| G. Vonitsanos, A. Kanavos, Ph. Mylonas |
| Spatio-Temporal Hybrid Deep Learning for Grid-Based Meteorological Time Series Forecasting |
| 22nd International Conference on Web Information Systems and Technologies (WEBIST 2026), Angers, France, 27-29 October 2026 |
|
ABSTRACT
|
| Weather forecasting is essential for several domains, including agriculture, transportation, energy management, and environmental monitoring. Recent advances in deep learning have enabled the development of data-driven forecasting models capable of capturing complex temporal dependencies, seasonal patterns, and nonlinear relationships in meteorological time series. This study investigates Long Short-Term Memory (LSTM) and Transformer-based architectures for multi-horizon average temperature forecasting using daily gridded observations from the JRC MARS Meteorological Database. A persistence model is also included as an informed reference model, allowing the deep learning approaches to be evaluated against the long-term annual behavior of temperature. The proposed framework incorporates multiple meteorological variables, including temperature-related features, precipitation, vapor pressure, wind speed, reference evapotranspiration, and solar radiation, to predict future average temperature values across 1-day, 3-day, and 7-day forecasting horizons. Forecasting performance is evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), while SHapley Additive exPlanations (SHAP) are employed to enhance interpretability by identifying the relative contribution of each meteorological variable to the model predictions.
|
| 27 October , 2026 |
| G. Vonitsanos, A. Kanavos, Ph. Mylonas, "Spatio-Temporal Hybrid Deep Learning for Grid-Based Meteorological Time Series Forecasting", 22nd International Conference on Web Information Systems and Technologies (WEBIST 2026), Angers, France, 27-29 October 2026 |
[  PDF] [
BibTex] [
Print] [
 Back] |
