Developing Enhanced Diagnostic Output for the UFS: Increasing Information Extraction with Minimal Expense

Revealing what has occurred in any numerical weather prediction model is challenging due to the ratio of time steps to output times, and the cost of data storage. How model data is output (and how frequent) determines whether or not problems (bugs or extremes) can be detected. Model developers and forecasters can both gain insight from information about extreme values: forecasters can use them as a sign of intensity while developers may use extremes as a sign of numerical instability or other anomaly. The goal of this project is to increase the information content in model output without adding more I/O and data storage for operational modeling centers.

We have proposed to find novel ways to aggregate key pieces of information (e.g., from a single variable and from all time steps) to yield more data in a compact form. Designing a schema that encapsulates what the user wants (e.g., an accumulation, max/min, number of threshold exceedances, etc.), and storing multiple key pieces of data in a single, decodable variable that won’t be corrupted by compression (i.e., lossless but not lossy) is the primary goal. This presentation will give a few examples of encapsulating key pieces of information from precipitation time series and their relevance to both model developers and forecasters.

The goals of this proposed work are to increase information content coming out of the models and reduce the number of output parameters. By providing better insight into variable behavior we hope to empower community members to achieve a better understanding of model behavior during development and better monitoring of extreme events in the forecast process.