Noel A. C. Cressie is an Australian-American statistician whose career has been devoted to developing the mathematical and computational foundations of spatial and spatio-temporal statistics. His textbook Statistics for Spatial Data defined the field for a generation, and his subsequent work with Christopher Wikle on hierarchical statistical models provided a powerful Bayesian framework for combining physical process models with observational data. His contributions have had profound impact on environmental science, remote sensing, climate modeling, and ecology.
Life and Career
Born in Adelaide, Australia. Studies mathematics at the University of Adelaide before pursuing graduate work in statistics.
Earns his Ph.D. from Princeton University, studying under Geoffrey Watson and focusing on spatial statistical methods.
Publishes Statistics for Spatial Data (revised edition), which becomes the definitive reference for geostatistics, lattice data, and spatial point processes.
Co-authors Statistics for Spatio-Temporal Data with Christopher Wikle, extending hierarchical Bayesian methods to dynamic spatial processes.
Joins the University of Wollongong as Distinguished Professor, continuing research on environmental data analysis and remote sensing.
Spatial Statistics and Hierarchical Models
Cressie's foundational insight was that spatial data require a fundamentally different statistical treatment from independent observations. Measurements at nearby locations are correlated, and this correlation contains valuable information about the underlying process. His taxonomy of spatial data into geostatistical (continuous spatial index), lattice (discrete spatial index), and point pattern (random locations) categories provided an organizing framework that remains standard.
Process model: Y(s) ~ [Y | θ_p]
Parameter model: (θ_d, θ_p) ~ [θ]
Posterior Inference [Y, θ | Z] ∝ [Z | Y, θ_d] · [Y | θ_p] · [θ]
The hierarchical framework separates the data model (how observations relate to the true process), the process model (the underlying spatial or spatio-temporal dynamics), and the parameter model (prior distributions on unknown quantities). Bayesian inference via MCMC then combines all sources of uncertainty into a coherent posterior distribution. This approach is especially powerful for environmental problems where physical models provide informative process-level structure but are subject to both measurement error and model misspecification.
Classical kriging, the optimal linear predictor in geostatistics, produces spatial predictions but treats covariance parameters as known. Cressie and colleagues showed how embedding kriging within a Bayesian hierarchical model properly accounts for parameter uncertainty, producing wider but more honest prediction intervals. The Bayesian approach also naturally handles non-Gaussian data, missing observations, and change-of-support problems that challenge classical geostatistics.
Environmental and Climate Applications
Cressie has applied his methods to some of the most important environmental datasets of the modern era, including satellite-based measurements of atmospheric CO2 from NASA's Orbiting Carbon Observatory, global temperature reconstructions, and ecological population dynamics. His work on fixed-rank kriging addressed the computational challenges of applying spatial methods to the massive datasets produced by remote sensing instruments, enabling Bayesian spatial analysis at scales previously thought intractable.
Legacy
Cressie's influence spans both the theoretical development and the practical application of spatial statistics. By showing how Bayesian hierarchical models could incorporate scientific knowledge about physical processes while honestly representing uncertainty, he provided environmental scientists with tools that respect both the complexity of natural systems and the limitations of observational data. His framework has become the standard approach for integrating multiple data sources in environmental monitoring and prediction.
"A statistical model for spatial data should reflect the scientific understanding of the process that generated the data. The hierarchical framework makes this possible." — Noel Cressie