corresponding one
r ?) , , , . for block estimates.
y(h)= 187-0+603-8 3 _ -
[2 101-2 2 \101-2 Stonecontentof the topsoilat Plas Gogerddan
4g^~ ~ for 0< h 101-2m The topsoil at Plas Gogerddan, the Welsh Plant
y(h) = 187-0 + 603-8 for h > 101-2 m
Breeding Station, was surveyed in detail at the inter-
sections of a 15-2 m x 15-2 m square grid. Each field
y(O)=0. (I) on the Station was surveyed in the same way, but we
have analysed the data from the largest field only,
The experimental variogram and model are given Cae Ruel. Several properties including stone content
in Figure 1. The range of spatial dependence is were measured on bulked samples from ten cores of
101-2 m. The nugget variance is approximately a 2-5 cm diameter and 15 cm deep chosen randomly
quarterof the total, i.e. there is considerablevariation within 15-2 m squares around the grid node. This
over distances less than 20 m. Such variation within gave 440 observations of stone content over the
20 m could arise from irregularitiesin the underlying field.
272 MARGARET
OLIVER, andJOHNGERRARD
RICHARDWEBSTER
FIGURE2. Estimates of the thickness of cover loam at Hole Farmproduced by punctual kriging and shown in (a) as a perspective diagram,
and in (b) as an isarithmicmap with thicknesses in cm
The variogram for the four principal directions, and its envelope given by setting 0= 2-57 rad and
along the rows and columns and parallel to the 0=2 57+7r/2 rad is shown by the heavy lines in
diagonals, is shown in Figure 5. The slopes are very Figure 5.
differentin the differentdirections,and an anisotropic This model was then combined with the obser-
linearmodel, equation (27) part I with a = 1, provided vations to estimate the content of stones in the soil of
the best fit in the least squaressense. The model is Cae Ruel. The nearest 20 observations were used to
y(O,h)= 9-06 + {0'4722cos2( - 2-57) + estimate values of 30-5 m x 30-5 m blocks centred at
0-1582sin2(0- 25 7)} h, (2) 7-6 m intervals. Isarithmswere threaded through the
inphysical
Geostatistics applications
geography: 273
FIGURE 3. Estimates of the thickness of cover loam at Hole Farm produced by block kriging and shown in (a) as a perspective
diagram, and in (b) as an isarithmic map with thicknesses in cm
fine figure field of estimates to give a map of the mining was in meteorology (Gandin, 1965). It is an
variation in stone content over the field, Figure 6. It obvious application. It is general knowledge that
shows clearly that the variation is greatest in the places near to one another experience similar
direction NE to SW and least perpendicularto this. weather, while the weather at distant places is more
Figure 7 shows the estimation variances. or less different. We also know that there are short-
range and apparently random fluctuations. Weather
Rainfallin Englandand Wales recording stations are sparse,yet people would like to
As we mentioned in Part I (Oliver et al., 1989) one of be able to estimate the values of rainfall,temperature,
the earliest attempts to use geostatistics outside insolation, etc. at places between stations. For rainfall,
274 MARGARET OLIVER,RICHARD WEBSTERandJOHN GERRARD
FIGURE4. Estimation variances for kriging the thickness of cover loam at Hole Farmshown as perspective diagrams in (a) for point kriging,
and in (b) for block kriging
060- TT
40- TT,- r
^440
200 -- --
0 50 100 150
Lag/m
o 0o
* 22.5?
o 45?
67.5?
Y/mm2 A 90?
* 112.5?
v 135?
100 000 v 157.5?
- , _-B
[] , "o v!
?,
o v., A
40 000-
20 000 o i *v v
Value Standard
error
AkaikeInformation
Criterion 2667
Stage 1 Centre 22
Centre Centre 33
Centre interval Numberof
Sampling
I I Stage (metres) sampling
points
Stage 2
1 600 9
2 190 18
Stage 3
3 60 36
4 19 72
Stage 4 5 6 108
Stage 5
ISampling points
Source offreedom
Degrees Parameters
estimated
bymeansquares
Stage 1 fi a+m
a., m 2 - 1+ '... -+ u ,2a,2
2 + U102 +
a2 + u2 2
Stage2 f-f, m_ +.. + u23a32q+u22,22
the variogram, Figure 15. These graphs show very soil of the Wyre Forest occurs within 60 m, and this
clearly that over 80 per cent of the spatial variation provided the basis for designing a sampling plan to
occurs within 60 m. Stages 1 and 2 account for very estimate the variogram precisely. The nested survey
little of the variation,and many of the estimated com- provided no information about directional depen-
ponents of variancefor Stage 2 were negative. These dence in the variation, and at this stage we needed to
negative estimates suggest that the variation at be able to detect any anisotropy if it were present.
