Day-ahead Forecasts of Air Temperature

TL;DR

This paper presents an exponential smoothing-based framework for accurate day-ahead air temperature forecasting using historical data, effectively capturing seasonal patterns with a reported RMSE of 4.62 K over three days.

Contribution

It introduces a novel application of exponential smoothing for temperature prediction with demonstrated accuracy on real-world data.

Findings

Achieved RMSE of 4.62 K for 3-day ahead forecasts

Effectively captures seasonal variability in temperature data

Validated on weather station data from Alpena, Michigan

Abstract

Air temperature is an essential factor that directly impacts the weather. Temperature can be counted as an important sign of climatic change, that profoundly impacts our health, development, and urban planning. Therefore, it is vital to design a framework that can accurately predict the temperature values for considerable lead times. In this paper, we propose a technique based on exponential smoothing method to accurately predict temperature using historical values. Our proposed method shows good performance in capturing the seasonal variability of temperature. We report a root mean square error of $4.62$ K for a lead time of $3$ days, using daily averages of air temperature data. Our case study is based on weather stations located in the city of Alpena, Michigan, United States.