INTERPOLATION METHODS

Morrison (1971) characterizes the production of the distribution of a property in two dimensional space as a two staged process. In the first stage, a point interpolation method is used to provide estimates of the property at nodes located either at the intersection of regularly spaced, orthogonal grid lines, or centrally between the grid lines. Isolines are then produced by linear interpolation between these nodal values. Morrison (1971) concluded that the error introduced by the second stage linear interpolation may be assumed to be insignificant. There are many methods of interpolating randomly spaced point data. Some of these method are global while others are local. Global methods utilize all the known values to evaluate an unknown value, while in local methods only a specified number of nearest neighbours are used to evaluate an unknown value. The selected methods are representative of the general classes of methods available. The general purpose of interpolation is to obtain an estimate of a property at a point i , Pi, based on values of the property at sampled or measured locations.

Inverse-distance weighted method (IDW)

The inverse-distance weighted procedure is versatile, easy to program and understand, and is fairly accurate under a wide range of conditions (Lam, 1983). Using this method, the property at each unknown location for which a solution is sought is given by:

Pi =

j=1 G

P j /D n ij
j=1

(1) 1/D n ij

where Pi is the property at location i; Pj is the property at sampled location j; Dij is the distance from i to j; G is the number of sampled locations; and n is the inverse-distance weighting power. The value of n, in effect, controls the region of influence of each of the sampled locations. As n increases, the region of influence decreases until, in the limit, it becomes the area which is closer to point i than to any other. When n is set equal to zero, the method is identical to simply averaging the sampled values. As n gets larger, the method approximates the

Voronoi tessellation procedure (Watson and Philip, 1985). Usually, the value of n is set arbitrarily. Values of 1.65 and 2 were used for n by Kelway (1974) and NOAA (1972), respectively, for interpolating rainfall, while the ARMOS model (Kaluarachchi et al., 1990) recommends the use of values ranging from 4 to 8 for interpolating oil pressure heads. Watson and Philip (1985) listed some of the limitations of the inverse-distance weighted procedure. The major limitation is that estimates are bounded by the extrema in the sampled values. Additionally, the radial symmetry which this procedure imparts to the data obscures the effect of linear features such as ridges or valleys. For n 1, the derivative of the interpolated surface is discontinuous at the sampled locations, while for n >1, the surface is flat at these sampled locations. Watson and Philip (1985) have presented a variation of the inverse-distance weighted procedure which removes the limitations mentioned above. In their procedure the weights are functions of slopes as well as of distance.

Punctual Kriging (PKG)

In the punctual kriging method, a weighted mean of the sampled values is evaluated for each node, such that the estimation error is minimized by solving the following set of equations simultaneously (Davis, 1986): Pi = WjPj
j=1 G

(2)

Wj = 1 j=1 W j (D kj ) + = (D ki ) j=1
G

(3)

k = 1,2,...G

(4)

where Wj is the weight for location j; is a variable added to minimize the estimation error; and (Dkj) is the semivariogram value between points k and j. In situations where the nature of the semivariogram is unknown, the set of equations listed above can still be solved by assuming that there is no nugget effect and that the semivariogram is given by:

(D kj ) = A f(D kj )

(5)

where f is a known function of Dkj and A is a constant. There is only a small group of functions that can be used to define a semivariogram. McBratney and Webster (1986) have listed these functions and explained why other functions are inappropriate. In the simplest case in which f(Dkj) = Dkj, equation 4 may be replaced by:
j=1

W j D kj + = D ki

k = 1,2,...G

(6)

where is a constant, since it is not necessary to determine the value of or in order to solve for Pi. This simplest case, which is essentially the assumption of a linear semivariogram, is useful in situations where the number of sampled locations is not sufficient to define the actual shape of the semivariogram. Research has shown that while the shape of the semivariogram influences the estimation variance [a statistic whose usefulness is questionable in any case (Philip and Watson, 1986; Shrivastava, 1986)] , it has little effect on the interpolated values ( Tevis et al., 1991). Cooke and Mostaghimi (1992) used a linear semivariogram to estimate areal mean rainfall and obtained values which were not significantly different from those obtained by more conventional methods.

Akima's polynomial method (AQP)

Akima (1978) developed a method of interpolation in two dimensional space, based on fitting local fifth-order polynomials to triangles whose vertices are located at the sampled locations. He used a triangulation method proposed by Lawson ( 1972) which maximizes the minimum interior angle of the triangles. Once the triangles are formed, coefficients of the local polynomials are chosen such that the partial derivatives of the surface is continuous across the sides of the triangles and at the vertices. The local polynomials are of the form: P i = q lk x l y k
l=0 k=0 5 5l

(7)

where q are the 21 coefficients of each local polygon, x and y are coordinate distances relative to some local origin, and l and k are dummy variables. Akima's method can also be used for extrapolation, though the accuracy decreases for such an application.

