A1342
Title: EEMD-ELN regression for multi-scale relationships: Application for rainfall prediction
Authors: Ahmed Alsayed - University of Bergamo (Italy) [presenting]
Abstract: The increase in extreme weather events, including unseasonal rainfall and floods due to climate change, encourages more accurate, timely rainfall forecasts. In general, raw environmental data are often multi-scale, non-stationary, and highly intercorrelated, leading to poor prediction accuracy and reliability. To deal with these gaps, a hybrid approach is proposed using the ensemble empirical mode decomposition (EEMD) combined with elastic net (ELN) penalized regression to overcome these challenges. The proposed method presents several advantages; firstly, the original predictors of rainfall time-series data are decomposed using EEMD into intrinsic mode functions (IMFs) and one residual component. Secondly, this approach detects the relationship between a response variable and the new predictors at different time scales. The performance of the proposed method is proven by using atmospheric variables for the city of Basel in Switzerland over the period from 1/1/2022 to 31/12/2023. The main finding demonstrated that the proposed ELN-EEMD model outperformed the other models. Moreover, the final estimated model shows that rainfall is negatively affected by the high degree of temperature and extreme waves represented by the high-frequency IMF while it is positively affected by wind gusts and humidity at various high degrees.
