Abstract
Rapid developments in satellite remote-sensing technology have enabled the collection of geospatial data on a global scale, hence increasing the need for covariance functions that can capture spatial dependence on spherical domains. We propose a general method of constructing nonstationary, locally anisotropic covariance functions on the sphere based on covariance functions in \(\mathbb {R}^3\). We also provide theorems that specify the conditions under which the resulting correlation function is isotropic or axially symmetric. For large datasets on the sphere commonly seen in modern applications, the Vecchia approximation is used to achieve higher scalability on statistical inference. The importance of flexible covariance structures is demonstrated numerically using simulated data and a precipitation dataset. Supplementary materials accompanying this paper appear online.
Similar content being viewed by others
References
Alegría A, Cuevas-Pacheco F, Diggle P, Porcu E (2021) The f-family of covariance functions: a matérn analogue for modeling random fields on spheres. Spat Stat 43:100512
Banerjee S, Gelfand AE, Finley AO, Sang H (2008) Gaussian predictive process models for large spatial data sets. J R Stat Soc Ser B 70(4):825–848
Bissiri PG, Peron AP, Porcu E (2020) Strict positive definiteness under axial symmetry on the sphere. Stoch Environ Res Risk Assess 34:723–732
Blake LR, Porcu E, Hammerling DM (2022) Parametric nonstationary covariance functions on spheres. Statistics 11(1):e468
Castruccio S, Genton MG (2014) Beyond axial symmetry: an improved class of models for global data. Statistics 3(1):48–55
Castruccio S, Genton MG (2016) Compressing an ensemble with statistical models: an algorithm for global 3d spatio-temporal temperature. Technometrics 58(3):319–328
Castruccio S, Stein ML et al (2013) Global space-time models for climate ensembles. Ann Appl Stat 7(3):1593–1611
Du J, Ma C, Li Y (2013) Isotropic variogram matrix functions on spheres. Math Geosci 45(3):341–357
Emery X, Porcu E, Bissiri PG (2019) A semiparametric class of axially symmetric random fields on the sphere. Stoch Environ Res Risk Assess 33:1863–1874
Emery X, Arroyo D, Mery N (2021) Twenty-two families of multivariate covariance kernels on spheres, with their spectral representations and sufficient validity conditions. Stoch Environ Res Risk Assess 66:1–21
Gneiting T (2013) Strictly and non-strictly positive definite functions on spheres. Bernoulli 19(4):1327–1349
Guinness J (2018) Permutation and grouping methods for sharpening Gaussian process approximations. Technometrics 60(4):415–429
Guinness J, Fuentes M (2016) Isotropic covariance functions on spheres: some properties and modeling considerations. J Multivar Anal 143:143–152
Heaton M, Katzfuss M, Berrett C, Nychka D (2014) Constructing valid spatial processes on the sphere using kernel convolutions. Environmetrics 25(1):2–15
Huang C, Zhang H, Robeson SM (2011) On the validity of commonly used covariance and variogram functions on the sphere. Math Geosci 43(6):721–733
Jeong J, Jun M (2015a) A class of matérn-like covariance functions for smooth processes on a sphere. Spat Stat 11:1–18
Jeong J, Jun M (2015b) Covariance models on the surface of a sphere: When does it matter? Statistics 4(1):167–182
Jeong J, Jun M, Genton MG (2017) Spherical process models for global spatial statistics. Stat Sci 32(4):501
Jones R (1963) Stochastic processes on a sphere. Ann Math Stat 34(1):213–218
Jun M (2014) Matérn-based nonstationary cross-covariance models for global processes. J Multivar Anal 128:134–146
Jun M, Stein ML (2007) An approach to producing space-time covariance functions on spheres. Technometrics 49(4):468–479
Jun M, Stein ML (2008) Nonstationary covariance models for global data. Ann Appl Stat 2(4):1271–1289
Kang M, Katzfuss M (2021) Correlation-based sparse inverse Cholesky factorization for fast Gaussian-process inference. arXiv:2112.14591
Katzfuss M (2011) Hierarchical spatial and spatio-temporal modeling of massive datasets, with application to global mapping of CO\(_2\). PhD thesis, The Ohio State University
Katzfuss M, Guinness J (2021) A general framework for Vecchia approximations of Gaussian processes. Stat Sci 36(1):124–141
Katzfuss M, Guinness J, Gong W, Zilber D (2020a) Vecchia approximations of Gaussian-process predictions. J Agric Biol Environ Stat 25(3):383–414
Katzfuss M, Jurek M, Zilber D, Gong W, Guinness J, Zhang J, Schäfer F (2020b) GPvecchia: fast Gaussian-process inference using Vecchia approximations. R package version 0.1.3
Katzfuss M, Guinness J, Lawrence E (2022) Scaled Vecchia approximation for fast computer-model emulation. SIAM/ASA J Uncertain Quantif 10(2):537–554
Kay JE, Deser C, Phillips A, Mai A, Hannay C, Strand G, Arblaster JM, Bates SC, Danabasoglu G, Edwards J, Holland M, Kushner P, Lamarque JF, Lawrence D, Lindsay K, Middleton A, Munoz E, Neale R, Oleson K, Polvani L, Vertenstein M (2015) The Community Earth System Model (CESM) large ensemble project: a community resource for studying climate change in the presence of internal climate variability. Bull Am Meteorol Soc 96(8):1333–1349
