J65.2 Visualizations to Facilitate Regression for CAMPS

Thursday, 16 January 2020: 10:45 AM
258C (Boston Convention and Exhibition Center)
Alison L. Reynolds, College of William and Mary, Williamsburg, VA; and E. Schlie, D. E. Rudack, S. R. Olson, and E. Engle

Statistical postprocessing is an important tool used to improve numerical weather forecasts by finding statistical relationships between model data and observed conditions. The Community Atmospheric Model Postprocessing System (CAMPS) is a new, flexible tool currently under development at the Meteorological Development Laboratory (MDL). CAMPS will perform the capabilities of Model Output Statistics-2000, MDL’s internal statistical postprocessing software, in a modern format with plans for community involvement and expansion to other capabilities. The regression program needs accurate information about the variables to build a strong model. By comparing predictions from models such as the Global Forecast System (the predictors) with the actual conditions at the time (the predictands), users can determine which predictors are most correlated with observational data, which allows them to distill the types and number of predictors offered to the regression program. An integral first step in finding this statistical relationship is visualizing the data to identify which variables would be most useful in prediction. One of the new features of CAMPS is the visualization suite, which utilizes Python modules to clean and graph the data. These visualizations, including scatter and correlation matrices, allow users to quickly check which predictors are correlated with the specified predictand, as well as compare these relationships in specific regions or time periods. This aids users in identifying which variables to include in the screening regression process that selects predictors to use in the multiple linear regression model for providing objective statistically postprocessed weather guidance. We present the CAMPS visualizations which use Python modules to provide clear comparisons of model and observational data. These visualizations provide insight into the changing distribution of individual variables across different time periods or geographic locations. In addition to aiding in understanding the data and identifying relationships to facilitate the regression process, they allow users to communicate their findings in a clear and informative manner.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner