Resumen:
Time series forecasting of meteorological variables, such as the hourly air temperature, has multiple benefits for industry, agriculture, and the environment. Due to the high accuracy required for the associated short-term predictions, traditional methods cannot satisfy the requirements and generally ignore spatial dependencies. This paper proposes a deep Graph Convolutional Long Short Term Memory Neural Network (GCN-LSTM) technique to tackle the time series prediction problem in air temperature forecasting. In the proposed methodology, temporal and spatial-based imputation approaches have been employed to recover the weather variables missing values. The proposed approach is validated using real, open weather data from 37 meteorological stations in Spain. Performed analysis indicates that GCN-LSTM showed superior performance when compared with various state-of-the-art Deep Learning based models found in the literature, resulting in a more robust and computationally efficient model for forecasting air temperature in many meteorological stations simultaneously.
Resumen divulgativo:
Este trabajo propone una Red Neuronal Convolucional LSTM de Grafos para predecir la temperatura del aire. El enfoque propuesto se valida utilizando datos reales de 37 estaciones meteorológicas de España, mostrando un mejor en comparación con otros modelos reportados en el estado del arte.
Palabras Clave: Air temperature forecasting; Short-term forecasting; Deep learning; Deep graph convolutional neural networks; Missing values imputation
Índice de impacto JCR y cuartil WoS: 3,900 - Q1 (2023)
Referencia DOI:
https://doi.org/10.1007/s00477-022-02358-0
Publicado en papel: Mayo 2023.
Publicado on-line: Diciembre 2022.
Cita:
L. García-Duarte, J. Cifuentes, G. Marulanda, Short-term spatio-temporal forecasting of air temperatures using deep graph convolutional neural networks. Stochastic Environmental Research and Risk Assessment. Vol. 37, nº. 5, pp. 1649 - 1667, Mayo 2023. [Online: Diciembre 2022]
