Jointly non-Gaussian modeling of Argo anomaly and nugget fields

Develop a statistical model for Argo oceanographic data in which both the anomaly field and the nugget effect are non-Gaussian, thereby extending existing non-Gaussian stochastic partial differential equation models whose nugget effects remain Gaussian.

Background

The paper describes Argo mapping models that decompose an oceanographic field into a mean field, a spatially and temporally correlated anomaly field, and a nugget effect representing unresolved small-scale variability and measurement noise. Existing work has introduced non-Gaussian stochastic partial differential equation models for the anomaly field, including models driven by normal-inverse Gaussian white noise, and has shown improvements in predictive-distribution characteristics for temperature and salinity mapping.

However, the reviewed models retain a Gaussian nugget effect. The authors identify the absence of Argo-data models that simultaneously represent non-Gaussian behavior in both the anomaly field and the nugget effect. Such a model would be relevant because unresolved mesoscale and submesoscale variability and sensor-specific errors can produce non-Gaussian behavior at small scales.