Spatial and Temporal Trends in Weather Forecasting and Improving Predictions with ARIMA Modeling

• Activity Number: 254 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Manasi Sheth* and Mahalaxmi Gundreddy and Vivek Shah and Pritam Barlota and Eric Suess
• Companies: California State University and California State University East Bay and Applied Materials, Inc. and California State University East Bay and CSU East Bay
• Keywords: weather forecasting; time series prediction; ARIMA

Abstract

The main purpose of the paper is to analyze and improve the accuracy of the weather forecasts made by National Weather Service for the years 2014-2017 for different cities in the United States of America. The accuracy of the forecasts has been evaluated from different perspectives such as spatial, temporal and the time gap between prediction and the actual date. Since the temperatures were found to be autocorrelated, time series based ARIMA models were implemented to improve the maximum and minimum temperature forecasts. Although, there was significant improvement in predictions of minimum temperature, improvement was observed for maximum temperature predictions only for few cities. To improve the maximum temperature predictions for all the cities, another ARIMA model was implemented to predict the residuals from the forecasted temperature. By predicting the residuals, all maximum temperature forecasts as well as minimum temperature forecasts were improved.