Flexible Infrastructure in the GSL Phenomena-Oriented Model Verification System

Verification is essential to the development and operational transition of Global Systems Laboratory (GSL) numerical weather prediction modeling systems, historically with the Rapid Refresh (RAP) and the High Resolution Rapid Refresh (HRRR), and presently with the experimental Rapid Refresh Forecast System (RRFS). For the last fifteen years, GSL has developed and maintained a verification system designed around a MySQL database with preprocessed aggregated sums, which were derived from model-obs pairs ingested from realtime and retrospective sources. This system allows GSL’s verification web application, the Model Analysis Tool Suite (MATS), to quickly derive statistical scores from the database in an interactive manner, providing timely feedback on design decisions to developers. While this approach has been successful in numerous R2O projects, the evolution of model development is facilitating the need for more in-depth analysis than this system can currently provide. As errors have decreased with model advancements, the underlying processes responsible for them have become increasingly difficult to determine with the predetermined statistical aggregation method. Recognizing the need for greater user control over statistical generation for model development, the GSL model verification team is developing a phenomena-oriented verification system that will enable dynamic modeler-prescribed regimes within the MATS user interface, covering geographical domains, atmospheric conditions, and observation networks. To that end, the limitations of the rigid schema in the legacy MySQL database proved inadequate to this challenge and necessitated a change in underlying infrastructure to the more flexible NoSQL document-style database technology, Couchbase. During this transition, a new data model was developed to optimize data pre-processing, allowing for a wider array of variables to be handled by the verification system, and middleware was added to improve the efficiency of data retrieval and statistical calculations from the database.

In this presentation, we will highlight the ongoing development of the phenomena-oriented model verification system at GSL, specifically the Couchbase NoSQL document database, data model, and data processing middleware. We will discuss the motivation, requirements, and methodologies utilized and future plans and uses for the system within numerical model development at GSL.