Parsimonious Compactly Supported Covariance Models in the Gauss Hypergeometric Class: Identifiability, Reparameterizations, and Asymptotic Properties

Abstract: We study the covariance model belonging to the Gauss hypergeometric ($GH$) class, a highly flexible and compactly supported correlation model (kernel). This class includes the well-known Generalized Wendland ($GW$) and Mat\'ern ($MT$) kernels as special cases. First, we provide necessary and sufficient conditions for the validity of the $\mathcal{GH}$ model. We then demonstrate that this family of models suffers from identifiability issues under both increasing and fixed-domain asymptotics. To address this problem, we propose a parsimonious version of the model that adheres to an optimality criterion. This approach results in a new class of compactly supported kernels, which can be seen as an improvement over the $\mathcal{GW}$ model. Additionally, we show that two specific compact support reparameterizations allow us to recover the $MT$ model, highlighting the advantages and disadvantages of each reparameterization. Finally, we establish strong consistency and the asymptotic distribution of the maximum likelihood estimator of the microergodic parameter associated with the proposed parsimonious model under fixed-domain asymptotics. The effectiveness of our proposal is illustrated through a simulation study exploring the finite-sample properties of the MLE under both increasing- and fixed-domain asymptotics and the analysis of a georeferenced climate dataset, using both Gaussian and Tukey-$h$ random fields. In this application, the proposed model outperforms the the $MT$ model.