Bayesian Spatiotemporal Wombling Authors: Aritra Halder, Didong Li, Sudipto Banerjee Published: 2026 ABSTRACT AbstractStochastic process models for spatiotemporal data underlying random fields find substantial utility in a range of scientific disciplines. Subsequent to predictive inference on the values of the random field (or spatial surface indexed continuously over time) at arbitrary space-time coordinates, scientific interest often turns to gleaning information regarding zones of rapid spatial-temporal change. We develop Bayesian modeling and inference for directional rates of change along a given surface. These surfaces, which demarcate regions of rapid change, are referred to as “wombling” surface boundaries. Existing methods for studying such changes have often been associated with curves and are not easily extendable to surfaces resulting from curves evolving over time. Our current contribution devises a fully model-based inferential framework for analyzing differential behavior in spatiotemporal responses by formalizing the notion of a “wombling” surface boundary using conventional multi-linear vector analytic frameworks and geometry followed by posterior predictive computations using triangulated surface approximations. We illustrate our methodology with comprehensive simulation experiments followed by multiple applications in neuroimaging; environmental and climate science and pollutant analysis in environmental health. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.Keywords: Gaussian processes, Parametric surfaces, Rates of change, Spatiotemporal boundary analysis, Spatiotemporal processes, Wombling Introduction Statistical inference for rates of change in spatial random fields is instrumental in data-driven scientific discoveries. Such inference is pursued in diverse scientific investigations and can be traced back to the origins of evolutionary biology that have resulted in the discovery of lurking inexplicable differences (see, e.g., Mayr 1944, on the “The Wallace, Weber and Lydekker Lines”). More generally, the study of spatial rates of change in the context of scientific experiments is often referred to as “wombling” (Womble 1951; Gleyze, Bacro, and Allard 2001). “Wombling” broadly refers to delineating zones of rapid change on spatial surfaces usually defined in terms of locations with statistically significant directional gradients (Banerjee, Gelfand, and Sirmans 2003), but also by testing paths or boundaries for rapid change on the spatial field (Banerjee and Gelfand 2006; Fitzpatrick et al. 2010; Qu, Bradley, and Niu 2021). The methods employed for “wombling” depend on whether the data features a continuous or discrete spatial indexing. For recent approaches to discrete (areal) data, see Wu and Banerjee (2026), and references therein. Here we will treat space and time as continuous when modeling the random field. We consider point-referenced spatiotemporal data, where variables of interest are mapped at locations within an Euclidean coordinate frame (e.g., using a planar map projection) and indexed over time as a realization of a smoothly varying stochastic process. Probabilistic inference on higher order rates of change relies on the smoothness of the parent process (see, e.g., Adler 1981; Greenwood 1984; Kent 1989; Morris, Mitchell, and Ylvisaker 1993; Mardia et al. 1996; Stein 1999; Banerjee and Gelfand 2003; Guindani and Gelfand 2006). Formal inference on rates of change over spatial random fields enjoys a significant literature in diverse scenarios (see, e.g., Banerjee, Gelfand, and Sirmans 2003; Majumdar et al. 2006; Liang, Banerjee, and Carlin 2009; Gabriel, Allard, and Bacro 2011; Heaton 2014; Terres and Gelfand 2015; Wang and Berger 2016; Terres and Gelfand 