Quantifying Confidence and Accuracy in Weather Forecasts and Numerical Prediction

Background

The results of accurate atmospheric models can prepare the general populace with vital information regarding weather events like severe thunderstorm outbreaks, flooding and blizzards. However, the large amount of physical processes controlling the weather requires models to continually integrate real-time observations to keep them valid.

Also when reviewing a forecast scenario it is important to understand the underlying risk entailed in forecasting.  Some scenarios may be considered by a professional to be "straight forward" and "low risk" while other scenarios may be complex, involve a rapidly developing weather system or otherwise be a challenging or "high risk" forecast.  

Here we use a conceptual model to show how forecast risk and potential accuracy can be graphically represented (below).  The plot below shows forecast risk on the X-axis based on viewable meteorological features that may impose complexity and risk on a given forecast (we call this metric the "Confidence Index" or "CI").  The resulting actual forecast error from the forecast is on the Y-axis.  The resulting scatterplot created from a year of forecasts creates a triangle or "wedge" shape where low-risk forecasts provide typically a low error while higher risk forecasts produce both low-error and higher-error forecasts, the upper limit of said error can be seen to increase with risk.

Project Details

We will work with a student on validating a forecast model used in the Western South Dakota for real-time forecasts and decision making and also associate forecast error to forecast risk using the Confidence Index. 

The student will also tour the Rapid City, National Weather Service where measurements are collected and forecasts are produced and also visit a remote South Dakota Mesonet station to understand what is measured and how the data is incorporated into a weather forecast model as well as the larger computational- and human-driven weather forecast process.

Skills Development

• Programming
• Statistics 
• Numerical Analysis
• Numerical Weather Prediction 
• Weather Research and Forecast Model 
• Unix environment

Research Duties

Students will be using scripts to determine the accuracy of numerical weather forecasts and also compare these forecast to "scores" of Confidence to asses if regional forecasts provide similar CI-Error pattens as regional forecasts.

Impact of Research

We hope that this project will improve regional numerical weather prediction and assessing the risk of forecasts which can then be used to assign workflow of the larger forecast process (both the human meteorologists and their numerical tools).  This project is also part of a larger project to assess the role of multiple regional forecasts in creating a multi-university/organization forecast ensemble.

Research Team

You will be working with both Dr. Capehart. 

Dr. Capehart has expertise in numerical weather prediction, regional climate modeling and hydrologic modeling, and will serve as the lead on the project.

For more details on this project, contact the lead investigator:

Dr. Bill Capehart

Atmospheric and Environmental Sciences

South Dakota School of Mines and Technology
Rapid City, SD  57702

Ph: +1 (605) 394-1994

Email:  William.Capehart@sdsmt.edu