Proceedings Article | 29 December 2008
KEYWORDS: Artificial neural networks, Lithium, Error analysis, Neurons, Geographic information systems, Algorithm development, Neural networks, Associative arrays, Remote sensing, Cartography
In this paper, we specially emphasize on BP Artificial Neural Networks (BP ANNs) in spatial interpolation of DEM, and
simulate one spatial interpolation case in the area where there are several discrete known levelling points using input
vector: (X, Y, XY, X2, Y2) or (X, Y, XY, X2, Y2, XY2, X2Y, X3, Y3)instead of (X,Y), where the X is the horizontal
coordinate and the Y is the vertical coordinate. The results show that the new input vectors are usually applicable and
better than the classic one. In the numerical experiment of this paper, the maximum error is 2.032m when the input
vector, (X, Y, XY, X2, Y2, XY2, X2Y, X3, Y3) is used, while it is 2.807m when (X, Y) is applied. Further this BP ANNs
method is better than the classic polynomial method in which the maximum error of polynomial method is 6.728m.