A neural method for geographical continuous field estimation

摘要

Geographical continuous fields are usually represented by means of sample points or zonal attribute sets, which leads to discontinuities in attribute values, or unrealistic homogeneities. The solution is to interpolate and extrapolate from known values, according to some constraints upon the estimation. Then, a new estimation method is proposed, based on finite difference method results, and the Hopfield neural network. This new method meets our statistical and morphological constraints.