Implementation of Neural Networks within ArcGIS for Spatial Interpolation

We implement artificial neural networks (ANNs) using advanced back-propagation algorithms and neural network residual kriging (NNRK) within ArcGIS for spatial data interpolation. The ANN model enables GIS users to call input-output examples, train and test ANNs, perform spatial analyses and display interpolation results as a single process, with the help of geoprocessor functions within GIS software; while the NNRK model uses Kriging Wizard provided within GIS software and the ANN model developed in this study. We apply these models to create a grid-based digital elevation model (DEM) from elevation points for four study areas, which have different types of elevation data. Simulation results show that the advanced back-propagations such as iRprop speed up learning while they can get struck at local minima. DEM using ANNs can be more accurate than those using IDW or Kriging, in particular when the number of data points is small, and ANNs perform much better than polynomial interpolation as a global interpolation method. In addition, we observe that the NNRK model can outperform conventional spatial interpolation methods such as Kriging, IDW, and polynomial interpolation. A key outcome of this work is that the ANN model created within the de facto standard GIS software is applicable to various spatial analyses including hazard risk assessment over a large area, in particular when there are multiple potential causes, the relationship between risk factors and hazard events is not clear, and the number of available data is small, given its performance of DEM generation.