2016; Wang, Bhattacharya, and Gelfand 2018; Qu, Bradley, and Niu 2021), including recent investigations into optimal plug-in models for gradients (see, e.g., Liu and Li 2026) and inference on spatial curvature processes (see, e.g., Halder, Banerjee, and Dey 2024). This manuscript develops statistical inference for “wombling” boundaries within a hierarchical spatiotemporal modeling framework (Cressie and Wikle 2015). The literature on rates of change on spatiotemporal random fields is considerably sparser than for spatial random fields. Quick, Banerjee, and Carlin (2013) explored temporal gradients in the context of dynamic Markov random fields, while Quick, Banerjee, and Carlin (2015) developed inference restricted to first-order and mixed spatial-temporal gradients on spatiotemporal random fields. More recently Yu et al. (2023) discussed the utility of temporal rates of change to conduct inference on stationary points in Gaussian process regression models. Wombling can be contrasted and compared with popular machine learning approaches to detecting clusters (see, e.g., Neal 2000; Birant and Kut 2007), image segmentation (see, e.g., Ronneberger, Fischer, and Brox 2015), hot-spot detection (see, e.g, Bivand and Wong 2018; Di Martino, Pedrycz, and Sessa 2018) and level set methods (see, e.g., Osher and Fedkiw 2001; Lie, Lysaker, and Tai 2006; Boledi et al. 2022). Wombling seeks predictive inference over curves (and surfaces) offering fully model-based uncertainty quantification leveraging rates of change. Clustering aims to find homogeneous regions in the reference domain, unable to offer inference over curves separating such regions. Image segmentation aims to locate boundaries in the interpolated surface images. It often requires careful and expensive training (see, e.g., Akkus et al. 2017) and seldom provides uncertainty quantification. Hot-spot detection techniques for point-referenced data are only able to track regions of rapid change. Level set approaches leverage finite differences to detect boundaries but are unable to offer any uncertainty quantification on them. Wombling yields statistical inference over arbitrary space-time coordinates as well as, on geographic features that may or may not be delineating zones of rapid change (e.g., mountains, rivers, administrative boundaries). Within a spatiotemporal setting, curves tracking rapid spatial change evolve over time to produce surfaces posing challenges for wombling which has been typically associated to curves. Furthermore, the extent of change may not be captured by first-order derivatives. For example, the rate of change in neural activity may show substantial increments over time—persistently increasing patterns in event related potentials (ERPs) over time, requiring inference on higher-order gradient processes like the spatiotemporal curvature. Here, we construct these processes by jointly modeling the parent Gaussian process and its derivatives to perform inference on higher-order rates of change for spatiotemporal fields. More pertinently for “wombling”, we seek to evaluate average higher-order rates of change not only at arbitrary space-time coordinates in a posterior predictive fashion, but we also extend inference to integrals over surfaces representing temporally evolving zones that are posited to be harboring rapid change, for example, large pockets of smog traversing a geographic space. In this context, spatiotemporal random fields offer significant challenges over purely spatial settings as we develop inference for integrals over surfaces resulting from evolving boundaries in contrast to boundaries that remain static. The article evolves as follows. Section 2 sets the stage for our methodological framework with an outline of the calculus of