190 m is less than would be expected; the most likely Transects are efficient for estimating the variogram
explanation is that they indicate repetition in values over one order of magnitude of distance to just
of soil properties at that distance. Negative com- beyond the limit of spatial dependence. So we
ponents of variance have also been interpreted as sampled the soil along transects at 5 m intervals to
estimates of zero, i.e. making no contribution to the ensure that the residualvariancewould be no greater
variance at that particular stage. There can be no than before, and by extending them to 100 m we
certainty about the interpretationbecause the confi- ensured that we had sufficientpairs of comparisons at
dence limits are wide with so few degrees of freedom. 60 m to estimate the range and sill if present. We
The residual variance, which represents the unre- aligned the transectsin three directions at 60? to each
solved variance over distances less than 6 m and any other to detect and estimate any anisotropy. The total
measurement error, is considerable in view of the transect length in each direction was 500 m to give a
small sampling interval at Stage 5. minimum of 100 comparisons at the first lag. In two
To summarize the results for the large number of directions five 100 m transects,with stratifiedrandom
variateswe computed their principalcomponents and starting points, were sampled. This sampling plan
then the components of variancefor the firstfive. The ensured fairly even coverage. In the third direction a
variograms for these leading principal components, single 500 m long transect was sampled. The soil
which account for over 70 per cent of the variance properties were recorded as for the nested survey.
in the sample, are shown in Figure 15 with the com- Samplevariogramswere computed conventionally
ponents of variance represented as percentages. The for each property in each directionby equation (15), in
first three principalcomponents representthe general Part I. They were sufficiently similarfor the variation
spatialdependence of the individualvariates,whereas to be judged isotropic and to be pooled for the three
components 4 and 5 largely comprise the residual directions. Those for a representative selection of
short-range variation, i.e. within 6 m. Webster and properties are given in Figure 16. Models were fitted
Butler(1976) and Nortcliff (1978) also found that the to the semi-variances by weighted least squares
first two or three principal components represented approximation as described by McBratney and
the spatially dependent variation well with the short- Webster (1986) using Ross's (1987) Maximum Likeli-
range variation being concentrated in the higher hood Program (MLP).
order ones. The variograms of the morphological properties
The results of the nested survey and analysis increase to a sill, indicating that these properties are
show clearly that most of the spatial variation in the second-order stationary within the region. They also
280 MARGARET OLIVER,RICHARD WEBSTERandJOHN GERRARD
5- 600-
Stones ???r??..
Sand
r
4-
400-
a<
3- -
c(
L. / ?
ca
2
200-
,'
....
-/ /#
1 . ^
n 0 I I I I I
400 15
Clay ...... Principal components
300 -
10 1
a)
0 - . -. -
c .*
.c 200 : .,
>
I..?
/.*
./
5
100- /
2
/3
4
5
I I I I I I l I! I
0 6 19 60 190 600 0 6 19 60 190 600
Spacing/m
have similar ranges or effective ranges, which around 40 m. The firstprincipalcomponent accounts
suggests that these variables are controlled by for most of the spatially dependent variation, and
some common genesis or soil-forming process. The this is reflected by the slope and sill variance of its
morphological properties are considerably more variogram.An exponential model provided the best
variable in the subsoil than in the topsoil. The expo- fit to the first three principal components. The
nential model generally provided the best fit, and the average range of spatial dependence for the first five
average range of spatial dependence indicated by the principalcomponents, which account for 77 per cent
distance parameters of the model variograms is of the variation,is 41-6 m.
inphysical
Geostatistics applications
geography: 281
Y Y Sand
600
Y
0.05-
pH
0O;, o 0.o06'
pH
0.04- 0 0 0 o
/ A
0.03
A..-, ?;3
AA-
0.02
0.01O
0 0 10 30 50 70
Lag/m
FIGURE16. Sample and model variograms from the linear survey of the soil of the Wyre Forest for (a) stone content, (b) sand content, (c)
clay content, and (d) pH. The units of variance in (a), (b) and (c) are (percent)2
By using this two-stage approachwe were able to sampling scheme to determine the variograms preci-
identify the spatial scale of variation precisely for sely. With these known we could then sample at an
individual variables and from the leading principal appropriateintensity to map, either by spatial classifi-
components an average for the region with very little cation or by kriging, and Oliver and Webster (1987)
existing information.The results of the nested survey show the results of applying both procedures to a
provided an effective framework for designing the part of the Wyre Forest.
