Professional Documents
Culture Documents
00/0
Printed in Great Britain. © 1981 Pergamon Press Ltd.
A. B. MCBRATNEY
Department of Soil Science, University of Aberdeen, Scotland
and
R. WEBSTER
Rothamsted Experimental Station, Harpenden, Hertfordshire, England
Abstract--A FORTRAN IV program, OSSFIM, is presented for calculating estimation variances when interpolat-
ing by kriging from regular rectangular and triangular grids of data and previously-determined semi-variogram. The
variances are computed for a range of grid spacings and block sizes, and the results graphed. The user chooses a
block size, and can read from the appropriate graph the sample spacing corresponding to any prescribed maximum
tolerable error. This is the optimal sampling scheme. Use of the program is illustrated with two examples showing
different types of variation in soil. In one, the pH of topsoil is isotropic with a spherical semi-variogram and
negligible nugget variance. An equilateral triangular grid is the best sampling scheme: it is approximately 10 per
cent more efficient than a square grid. In the other example, variation is linear but anisotropic with a large nugget
variance. In these circumstances, a triangular grid has no advantage over a rectangular one, which should be
elongated in the ratio 1.88 to 1 in the direction of minimum variation.
Key Words: Geostatistics, Kriging, Regionalized variables, Sampling, Soil Survey, Soil Science.
can be diminished by specifying a large search radius or that for any one direction, xi and h are scalar quantities,
large numbers of points. In general, 40 points is sufficient and we have changed the notation accordingly. Models
for comparing different configurations. Coordinates of were fitted then to the experimental semi-variogram
the n points on the neighbourhood grid are transformed using weighted, non-linear least-squares. The weights
to correspond to real grid spacings. were equal to n(h), thereby obtaining the best fit near
SUBROUTINE KRGERR solves the kriging equation the origin (Journel and Huijbregts, 1978). Further
A-~ b = c , where A is the symmetric matrix of semi- examples and discussion are given by Burgess, Webster
variances between grid points calculated using FUNC- and McBratney (in press).
TION GAMMAH, and b contains the average semi-
variance y(x, 1,0 calculated by FUNCTION SEMAV Example I: pH at Broom's Barn
using an equally-weighted discrete approximation. The first example is for the pH of the top 15 cm of soil
Matrix A is inverted in SUBROUTINE INVERS and at Broom's Barn Experimental Station, Suffolk, England.
multiplied by b to give the vector c of weights a, i= Figure 1 shows the estimated semi-variances for 8 direc-
1, 2 . . . n, for each of the n points. tions up to a lag of 400 m. The solid line is the model of
A Lagrange variable is addded to the set of simul- best fit over the first 8 lag intervals, 320 m. It is isotropic
taneous linear equations to minimize the kriging variance and spherical with the following equation:
with respect to each X, subject to the constraint that
Iy(h)=0.021+0.330{~ h 2 ~ for h ~<196.4
196.4
iz: [3'(h) = 0.021 + 0.330 for h > 196.4,
(2)
and a Lagrange parameter ~, is contained also in vector c
(Journel and Huijbregts, 1978). The transpose of vector b where h is in meters. This type of semi-variogram is rare
is multiplied by vector e to give for soil properties in that it has a small nugget variance.
The pH at Broom's Barn seems to be continuous with
~, ,~,~,(x,, E) + ¢.
i~l
little or no short-range variation. The reasons for this are
twofold. Firstly, the measurements were obtained from
25 bulked cores 15cm deep taken over a 16xl6m
The within-block variance 3'(V, V) is subtracted and o'r ~ square. Secondly, a property such as pH can be smoothed
calculated. The kriging variance is obtained for several through chemical equilibration, and this process may
grid spacings and block sizes for either square or trian- even explain the shape of the semi-variogram.
gular grids or both in one run of the program. Figure 2 shows the maximum kriging variance for
The input to the program is described on lines I 11-237 point (16 m x 16 m x 0.15 m cuboid) estimates as a function
of the listing (Appendix I). The output consists of tables of grid spacing. The error increases with an approximately
of tyK2 and o-g for specified spacings and block sizes for sigmoid shape with increased grid spacing up to about
each type of grid. There is an option for graphic output 280 m, at which distance the sample points involved in
using the GHOST system (Calderbank and Prior, 1977), the interpolation become effectively independent; that is
and the graphs shown in Figures 3-8 were produced in are outside the range of the semi-variogram. The value of
this manner. All calls to the graphics system are in 0.360 for trr 2 at grid spacings greater than 280 m is equal
SUBROUTINE OSSGRF; users who do not have access to the sill variance plus the sill variance/40 which is
to GHOST therefore need change only this subroutine. equivalent to the estimation variance of classical statis-
Sample input and output files are listed in Appendix 2. tics. The triangular grid is always better than the square
grid up to this distance. The differences between the two
EXAMPLES compare closely with those found by Matrrn (1960) for
The two examples that follow derive from surveys of global estimates from square and triangular grids.
