Spatial Statistics

Spatial statistics is the branch of data analysis concerned with observations that are tied to geographical or abstract spatial locations. Central to the field is the characterisation and modelling of spatial dependence, whereby measurements taken at nearby locations tend to be more similar than those taken farther apart. Core tools include variogram and covariance functions, which summarise how variability changes with separation distance, and kriging, a best-linear unbiased interpolation technique that predicts values at unobserved sites from neighbouring data. Beyond point-referenced (geostatistical) methods, spatial statistics encompasses areal data models, spatial point processes and spatial functional data analysis, in which entire curves or surfaces are observed at each location. Extensions to high-dimensional or non-Euclidean settings have brought ideas from machine learning and differential geometry into the field, allowing covariance structures on manifolds of positive-definite matrices or functions in infinite-dimensional (Hilbert) spaces. Applications range from environmental monitoring and natural-resource mapping through epidemiology and public health surveillance to urban planning, climate modelling and medical imaging, reflecting the universal need to quantify, predict and understand processes that vary across space.

Research from Nature Portfolio

A novel physics-driven spatiotemporal regularization method has been proposed for inverse problems in high-dimensional predictive modelling. By coupling partial-differential-equation-based constraints with spatial and temporal smoothing penalties, this framework achieves markedly improved reconstructions of dynamic fields on complex geometries compared to classical Tikhonov or L1 regularisation methods. Another advance introduces ultra-flexible, compactly supported non-stationary covariance kernels that enable exact Gaussian process inference on datasets exceeding five million points. These kernels learn inherent sparsity patterns in the data rather than imposing them by approximation, thus preserving the accuracy and uncertainty quantification of full Gaussian processes while scaling to massive spatial datasets.

Research from all publishers

One study decomposes the functional trace-variogram into amplitude and phase components to improve kriging and clustering of spatially misaligned functional data. By separately modelling magnitude and alignment variability, it achieves more interpretable clusters and greater predictive accuracy for daily spectral curves and simulated surfaces. Another contribution derives parametric rates of convergence for empirical Fréchet means (barycentres) in a broad class of curved spaces, including Alexandrov spaces with curvature bounds and the Wasserstein space of probability measures. Under natural geodesic extendibility conditions, this work guarantees fast, dimension-independent convergence even in infinite dimensions. A third line of research generalises principal component analysis to space forms of constant curvature, yielding optimal low-dimensional Riemannian affine subspaces via a closed-form eigenequation. This Space Form PCA offers nested solutions with superior accuracy and convergence compared to iterative methods on spheres and hyperbolic manifolds.

Spatial Statistics publication trend

The graph below shows the total number of articles in spatial statistics across all publications each year (not limited to Nature Index journals).

Technical terms

Spatial autocorrelation: The tendency for observations at nearby locations to exhibit similar values more than those farther apart.

Semivariogram: A function γ(h) estimating half the expected squared difference between values separated by vector h, summarising spatial variability.

Kriging: An interpolation method that predicts values at unobserved sites by optimally weighting neighbouring observations according to their covariance.

Gaussian process: A distribution over functions such that any finite collection of function values has a joint multivariate normal distribution, used as a flexible prior for spatial fields.

Hilbert space: A complete inner-product space generalising Euclidean space to infinite dimensions, used to represent functional observations.

Trace-variogram: A scalar variogram for functional data defined by the trace of covariance operators at different spatial lags.

Riemannian manifold: A smooth, curved space equipped with an inner product on each tangent space, defining geodesics as shortest paths.

Fréchet mean (barycentre): The point on a manifold minimising expected squared geodesic distance to a distribution of data points.

References

1. Physics-driven Spatiotemporal Regularization for High-dimensional Predictive Modeling: A Novel Approach to Solve the Inverse ECG Problem. Scientific Reports (2016).
2. Exact Gaussian processes for massive datasets via non-stationary sparsity-discovering kernels. Scientific Reports (2023).
3. Variograms for kriging and clustering of spatial functional data with phase variation. Spatial Statistics (2022).
4. Fast convergence of empirical barycenters in Alexandrov spaces and the Wasserstein space. Journal of the European Mathematical Society (2022).
5. Principal Component Analysis in Space Forms. IEEE Transactions on Signal Processing (2024).