Testing Artificial Neural Network (ANN) for Spatial Interpolation

The aim of this research is to test Artificial Neural Network (ANN) package in GRASS 6.4 software for spatial interpolation and to compare it with common interpolation techniques IDW and ordinary kriging. This package was also compared with neural networks packages nnet and neuralnet available in software R Project. The entire packages uses multi-layer perceptron (MLP) model trained with the back propagation algorithm. Evaluation methods were based mainly on RMSE. All the tests were done on artificial data created in R Project software; which simulated three surfaces with different characteristics. In order to find the best configuration for the multilayer perceptron many different settings of network were tested (test-and-trial method). The number of neurons in hidden layers was the main tested parameter. Results indicate that MLP model in the ANN module implemented in GRASS can be used for spatial interpolation purposes. However the resulting RMSE was higher than RMSE from IDW and ordinary kriging method and time consuming. When compared neural network packages in GRASS GIS and R Project; it is better to use the packages in R Project. Training of MLP was faster in this case and results were the same or slightly better.