spatial-temporal derivative processes and the related distribution theory for legitimate probabilistic inference. Section 3 develops measures of higher-order rates of change and their averages over evolving boundaries (Section 3.3). Section 4 develops a Bayesian inferential framework underscoring computation using surface triangulation to avoid high-dimensional quadrature. Sections 5 and 6 provide simulation experiments and spatiotemporal data analysis to illustrate inference on rates of change at points and over surfaces arising from neuroimaging as ERPs in electroencephalography (EEG) sessions to detect predisposition to alcoholism. Applications featuring analysis of precipitation in Northern California and PM2.5 levels during the Canada wildfires are housed in the supplement. Section 7 concludes with a discussion. Spatiotemporal Derivative Processes Calculus for Spatiotemporal Derivatives Let Z(s,t):(s,t)∈Rd×R+ be a spatial-temporal process, where s∈Rd (usually d=2) denotes a spatial location and t is the temporal reference on R+=[0,∞). We require formal spatiotemporal derivative processes induced by Z(s,t). Such derivatives can be purely spatial, temporal or jointly spatial and temporal underscoring the need for higher-order derivative processes to achieve probabilistic inference. The dimensions for such processes increase with the order of derivatives—the kth-order derivative is a k-fold tensor array—so we use conventional multi-linear algebra in developing higher-order derivatives. Our subsequent developments rely heavily upon the gradient operator applied to Z(s,t) to produce the (d+1)×1 vector of partial derivatives with respect to each coordinate, that is, ∇Z(s,t)≔∂1Z(s,t),∂2Z(s,t),…,∂dZ(s,t),∂tZ(s,t)T=∂sZ(s,t)T,∂tZ(s,t)T, where ∂j denotes the partial derivative with respect to the jth coordinate of s,∂sZ(s,t) is the d×1 spatial gradient and ∂t is the partial derivative with respect to time. Following Banerjee, Gelfand, and Sirmans (2003), one can obtain spatial-temporal derivatives as limits of finite difference processes. We define the shift operator Dh,u,VZ(s,t)=Z(s+hu,t+hv), where u∈Rd is a unit vector representing a displacement in spatial location, v∈[0,1] is a scaled temporal shift and h is a scalar denoting the absolute value of the displacements. Then, ∇≔limh→0h-1Dh,e1,0-1,…,Dh,ed,0-1,Dh,0,1-1T, where e1,…,ed are the d orthonormal vectors in Rd. We will specifically use the tensor product ∇⊗2≔∇⊗∇, which yields the (d+1)2×1 vector ∇⊗(∇Z(s,t)) by applying ∇ to each element of ∇Z(s,t). Higher-orders are defined recursively as ∇⊗r≔∇⊗∇⊗(r-1). The (d+1)r vector ∇⊗rZ(s,t) contains all partial derivatives of order r in both space and time. Section S1 provides explicit constructions for these tensor products. The process Z(s,t) is kth order mean-squared differentiable if the elements of ∇⊗rZ(s,t) are continuously differentiable for r0 for any finite collection of spatiotemporal coordinates si,ti:i=1,…,N}, where 𝚫ij=si-sj,δij=ti-tj and ai,aj∈R,i,j=1,2,…,N are not all zero. The process Z(s,t) is mean-squared continuous at (s,t) if E[Z(s+𝚫,t+δ)-Z(s,t)]2→0 as (𝚫,δ)→0d+1. Under isotropy, we have K=K(𝚫,δ)=K~(‖𝚫‖,|δ|)=K~. Our subsequent developments rely upon joint processes consisting of the parent and derivatives of different orders. To that end, let Lr:R→Rmr be the differential operator of order r, where mr=∑i=1r(d+1)i, which maps Z(s,t) to the derivative processes up to order r. Thus, LrZ(s,t)=∇⊗1Z(s,t)T,…,∇⊗rZ(s,t)TT is mr×1. For LrZ(s,t) to be a valid process we require unique pure and partial derivative processes. In general, LrZ(s,t) includes redundancies. For example, L2Z(s,t) includes ∂2∂s∂t and ∂2∂t∂s, while L4Z(s,t) includes ∂4∂s3∂t which is not required. We use elimination and permutation matrices