Space-time series forecasting by artificial neural networks
Tao Cheng, Jiaqiu Wang, Xia Li · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2008
Spatio-Temporal Autoregressive Integrated Moving Average (STAIRMA) model family is a very useful tool in modeling space-time series data. It assumes that space-time series data is correlated linearly in space and time. However, in reality most space-time series contains nonlinear space-time autocorrelation structure, which can't be modeled by STARIMA. Artificial neural networks (ANN) have shown great flexibility in modeling and forecasting nonlinear dynamic process. In the paper, we developed an architecture approach to model space-time series data using artificial neural network (ANN). The model is tested with forest fire prediction in Canada. The experimental result demonstrates that STANN achieves much better prediction accuracy than STARIMA model.