Space-time data analysis and assimilation techniques in atmospheric sciences typically consider input from monitoring measurements. The input is often processed in a manner that acknowledges characteristics of the measurements (e.g., underlying patterns, fluctuation features) under conditions of uncertainty, leads to the derivation of secondary information that serves study-oriented goals, and provides input to space-time prediction techniques. We present a novel approach that blends a rigorous space-time prediction model (Bayesian Maximum Entropy, BME) with a cognitively informed visualization of high-dimensional data (spatialization). The combined BME and spatialization approach (BME-S) is used to study monthly-averaged NO 2 and mean annual SO 4 measurements in California over the 15-year period 1988-2002. Using the original scattered measurements of these two pollutants BME generates spatiotemporal predictions on a regular grid across the state. Subsequently, the prediction network undergoes the spatialization transformation into a lower-dimensional geometric representation, aimed at revealing patterns and relationships that exist within the input data. The proposed BME-S provides a powerful spatiotemporal framework to study a variety of air pollution data sources.