the soil on experiment stations, one devoted to sugar The comparison in this and subsequent figures are for
beet production in lowland agriculture, the other con- square and triangular grids with the same sampling in-
cerned with hill grazing. In both surveys, measurements tensity and the actual distances between points on the
previously had been made on square grids. To conform triangular grid are 1.0746 times greater. The grid spacings
to the reconnaissance procedure that is recommended in for the triangular grids on the graphs are what we shall
Part I, the semi-variograms were computed for the two term the equivalent square grid spacings. In this situa-
stations by treating the grids as a series of parallel tion, reasonable point estimates of pH could be made,
transects in several directions. For each direction, there- since the 'points' are 16mx 16m squares. However,
fore, the semi-variance is plots in field experiments are often larger--0.04 ha--and
we have chosen to design a scheme that will estimate
i n(h) means of blocks 20m x20m with a maximum error,
~(h) = ~ ~ ~z(x,)- z(x, + h)) 2, (1) trKmax, of 0.2 pH unit. Figure 3 shows a graph of trr 2
against grid spacing for blocks of 20m x 20m x0.15 m,
where z(x~) are values at points x, such that data are and for cell-centered blocks is similar in appearance to
available both at x~ and x~ + h, and n(h) is the number of Figure 2. The kriging variance reaches a plateau at
pairs of data points separated by a gap equal to h. Note approximately 280m with a value of 0.313, which is
Design of optimal samplingschemes for local estimation--II 337
0.5
I o •
o,, ~ • ~o ~ I~g
o ,
~
:t ~ ' o"f ~
°//o°
6
A A "116 & v ~ ADO eOA
[] ~ v'~
o
0.3 17 ~ v
o
y(h)
o" oO
e÷ o
0.2
I I I I I
0 2 4 6 8 10
h (distance / 40m)
Figure 1. Semi-variogramof pH of top soil over Broom's Barn ExperimentalStation with estimates for 8 directions
and fitted model shown by solid line.
equal to the sill variance plus the sill variance/40 minus similar to that near the ordinate in Figure 3 caused by
the within-blocks variance for 0.04 ha blocks. The grid- using only the nearest 40 points to the block center for
centered blocks have smaller estimation errors, but the interpolation. Therefore, with this size of block, not all
errors become larger at small grid spacings. There is an grid points within the block are employed and the esti-
apparent increase in the kriging error for very small grid mation variance increases accordingly.
spacings, because the few points, from which estimates
are made, are concentrated in the centers of the blocks. Example 2: thickness of peat at Lephinmore
This practice is not prudent. We require a grid spacing The example derives from a soil survey of Lephinmore
that corresponds to a maximum kriging error of 0.2 pH farm, Argyll, Scotland. Blanket peat is extensive and its
units, or ~g2 of 0.04. This is 40.8 m for the square grid thickness (in cm) has been recorded at points 100 m apart
and 42.8 m equivalent square grid spacing for the trian- on a square grid. Figure 5 illustrates the estimated semi-
~42.8"[2 variances for 8 directions up to a lag of 1000 m. Spatial
gular grid. This represents a saving of 1.40.8J or 10.0%
dependence is linear and geometrically anisotropic. The
for the triangular over the square grid, which seems following model was fitted to the sample semi-variances:
valuable. Hence, a practical sampling recommendation is
an equilateral grid with sample supports 45 m apart
which has an equivalent square grid spacing of 41.8 m. 7(h, 0) = 504 + {2202 cos2(0 - I. 194) +
1172 sin2(0 - 1.194)}'12h (3)
The same information might be required for blocks of a
different size, and, if so, can be derived from Figure 4 in
which the kriging variance is plotted against block size where h is in kilometers, 0 is in radians and 0 - - 0 is
for the 45 m triangular grid, and the 41.8 m square grid west-east. The two solid lines in Figure 5 define the
for the sake of comparison. Thus, estimates for l ha envelope within which the fitted model lies. The semi-
might be desired for advisory purposes. For these, the variogram has a large nugget variance, and is typical of
error is 0.08 pH units. In this survey, because the 'point' soil variables that are measured on small (true point)
estimates cover an area of 16 m x 16 m, their estimation supports and show rapid fluctuations over small dis-
error is only 0.28 pH units. In Figure 4, the square grid is tances. In this situation, the variation in peat thickness is
always worse in terms of O'gm,,
2 up to 90 m, including the largely a function of the microtopography of the land
block sizes between 50m and 90m where the grid- surface before the peat began to form. The anisotropy is
centered blocks have higher estimation variances. Beyond caused by different variations in thickness across and up
90 m, there seems to be a negligible difference between the slopes.
the grids at this sampling intensity. The apparent in- For the sake of this analysis, we require the semi-
crease in cry:2 above 260 m is a computational artefact variogram model only for the direction of maximum
338 A.B. McB~TNEY and R. WEBs'r~
~OINTS
/I /
//
0.3
I III/
II
I II
0.2
2
OK
/1/!
// I/
0.1
o.o L I J A I I I l
0 .5O tO0 150 20O 25O 3OO 350 400
CELA.-CENTBED
. . . . . . . . . . . . CEL~--C~I'BED
Figure 2. Topsoil pH at Broom's Barn. Graph of maximum kriging variance of points as function of grid spacing for
square and equilateral triangular grids, in this and subsequent figures, shape of function is interpolated from 20
estimates of kriging variance.
Design of optimal sampling schemes for local estimation--lI 339
2o°oo m
0°3
0.2 /
2
OK
0°I
20°00 m
0.0
/ I i I i I I I
0 ,~0 "100 "150 2OO 25O 3OO 350 /.00
GRID SPACING ( m)
Figure 3. Topsoil at Broom's Barn. Graph of kriging variance of 20 m x 20 m blocks centered at middle of grid
cells and at grid points as function of grid spacing for "square and equilateral triangular grids.
