Assessing Prediction Error in Traditional Weather Forecasts vs. a Data-Centric Approach

• Activity Number: 121 - SPEED: Data Expo
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Robert Garrett* and Ryan Estep and Nichole Rook and Benjamin William Schweitzer and Thomas Fisher
• Companies: Miami University and Miami University and Miami University and Miami University and Miami University
• Keywords: Predictive Modeling; Weather Forecast

Abstract

Millions of people depend on weather forecasts for everything from storm safety, to aviation, to planning a fishing trip. In light of these applications, we aim to assess the accuracy of weather forecasts in the United States. First, we identify regions of high temporal correlation between the 113 given weather stations. Then, we use these regions to leverage data from the National Climatic Data Center to create a purely data-driven forecast. Finally, we can compare the forecast error in our naive approach to that of the given forecasts to see if there are any temporal or regional biases in forecasting behavior.