Atmospheric Physics

Featured Study: Total Ozone Mapping With Multi-Source Information Integration

Atmospheric studies of ozone often require high-resolution maps of its distribution in space and time. At the same time, modeling and visualization of ozone distributions can involve a considerable level of uncertainty due to (i) the high natural variability of ozone concentrations, and (ii) the different levels of accuracy of the algorithms used to generate data from remote sensing instruments.

We applied space-time analysis techniques from the Knowledge Synthesis framework to integrate efficiently salient physical knowledge bases of varying degrees of uncertainty about atmospheric ozone. Our analysis generated and updated realistic pictures of ozone distribution across space and time.

About the Analysis

We examine a numerical study of total ozone (TO; that is, tropospheric and stratospheric concentrations along the atmospheric column) with primary data sets generated by measuring instruments onboard the Nimbus 7 satellite for different periods in 1988. The animated maps in the following figure display the predicted TO concentration, measured in Dobson units (DU), across the USA in the 5-day period of July 6-10, 1988. In the maps, TO concentrations range between 200 DU (light blue) and 450 DU (dark red). Spatiotemporal kriging was used to produce the maps on the left hand side, which can only process the hard data observations from the satellite, shown as circles on the maps. The maps on the right hand side illustrate the corresponding output produced by Bayesian Maximum Entropy (BME) that also accounts for uncertain and secondary soft information, as explained in the following. The ability to incorporate additional information in BME prediction leads to significantly more detailed and informative maps without artifacts.   

In addition to exact observations from the satellite measurements, TO concentration can be estimated by measurements from the Total Ozone Mapping Spectrometer (TOMS) and Solar Backscatter Ultraviolet (SBUV) remote sensing system. Observations from these systems yield the tropospheric ozone residual; however, major sources of error add significant uncertainty to those additional TO estimates. Moreover, TO concentration can be modeled indirectly by tropopause pressure empirical relationships. In all, tropopause pressure models as a secondary source and the TOMS/SBUV uncertain TO observations are available sources of valuable extra information for spatiotemproal TO prediction.

Nonlinear predictors in the Knowledge Synthesis framework can account rigorously for the above additional informative content and the associated uncertainty. At the same time, conventional interpolation techniques like kriging are unable to use these sources satisfactorily. Our analysis took advantage of all available, useful TO knowledge bases for this study to predict the TO concentration by using the BME-based technique from the Knowledge Synthesis framework. In parallel, we obtained the kriging prediction, too. The output comparison concludes that Knowledge Synthesis techniques produce more informative results than conventional techniques when the input of a study extends beyond hard data observations.

Reference

Christakos G., Kolovos A., Serre M.L., and F. Vukovich. 2004. Total Ozone Mapping by Integrating Data Bases From Remote Sensing Instruments and Empirical Models. IEEE-Trans on Geosci and Remote Sensing, 42(5), pp.991-1008.