340 A.B. McBP.ATNEYand R. WEBSTER
0.06
0o0~
2
OK
0.(~
(-.lo88 m
0 . 0 0 t.. I ~ t I t --
0 50 "100 150 200 250 30O
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F..EL~-CEN'r~IED GRIO--CENT~
Figure 4, Topsoil pH at Broom's Barn. Graph of kriging variance of square blocks centered at middle of grid cells
and at grid points on 45 m equilateral triangular grid and 41.8 m square grid as function of length of block size.
Design of optimal sampling schemes for local estimation--II 341
1000
800 A
AA A
L o
Y(h) V
400
200
• N-S A SSE-NNW
o W-E + ENE-WSW
n NW-SE • WNW-ESE
v NE-SW x NNE-SSW
I I I 1 |
0 2 4 6 8 10
h (distance/lOOm)
Figure 5. Semi-variogram of thickness of peat at Lephinmore with estimates for 8 directions. Two solid lines show
envelope of fitted model.
342 A.B. MCBRATNEYand R. WEBSTER
800
J
/
50O
400
2
OK
3OO
200
100
0 t I I I I
0°0 0oi 0°2 0.3 0,4 0.5
............ CE]-~-CE-N~
Figure 6. Depth of peat (cm) at Lephinmore in direction of maximum variation. Graph of maximum kriging
variance of points as function of grid spacing for square and equilateral triangular grids.
Design of optimal sampling schemes for local estimation--II 343
12,0
100
/." ...4/"
..'" /7
80 //
2
60
0K
40
20
0 I I I I I
0.0 0.I 0.2 0°3 0°¢ 0.5
GRID SPACING ( k l n )
Figure 7. Depth of peat (cm) at Lephinmore in direction of maximum variation. Graph of kriging variance of I ha
(100 m × 100 m) blocks centered at middle of grid cells and at grid points as function of grid spacing in direction of
maximum variation for rectangular and isosceles triangular grids.
344 A. R. MCBRATNEY and R. WEBSTER
30
20
2
OK
0o06
10
L , I |
0,0 0 1 0.2 0=3 0.4
S ] ZE OF BLOCK S I DE ~kin)
CEU.,-CENTC~r_.D . . . . . . . . . . . . O:~ID- ~
Figure 8. Depth of peat (cm) at Lephinmore in direction of maximum variation. Graph of kriging variance of square
blocks centered at middle of grid cells and at grid points for 60 m x I IO m rectangular grid as function of length of
block side.
Design of optimal sampling schemes for local estimation--II 345
variation and the anisotropy ratio. The model is block width for the 6 0 m x l l0m grid. The estimation
errors are small, and do not decrease rapidly with in-
y(h)= 504 + 220 h (4) creasing positions of the block relative to the grid. It is
interesting to note that although the kriging error for a
and the anisotropy r is 220/117 = 1.88. I ha block on this grid is approximately 5 cm the error of
Figure 6 shows the effect of the large nugget variance a point estimate is 23 cm?
in that for point estimates there is only a small increase
in error as the grid spacing is increased, and this error is Acknowledgments--Wethank Mr. J. S. Bibby and Mr. G. Hud-
very large. There is no detectable difference between the son of the Soil Survey of Scotland for access to the Lephinmore
data and Mrs. J. W. Munden for help in developing the program,
square and equilateral triangular grids because of the A. B. McBratney was supported by a studentship from the
large nugget variance. The errors for grid spacings less Department of Agriculture and Fisheries for Scotland.
than 0.1 km are probably over-estimated because the
extrapolation of the semi-variogram to the ordinate takes
no account of the (unknown) semi-variances in these
REFERENCES
small intervals.
Burgess, T. M., Webster, R., and McBratney, A. B., in press,
For such a variable property, point estimates are of Optimal interpolation and isarithmic mapping of soil proper-
little consequence, and for management purposes, inter- ties. IV. Sampling strategy: Jour. Soil Science, v. 32.
est is likely to be in I ha (0.1 km x 0.1 km) blocks. A reason- Calderbank, V. J., and Prior, W. A. J., 1977, The GHOST
able maximum error would be 5 cm corresponding to a graphical output system. User Manual: UKAEA Culham
Laboratory, Abingdon, 125 p.
kriging variance of 25 cm 2. Figure 7 shows the kriging Clark, l., 1976, Some practical computational aspects of mine
variance for 100 m x 100 m blocks plotted against sample planning: In advanced geostatistics in the mining industry.
spacing a in the direction of maximum variation. Guarascio M., David, M., and Huijbregts, C., eds.,: Reidel,
However, in order to take account of the geometric Dordrecht, 461 p.
anisotropy present, an optimal sampling scheme will Journel, A. G., and Huijbregts, C. J., 1978, Mining geostatistics:
Academic Press, London, 600 p.
have a sample spacing b equal to ra in the direction of McBratney, A. B., Webster, R., and Burgess, T. M., 1981, The
minimum variation. The variances shown in Figure 7 are design of optimal sampling schemes for local estimation of
much smaller than those for point estimates, because the regionalised variables. I. Theory and Method: Computers &
nugget variance is so large. We also see that to achieve a Geosciences, v. 7, no. 4, p. 331-334.
kriging variance of approximately 25 cm: there is no Mat~rn, B., 1960, Spatial variation: Meddelanden frAn Statens
Skogiorskningsinstitut, v. 49, no. 5, 144 p.
perceptible merit in the triangular grid, for the same Matheron, G., 1965, Les variables r~gionalis~eset leur estima-
reason. Therefore, we should recommend survey on a tion: Masson, Paris, 305 p.
rectangular grid, with an internode spacing of 63 m in the Matheron, G., 1965, The theory of regionalised variables and its
direction of maximum variation, or 60 m if a round figure applications: Les Cahiers du Centre de Morphologie Math~m-
atique, no. 5, Centre de Gc~ostatistique,Fontainebleau, 211 p.
is desired, and an internode spacing of 60 r or l l0m in Webster, g., and Burgess, T. M., 1980,Optimal interpolation and
the perpendicular direction. isarithmic mapping of soil properties. Ill. Changing drift and
Figure 8 shows the kriging variance as a function of universal kriging: Jour. Soil Science, v. 31, no. 3, p. 505-524.