to transform LrZ(s,t) to 𝓛rZ(s,t):R→Rmr*,mr*=∑i=0ri+di, comprised of unique partial and pure derivatives. Let ∂sr=∂r/∂1i1…∂did with ∑k=1dik=r and ik≥0. If ∂tj1+j2∂sr1+r2-j1-j2K(𝚫,δ) exists, then cov∂tj1∂sr1-j1Z(s,t),∂tj2∂sr2-j2Z(s+𝚫,t+δ)=(-1)r1∂tj1+j2∂sr1+r2-j1-j2K(𝚫,δ) is the covariance between the processes, ∂tj1∂sr1-j1Z(s,t) and ∂tj2∂sr2-j2Z(s+𝚫,t+δ) (see Section S2). The partial derivatives operate element-wise on K(⋅,⋅). Also, 𝓛rZ(s,t) is a valid GP if ∂t2j∂s2(r-j)K0d,0 exists for every j=0,1,…,r. In particular, the case j=r-j=2 requires ∂t4∂s4K0d,0 to exist. A GP assumption on Z(s,t) has immediate consequences. If Z1(s,t)~GP0,K1(⋅,⋅) and Z2(s,t)~GP0,K2(⋅,⋅) independently, then 𝓛r1Z1(s,t) and 𝓛r2Z2(s,t) are independent. For a1,a2∈R,𝓛ra1Z1(s,t)+a2Z2(s,t)=a1𝓛rZ1(s,t)+a2𝓛rZ2(s,t) is a GP. Any sub-vector of 𝓛rZ(s,t) is a GP. This is useful if selected derivative processes are required for inference as is seen in the next paragraph. In addition, inference extends to differential geometric operators like the divergence and the Laplacian (see Section S3). We estimate 𝓛*Z(s,t)=𝓛sZ(s,t)T,∂tZ(s,t),∂t𝓛sZ(s,t)T,∂t2Z(s,t),∂t2𝓛sZ(s,t)TT, where 𝓛sZ(s,t)=∂sZ(s,t)T,∂s~2Z(s,t)TT and ∂s~2 is the operator producing the d2+d vector of unique second order spatial derivatives. The elements of 𝓛*Z(s,t) capture spatial and temporal rates of change in Z(s,t). The process 𝓛sZ(s,t) captures spatial gradients and curvature in Z(s,t) while, ∂t𝓛sZ(s,t) and ∂t2𝓛sZ(s,t) capture temporal gradients and curvature in 𝓛sZ(s,t) respectively. Casting 𝓛* as a spatiotemporal differential operator, 𝓛*:R→Rm*,m*=31+d+d(d+1)2-1.Z(s,t)𝓛*Z(s,t) is a zero-centered stationary process. Its cross-covariance function VZ,𝓛*Z(𝚫,δ)=covZ(s,t)𝓛*Z(s,t),Zs′,t′𝓛*Zs′,t′ is a matrix-valued function from Rd×R to Rm*+1×m*+1 with l,l′th element being the covariance between the lth element of Z(s,t)𝓛*Z(s,t) and l′th element of Zs′,t′𝓛*Zs′,t′. The cross-covariance VZ,𝓛*Z(𝚫,δ) is expressible as a block matrix in terms of K(𝚫,δ) and its derivatives with blocks ∂ti-1∂tj-1K(𝚫,δ)∂ti-1∂tj-1∂sK(𝚫,δ)T∂ti-1∂tj-1∂~s2K(𝚫,δ)T-∂ti-1∂tj-1∂sK(𝚫,δ)-∂ti-1∂tj-1HK11(𝚫,δ)-∂ti-1∂tj-1HK12(𝚫,δ)∂ti-1∂tj-1∂~s2K(𝚫,δ)∂ti-1∂tj-1HK21(𝚫,δ)∂ti-1∂tj-1HK22(𝚫,δ),i,j=1,2,3. where ∂t acts element-wise on matrices. The entries HK11(𝚫,δ)=var∂sZ(s,t),HK12(𝚫,δ)=cov∂sZ(s,t),∂s~2Zs′,t′=HK21(𝚫,δ)T and HK22(𝚫,δ)=var∂s~2Z(s,t) (see Section S4). Under isotropy, we have some simplifications when obtaining the entries of VZ,𝓛*Z(𝚫,δ). We write the selected entries as follows: ∂sK=dK~d‖𝚫‖𝚫‖𝚫‖,∂tK=dK~d|δ|δ|δ| and ∂t2∂~s2K=d2d|δ|2dd‖𝚫‖K~P2,1‖𝚫‖+d2d|δ|2d2d‖𝚫‖2K~-1‖𝚫‖d2d|δ|2dd‖𝚫‖K~P2,2‖𝚫‖2 where P⋅,⋅ are permutation matrices (see Section S5.1 for details). Simplifications owing to isotropy also arise in the expressions for the covariance of directional spatiotemporal derivative processes. For instance, the variance for the directional spatiotemporal curvature process has the following expression: 3‖𝚫‖21-5uT𝚫2‖𝚫‖21-uT𝚫2‖𝚫‖2d4d|δ|4d2d‖𝚫‖2K~-1‖𝚫‖d4d|δ|4dd‖𝚫‖K~+6‖𝚫‖uT𝚫2‖𝚫‖21-uT𝚫2‖𝚫‖2d4d|δ|4d3d‖𝚫‖3K~+uT𝚫‖𝚫‖4d4d|δ|4d4d‖𝚫‖4K~. Similar results can be obtained for all spatiotemporal directional derivatives (see Section S5.2). We observe that 𝓛*Z(s,t) is no longer isotropic. We use spectral theory to characterize covariance functions that admit such processes. The existence of fourth-order spectral moments for space and time ensures admittance of the differential processes within 𝓛*Z(s,t). We adopt non-separable spatiotemporal covariance functions of the form, K(𝚫,δ)=K~(‖𝚫‖,|δ|)=σ2ψ|δ|2d/2φ‖𝚫‖2ψ|δ|2, where φ(x) and ψ(x) are complete monotone, and positive with complete monotone derivative on [0,∞), respectively (see, e.g., Gneiting 2002). Specifically, we use the Matérn kernel, (x)=2ν-1Γ(ν)-1ϕsx1/2νKνϕsx1/2,ϕs>0,ν>0, where