A Neural Network-based Method for Solving “Nested Hierarchy” Areal Interpolation Problems

Abstract

This study proposes a neural network approach to solving areal interpolation scenarios, specifically the “nested hierarchy” problem. The neural network method presented adopts the approach taken by intelligent interpolation methods where ancillary spatial information is presented to assist in achieving more accurate results. For this study, the data to be estimated are total populations for census tracts and block groups in Hartford County, Connecticut. A number of neural network models are generated containing various combinations of ancillary spatial information. The neural-network-derived predictions are compared with the predicted populations derived from three existing interpolation methods: areal weighting, a dasymetric areal weighting approach using remote sensing data, and ordinary least squares (OLS) regression. For each scenario presented, the proposed neural network approach outperforms each of the existing methods.