APPENDIX 1
0001C O S S F I M
0002C
0003C Optimal Sampling Scheme For Isarithmic Mapping
0004C (of r e g i o n a l i s e d variables)
0005C
0006C
0007C Program to find the standard error of a kriged estimate
0008C for points or square blocks of a given size from a square
0009C or equilateral triangular grid of a given spacing from
0010C knowledge of the s e m i - v a r i o g r a m of the property. The
O011C m a x i m u m kriging standard error is a measure of the
0012C g o o d n e s s of the sampling grid. For point kriging the
0013C m a x i m u m standard error of e s t i m a t i o n occurs at the
0014C m i d d l e of a grid cell. For block kriging the m a x i m u m
0015C standard error of estimation occurs either at a d a t u m
0016C point or at the middle of a grid cell. Hence a
0017C sampling scheme can be chosen after weighing the
0018C a c c e p t a b l e error against the effort or cost.
0019C
0020C
0021C
0022 REAL XSQGRD(100),YSQGRD(100),XTRGRD(100),YTRGRD(100)
346 A. B. M c B ~ E Y and R. WEBSTER
0099 6 ,5.5836,5.5836,5.5836,5.5836,5.5836,5.5836,5.5836
0100 & ,5.5836,5.5836,5.5836
0101 7 ,6.5142,6.5142,6.5142,6.5142,6.5142,6.5142,6.5142
0102 & ,6.5142,6.5142
0103 8 ,7.4448,7.4448,7.4448,7.4448,7.4448,7.4448,7.4448
0104 & ,7.4448
0105 9 ,8.3754,8.3754,8.3754,8.3754,8.3754,8.3754.,8.3754
0106 T ,9.3060,9.3060,9.3060,9.3060,9.3060,9.3060
0107 E ,1.8612,1.8612,10.2366/
0108C
0109C ..........................................
0110C
0111C INPUT
0112C
0113C
0114C Records 1-2 FORMAT(16A4)
0115C
0116C ITITLE two lines of title each of
0117C 64 c h a r a c t e r s
0118C
0119C
0120C Record 3 FORMAT(80Y) free format
0121C
0122C TYPKRG 0 for point kriging
0123C 1 for block kriging
0124C UNITKM a value which relates the
0125C distance units in the semi-
0126C v a r i o g r a m model to the
0127C kilometre, e.g. if model is in
0128C units of 50m then UNITKM=0.05.
0129C NOTE. The program switches from
0130C metres to kilometres if
0131C UNITKM.gt.0.099.
0132C NSPACE number of grid spacings required
0133C in one run of program (max 20)
0134C NBLOCK number of block sizes required
0135C in one run of program (max 20)
0136C
0137C
0138C Record 4 FORMAT(80Y) free format
0139C
0140C TYPGRD 3 for an equilateral triangular
0141C grid
0142C 4 for a square grid with the same
0143C sampling density as the
0144C triangular
0145C 5 for both
0146C PNTFND 0 to find points by specifying
0147C radius
0148C i to find points by specifying
0149C number of points to be found
0150C RADIUS the search radius for points in
0151C units of square grid intervals.
0152C 2.0< RADIUS >3.5 p r o b a b l y
0153C (ignored if PNTFND.eg.I)
0154C NREQRD number of points required for
0155C i n t e r p o l a t i o n (max set to 50).
0156C (ignored if PNTFND.eg.0)
0157C
0158C
0159C Record 5 FORMAT(80Y) specifies s e m i - v a r i o g r a m model in
0160C direction of m a x i m u m variation
0161C
0162C MODTYP I linear model (no sill)
0163C 2 linear with sill model
0164C 3 spherical model
0165C 4 exponential model
0166C 5 g a u s s i a n model
0167C CO value for nugget variance
0168C GRAD gradient of linear part of linear
0169C models. For other models
0170C GRADs0.0
0171C CSLMCO value for sill v a r i a n c e - n u g g e t
0172C variance. Input 0.0 for
0173C linear models.
