Abstract
With the rapid advances of data acquisition techniques, spatio-temporal data are becoming increasingly abundant in a diverse array of disciplines. Here, we develop spatio-temporal regression methodology for analyzing large amounts of spatially referenced data collected over time, motivated by environmental studies utilizing remotely sensed satellite data. In particular, we specify a semiparametric autoregressive model without the usual Gaussian assumption and devise a computationally scalable procedure that enables the regression analysis of large datasets. We estimate the model parameters by maximum pseudolikelihood and show that the computational complexity can be reduced from cubic to linear of the sample size. Asymptotic properties under suitable regularity conditions are further established that inform the computational procedure to be efficient and scalable. A simulation study is conducted to evaluate the finite-sample properties of the parameter estimation and statistical inference. We illustrate our methodology by a dataset with over 2.96 million observations of annual land surface temperature, and comparison with an existing state-of-the-art approach to spatio-temporal regression highlights the advantages of our method.
Supplementary materials accompanying this paper appear online.
Similar content being viewed by others
References
Agirbas E, Koca L, Aytan U (2017) Spatio-temporal pattern of phytoplankton and pigment composition in surface waters of south-eastern Black Sea. Oceanologia 59(3):283–299
Anderson E et al (1999) LAPACK users’ guide, 3rd edn. SIAM, Philadelphia
Anselin L (2013) Spatial econometrics: methods and models. Springer, Cham
Asseng S et al (2015) Rising temperatures reduce global wheat production. Nat Clim Chang 5(2):143–147
Banerjee S, Gelfand AE, Finley AO, Sang H (2008) Gaussian predictive process models for large spatial data sets. J Roy Stat Soc B 70(4):825–848
Belgiu M, Stein A (2019) Spatiotemporal image fusion in remote sensing. Remote Sens 11(7):818
Blackford LS et al (2002) An updated set of basic linear algebra subprograms (blas). ACM Trans Math Softw 28(2):135–151
Brynjarsdóttir J, Berliner LM (2014) Dimension-reduced modeling of spatio-temporal processes. J Am Stat Assoc 109(508):1647–1659
Buluc A, Gilbert JR (2011) The combinatorial BLAS: design, implementation, and applications. Int J High Perform Comput Appl 25(4):496–509
Case AC (1991) Spatial patterns in household demand. Econometrica 59(4):953–965
Chakraborty T, Hsu A, Manya D, Sheriff G (2020) A spatially explicit surface urban heat island database for the United States: characterization, uncertainties, and possible applications. ISPRS J Photogramm Remote Sens 168:74–88
Chi G, Zhu J (2019) Spatial regression models for the social sciences. SAGE, New York
Chu T, Zhu J, Wang H (2019) Semiparametric modeling with nonseparable and nonstationary spatio-temporal covariance functions and its inference. Stat Sin 29(3):1233–1252
Coppersmith D, Winograd S (1990) Matrix multiplication via arithmetic progressions. J Symb Comput 9(3):251–280
Cressie N (1993) Statistics for spatial data. Wiley, New York
Cressie N, Shi T, Kang EL (2010) Fixed rank filtering for spatio-temporal data. J Comput Graph Stat 19(3):724–745
Cressie N, Wikle CK (2011) Statistics for spatio-temporal data. Wiley, New York
Diffenbaugh NS, Davenport FV, Burke M (2021) Historical warming has increased U.S. crop insurance losses. Environ Res Lett 16(8):084025
Dutilleul PRL (2011) Spatio-temporal heterogeneity: concepts and analyses. Cambridge University Press, Cambridge
Fernández C, Steel MF (1998) On Bayesian modeling of fat tails and skewness. J Am Stat Assoc 93(441):359–371
Finley AO, Banerjee S, Gelfand AE (2012) Bayesian dynamic modeling for large space-time datasets using gaussian predictive processes. J Geogr Syst 14(1):29–47
Fu P, Weng Q (2016) A time series analysis of urbanization induced land use and land cover change and its impact on land surface temperature with landsat imagery. Remote Sens Environ 175:205–214
Gao Z, Ma Y, Wang H, Yao Q (2019) Banded spatio-temporal autoregressions. J Econom 208(1):211–230
Gasparrini A et al (2017) Projections of temperature-related excess mortality under climate change scenarios. Lancet Planet Health 1(9):e360–e367
Gilbert JR, Moler C, Schreiber R (1992) Sparse matrices in MATLAB: design and implementation. SIAM J Matrix Anal Appl 13(1):333–356
