|
Abstract:
|
For binary spatial data, a hierarchical statistical model is proposed, where the probability map of responses is determined by a latent Gaussian random field. However, the predictive distribution of this latent field will be non-Gaussian. Since Kernel Principal Component Analysis (KPCA) has the capability of preserving high-order statistics of non-Gaussian, non-stationary data of complex structures, we use a KPCA algorithm to parameterize the predictive distribution of the latent field from which optimal predictions are obtained. The effect of kernel choice on inferences will be investigated.
|