0174C ARANGE a value for the range of the
0175C model. This is a distance
0701 DINCX=R2NEST*DX
0702 DINCYsR2NEST*DY
0703 WBLVAR=0.0
0704 DO I0 I=I,NEST
0705 DO lO J=I,NEST
0706 XBLCOs(FLOAT(2*I)-I.O)*DINCX
0707 YBLCO=(FLOAT(2*J)-I.O)*DINCY
0708 DO i0 K~I,NEST
0709 DO i0 L'I,NEST
0710 XXBLCOz(FLOAT(2*K)-I.0)*DINCX
0711 YYBLCO~(FLOAT(2*L)-I.0)*DINCY
0712 WBLVAR=WBLVAR+GAMMAH(XBLCO,YBLCO,XXBLCO,YYBLCO)
0713 i0 CONTINUE
0714 WBLVARsWBLVAR*RNSTSQ*RNSTSQ
0715 RETURN
0716 END
0774C
0775 20 IF(H.LT.ARANGE) GOTO 25
0776 GAMMAH-SILL
0777 RETURN
0778 25 GAMMAH-CO+GRAD*H
0779 RETURN
0780C
0781C isotropic spherical model
0782C
0783 30 IF(H.LT.ARA~GE) GOTO 35
0784 GAMMAB - SILL
0785 RETURN
0786 35 HOVERA-H/ARANGE
0787 GAMMAH-CO+CSLMCO*(I.5*HOVERA-0.5*HOVERA*HOVERA*HOVERA)
0788 RETURN
0789C
0790C isotropic exponential model
0791C
0792 40 HOVERA--H/ARANGE
0793 GAMMAH - CO+CSLMCO*(I.0-EXP(HOVERA))
0794 RETURN
0795C
0796C isotropic gsussian model
0797C
~798 50 HOVERAsH/ARANGE
0799 HOVERA--HOVERA*HOVERA
0800 GAMMAH s CO+CSLMCO*(I.0-EXP(HOVERA))
0801 RETURN
0802 END
0803 SUBROUTINE INVERS(SYMMAT,ISDIM,ISIZE,IFAULT)
0804C
0805C Subroutine for inverting a symmetric matrix, SYMMAT.
0806C It requires a small (machine dependent) value for TOL
0807C as a criterion of singularity.
0808C
0809C Arguments
0810C SYMMAT the symmetric matrix, its inverse is returned
0811C in SYMMAT
0812C ISDIM the dimensions of SYMMAT (max set to 51)
0813C ISIZE the order of SYMMAT (max ISDIM)
0814C IFAULT error indicator. A value of 0 is returned
0815C if inversion is successful, 1 if
0816C unsuccessful (matrix non positive definite).
0817C
0818C
0819 DOUBLE PRECISION SYMMAT(ISDIM, I S D I M ) , D ( 5 1 ) , E ( 5 1 ) , Y , W , T O L
0820 INTEGER Z(51),P
0821C
0822 DATA TOL/I.0D-12/
0823 DO i00 J-I,ISIZE
0824 i00 Z(J) -J
0825 DO 160 I-I,ISIZE
0826 K=I
0827 Y=SYMMAT(I, I)
0828 L=I-I
0829 P-I+1
0830 IF(P.GT.ISIZE) GOTO 120
0831 DO 105 J-P,ISIZE
0832 W=SYMMAT ( I ,J)
0833 IF(DABS(W) .LE.DABS(Y)) GOTO 105
0834 K~J
0835 Y=W
0836 105 CONTINUE
0837 120 IF(DABS(Y) .LT.TOL) GOTO 200
0838 Y-I. 0D0/Y
0839 DO 130 J-I,ISIZE
0840 E(J) -SYMMAT (J, K)
0841 SYMMAT (J, K) -SYMMAT (J, I)
0842 SYMMAT (J, I) --E(J) *Y
0843 SYMMAT (I ,J) -SYMMAT (I ,J) *Y
0844 130 D(J) =SYMMAT ( I ,J)
0845C
0846 SYMMAT(I, I) -Y
0847 J-Z (1)
0848 Z(1)=Z(K)
0849 Z(K)=J
0850 DO 150 KsI,ISIZE
Design of optimal sampling schemes for local estimation--ll 357
0926 SIZE=SCALE*7.0+1.0
0927 IMAG=INT(SIZE)
0928 CALL CTRMAG(IMAG)
0929 CALL BLKPEN
0930 CALL AXESSI(XINC,YINC)
0931 CALL BOX (XMIN,XMAX,YMIN,yMAX+RYHALF)
0932 121 CALL MAP(XMIN,XMAX,YMIN-RY2P6,YMIN)
0933 CALL PSPACE(XL,XH,YLL,YL)
0934 IF(BOTH.AND..NOT.POINTK) GOTO 124
0935 IF(CENCEL) GOTO 122
0936 IF(COLOUR) CALL GRNPEN
0937 IF(.NOT.COLOUR) CALL BROKEN(2,3,2,3)
0938 GOTO 134
0939 122 IF(BOTH) GOTO 123
0940 IF(COLOUR) CALL REDPEN
0941 GOTO 133
0942 123 IF(COLOUR.AND.SQGRID) CALL REDPEN
0943 IF(COLOUR.AND..NOT.SQGRID) CALL GRNPEN
0944 IF(.NOT.COLOUR.AND..NOT.SQGRID) CALL BROKEN(2,3,2,3)
0945 IF(SQGRID) GOTO 133
0946 GOTO 135
0947 124 IF(.NOT.CENCEL) GOTO 126
0948 IF(.NOT.SQGRID) GOTO 125
0949 IF(COLOUR) CALL REDPEN
0950 GOTO 133
0951 125 IF(COLOUR) CALL GRNPEN
0952 IF(.NOT.COLOUR) CALL BROKEN(I,1,1,1)
0953 GOTO 135
0954 126 IF(.NOT.SQGRID) GOTO 127
0955 IF(COLOUR) CALL REDPEN
0956 CALL BROKEN(I,3,1,3)
0957 GOTO 134
0958 127 IF(COLOUR) CALL GRNPEN
0959 CALL BROKEN(5,2,5,2)
0960 GOTO 136
0961 133 CALL POSITN(XMIN+RX,YMIN-RYIPL)
0962 IF(CENCEL) GOTO 137
0963 GOTO 138
0964 134 CALL POSITN(XHALF+RX,YMIN-RYIPL)
0965 IF(CENCEL) GOTO 137
0966 GOTO 138
0967 135 CALL POSITN(XMIN+RX,YMIN-RY2P5)
0968 IF(CENCEL) GOTO 137
0969 GOTO 138