Kν(⋅) is the modified Bessel function of the second kind, and ψ(x)=ϕt2xα+1β,a>0,0<α≤1,0≤β≤1. We fix α=β=1, producing the covariance kernel, Existence of the fourth spectral moment requires the fractal parameter, v>2. Although irregularities exist in smoothness for separable covariance functions (see, e.g., Stein 2005), characterizing smoothness for such functions is relatively simpler. Section S6 provides further details. Establishing the validity of 𝓛*Z(s,t), requires no further assumptions. Inference on 𝓛*Zs0,t0 at an arbitrary space-time location s0,t0 is sought through the joint distribution, P𝓩,ℒ*Zs0,t0, where 𝓩=Zs1,t1,…,ZsN,tNT is an N×1 vector of space-time measurements. Let ∑𝓩=covZsi,ti,Zsi′,ti′=K𝚫ii′,δii′ be the N×N covariance matrix associated with our measurements where, 𝚫ii′=si-si′,δii′=ti-ti′,i,i′=1,…,N. Let 𝓚0=cov𝓩,𝓛*Zs0,t0=𝓚10𝚫10,δ10T⋮𝓚N0𝚫N0,δN0T be a N×m* matrix with columns 𝓚i0𝚫i0,δi0=covZsi,ti,𝓛*Zs0,t0 for i=1,…,N. The joint distribution is where V𝓛*Z0d,0 is the cross-covariance matrix of 𝓛*Z(s,t) evaluated at 0d,0. Section 4 develops posterior predictive inference for 𝓛*Zs0,t0. Wombling on Spatiotemporal Surfaces Bayesian spatiotemporal wombling requires inference for integrals over surfaces that enclose regions of rapid change in space-time coordinates. Such surfaces are defined from planar curves tracking rapid change in space and across time. In traditional wombling, which hitherto has been purely spatial, the curves remain fixed over time (e.g., geographic features like rivers, mountain ranges or administrative boundaries) and we seek to measure change along these curves. Alternatively, the regions of interest can vary over time, for example, scalp regions with ERPs over a certain threshold or, tracking smog arising from wildfires (see the supplement, S15.2) in the interest of public health, regions evolve with changing wind patterns. This will be the focus of our discussion. We focus on real valued processes, Z(s,t), over H3=R2×R+⊂R3 using familiar concepts in differential geometry (see, e.g., Spivak 1979) for the ensuing developments. We explore spatiotemporal derivatives and curvatures over smooth and regular surfaces with boundaries. This extends the development of derivatives and curvatures over planar curves required for curvilinear wombling within a purely spatial context (see, e.g., Banerjee and Gelfand 2006; Halder, Banerjee, and Dey 2024) to surfaces in space-time. Wombling Measures for Surfaces In general, we assume that 𝒞 is a surface in H3. Our aim is to statistically estimate the average directional gradient of Z(s,t) along 𝒞 from the realized data. We define such measures of interest on parametric surfaces. Let 𝒞=sx(ω,v),sy(ω,v),t(v):(ω,v)∈𝒟ω×𝒟v be a smooth and regular surface with domain, 𝒟ω×𝒟v⊂H2=R×R+ and let t(v)=v for convenience. The outward pointing normal and unit normal to 𝒞 at (ω,v) are respectively, where ∂sx,sy∂(ω,v)=∂sx(ω,v)∂ω∂sy(ω,v)∂v-∂sy(ω,v)∂ω∂sx(ω,v)∂v. Let n¯s(ω,v)=∂sy(ω,v)∂ω,-∂sx(ω,v)∂ωT and n‾t(ω,v)=∂sx,sy∂(ω,v). If sx,sy are injective functions and all partial derivatives in (3) exist, then the surface 𝒞 is smooth. Also, 𝒞 is regular if n¯s(ω,v)≠0 and the unit normal n(s(ω,v),v) in (3) is well-defined. We denote the surface area of 𝒞, as 𝒜(𝒞) given by If 𝒟v=v0,v1 and 𝒟ω=ω0,ω1, then 𝒜(𝒞)=𝒜ω0,v0ω1,v1=∫v0v1∫ω0ω1‖n¯(s(ω,v),v)‖dωdv and, hence, d𝒜v0,ω0(ω,v)=‖n¯(s(ω,v),v)‖dωdv. We denote, nt(ω,v)=nt and ns(ω,v)=ns for brevity of ensuing expressions. For any point (s(ω,v),v) on 𝒞, let 𝓛nt,ns*Z(s(ω,v),v)=Nst𝓛*Z(s(ω,v),v), be the 8 × 1 directional differential process along the normal direction matrix, Nst=n~s,-1⊕ntn~s⊕nt2n~s, where n~s=1⊕n~s,-1,n~s,-1=nsT⊕n~s⊗2T and n~s⊗2=100001100001ns⊗2. We define wombling measures as the total and average differential over 