Title: Image forecasting using dynamical functional time series models
Authors: Julian Austin - Newcastle University (United Kingdom) [presenting]
Jian Qing Shi - Southern Univesity of Science and Technology (China)
Zhenhong Li - Newcastle University (United Kingdom)
Abstract: The advancement of remote sensing technologies over the last few decades has meant a plethora of earth observation data is becoming readily available. Such data is often viewed as imagery and can be used for climate analysis, land monitoring or natural hazard studies. However, the data sets are often limited due to constraints on acquisition times. Forecasting or interpolation of these images would provide additional information which can be fed back into management scenarios. Forecasting of remotely sensed images is often difficult due to the high dimensionality of the data coupled with the spatial and temporal dependency among images. We consider the approach of treating imagery from a functional point of view. We utilise functional decompositions to reduce the dimensionality of the data before forecasting the simpler series of component scores. We compare functional principal component analysis (fPCA) and functional maximal correlation factors (fMAF) methods of decomposition to see the impact of forcing correlation in our functional components on our forecast results. We find that both fPCA and fMAF are capable of producing reasonable results, and forcing temporal correlation in fMAF decomposition causes more stable component scores for forecasting and interpolation.