0970 136 CALL POSITN(XHALF+RX,YMIN-RY2P5)
0971 IF(.NOT.CENCEL) GOTO 138
0972 137 CALL LINE(RX2P5,0.0)
0973 CALL TYPECS(' CELL-CENTRED',14)
0974 GOTO 139
0975 138 CALL LINE(RX2P5,0.0)
0976 CALL TYPECS(' GRID-CENTRED',14)
0977 139 CALL MAP(XMIN,XMAX,YMIN,YMAX+RYHALF)
0978 CALL PSPACE(XL,XH,YL,YH)
0979 IF(NBLOCK.EQ.I) GOTO 160
0980 DO 150 J=I,NSPACE
0981 DO 140 K=I,NBLOCK
0982 GSTVEC(K)=SERROR(J,K)
0983 140 CONTINUE
0984 CALL POSITN(BLKLEN(1),GSTVEC(1))
0985 CALL CURVEO(BLKLEN,GSTVEC,I,NBLOCK)
0986 IF(BOTH.AND.SQGRID.AND.CENCEL) GOTO 145
0987 IF(.NOT.BOTH.AND.CENCEL) GOTO 145
0988 GOTO 150
0989 145 CALL SPACE(3)
0990 CALL TYPENF(GRDSPC(J),2)
0991 IF(.NOT.KMETRE) GOTO 148
0992 CALL TYPECS(' km',3)
0993 GOTO 150
0994 148 CALL TYPECS(' m',2)
0995 150 CONTINUE
0996 GOTO 200
0997 160 DO 170 J=I,NSPACE
0998 GSTVEC(J)=SERROR(J,I)
0999 170 CONTINUE
1000 CALL POSITN(GRDSPC(1),GSTVEC(1))
1001 CALL CURVEO(GRDSPC,GSTVEC,I,NSPACE)
1002 CALL SPACE(3)
Design of optimal sampling schemes for local estimation--ll 359
1003 I F ( B O T H . A N D . . N O T . S Q G R I D ) GOTO 200
1004 IF(POINTK) GOTO 175
1005 CALL TYPENF(BLKLEN(1),2)
1006 GOTO 180
1007 175 CALL TYPECS('POINTS',6)
1008 GOTO 200
1009 i80 IF(.NOT.KMETRE) GOTO 190
i010 CALL TYPECS(' km',3)
i011 GOTO 200
1012 190 CALL TYPECS(' m',2)
1013 200 IF(.NOT.CENCEL) GOTO 280
1014 CALL CTRMAG(IMAG+2)
1015 IF(COLOUR) CALL REDPEN
1016 CALL P L O T C S ( X M I N , Y M A X + R Y , T I T L E , 6 4 )
1017 CALL BLKPEN
1018 CALL CTRMAG(IMAG+I)
1019 CALL P O S I T N ( X H A L F - R X , Y M I N - R Y H A L F )
1020 IF(NBLOCK.EQ.I) GOTO 220
1021 CALL T Y P E C S ( ' S I Z E OF BLOCK SIDE',IS)
1022 GOTO 230
1023 220 CALL T Y P E C S ( ' G R I D SPACING',12)
1024 230 IF(.NOT.KMETRE) GOTO 240
1025 CALL TYPECS(' (km)',5)
1026 GOTO 250
1027 240 CALL TYPECS( ' (m)',4)
1028 250 CALL CTRSET(4)
1029 CALL CTRMAG(IMAG+2)
1030 CALL POSITN(XMIN-RX,YHALF)
1031 CALL TYPENC(29)
1032 CALL CTRSET(8)
1033 CALL SUFFIX
1034 CALL TYPECS('K',I)
1035 CALL NORMAL
1036 CALL P O S I T N ( X M I N - R X , Y H A L F )
1037 CALL CTRSET(4)
1038 CALL TYPENC(29)
1039 CALL CTRSET(8)
1040 CALL SUPFIX
1041 CALL TYPECS('2',I)
1042 CALL NORMAL
1043 CALL P O S I T N ( X M I N + R X , Y M I N - R Y )
1044 I F ( B O T H . A N D . . N O T . S Q G R I D ) CALL P O S I T N ( X M I N + R X , Y M I N - R Y 2 )
1045 CALL CTRMAG(IMAG)
1046 IF(SQGRID) GOTO 260
1047 CALL TYPECSq ' E Q U I L A T E R A L T R I A N G U L A R GRID ',28)
1048 GOTO 270
1049 260 CALL TYPECSq 'SQUARE GRID ',12)
1050 270 CALL CTRMAG~IMAG-I)
1051 CALL TYPECS ' radius = ',I0)
1052 CALL CTRMAG IMAG)
1053 CALL TYPENF RADIUS,2)
1054 CALL CTRMAG IMAG-I)
1055 CALL TYPECS ' number of points used for i n t e r p o l a t i o n =',44)
1056 CALL CTRMAGqIMAG)
1057 CALL TYPENI(NFOUND)
1058 280 I F ( B O T H . A N D . . N O T . P O I N T K . A N D . . N O T . S Q G R I D . A N D . . N O T . C E N C E L )
1059 & CALL GREND
1060 IF(.NOT.BOTH.AND..NOT.POINTK.AND..NOT.CENCEL) CALL GREND
1061 I F ( B O T H . A N D . P O I N T K . A N D . . N O T . S Q G R I D ) CALL GREND
1062 I F ( . N O T . B O T H . A N D . P O I N T K ) CALL GREND
1063 CALL FULL
1064 CALL BLKPEN
1065 RETURN
1066 END
360 A.B. McBI~TNEYand R. WEBSTER
APPENDIX 2
SPHERICAL MODEL
t
GAMMA(H) = 0.021000 + 0.330000(1.5H/ 196.399830 + 0.5 (H**3/ 196.399830"'3)) ; H < 196.399830
GAMMA(H) = 0.351000 ; H >= 196.399830 a
DISTANCES (H) ARE IN METRES
WITHIN-BLOCK VARIANCES
o
T
Size of block side (m)
20.00 30.00 40.00 50.00 60.00 80.00 I00.00 150.00 200.00 300.00
0.0470 0.0600 0.0728 0.0855 0.0980 0.1226 0.1461 0.1992 0.2408 0.2893
Topsoil pH at Broom's Barn Experimental Station Isotropic spherical semi-variogram model.