𝒞, respectively, where the double integral acts element-wise. Therefore, 𝚪 and 𝚪¯ are 8 × 1 vectors. A large value for any entry of 𝚪(𝒞) (or 𝚪¯(𝒞)) indicates that 𝒞 tracks a zone of rapid spatiotemporal change. More generally, 𝓛nt,ns*Z(s,t) can be replaced with, g𝓛*Z(s,t), where g(⋅) is any linear functional. Section S7 offers more details on parameterized surfaces. Specifically, we are interested in surfaces with a boundary (in the topological sense). In the subsequent examples, the boundary for 𝒞 are open or closed parametric planar curves, Cvi=sxω,vi,syω,vi:ω∈ω0,ω1,i=0,1. If, in particular, Cv0=Cv1=Cv=C, for every v∈v0,v1, the curve stays fixed over time so sx and sy are functions of ω only. From (4), 𝒜(𝒞)=v1-v0∫ω0ω1∂sy(ω)∂ω2+∂sx(ω)∂ω212dω=v1-v0ℓ(C), where ℓ(C) is the arc-length of the parametric curve C=sx(ω),sy(ω):ω∈ω0,ω1. Curves on Surfaces Curves of interest on 𝒞 are planar curves. They mark spatial regions of rapid change at time, v=v′. Let Cv′=sxω,v′,syω,v′:ω∈ω0,ω1,ω0≠ω1 denote an open parametric planar curve. The wombling measures in (5) are defined with respect to the normal, nsω,v′, to Cv′. One could, in principle, consider defining wombling measures with respect to the tangential direction to Cv′. The tangent is u¯ω,v′=∂sxω,v′∂ω,∂syω,v′∂ωT, with corresponding unit tangent, uω,v′=u¯ω,v′u¯ω,v′. However, wombling boundaries are, in general, representative of geographic features that delineate zones of rapid change. This suggests that such boundaries should be identified based upon rates of change orthogonal, and not tangential, to the curve. Furthermore, replacing normals with tangents in (5) for closed curves yields the value of 0 in each element of 𝚪(𝒞). For example, if Cv′ is an administrative or natural boundary of a closed region (e.g., zip-code, county, boundary of water body), then the corresponding wombling measures along the tangential direction will be zero and, hence, lack interpretation. On the other hand, the wombling measures defined with normals, as in (5), for closed curves correspond to the notion of spatiotemporal flux associated with the enclosed region (see, e.g., Banerjee and Gelfand 2006, for spatial flux). In particular, if for every v∈v0,v1,Cv is a closed curve, then the surface 𝒞 encloses a topologically compact region W with a piece-wise smooth boundary ∂W. We denote the integral over the boundary by ∫∂W. The total wombling measure is 𝚪(𝒞)=∫∂W𝓛nt,ns*Z(s,t)d𝒜. For such volumes the spatiotemporal flux and rate of change in spatiotemporal flux represent rates of change in the spatiotemporal process enclosed by W. If Z(s,t) is three times differentiable, then the closed surface integral can be expressed as a volume integral over W without requiring explicit parametric representation. Let F1(s,t)=∂sZ(s,t)∂tZ(s,t) and F2(s,t)=F21⋮F26=∇⊗2Z(s,t). The spatiotemporal flux and the rate of change in spatiotemporal flux satisfy the identities, ∫𝒞=∂Wn(s,t)TF1(s,t)d𝒜=∭WdivF1dW and ∫𝒞=∂Wn(s,t)⊗2TF2(s,t)d𝒜=∭WdivnsxF21F22F23+divnsyF22F24F25+divntF23F25F26dW, where n(s,t)=nsxnsynt and dW=dsxdsydt is the volume element. The integrand on the right consists of elements from ∇⊗3Z(s,t). The identities are obtained using the Divergence Theorem due to Gauss (see, e.g., Spivak 1979, vol. 4, chap. 7, p. 192). The first identity concerning spatiotemporal flux requires Z(s,t) to be twice differentiable (see Section S8). Turning to practical computation of the spatiotemporal flux (or its rate of change) ∭W can be evaluated using a customary Riemann approximation (with cubes inside the region W), which, while simple in concept, can be demanding on resources and requires smoother process specifications. On the other hand, evaluating ∫𝒞=∂W is conceptually harder and benefits from the triangulation procedure