Stochastic simulation on reproducing long-term memory of hydroclimatological variables using deep learning model

Abstract

Stochastic simulation has been employed for producing long-term records and assessing the impact of climate change on hydrological and climatological variables in the future. However, traditional stochastic simulation of hydroclimatological variables often underestimates the variability and correlation structure of larger timescale due to the preservation of long-term memory. However, the Long Short-Term Memory (LSTM) model, one type of recurrent neural network (RNN), employed in different fields, exhibits a remarkable long-term memory characteristic owing to the recursive hidden and cell states. The current study, therefore, applied the LSTM model to the stochastic simulation of hydroclimatological variables to examine how good the LSTM model can preserve the long-term memory and overcome the drawbacks of conventional time series models. The simulation involved a trigonometric function and the Rossler system as well as real case studies for hydrological and climatological variables. Results showed that the LSTM model reproduced the variability and correlation structure of the larger timescale as well as the key statistics of the original time domain better than the traditional models. The hidden and cell states of the LSTM containing the long-memory and oscillation structure following the observations allows better performance compared to the other tested conventional models. This better representation of the long-term variability can be critical in water manager since future water resources planning and management is highly related with this long-term variability. Thus, it is concluded that the LSTM model can be a potential alternative for the stochastic simulation of hydroclimatological variables. Also, note that another long-term memory model such as Gated Recurrent Unit can be also applicable.