KRIGING VARIANCES FOR AN EQUILATERAL TRIANGULAR GRID WITH INIERPOLATION POINTS A~ THE CENTRE OF G R I D CELLS
20.00 30.00 40.00 50.00 60.00 80.00 ]00.00 150.00 200.00 300.00
20.00 0.0123 0 0076 0.0053 0.0039 0.0029 0.00]9 0.0013 0.0017 0.0087 0.0387
S 40.00 0.0363 0 0273 0.0201 0.0148 0.0115 0.0079 0.0057 0.0029 0.0018 0.0027
p 60.00 0.0619 0 0515 0.0423 0.0342 0.0275 0.018] 0.0132 0.0076 0.0044 0.0023 t~
~m
a 80.00 0.0885 0 0775 0.0673 0.0580 0.0496 0.0355 0.0255 0.0142 0.0089 0.0045
c 100.00 0 ]167 0 ]052 0.0943 0.0842 0.0747 0.0579 0.0440 0.0235 0 0152 0.0078
i ]20.00 0 1490 0 1370 0.1255 0.1145 0.]040 0.0846 0.0675 0.0366 0 0221 0.0122 -¢
n 140.00 0 1857 0 1733 0.1612 0.]494 0.1379 0.1162 0.0961 0.0552 0 031l 0.0152
g 160.00 0 2206 0.2078 0.1953 0.]830 0.1709 0.1475 0.1254 0.0775 0 0441 0.0188
180.00 0 2517 0.2388 0.2260 0.2134 0.2008 0.1764 0.1530 0.1004 0 0604 0.0250
(m) 200.00 0 2763 0.2633 0.2504 0.2375 0 2247 0.1997 0.1755 0.1202 0 0764 0.0328
220.00 0.2927 0.2797 0.2666 0.2537 0 2408 0.2155 0.1909 0.1343 0.0889 0.0395
240.00 0.3028 0.2897 0.2767 0.2637 0 2508 0.2254 0.2007 0.1440 0.0983 0.0454
260.00 0.3084 0.2953 0.2823 0.2693 0 2565 0.2312 0.2066 0.]504 0.]050 0.0507
280.00 0.3110 0.2980 0.2850 0.2721 0 2593 0.2342 0.2099 0.1545 0.1098 0.0553
300.00 0.3121 0.2991 0.2861 0.2733 0 2606 0.2357 0.2117 0.1570 0.1130 0.0591
KRIGING STANDARD ERRORS FOR AN EQUILATERAL TRIANGULAR GRID WITH INTERPOLATION POINTS AT THE CENTRE OF G R I D CELLS
20.00 30.00 40.00 50.00 60.00 80.00 100.00 150.00 200.00 300.00
20.00 0.1110 0.0873 0.0731 0.0625 0.0541 0.0435 0.0364 0.0418 0.0932 0.1967
S 40.00 0 1906 0.1652 0.1416 0.1218 0 1073 0.0887 0.0753 0 0541 0.0430 0.0519
P 60.00 0 2487 0.2270 0.2057 0.1851 0 1657 0.1345 0.1148 0 0870 0.0666 0.0484
G a 80.00 0 2975 0.2784 0.2595 0.2409 0 2227 0.1883 0.1597 "0 1192 0.0945 0.0670
r c i00.00 0 3415 0.3243 0.3071 0.2901 0 2733 0.2406 0.2099 0 1535 0.1232 0.0881
i i 120.00 0 3860 0.3701 0.3543 0.3384 0 3225 0.2909 0.2598 0 1912 0.1487 0.1104
d n 140.00 0 4309 0.4163 0.4015 0.3865 0 3714 0.3408 0.3100 0 2350 0.1764 0.1232
g 160.00 0 4696 0.4559 0.4419 0.4278 0 4134 0.3841 0.3542 0 2785 0.2099 0.1372
180.00 0 5017 0.4887 0.4754 0.4619 0 4482 0.4200 0.3911 0 3169
(m) 0.2458 0.1582
200.00 0 5257 0.5131 0.5004 0.4873 0 4741 0.4469 0.4189 0.3467 0.2765 0.1811 o
220.00 0 5410 0.5288 0.5164 0.5037 0 4907 0.4642 0.4369 0.3665 0.2982 0.1987
240.00 0 5503 0.5382 0.5260 0.5135 0 5008 0.4748 0.4480
E
0.3795 0.3135 0.2131
260.00 0.5553 0.5434 0.5313 0.5190 0.5064 0.4808 0.4546 0.3878 0.3241 0.2252
280.00 0.5577 0.5459 0.5339 0.5217 0.5092 0.4840 0.4582 0.3931 0.3314 0.2352
300.00 0.5586 0.5469 0.5349 0.5228 0.5105 0.4855 0.4601 0.3962 o
0.3362 0.2432
T
**NOTE for c o m p a r a t i v e purposes the grid spacings are e q u i v a l e n t to spacings on square grids with
the same sampling intensity. The actual d i s t a n c e s between grid points are 1.0746 times greater.