Guinness J (2018) Permutation and grouping methods for sharpening Gaussian process approximations. Technometrics 60(4):415–429
Guinness J (2021) Gaussian process learning via Fisher scoring of Vecchia’s approximation. Stat Comput 31(3):25
Guo S, Wang Y, Yao Q (2016) High-dimensional and banded vector autoregressions. Biometrika 103(4):889–903
Hanewinkel M et al (2013) Climate change may cause severe loss in the economic value of european forest land. Nat Clim Chang 3(3):203–207
Hu H-W et al (2016) Effects of climate warming and elevated CO2 on autotrophic nitrification and nitrifiers in dryland ecosystems. Soil Biol Biochem 92:1–15
Huang H-C, Cressie N (1996) Spatio-temporal prediction of snow water equivalent using the Kalman filter. Comput Stat Data Anal 22(2):159–175
IPCC (2021) Climate Change 2021: the physical science basis. Contribution of Working Group I to the Sixth assessment report of the intergovernmental panel on climate change, Volume In Press. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press
Johannesson G, Cressie N, Huang H-C (2007) Dynamic multi-resolution spatial models. Environ Ecol Stat 14(1):5–25
Katzfuss M, Guinness J (2021) A general framework for vecchia approximations of gaussian processes. Stat Sci 36(1):124–141
Katzfuss M, Stroud JR, Wikle CK (2016) Understanding the ensemble Kalman filter. Am Stat 70(4):350–357
Kilic E, Stanica P (2013) The inverse of banded matrices. J Comput Appl Math 237(1):126–135
Kressner D (2005) Numerical methods for general and structured eigenvalue problems. Springer, Cham
Lee L-F, Yu J (2015) Estimation of fixed effects panel regression models with separable and nonseparable space-time filters. J Econ 184(1):174–192
Lesk C et al (2017) Threats to North American forests from southern pine beetle with warming winters. Nat Clim Chang 7(10):713–717
Li L, Yang Z (2021) Spatial dynamic panel data models with correlated random effects. J Econ 221(2):424–454
Lu Z, Steinskog DJ, Tjøstheim D, Yao Q (2009) Adaptively varying-coefficient spatiotemporal models. J Roy Stat Soc B 71(4):859–880
Luszczek P (2009) Parallel programming in MATLAB. Int J High Perform Comput Appl 23(3):277–283
Mafteiu-Scai LO (2015) The bandwidths of a matrix: a survey of algorithms. Ann West Univ Timisoara Math Comput Sci 52(2):183–223
Mariella L, Tarantino M (2010) Spatial temporal conditional auto-regressive model: a new autoregressive matrix. Aust J Stat 39(3):223–244
Mueller SE et al (2020) Climate relationships with increasing wildfire in the southwestern US from 1984 to 2015. For Ecol Manag 460:117861
NOAA (2021). State of the climate: global climate report for annual 2020
Nocedal J, Wright SJ (2006) Numerical optimization, 2nd edn. Springer, Cham
Nordman DJ, Lahiri SN (2004) On optimal spatial subsample size for variance estimation. Ann Stat 32(5):1981–2027
Oleson KW et al (2018) Avoided climate impacts of urban and rural heat and cold waves over the U.S. using large climate model ensembles for RCP85 and RCP45. Clim Change 146(3):377–392
Omernik JM, Griffith GE (2014) Ecoregions of the conterminous United States: evolution of a hierarchical spatial framework. Environ Manag 54(6):1249–1266
Rue H et al (2017) Bayesian computing with INLA: a review. Ann Rev Stat Appl 4(1):395–421
Rue H, Martino S, Chopin N (2009) Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. J R Stat Soc B 71(2):319–392
Saad Y (2003) Iterative methods for sparse linear systems (Second ed.). SIAM
Sarangi C et al (2021) Urbanization amplifies nighttime heat stress on warmer days over the US. Geophys Res Lett 48(24):e2021GL095678
Shen X et al (2022) Effect of shrub encroachment on land surface temperature in semi-arid areas of temperate regions of the northern hemisphere. Agric For Meteorol 320:108943
Sherman M (1996) Variance estimation for statistics computed from spatial lattice data. J Roy Stat Soc B 58(3):509–523
Shi W, Lee L-F (2017) Spatial dynamic panel data models with interactive fixed effects. J Econom 197(2):323–347
Stewart GW (2002) A Krylov-Schur algorithm for large eigenproblems. SIAM J Matrix Anal Appl 23(3):601–614
Thompson R, Hornigold R, Page L, Waite T (2018) Associations between high ambient temperatures and heat waves with mental health outcomes: a systematic review. Public Health 161:171–191