282 MARGARET
OLIVER, andJOHNGERRARD
RICHARDWEBSTER
INTERPRETING THE UNDERLYING continuous curves. Cox (1978) also pointed out that
FACTORS CONTROLLING SPATIAL such an analysis will produce boundaries on request,
VARIATION and these may have little geomorphological signifi-
Soil variation cance.
In the Wyre Forest the average range of spatial An impasse has clearlybeen reachedin the analysis
of hillslope form, and a differentapproachis required
dependence for the properties of the soil corresponds
that neither pre-judges the nature of slope form nor
reasonably with the average spacing between the
relies on arbitrary decisions. Regionalized variable
lithological units of about 50m, indicating the
control of parent materialon the pattern of variation theory seems likely to provide a solution since it can
of the morphological properties of the soil. The treat slope as a continuous variable.Cox (1983) used
variation of pH, however, is quite different: it is the autocorrelation coefficient, but this measure is
more variable in the topsoil than in the subsoil, and limited to second-order stationarity. The variogram
the variograms for depths 1 and 3 are unbounded, being less demanding in its assumptions provides
more scope, and so we used it for our analysis of
Figure 16. They are fitted best by an isotropic power
function of the form slopes. Our examples are from Dartmoor.
Dartmoor
y(h)= c, + wha (4)
Two directly opposing hillslope profiles were
where a the exponent, which must lie between 0 and measured in the north-south trending valley of
2, describes the shape of the variogram.The quantity the River Cowsic in Central Dartmoor where the
c0 is the nugget component. These differences underlying rock is well-jointed mega-crystic granite
suggest that the variation in pH is not controlled by with primary and secondary joint partings. It is
parent material. The more intense variation in the covered with a variable thickness of soliflucted
topsoil was interpreted,following the suggestions of material. One hillslope, SLW, faced West and the
Skeffington(1983), as a consequence of the way rain- other, SLE, East. Both slopes occur in the widest
fall reachesthe ground under trees. Much of the water portion of the valley. The West-facing slope is 801 m
reaches the soil as stemflow, and so leaching is con- long with a mean angle of 5-9 degrees and the East-
centrated next to the trunks of trees. This explains facing one is 492 m long with a mean angle of 7-0
why the pH in the topsoil varies more than that in the degrees.
subsoil. However, it does not explain why the vari- Slope angles were measured with an inclinometer
ation is unbounded:this remainsunanswered. to the nearest degree over consecutive 3 m lengths
orthogonal to the contours. It is already well-estab-
Valleysideslopeprofiles lished that the distance over which slope angle is
Geomorphologists have been trying to describe measured affects the analysis of slope form (Gerrard
hillslope profiles accuratelyfor many years, but have 1978, Gerrardand Robinson 1971). Furthermore,in
encountered many difficulties. Traditional methods Part I we stressed that the variogram is scale-
have attempted either to subdivide profiles into dis-
crete components, or to model the slope by some dependent. Thus in addition the E-facing slope was
also measured over 1-5 m units to examine the effect
form of polynomial function. Methods of profile sub- of changing the support of the sample on the
division involve partitioning the slope into segments
interpretationof slope form.
or elements. A slope segment is a rectilinearportion
Variogramswere computed from the slope data for
of a hillslope profile that differs distinctly in mean both hillslopes and models were fitted using MLP
inclination from segments above or below it. More
(Ross, 1987), Figure 17a, b, and c. These indicate
generally a slope element may be defined as an several important spatial characteristics of these
approximatelysmoothly curvedportion of a hillslope, slopes. A periodic model without damping provided
either convex or concave. Two computer-based the best fit in the least squares sense to all the
methods exist for delimiting such units; the 'best
variograms.The variogramsfor the long transectson
units' analysis of Young (1971) and the linear both slopes appear to increase without bounds, and
regression method of Ongley (1970). Both methods, so in addition to the periodic function we have added
however, involve arbitrarydecisions that limit their a power term to describe them. The full model is
general applicability.An additional criticism of such
methods is that they assume hillslopes are composed /27rh)
=s27snh\
y(h)=c + cl cos- +C2 sin
- + c3h, (5)
of distinct components ratherthan regarding them as
inphysical
Geostatistics geography:
applications 283
3.5- forthe phaseangleand
*.
