294
The Model Spectrum: a new approach to visualizing probabilistic weather forecasts

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 24 January 2011
The Model Spectrum: a new approach to visualizing probabilistic weather forecasts
Washington State Convention Center
Jonathan P. Wolfe, NOAA/NWSFO, Charleston, WV

Forecasters today are challenged with conveying the information acquired from analyzing an ever increasingly large dataset of numerical weather models and their fields into a deterministic forecast. A limitation of deterministic forecasts is that they do not encompass the range of possibilities that exist and instead present a single most likely outcome. However, some of the lower probability but higher impact outcomes can be critical to decision makers. An appealing alternative is to use a probabilistic forecast; however, these are often hard to interpret and can lead to confusion. The Model Spectrum attempts to overcome many of the pitfalls inherent to visualizing probabilistic forecasts.

The Model Spectrum is an interactive web-based probabilistic forecast tool. This tool allows a forecaster to see how their forecast fits into the range of model guidance available as well as climatological normals and records. It uses a box and whisker plot to display the model data. Data is available for a variety of fields and can be accessed for a point or area. The box and whisker plot bins the data into five categories: the maximum and minimum values, the upper and lower quartiles and the median. This arrangement shows the potential outcomes, outliers, and spread of the data. In general, the tighter the distribution of the data around a value, the more confidence one has in forecasting that value. If the spread is large, confidence is low for specifying a particular value but the possibilities are still shown alerting the user to the potential range of values that might occur. The simple, robust presentation provides a smooth interface to access the information needed to make decisions. The amount of data folded into the graph is enormous, exceeding 5000 data points for a typical forecast package issued from the Portland National Weather Service Office, yet the options available enable the user to quickly find what they are looking for. One option allows the user to apply a 30 day bias correction to the model data to correct for systematic numerical model output biases further narrowing the spread of the data.

From both the forecaster and user's standpoint, this tool is extremely useful and is envisioned to become an integral component in communicating uncertainty in forecasts.