Major limitations of this method are that higher order derivatives are not continuous across the sides of the triangles, and that if the property changes drastically over a short distance, there tends to be some oscillations in the estimated values close to the vertices. Preusser (1990 ) developed a modified method which has continuous second derivatives across the boundaries and Montefusco and Casciola (1989) developed a modified method which eliminates any tendencies to oscillate.

Minimum surface curvature method (MSC)

The minimum surface curvature method was developed by Briggs (1974). He showed that if the total curvature of a surface is minimized under the constraint that the surface honors the values at sampled locations then the surface satisfies the equation: 4P + 2 4P + 4P = 0 x 4 x 2 y 2 y 4 (8)

at all unsampled locations. On a discrete, gridded surface, this corresponds to solving simultaneously the finite difference equation: Pi+2,j + Pi,j+2 + Pi-2,j + Pi,j-2 + 2(Pi+1,j+1 + Pi-1,j+1 + Pi+1,j-1 + Pi-1,j-1) - 8(Pi+1,j + Pi-1,j + Pi,j-1 + Pi,j+1) + 20 Pi,j = 0 (9)

for each grid point which is not sampled or which is not at nor immediately adjacent to the boundary of the domain. Similar difference equations were presented for those grid intersection points not included in Equation 9. Sampled locations are excluded from the solution as their values are always honored. These simultaneous equations can be solved iteratively.

REFERENCES
Akima, H. 1978. A method of bivariate interpolation and smooth surface fitting for irregularly distributed data points. ACM Trans. on Math. Software 4: 148-159. Briggs, I.C. 1974. Machine contouring using minimum curvature. Geophysics 39: 39-48. Cooke, R.A. and S. Mostaghimi. 1992. A microcomputer based routine for obtaining mean watershed precipitation from point values. Computers and Geosciences 18: 823-837. Davis, J. C. 1986. Statistics and data analysis in geology (2nd ed.). New York: John Wiley & Sons. Kaluarachchi, J.J., J.C. Parker and R.J. Lenhard. 1990. A numerical model for areal migration of water and light hydrocarbon in unconfined aquifers. Adv. Water Resour. 13: 29-40 Kelway, P. S. 1974. A scheme for assessing the reliability of interpolated rainfall estimates. Jour. Hydrology 21: 247-267. Lam, N. S. 1983. Spatial interpolation methods review. The American Cartographer 10: 129-149. Lawson, C.L. 1972. Generation of a triangular grid with application to contour plotting. Tech. Memo. 299, Sect. 914, Jet Propulsion Lab., Caltech, Pasadena, California. McBratney, A.B., and R. Webster 1986. Choosing functions for semi-variograms of soil properties and fitting them to estimates. Jour. Soil Science 37: 617-639. Montefusco, L.B. and G. Casciola. 1989. C1 surface interpolation. ACM Transactions on Mathematical Software 15: 365-374. Morrison, J. 1971. Method-produced errors in isarithmic mapping. TM CA-5, American Congress on Surveying and Mapping, Washington DC.

N.O.A.A. 1972. National Weather Service River Forecast System forecast procedures. TM NWS HYDRO-14, US Department of Commerce, Washington DC. Philip, G.M. and D.F. Watson. 1986. Matheronian geostatistics - Quo vadis?. Math. Geol. 18: 93-117. Preusser, A. 1990. C1- and C2- interpolation on triangles with quintic and nonic bivariate polynomials. ACM Transactions on Mathematical Software 16: 253-257. Srivastava, R.M. 1986. Philip and Watson - Quo vadunt?. Math. Geol. 18: 141-149. Tevis, J.W., A.D. Whittaker and D.J. McCauley. 1991. Efficient use of data in the kriging of soil pH. ASAE paper 91-7074. St. Joseph, MI: ASAE. Watson, D.F. and G.M. Philip. 1985. A refinement of inverse distance weighted interpolation. Geo-Processing 2: 315- 327.

(7) C

A (4)

8 X 3

12

B (11)
Arithmetic Mean : (A+B+C)/3

Inverse Distance: Pi =
j=1 G

P j /D n ij
j=1

1/D n ij

x=

4 8n 1 8n

7 + 11 + n 3 12 n 1 + 31n + 12 n

n = 1:

x=
n=2 n=8

4 7 + 11 + 12 8 3 1 1 1 + + 8 3 12

x=

4 82 1 82

7 + 11 + 2 3 12 2 1 + 312 + 12 2

x=

4 88 1 88

7 + 11 + 8 3 12 8 1 + 318 + 12 8

n = 0.5
4

x=

8 0.5 1 8 0.5

7 + 311 0.5 + 12 0.5 1 + 31 0.5 + 12 0.5