discussed in Section 4.2. Finally, we remark that one could possibly use Stokes’ theorem on surfaces 𝒞=sx(ω(v)),sy(ω(v)),v:v∈𝒟v. This is a parameterization from R+→R3. Wombling measures would then be line integrals defined on curves bounding a surface ∮Cn(s,t)TF1(s,t)dℓ=∬𝒞 curl F1d𝒜. These would then be defined on curves and be incongruous with our aim of associating them with surfaces. We do not pursue such developments here. Spatiotemporal Wombling Our scientific applications include, for example, temporally evolving wombling measures for neural activity over regions of the human scalp. We consider temporally indexed curves, say Ct, which generate surfaces. For example, a line segment is described as Ct=st+ωut:ω∈[0,1] on t∈t0,t1. We write C0=s0+ωu:ω∈[0,1] and C1=s0+v+ω(w-v):ω∈[0,1]} for t=t0 and t=t1, respectively, where st0=s0,ut0=u,v=st1-s0 is the displacement vector for s0 and w=ut1+v is the corresponding displacement of ut1. Hence, Ct0 evolves continuously to Ct1 generating planes from two sets of non-collinear points, s0t0,s0+ut0,s0+vt1 and s0+ut0,s0+vt1,s0+wt1. The resulting surface, 𝒞, is a union of two triangular planes given by 𝒞0=s0t0+ωu0+vv1:(ω,v)∈T and 𝒞1=s0+wt1+ωv-w0+vu-w-1:(ω,v)∈T, where T={(ω,v):ω,v∈[0,1],ω+v≤1} is a triangular region in R2 (left panel of Figure 1). The unit normals (independent of (ω,v)) are ni=n¯in¯i for i=0,1, where n¯0=v1×u0=n¯s,0n‾t,0 and n¯1=v-w0×w-u1=n‾s,1n‾t,1, respectively, and × is the vector product. The areas are 𝒜𝒞i=∬Tn¯idωdv=n¯i2, and wombling measures in (5) are 𝚪𝒞i=n¯i∬Tℒnt,i,ns,i*Z(s(ω,v),v)dωdv, for i=0,1. The total and average wombling measures over the plane 𝒞=𝒞0∪𝒞1 with area 𝒜(𝒞)=𝒜𝒞0+𝒜𝒞1 are 𝚪(𝒞)=𝚪𝒞0+𝚪𝒞1 and 𝚪¯(𝒞)=𝚪(𝒞)𝒜(𝒞), respectively. Illustrative examples for wombling: (left) parametric planes (right) parametric circles. Normals to the surface are indicated using red arrows and × denotes the vector product. The right panel of Figure 1 displays another example: a surface generated from arcs of parametric semi-circles, 𝒞=(rvcosω,rvsinω,v):ω∈[0,π],v∈t0,t0+1,r>0. The radii of the arcs increase linearly with time. The normal to 𝒞 is n¯(s(ω,v),v)=rvcosω,rvsinω,-r2vT. The unit normal is n(s(ω,v),v)=11+r2(cosω,sinω,-r)T and the area is (𝒞)=∫t0t0+1∫0π‖n¯(s(ω,v),v)‖dωdv=t0+12πr1+r2. The average wombling measure is 𝚪¯(𝒞)=1t0+12π∫t0t0+1∫0π𝓛nt(v),ns(ω,v)*Z(s(ω,v),v)vdωdv. Section S14 discusses static wombling, where the curve stays fixed over time. Distribution Theory for Wombling Measures We extend the distribution theory in Section 2 to derive the joint distribution for the N×1 process 𝓩 and the wombling measures 𝚪(𝒞) in (5) for conducting predictive inference. Suppose 𝒞 is generated over intervals, 𝒟v0=0,v0 and 𝒟ω0=0,ω0. For any u*∈𝒟v0 and ω*∈𝒟ω0, let 𝒞* denote the surface restricted to 𝒟*=𝒟v*×𝒟ω*=0,v*×0,ω*. The wombling measures in (5) are 𝚪𝒞*=∬𝒟*𝓛nt,ns*Z(s(ω,v),v)‖n¯(s(ω,v),v)‖dωdv. The directional differential processes, 𝓛nt,ns*Z(s,t), are valid zero-mean Gaussian processes (GPs). Hence, 𝚪𝒞*~𝒩808,K𝚪𝒞*,𝒞*, where 𝚫ω1,ω2,v1,v2=sω2,v2-sω1,v1 and v1,v2=v2-v1. The repeated integral acts element-wise on matrices. 𝚪𝒞* is a dependent GP on 𝒟v0×𝒟ω0—for any two points, (ω1*,v1*) and (ω2*,v2*), the dependence is specified through, 𝚪𝒞1*𝚪𝒞2*~𝒩16016,K𝚪𝒞1*,𝒞1*K𝚪𝒞1*,𝒞2*K𝚪𝒞2*,𝒞1*K𝚪𝒞2*,𝒞2*, where 𝚪𝒞i* is the wombling measure computed over 0,ωi*×0,vi* and each K𝚪𝒞i*,𝒞j* is 8 × 8 obtained from (6) for i,j=1,2.