2026 Distinguished Lecturer: douglas nychka

is data science a science?: spatial prediction of air quality over italy. 

The use of data science to solve problems through machine learning algorithms, and more recently deep learning, has had undisputed success. But is this an application of science? By this we mean methods based on a foundational theory with known properties. In particular, theoretical grounding provides guidance on how a method generalizes beyond the data on which it was trained.

This talk explores this question through the problem of inferring daily NO₂ concentrations at a fine spatial scale for Italy. A major success of statistical science—particularly in environmental applications—is the richness of probability models for describing spatial dependence, incorporating additional variables, and quantifying uncertainty at unobserved locations. Although attractive, these models are computationally intensive and difficult to implement for large datasets.

Computational technology developed for AI, however, can be repurposed for statistical analysis. In this way, we blend the efficiency of machine learning algorithms with statistical models that rigorously quantify uncertainty. For air quality, a key feature is the spatially varying correlations among a pollutant at different locations. We train a deep learning model (LatticeVision) to estimate these spatial correlations by simulating Gaussian fields with diverse correlation structures. This tool is then applied to a chemical transport model for Italy, and the resulting correlation function provides the basis for Gaussian process spatial prediction using observations from a monitoring network.

Is this an application of science? Although rooted in state-of-the-art statistics, our analysis also has empirical elements that suggest new research directions.

Douglas Nychka is a data scientist whose areas of research include the theory, computation, and application of curve and surface fitting with a focus on geophysical and environmental applications. Before coming to Mines he directed the Applied Mathematics Institute at the National Center for Atmospheric Research (1997-2018). His current  research is on the  computation of spatial statistics methods for large data sets, leveraging the efficiency of deep learning for data analysis, and the migration of these methods into easy-to-use R packages. He is a Fellow of the American Statistical Association and the Institute for Mathematical Statistics.