|
Abstract:
|
Data from geophysical applications often cover a large portion of the Earth, so that there is a growing interest in developing covariance structures for processes on the surface of a sphere. A common assumption for global processes on the sphere is that they exhibit stationarity with respect to longitude, but they have nonstationarity with respect to latitude. However, this axially symmetric assumption can be restrictive for quantities such as surface temperature because its statistical behavior is strongly affected by geographical indicators such as land and ocean. We introduce nonstationary covariance models that can account for different geographical indicators over the Earth. The idea is that given a nonstationary process on the sphere we consider the spatial deformation approach to model the nonstationary covariance structure while incorporating large scale geographical indicators on the process. We use surface temperature from climate model outputs to compare our new nonstationary model with an axially symmetric model and a spatial deformation model. The new nonstationary model results in more flexible covariance structures and outperforms its competitors.
|