𝚪𝒞* is mean-square continuous. Although Z(s,t) is stationary, 𝚪𝒞* need not be. Let G𝒞*=cov𝓩,𝚪𝒞*=γ1𝒞*T⋮γN𝒞*T as the N×8 matrix, where γi𝒞*=covZsi,ti,𝚪𝒞*,𝚫i(ω,v)=si-s(ω,v),δi(v)=ti-v, and the integrals act element-wise on the vector. The desired distribution P𝓩,𝚪𝒞* is Bayesian Inference and Computation Spatiotemporal Hierarchical Model Let 𝓨=ys1,t1,…,ysN,tNT be the N×1 vector of observations over spatiotemporal coordinates s1,t1,…,sN,tN. We construct the spatiotemporal hierarchical model, where μsi,ti=xsi,tiTβ, each xsi,ti is a p×1 vector of spatiotemporal covariates, Zsi,tis are realizations of the process Z(s,t)~GP(0,K(⋅,⋅,θ)) and ϵsi,ti~iidN0,τ2 accounts for white noise. The process parameters are Θ=θ,τ2,β, where θ=σ2,ϕs,ϕt. With customary prior specifications we obtain the posterior, where πθ,τ2=Uϕs∣aϕs,bϕs×Uϕt∣aϕt,bϕt×IGσ2∣aσ,bσ×IGτ2∣aτ,bτ,∑𝓩=σ2Rϕs,ϕt is the N×N spatiotemporal covariance matrix constructed from the Matérn covariance kernel in (1), IG denotes the inverse-gamma distribution, 𝒩 denotes univariate Gaussian and 𝒩N denotes N-variate Gaussian distribution. We compute P(Θ∣𝓨) by marginalizing out 𝓩, thereby producing a “collapsed”, more efficient alternative. In a more general setting, if learning smoothness of the process from available data is of interest, then v~Uav,bv. Depending on the posterior estimate for ν, the spatiotemporal differential processes are sampled using closed-form Matérn kernels with half-integer values for ν as follows, ν=1+12 or, 2+12. For instance, if 2≤ν≤3, then a Matérn with ν=52 is used to generate posterior samples for the differential processes. This features in our implementation. Alternatively, models can be fit simultaneously with v=12,32 and 52 and a model selection criteria can be used to select v for the best fit. The posterior distribution in (10) is sampled using a combination of Gibbs and Metropolis-Hastings (MH) steps. We first sample θ and τ2 using MH steps, followed by Gibbs samples for β and 𝓩, one-for-one corresponding to samples of θ and τ2. Predictive inference on spatiotemporal differential processes and wombling measures for the latent surface is achieved by sampling from P𝓛*Zs0,t0∣𝓨 and P(𝚪(𝒞)∣𝓨), respectively, where 𝒞 is a surface of interest. The posterior predictive distribution of the derivative processes is 𝓛*Zs0,t0∣𝓨=∫P𝓛*Zs0,t0∣𝓩,ΘP(𝓩∣𝓨,Θ)P(Θ∣𝓨)dΘ. Posterior inference for any sub-vector, say, ∂t2∂s~2Zs0,t0 is obtained as ∂t2∂s~2Zs0,t0∣𝓨=∫⋯∫P∂t2∂~s2Zs0,t0∣𝓛~*Zs0,t0,𝓩,ΘP𝓛~*Zs0,t0∣𝓩,ΘP(𝓩∣𝓨,Θ)P(Θ∣𝒴)dΘd𝓛~*Z, where 𝓛~*Z is the vector of spatiotemporal derivative processes excluding ∂t2∂~s2Zs0,t0. Posterior sampling from the above conditionals is done one-for-one using (2). One instance of 𝓛*Zs0,t0 is sampled for each posterior sample of θ obtained from the posterior, P(θ∣𝓩). The conditional predictive distribution of the spatiotemporal differential process, 𝓛*Zs0,t0∣𝒵,θ~𝒩m-1μ𝓛,∑𝓛, where μ𝓛=-𝓚0∑𝓩-1𝓩 and ∑𝓛=V𝓛*Z0d,0-𝓚0∑𝓩-1𝓚0T. Similar arguments follow for any element or sub-vector of 𝓛*Zs0,t0. Posterior samples for the wombling measures are obtained similarly. The posterior, P(𝚪(𝒞)∣𝓨)=∫P(𝚪(𝒞)∣𝓩,Θ)P(𝓩∣𝓨,Θ)P(Θ∣𝓨)dΘd𝓩 is used to sample wombling measures. For each posterior sample of 𝓩 and θ, we draw 𝚪(𝒞)∣𝓩,θ~𝒩8-G(𝒞)T∑𝓩-1𝓩,K𝚪(𝒞,𝒞)-G(𝒞)T∑𝓩-1G(𝒞), where, K𝚪(𝒞,𝒞),G(𝒞) are computed using (6) and (7) using an appropriate parameterization (see Section 4.2). Posterior samples for 𝓩,𝓛*Z and 𝚪(𝒞) are generated together to avoid repeated computation of ∑𝓩-1. We leverage parallel computation to improve algorithmic complexity. Triangulation of Surfaces For a surface (compact and connected in the topological sense) to be identified as a wombling boundary, posterior sampling of wombling measures requires a parametric form. We choose to perform spatiotemporal wombling using simple parametric regions, specifically triangular planes as shown in Figure 1 (left). The spatiotemporal wombling measures in (5) are computed on each region followed by an aggregation to produce the wombling measure for the surface. The surface 𝒞 is triangulated using a partition. We partition 𝒟ω=ω0,ω1 using ω0=ω1′<⋯<ωnω′=ω1 and 𝒟v=tj,tj+1 using tj=v0=v1′<⋯