CM 3.0 A = cl/cosp
a)
a0
2.5
0)
-0
fortheamplitude.Theparameters of thesemodelsare
2.0 "-
given in TableIV.
?^
(a)
i I I
1
Note thatthismodelis authorizedfor one dimen-
0 20 40 60 80 100 120 Lag/m
sion only. It describesa patternof variationthat
8
repeatsitself regularly.We interpretedthe periodic
7 modelasindicatingthepresenceof terracesseparated
6
* X
by a moresteeplyslopingsectionwhicharerepeated
a)
0)
fairlyregularlyalong the slope transects.Figure17a
-
5 gives the variogramfor the West-facingslope,SLW.
4 (b) The wavelengthof the periodicfunctionis 25-2m
whichrepresentsthe averagerepeatdistancein the
0 20 40 60 80 100 120 Lag/m gradient.Thiswas not evidentfromGerrard's (1982)
10- earlieranalyses,Figure18a.The fairlysmoothsmall
9 terracesseparatedby roughermore steeply sloping
N
8
sectionsarediscernibleon the graphsof slope angle
00
(1)
cn 7
againstdistance(Figs.18a,b andc),but they aretoo
- a
V1
erraticand ill-definedto estimatethe periodicityby
C
eye. The amplitudeof the model is modest for this
5 (C) slope becausethe contrastsin slope are fairlysmall.
The AkaikeInformationCriterionfor the periodic
0 20 40 60 Lag/m
model is largerthan that for the linearmodel (the
FIGURE17. Experimentaland model variogramsfor hillslope
modelthatfittedthenextbestin termsof theresidual
measurements in the CowsicValley,Dartmoor(a)SLWmeasured sumof squares).This demonstratesthe effectof the
over 3 m intervals,(b)SLEImeasuredover 3 m intervals,and (c) addedcomplexityof thismodel.Itwasworthwhilein
SLE2measuredover 1-5m intervals thisinstance,however,sincethemodelconfirmedour
fieldobservationsandimprovedinterpretation.
Figure 17b gives the variogramand model for
wherecois the nugget variance,r is the wavelength, slope angleof the East-facingslope,SLEI,measured
over a length of 3 m intervals.The AkaikeInfor-
and c3 is the coefficient of the power term. The mationCriterionfor the periodicmodelis muchless
coefficientsof the cosineandthe sineterms,cl andc2, thanforthe linearmodel(thenext best fittingmodel)
definethe amplitudeandthe phaseangleof the sinu-
soidalcomponentof the variogram.Thiscomponent justifyingits greatercomplexity.This slope is also
canbe expressedin the form
characterized by a regularrepetitionof features,but
at 58 m whichis muchlargerthanthatfor the West-
Y=A cos (-
facingslope. The terracesare fairlysmoothwith an
p), (6) averageslope of 8? to 9? and they are separatedby
more irregularand steeply sloping sections. The
where A is the amplitude and (p is the phase angle.
amplitudeof the modelis also muchlargerthanthat
From trigonometry this can be rewritten as for the West-facingslope:the contrastsare greater
andtheterracesaremoreclearlydefinedon thisslope.
Y=A coso cos0+ A sin(psinO. (7) Thisexpressesquantitativelythe impressionsgained
by comparingthe graphsof slopeprofile.
This has the same form as equation (5) in which The model variogramsfor both slopes show a
A cos(pequals cl and A sin( equals c2.Since we know nuggeteffect,whichwe interpreted in PartI as arising
cl and c2from fitting the model we can solve the two from measurementerror,spatiallydependentvari-
equations in the two unknowns, giving: ation over distancesless thanthe samplinginterval
andpurelyrandomvariation.In theseexamplesonly
(p= arctan (c2/c1) variation within the support is significant.This
reuait ftesoewihntespot fwt h au
10 20 30 40 50 60 270 80 90 10
20- S
16 -
12-
* ? ? ?- *
8 . . ***.*...... * *. e . .. .. . *. * *.*
20-
4- ?
