Simulation of the projected river flow changes using integrated downscaling and Bayesian optimization-tuned kernel-based models

In terms of having a comprehensive vision toward investigating the impact of future climate changes, LARS-WG, a weather generator model, was employed in the present study to project future changes in precipitation and maximum and minimum temperature. Considering the significant uncertainties linked to future climate projections, three general circulation models (GCMs), HadGEM2-ES, BCC-CSM1, and CanESM2, were utilized to project the future climate depending on three important scenarios of emissions, RCP2.6, RCP4.5, and RCP8.5, covering the period of from 2020 to 2041. Two local synoptic stations in Iran, representing distinct climatic characteristics, were utilized to validate the model, by adopting the historical data spanning twenty years (2001-2020). The results indicated a decrease in mean runoff during spring for semiarid areas (maximum of 19.95 m3/s in May) and in both spring and winter for humid areas (maximum of 8.7 m3/s in May) over the next twenty years. Moreover, under the RCP8.5 scenario, the humid region experienced a maximum rainfall increase of 66.66 mm compared to the base period. Additionally, the semi-arid region observed a maximum temperature increase of 1.18 degrees C under the same RCP8.5 scenario. The next step involved using downscaled parameters as inputs for the Bayesian optimization-based kernel-based models to simulate future runoff considering the impact of climate change. The results obtained demonstrate that the semi-arid region exhibits better forecasting capabilities, achieving evaluation criteria values of R = 0.973, NSE = 0.946, and RMSE = 0.020. The result emphasizes the need for implementing sustainable adaptation strategies to protect future water resources in the basin.