e-Archivo. Repositorio Institucional de la Universidad Carlos III de Madrid
Publicación
Artículo científico (JournalArticle)
Modelling time series with multiple seasonalities: an application to hourly NO2 pollution levels
Multiple seasonalities often appear in high-frequency data such as hourly measurements of air pollutants. These multiple seasonalities are due to human activities, with daily and weekly cycles, and climatic conditions, with daily and annual cycles. Multiple seasonal components were in the past often modelled in a deterministic way by trigonometric functions or dummy variables. Since pollution seasonalities vary over time, the deterministic assumption is very strict and a more flexible model is needed. We propose to allow seasonality to slowly change as a seasonal autoregressive integrated moving average (ARIMA) model, where the seasonality is modelled as a stochastic process (combining different seasonal ARIMA models). We apply the proposed methodology to hourly measurements of pollutants in Madrid with daily, weekly, and annual seasonalities, and demonstrate the usefulness of our approach by comparing it with other methodological approaches proposed for this type of data. In an extensive exercise involving 15-year hourly forecasts, we show that the proposed procedure performs very well in predicting hourly pollution over a 24-h horizon and improves on deep neural network procedures. Unlike most papers on pollution forecasting, we incorporate a clear exploratory analysis of Madrid's pollution levels based on a deep understanding of the underlying meteorological phenomena. Additionally, the impact on the predictions of covariates such as wind speed, temperature, and festivities was evaluated., The authors gratefully acknowledge the financial support from the Spanish government through the Ministry of Science and Innovation projects PID2019-109196GB-I00 and PID2022-138114NB-I00. Daniel Peña also thanks FUNCAS for the support provided for his research. A.M. Alonso also acknowledges the funding by Grant PID2023-151013NB-I00, funded by MCIN/AEI/10.13039/501100011033 and by FEDER/UE.
Proyecto:
AEI / Plan Estatal de Investigación Científica y Técnica y de Innovación 2017-2020 PID2019-109196GB-I00
,
AEI / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 PID2022-138114NB-I00
,
AEI / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 PID2023-151013NB-I00