KRIGING VARIANCES FOR AN EQUILATERAL TRIANGULAR GRID WITH INTERPOLATION POINTS AT GRID POINTS
20.00 30.00 40.00 50.00 60.00 80.00 100.00 150.00 200.00 300.00
20.00 0.0109 0.0080 0.0054 0.0038 0.0029 0.0019 0.0013 . 0.0027 0.0121 0.0474
S 40.00 0.0188 0.0181 0.0167 0.0147 0.0124 0.0081 0.0055 0.0030 0.0019 0 0042
p 60.00 0.0226 0.0235 0.0236 0.0231 0.0220 0.0187 0.0146 0.0073 0.0046 0 0024
G a 80.00 0.0249 0.0268 0.0281 0.0287 0.0288 0.0276 0.0250 0.0154 0.0087 0 0047
r c 100.00 0.0270 0.0298 0.0321 0.0338 0.0349 0.0357 0.0346 0.0264 0.0160 0 0077
i i 120.00 0.0282 0.0316 0.0345 0.0368 0.0386 0.0407 0.0409 0.0349 0.0244 0 0117
d n 140.00 0.0286 0.0323 0.0355 0.0381 0.0402 0.0431 0.0440 0.0399 0.0309 0 0154
g 160.00 0.0288 0.0325 0.0358 0.0385 0.0408 0.0440 0.0454 0.0432 0.0364 0 0214 =-
180.00 0.0288 0.0326 0.0359 0.0387 0.0411 0.0446 0.0466 0.0461 0.0416 0 0294
(m) 200.00 0.0288 0.0326 0.0360 0.0389 0.0413 0.0451 0.0474 0.0482 0.0454 0 0366
220.00 0.0288 0.0326 0.0360 0.0389 0.0414 0.0451 0.0475 0.0488 0.0470 0.0409
240.00 0.0288 0.0326 0.0360 0.0389 0.0414 0.0451 0.0475 0.0490 0.0476 0.0433
260.00 0.0288 0.0326 0.0360 0.0389 0.0414 0.0451 0.0475 0.0490 0.0478 0.0445 ~m
280.00 0.0288 0.0326 0.0360 0.0389 0.0414 0.0451 0.0475 0.0490 0.0478 0.0451
300.00 0.0288 0.0326 0.0360 0.0389 0.0414 0.0451 0.0475 0.0490 0.0478 0.0453
**NOTE for c o m p a r a t i v e purposes the grid spacings are e q u i v a l e n t to spacings on square grids with
the same sampling intensity. The actual d i s t a n c e s between grid points are 1.0746 times greater.
KRIGING STANDARD ERRORS FOR AN EQUILATERAL TRIANGULAR GRID WITH INTERPOLATION POINTS AT GRID POINTS
20.00 30.00 40.00 50.00 60.00 80.00 ]00.00 150.00 200.00 300.00
20.00 0.1045 0.0896 0.0737 0.0620 0.0540 0.0436 0.0366 0.0524 0.ii00 0.2177
S 40.00 0.1371 0.1345 0.1290 0 1211 0.1113 0.0899 0.0744 0.0551 0.0434 0.06~2
P 60.00 0.1504 0.1533 0.1537 0 1519 0.1483 0.1366 0.1208 0.0854 0.0678 0.0490
G a 80.00 0.1577 0.1637 0.1675 0 1694 0.1697 0.1662 0.1580 0.1240 0.0935 0.0685 --.
r c I00.00 0.1642 0.1727 0.1792 0 1838 0.1869 0.1889 0.1861 "0.1626 0.1264 0.0875
i i 120.00 0.1679 0.1778 0.1857 0 1919 0.1965 0.2018 0.2023 0.1868 0.1562 0.1081
d n 140.00 0.1692 0.1797 0.1883 0 1952 0.2006 0.2075 0.2098 0.1997 0.1758 0.]240
g 160.00 0.1697 0.1804 0.1892 0 1963 0.2020 0.2097 0.2131 0 2077 0.1907 0.1463
180.00 0.1698 0.1805 0.1895 0 1968 0.2028 0.2113 0.2158 0 2148 0.2039 0.1715
(m) 200.00 0.1698 0.1807 0.1897 0 1972 0.2033 0.2123 0.2176 0 2195 0.2131 0 .]914
220.00 0.1698 0.1807 0.1897 0 1972 0.2034 0.2124 0.2179 0 2210 0.2168 0.2023
240.00 0.1698 0.1807 0.1897 0 1972 0.2034 0.2124 0.2180 0 2213 0.2182 0.2081
260.00 0.1698 0.1807 0.1897 0.1972 0.2034 0.2124 0.2180 0 2214 0.2186 0.2110
280.00 0.1698 0.1807 0.1897 0.1972 0.2034 0.2124 0.2180 0.2214 0.2187 0,2123
300.00 0.1698 0.1807 0.1897 0.1972 0.2034 0.2124 0.2180 0.2214 0.2187 0.2129
T
**NOTE for c o m p a r a t i v e purposes the grid spacings are equivalent to spacings on square grids with
the same sampling intensity. The actual d i s t a n c e s b e t w e e n grid points are 1.0746 times greater.