Verification, Diagnosis, and Adjustment of Current Temperature Forecasting System in the United States

• Activity Number: 199 - SPEED: Data Expo
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 11:35 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Han-Yueh Lee* and Hsiao-Ting Lin
• Companies: National Tsing Hua University and National Tsing Hua University
• Keywords: quality assessment; model ensemble; temporal-spatial modeling

Abstract

Since the climate changes frequently in recent years and the extreme weather conditions may cause huge loss in different fields, continuously monitoring and improving the quality of current forecast system is definitely crucial.Therefore, this study retrieves the temperature forecast data from National Weather Service website, and try to meet the following objectives: (1)To assess the quality of the daily temperature forecast in 113 cities in the United States from 2014 to 2017. (2)To introduce an adjusted forecast system which may improve the current forecast accuracy. In order to achieve the first goal, we will begin with using some standard verification methods to assess the overall accuracy, and then adopting several diagnostic verification methods to identify some possible reasons of the inaccuracy of the present forecast system. Moreover, we aim to adopt the results of the first goal to achieve the second research goal. We will firstly construct several locally models under different conditions, and then use model ensemble idea to combine our locally models with the current forecast model and get a new aggregation model. By doing so, we could reach higher forecast accuracy.