Analysis of High Temperature Forecast Accuracy of Consumer Weather Forecasts from 2005-2016 (331)

Eric Floehr, ForecastWatch, Dublin, OH

Bruce Rose, ForecastWatch

 

Abstract:

In 2005, Eric Floehr began collecting weather forecasts across the United States from several national weather forecast providers. What started as a personal curiosity about the accuracy of weather forecasts, turned into a full-time business. After 12 years of collection over 800 locations within the United States, ForecastWatch has accuracy data that show clear trends in weather forecast accuracy and how weather forecasts are becoming more accurate.This paper, written by Eric Floehr and Bruce Rose, analyzes forecast accuracy and trends over a twelve-year span of time from ten different forecast providers. It presents an in-depth analysis of the accuracy of weather forecasts, focusing on high temperature predictions. The goal of the analysis was to assess the forecast accuracy, and to determine the degree to which forecasts are improving with time. The data confirm what experts already know: weather forecasts are highly accurate, and are improving dramatically.High temperature forecasts from one- to ten-days-out were compared with observed high temperatures. Pairs of forecast and observed data were assessed via root mean square error (RMSE). Pairs were also categorized by absolute error as &ldquoPerfect&rdquo (&lt1&degF error) &ldquoGood&rdquo (&lt=3&degF error) and &ldquoBust&rdquo (&gt=10&degF error). Analyses were performed to assess accuracy over time, bias, and differences in accuracy between near- and long-term forecasts.ForecastWatch data show that high temperature forecasts are generally extremely accurate, and continue to measurably improve. There was considerable improvement in the five-day-out forecast accuracy and there are several reasons why. The weather models are getting better, techniques are improving, and the human forecast is improving. Meteorologists are now more skilled, well-trained, or have more experience than ever before. Moreover, the forecasts are more reliable and are updated more frequently due to the continued decrease in technology costs, which allows for more computational power per dollar spent.