16-
12-
1 I l *
e ?e
? .. ?. ?
LO 4-
4
12-
FIG 8. Spea . le a? p i? o e i h V ey
10 20 30 40 50 60 70 80 90 100
20- --
20
20 con t.
0 16-
16-
16-
t 12-.
el . ? ?
FIGURe Selangl i
SL E2
4-
110 120 130 140 150 160 1;0 180 190 200
slope measurement. This is particularly useful since We also measured slope angle over 15 m intervals
Successive
pointstrorn thebase of the s[opet
sugetog-thtteng
vainc t fo aSLs23wihissalcmae
hes fetqatfe h
neso o f 5 .4frSE.
slop mesreet
Thi is pat u cs s fuualic We als mesue sl o p a ovr n l mnevl
gemrhooit hav fontdfiutt
uniyo
hat-aigsoe
haiga
ntoe
198.I
roughness~? (Pros hsar
prvoul ie nFgreIc eidi oe ihu
example t h tw slp h qut
aeifrnugtdmigaanprvdstebsi.Tewvlnt
v
geomorphologists have found it difficult to quantify on the East-facing slope. The variogram and its model
roughness previously (Parsons, 1978). In this are given in Figure 17c. A periodic model without
example the two slopes have quite different nugget damping again provides the best fit. The wavelength
Geostatisticsin physicalgeography:applications 285
TABLEIV.Parameter
estimatesof theperiodic
modelsfittedto variograms
of theslopes
Variances
SLW 2-3042 -0-1013 0-1498 25-23 0-00318 1-205 0-1013 0-4484 44-45 (38-42)*
SLE1 5-4607 -0-5889 -1-0440 54-32 0-0137 1 0-7913 0-6589 59-11 (72-01)*
SLE2 6-8824 -0-7084 -0-7084 65-18 - 0-7084 0
Note: *The value in brackets is the Akaike Information Criterion for the linear model
suggests repetition of the features on average every and as Oliver and Webster (1986b) showed, modell-
62 m, which compares closely with the results for the ing the variogram is quite as effective in describing
3 m spacing. The nugget variance of 6-9 is slightly the variation quantitatively.
larger than that for the same slope measured over a
length of 3 m. This indicates the effect of the size of CONCLUSIONS
the support on the measurement. The shorter the
distance over which the slope is measured, i.e. the The examples described above illustrateseveral ways
smaller the support, the greater the effect of inter- in which regionalized variable theory can be used
vening irregularities resulting in a larger nugget profitablyin physical geography. The principalaim in
variance. In other words the larger the support the mining was the estimation of concentrations of metal
greater the smoothing effect. in ores and of mineralreserves. It is equally applicable
We think that the periodic step-like features for estimating the thickness of surficial deposits,
result from periglacialsolifluction,and that the major soil properties, and meteorological phenomena, for
joints in the granite have a strong control on their example. Furthermore, the theory can be used to
spacing. The difference between opposing valley plan rational sampling schemes, and a geostatistical
sides possibly arose because of differences in the analysis, by describing quantitatively the underlying
micro-climate. The East-facing slope is a leeward structureof variation,may help us to understandthat
slope. It would have had more snow on it resulting structure.
in more meltwater to penetratejoints thereby increas-
ing hydraulic pressures in the joint system and ACKNOWLEDGEMENTS
weathering, and enhancing solifluction during the We thank Mrs
late Pleistocene. This could explain the larger, better Joyce Munden and Mrs JeanDowling
for preparingthe figures, and Mrs Joyce Munden for
developed features on this slope.
help with the computing.
Spatial analysis of this kind can characterize the
forms of slopes and provide quantities by which
they and related attributes of the landscape can be REFERENCES
compared,and physical explanation can be sought for BURGESS,T. M. and WEBSTER,R. (1980). 'Optimal inter-
them. In the Dartmoor example informationfrom the polation and isarithmic mapping of soil properties. I.
variogramshas led to a reassessment of the processes Thesemi-variogram andpunctualkriging',J.SoilSci.31:
responsible for shaping the slopes. 315-31
In situations like this where periodicity is sus- CORBETT,W. M. and TATLER,W. (1974) Soils in Norfolk.
