Activity: Talk or presentation › Oral presentation
Climate change is one of the key factors impacting various natural and physical processes. These processes are often studied as time series, i.e. in the form of dynamic sequences of observations recorded in chronology. This talk presents the performance of a novel ‘climatic module’ developed for associating climatic trends within a hybrid system of stochastic modelling approaches (referred to as STL_HMM_GP) developed for generating synthetic time series [1, 2]. The coupled system (that integrates ‘climatic module’ and ‘STL_HMM_GP’) is aimed at generating climate impacted synthetic time series. The performance of the unified system is demonstrated for two applications using high-resolution energy demand and long-term streamflow sequences.
The original STL_HMM_GP system integrates three conventional statistical techniques a) time-series decomposition technique (STL: A Seasonal –Trend Decomposition procedure based on Loess processes ); b) a hidden Markov model (HMM) and c) a generalised Pareto (GP) distribution for simulating highly stochastic time series. The application of the STL approach decomposes complex time series into trend, seasonal and random components. The HMM model is applied to stationary random components for simulating the patterns and statistical dynamics of underpinning time series and an extreme value distribution (e.g. Generalised Pareto) is applied for attaining high efficiencies in simulating extremes.
The ‘Climatic module’ exploits the potential of Principal Component Regression (PCR) and Partial Least Square regression (PLSR) approaches for associating the trend components of time series of climatic variables with the corresponding trend component of the time series of process (e.g. energy demand and streamflow). The climate associated trend component of the process time series is recombined with seasonal and HMM simulated random components to project the impact of climate change.
14 Sep 2022
RSS International Conference 2022: FOR ALL STATISTICIANS AND DATA SCIENTISTS.