Vecchia AV (1988) Estimation and model identification for continuous spatial processes. J Roy Stat Soc B 50(2):297–312
Vose RS et al (2017) Temperature changes in the United States. pp. 185–206. Climate Science Special Report: Fourth National Climate Assessment, Volume I. U.S. Global Change Research Program
Wan Z (2014) New refinements and validation of the collection-6 MODIS land-surface temperature/emissivity product. Remote sensing of Environment 140:(36–45)
Westerling AL, Hidalgo HG, Cayan DR, Swetnam TW (2006) Warming and earlier spring increase western U.S. forest wildfire activity. Science 313(5789):940–943
Wikle CK, Zammit-Mangion A, Cressie N (2019) Spatio-temporal statistics with R. Chapman and Hall/CRC, London
Xu G, Liang F, Genton MG (2015) A Bayesian spatio-temporal geostatistical model with an auxiliary lattice for large datasets. Stat Sin 25(1):61–79
Yan Y et al (2020) Driving forces of land surface temperature anomalous changes in North America in 2002–2018. Sci Rep 10(1):6931
Yu J, De Jong R, Lee L-F (2008) Quasi-maximum likelihood estimators for spatial dynamic panel data with fixed effects when both n and t are large. J Econom 146(1):118–134
Zhang B, Cressie N (2020) Bayesian inference of spatio-temporal changes of arctic sea ice. Bayesian Anal 15(2):605–631
Zhang B, Sang H, Huang JZ (2015) Full-scale approximations of spatio-temporal covariance models for large datasets. Stat Sin 25(1):99–114
Zhang W, Yao Q, Tong H, Stenseth NC (2003) Smoothing for spatiotemporal models and its application to modeling muskrat-mink interaction. Biometrics 59(4):813–821
Zhao C et al (2017) Temperature increase reduces global yields of major crops in four independent estimates. Proc Natl Acad Sci 114(35):9326–9331
Acknowledgements
This material is based upon work supported by the National Aeronautics and Space Administration (NASA) under AIST-80NSSC20K0282 and the National Science Foundation (NSF) under DMS-2245906. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of NASA and NSF.
Author information
Authors and Affiliations
Corresponding author
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.
Appendix A: Notation and Assumptions
Appendix A: Notation and Assumptions
We first introduce some notation and conventions. Given an \(n\times n\) matrix \({\varvec{P}}= (p_{ij})_{n\times n}\), we use \(\mathrm{{tr}}({\varvec{P}})\) and \(\mathrm{{det}}({\varvec{P}})\) to denote the trace and determinant of a square matrix \({\varvec{P}}\), and we let \(vec_D({\varvec{P}})\) denote the column vector formed by the diagonal elements of \({\varvec{P}}\). The (i, j)th element of a matrix \({\varvec{P}}\) is denoted by \(\mathrm{{ent}}_{ij}({\varvec{P}})\). We define \(\Vert {\varvec{P}}\Vert _{1} = \max _{1\le j\le n}\sum _{i=1}^n|p_{ij}|\) and \(\Vert {\varvec{P}}\Vert _{\infty } = \max _{1\le i\le n}\sum _{j=1}^n|p_{ij}|\). We also let \(\Vert {\varvec{P}}\Vert _2=\{\lambda _{\max }({\varvec{P}}'{\varvec{P}})\}^{1/2}\) and \(\Vert {\varvec{P}}\Vert _F=\{\mathrm{{tr}}({\varvec{P}}'{\varvec{P}})\}^{1/2}\) denote the spectral norm and the Frobenius norm, respectively. Let \(abs({\varvec{P}}) = (|p_{i,j}|)_{n\times n}\). A sequence of \(n\times n\) matrix \({\varvec{P}}_n\) is said to be uniformly bounded in row and column sums (UB), if \(\sup _{n\ge 1}\Vert {\varvec{P}}_n\Vert _{1} <\infty \) and \(\sup _{n\ge 1}\Vert {\varvec{P}}_n\Vert _{\infty } <\infty \). We also use \({{\varvec{0}}}\) and \({{\varvec{1}}}\) to denote a matrix or a vector with all elements equal zero and one, respectively. For a real-valued function \(f({\varvec{x}})\), \({\varvec{x}}= ({\varvec{X}}_1, \ldots , x_k)'\in {\mathbb {R}}^k\), we let \(\nabla f({\varvec{x}})\) denote the gradient vector and let \(\nabla ^2 f({\varvec{x}})\) denote the Hessian matrix. The partial derivative of f with respect to \(x_j\) is denoted by \(\partial _{x_j} f({\varvec{x}})\) or \(\frac{\partial f({\varvec{x}})}{\partial x_j}\), whereas the second partial derivative with respect to \(x_j\) is denoted as \(\partial _{x_jx_j} f({\varvec{x}})\) (or \(\frac{\partial ^2 f({\varvec{x}})}{\partial x_j^2}\)).
In the following, we provide the regularity conditions for establishing the large sample properties of the PMLE \({\widehat{{\varvec{\delta }}}}\).
A.1 The spatial weight matrix \({\varvec{W}}\) is nonstochastic and symmetric, with zero diagonal elements.
