Modeling and forecasting of the high-dimensional time series data with functional data analysis and machine learning approaches

Temperature impacts every part of the world. Meteorological analysis and weather forecasting play a crucial role in sustainable development by helping reduce the damage caused by extreme weather events. A key indicator of climate change is the change in surface temperature. This research focuses on developing and testing a lightweight and innovative weather prediction system that uses local weather stations and advanced functional time series (FTS) techniques to forecast air temperature (AT). The system is built on the latest functional autoregressive model of order one [FAR(1)]. Our results show that the proposed model provides more accurate forecasts than machine learning techniques. Additionally, we demonstrate that our model outperforms several benchmark methods in predicting AT.