Knapp A (2012) Global Bayesian nonstationary spatial modeling for very large datasets. Bachelor Thesis
Lindgren F, Rue H, Lindström J (2011) An explicit link between gaussian fields and gaussian Markov random fields: the stochastic partial differential equation approach. J R Stat Soc Ser B Stat Methodol 73(4):423–498
Ma C (2012) Stationary and isotropic vector random fields on spheres. Math Geosci 44(6):765–778
Ma C (2015) Isotropic covariance matrix functions on all spheres. Math Geosci 47(6):699–717
Menegatto V (2020) Positive definite functions on products of metric spaces via generalized stieltjes functions. Proc Am Math Soc 148(11):4781–4795
Paciorek C, Schervish M (2006) Spatial modelling using a new class of nonstationary covariance functions. Environmetrics 17(5):483–506
Schäfer F, Katzfuss M, Owhadi H (2021) Sparse Cholesky factorization by Kullback–Leibler minimization. SIAM J Sci Comput 43(3):A2019–A2046
Stein ML (2005) Nonstationary spatial covariance functions. Technical Report No. 21, University of Chicago
Stein ML et al (2007) Spatial variation of total column ozone on a global scale. Ann Appl Stat 1(1):191–210
Vecchia A (1988) Estimation and model identification for continuous spatial processes. J R Stat Soc Ser B 50(2):297–312
Vihola M (2012) Robust adaptive metropolis algorithm with coerced acceptance rate. Stat Comput 22(5):997–1008
Yaglom A (1987) Correlation theory of stationary and related random functions, vol 1. Springer, New York
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
All authors declare that they have no conflicts of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
A Proofs
A Proofs
Proof of Theorem 1
If \(\gamma _1 ({\textbf{s}})=\gamma _2 ({\textbf{s}}) \equiv \gamma \) is constant, then
To compute \(\mathbf{\Sigma (s)}=\mathcal {R}_z (l) \mathcal {R}_y (L) {\tilde{\mathbf{\Sigma }}(s)} \mathcal {R}_y (L)' \mathcal {R}_z (l)'\), we first compute
where \(x=\cos (L)\cos (l)\), \(y=\cos (L)\sin (l)\), \(z=\sin (L)\) are the (x, y, z)-coordinates of a three-dimensional Cartesian coordinate system. Then
does not depend on \(\textbf{s}\). And for \(i \ne j\),
WLOG, we ignore the constant coefficients inside the inverse, and then
Let \(\textbf{B}:= ({\tilde{\textbf{s}}}_j \tilde{s}_j' + \textbf{I}_3)^{-1}\), and so
So, computation of \(q({\textbf{s}_i,s_j})\) only involves terms \({\tilde{\textbf{s}}_i' B \tilde{s}_i}\), \({\tilde{\textbf{s}}_j' B \tilde{s}_j}\) and \({\tilde{\textbf{s}}_i' B \tilde{s}_j}\). Because
we have
Further,
So, \(q({\textbf{s}_i,\textbf{s}_j})\) just depends on the distance \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\). For the normalization term \(c({\textbf{s}_i, \textbf{s}_j})\), since we have proved that \(|\mathbf{\Sigma (s_i)}|=|\mathbf{\Sigma (s_j)}| \equiv \gamma ^2\),
We have proved that \(q({\textbf{s}_i,\textbf{s}_j})\) just depends on \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\), so \(c({\textbf{s}_i, \textbf{s}_j})\) also only depends on the distance \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\).
Overall, we can show
only depends on the distance \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\), where \(\rho (q)\) is a valid isotropic correlation function. So, \(\rho _{NS}({\textbf{s}_i,\textbf{s}_j})\) is isotropic. \(\square \)
Proof of Theorem 2
If \(\kappa ({\textbf{s}}) \equiv 0\) and \(\gamma _1(\cdot )\), \(\gamma _2(\cdot )\) depend on L only, then \(\mathcal {R}_x (\kappa (s)) \equiv \mathcal {R}_x (0) = \textbf{I}_3\). Then
Due to the results in Theorem 1, we have
where
Thus,
only depend on L. WLOG, ignore \(\gamma _1(L)\), \(\gamma _2(L)\) again (they only depend on L),
where
Then
Because
we can figure out that the computation of \(q^2({\textbf{s}_i, \textbf{s}_j})\) only involves the following types of terms
We can change the index i to j for the first three terms, and they are still valid. Thus, these values only depend on \(s_{i2}\), \(s_{j2}\) and \({\tilde{\textbf{s}}_i}' {\tilde{\textbf{s}}_j}\), and \({\tilde{\textbf{s}}_i}' {\tilde{ \textbf{s}}_j}\) can be expressed in terms of the distance \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\). The computation of \(q^2({\textbf{s}_i,\textbf{s}_j})\) only depends on the distance \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\) and the longitudes \(s_{i2}\), \(s_{j2}\). Similar to (10) in the proof of Theorem 1, we can also show that \(c({\textbf{s}_i,\textbf{s}_j})\) is a function of \(({\tilde{\textbf{s}}_i - \tilde{s}_j})'({\tilde{\textbf{s}}_i - \tilde{s}_j})\), \(s_{i2}\) and \(s_{j2}\). Then \(\rho _{NS}({\textbf{s}_i,\textbf{s}_j})=c({\textbf{s}_i,\textbf{s}_j})\rho (q({\textbf{s}_i,\textbf{s}_j})):=\rho _A ({\tilde{\textbf{s}}_i - \tilde{s}_j}, s_{i2}, s_{j2})\), so it is axially symmetric. \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Cao, J., ZHANG, J., SUN, Z. et al. Locally Anisotropic Nonstationary Covariance Functions on the Sphere. JABES 29, 212–231 (2024). https://doi.org/10.1007/s13253-023-00573-y
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13253-023-00573-y