A.2 The parameter space \(\mathbf {\Theta }_{{\varvec{\delta }}}\) of \({\varvec{\delta }}= ({\varvec{\beta }}', {\varvec{\theta }}', \sigma ^2)'\) is compact and is the product space of \(\mathbf {\Theta }_{{\varvec{\beta }}}\), \(\mathbf {\Theta }_{{\varvec{\theta }}}\), and \([{\underline{\sigma }}^2, {\bar{\sigma }}^2]\), where \(\mathbf {\Theta }_{{\varvec{\theta }}}\) is a compact set such that the matrix \({\varvec{{\mathcal {I}}}}_N -\lambda {\varvec{W}}\) is nonsingular and the eigenvalues of \(A({\varvec{\theta }})\) are less than one in magnitude, while \(\mathbf {\Theta }_{{\varvec{\beta }}}\) is a compact subset of \({\mathbb {R}}^k\). The true value \({\varvec{\delta }}_0= ({\varvec{\beta }}_0', {\varvec{\theta }}_0', \sigma _0^2)'\) lies in the interior of \(\mathbf {\Theta }_{{\varvec{\delta }}}\).
A.3 The vector of innovations \( {\varvec{V}}_t = (v_{1,t}, \ldots , v_{N,t})'\) \(\sim iid(0, \sigma _0^2 {\varvec{{\mathcal {I}}}}_{N})\) and \(E(|v_{j,t}|^{4+\eta }) <\infty \) for some \(\eta >0\) for all j, t.
A.4 The precision matrix, infinite sum of power of \({\varvec{A}}({\varvec{\theta }}_0)\), and the design matrix are UB. Namely,
-
(i)
\({\varvec{\Sigma }}({\varvec{\theta }})^{-1}={\varvec{B}}({\varvec{\theta }})'({\varvec{\Omega }}({\varvec{\theta }}))^{-1}{\varvec{B}}({\varvec{\theta }})\) and \({\varvec{S}}(\lambda )^{-1}\) are UB, \(\forall {\varvec{\theta }}\in \mathbf {\Theta }\).
-
(ii)
\(\sum _{h=1}^{\infty }abs({\varvec{A}}({\varvec{\theta }}_0)^h)\) is UB.
-
(iii)
The \(N\times k\) design matrix \({\varvec{X}}_{t}\) is nonstochastic with elements UB in N and t.
A.5 \(\lim _{N\rightarrow \infty } \frac{1}{N} {\varvec{X}}'{\varvec{\Sigma }}({\varvec{\theta }})^{-1}{\varvec{X}}=\lim _{N\rightarrow \infty } \frac{1}{N} {\varvec{X}}'{\varvec{B}}({\varvec{\theta }})'({\varvec{\Omega }}({\varvec{\theta }}))^{-1}{\varvec{B}}({\varvec{\theta }}){\varvec{X}}\) is nonsingular, \(\forall {\varvec{\theta }}\in \mathbf {\Theta }\).
A.6 \(\liminf _{N \rightarrow \infty } N^{-1}\sum _{j=1}^{NT} \nabla ^2 f_j({\varvec{\alpha }})\) is nonsingular, where \(f_j({\varvec{\alpha }})\) \(=\) \(-\log (\lambda _j({\varvec{\theta }})\sigma ^{-2}\sigma _0^2)\) \(+\) \(\lambda _j({\varvec{\theta }})\sigma ^{-2}\sigma _0^2\), \({\varvec{\alpha }}= ({\varvec{\theta }}', \sigma ^2)'\), and \(\lambda _j({\varvec{\theta }})\), \(j=1, \ldots , NT\), are the distinct eigenvalues of \({\varvec{\Sigma }}({\varvec{\theta }})^{-1}{\varvec{\Sigma }}({\varvec{\theta }}_0)\) in nonincreasing order.
A.7 \({\varvec{\Sigma }}({\varvec{\theta }})\), \(\partial _{\theta _i} ({\varvec{\Sigma }}({\varvec{\theta }})^{-1})\), \(\partial _{\theta _i\theta _j}^2 ({\varvec{\Sigma }}({\varvec{\theta }})^{-1})\), and \(\partial _{\theta _i\theta _j\theta _k}^3 ({\varvec{\Sigma }}({\varvec{\theta }})^{-1})\) are UB in \({\varvec{\theta }}= (\theta _1, \theta _2, \theta _3)' \in \mathbf {\Theta }\).
A.8 \(\lim _{N\rightarrow \infty }N^{-1}{\varvec{\Omega }}_{N} \) is nonsingular, where
with \({\varvec{m}}_{\lambda }\), \({\varvec{m}}_{\gamma }\), and \({\varvec{m}}_{\rho }\) defined in (S.14) in Supplementary Materials.
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
Ma, T.F., Wang, F., Zhu, J. et al. Scalable Semiparametric Spatio-temporal Regression for Large Data Analysis. JABES 28, 279–298 (2023). https://doi.org/10.1007/s13253-022-00525-y
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13253-022-00525-y