Future precipitation projections using CMIP6 multimodel ensemble: a deep learning-based bias correction approach

ABSTRACT The water cycle is significantly impacted by global warming and has an impact on hydrological systems all around the world. This study examines the seasonal time scale spatiotemporal variability of precipitation at Iran's northern stations for two climate zones: the near-future (NF) and the mid-future (MF). Therefore, to provide a more robust and reliable prediction, various multimodel ensembles (MMEs) of general circulation models are employed. The study focused on using some techniques, namely long short-term memory (LSTM), decision tree, multivariate linear regression, and artificial neural network to develop MMEs to simulate this and to project the precipitation patterns in these areas. The study indicated that the MMEs created by LSTM and empirical quantile mapping (QM-LSTM) had more consistent performance compared with other methods. Under SSP126 and SSP585, the projection of seasonal precipitation indicated that a slight wetting pattern could occur at most proposed stations at different times of the year, except in autumn. While a significantly decreasing pattern of precipitation intensities is observed in autumn in these regions for the NF and MF climate in scenarios SSP126 and SSP585, the range of precipitation change can reach 8.31% compared to the historical period.