II.Sheets-G 13/14 (Barningham/Sheringham) (Soil Survey
pected the analysis could be taken a stage furtherby Record No. 21, Harpenden)
transformingthe variograminto the power spectrum. COX, N. J. (1978)
'Hillslope profile analysis', Area 10: 131-
Regularrepetition would then show as distinct peaks 33
in the spectrum, and the frequencies at which they COX, N. J.
(1983) 'On the estimationof spatialauto-
occur correspond to the wavelengths in the original correlation in geomorphology', Earth Surface Processes
data. The spectrum contains no more information, and Landforms8: 89-93
286 MARGARET OLIVER,RICHARD WEBSTERandJOHN GERRARD
GANDIN, L. S. (1965) Objectiveanalysis of meteorological OLIVER,M. A. and WEBSTER,R. (1987) 'The elucidation
fields(IsraelProgramfor ScientificTranslations,Jerusalem) of soil pattern in the Wyre Forest in the West Midlands,
GERRARD,A. J.W. (1978) 'Hillslope profile analysis',Area England. II.Spatial distribution',J. Soil Sci. 38: 293-307
10:129-30 OLIVER, M., WEBSTER, R. and GERRARD, J. (1989)
GERRARD, A. J. W. (1982) Slope form, soil and regolith 'Geostatistics in physical geography. Part I: theory',
characteristicsin the basin of the River Cowsic, Central Trans.Inst. Br. Geogr.N.S. 14(3): 259-69
Dartmoor,Devon (Unpubl. PhD Thesis, Univ. of London) ONGLEY, E. D. (1970) 'Determination of rectilinearprofile
GERRARD, A. J. W. and ROBINSON, D. A. (1971) segments by automatic data processing', Zeitschriftfir
'Variabilityin slope measurements',Trans.Inst.Br.Geogr. Geomorphologie 14: 383-91
54: 45-54 PARSONS, A. J.(1978) 'A technique for the classificationof
McBRATNEY, A. B. and WEBSTER,R. (1986) 'Choosing hill-slope forms', Trans.Inst. Br. Geogr.N.S. 3: 432-43
functions for semi-variograms of soil properties and ROSS, G. J. S. (1987) MLP User manual (Numerical
fitting them to sampling estimates',J. Soil Sci. 37: 617-39 Algorithms Group, Oxford)
McBRATNEY, A. B., WEBSTER,R. and BURGESS,T. M. SAMPSON, R. J. (1978) The Surface II graphics system
(1981) 'The design of optimal sampling schemes for local (Revisionone),(Series on Spatial Analysis, No. 1., Kansas
estimation and mapping of regionalized variables. I. Geological Survey, Lawrence)
Theory and Method', Computers and Geosciences7: SKEFFINGTON. R. A. (1983) 'Soil properties under
331-34 three species of tree in southern England in relation to
McCULLAGH, M. J. (1975) 'Estimating by kriging the acid deposition in throughfall', in ULRICH, B. and
reliability of the proposed Trent telemetry network', PANKRATH, J. (eds) Effects of accumulation of air
Comp.Appl. 2: 35 7-74 pollutants in forest ecosystems (Reidel, Dordrecht)
NORTCLIFF,S. (1978) 'Soil variability and reconnaissance pp. 219-31
soil mapping: a statistical study in Norfolk', J. Soil Sci.29: WEBSTER,R. (1977) Quantitativeand numericalmethodsin
403-18 soil classificationand survey(ClarendonPress, Oxford)
OLIVER,M. A. (1984) Soil variationin the Wyre Forest:its WEBSTER,R. and BUTLER,B. E. (1976) 'Soil survey and
elucidationand measurement(Unpubl. PhD Thesis, Univ. classificationstudies at Ginninderra',Aust. J. Soil Res. 14:
of Birmingham) 1-24
OLIVER, M. A. and WEBSTER,R. (1986a) 'Combining YOUDEN, W. J. and MEHLICH,A. (1937) 'Selection of
nested and linear sampling for determining the scale and efficient methods for soil sampling', Contr. Boyce
form of spatial variation of regionalized variables',Geogr. ThompsonInst.for PlantRes. 9: 59-70
Anal. 18: 227-42 YOUNG, A. (1971) 'Slope profile analysis: the system of
OLIVER, M. A. and WEBSTER,R. (1986b) 'Semi-vario- best units' Inst.Br. Geogr.Spec.Publ.3: 1-13
grams for modelling the spatial pattern of land form and
soil properties', EarthSurfaceProcessesand LandformsII:
491-504