Visualization for Improving Algorithmic Measures of Morphometry or Budget Events

Statistical data summaries in atmosphere and ocean dynamics flow from algorithms rooted either in the governing equations (budget terms), or in descriptors of flow structures (morphometrics). In either case, these algorithms are usually designed to screen or filter data for evidence about a pre-existing conceptual narrative or argument. Algorithm design often begins from such preconceptions, but fluids (and the big data sets describing them) invariably contain complex situations with unforeseen projections onto that algorithmic filter. Such situations are "interesting", whether as evidence, counter-evidence, or a narrative-orthogonal indication that the algorithm isn't measuring what it was designed to measure. "Drilling down" to rich visualizations of the full-detail data in these "interesting" regions, in the context of the algorithmic measure or filtered view, can lend either punch to the argument, or depth of understanding to possible refinements of the underlying concepts and algorithms in a next iteration of the work.

To facilitate such visualization, the DRILSDOWN project provides a science-driven suite of tools, many of them centered around the Interactive Data Viewer (IDV). The IDV is an advanced, free, all-platform, click-to-install Java application which we have linked to the Python ecosystem by writing extensions to Jupyter-ipython notebooks and the popular xarray package for labeled gridded data. This talk will illustrate the DRILSDOWN toolkit with leading use cases involving eddy momentum flux in the atmosphere (based on a budget measure), and Loop Current separation events in the ocean (based on a flow configuration event). In both cases, visualization-driven insights in thoughtfully selected "interesting" cases advance the science, by spurring (1) refinements of the algorithmic measures of "interestingness", (2) reinterpretation of the phenomenology underlying budget terms, and/or (3) illustrative synthetic-data partial overwrites or short-term simulation-launching exercises, with the end result of better communicating and enriching the scientific discourse driving the work.