Integrating uncertainty decomposition and CycleGAN bias correction in enhancing future hydrologic projections in a snow-dominated alpine watershed

Given the sensitivity of snow to climate change and its critical role in the hydrological cycle of alpine regions, it is essential to accurately project future snow processes in mountainous areas. This study, taking the Manas River Basin (MRB) in Xinjiang China as the test bed, aims to quantify the uncertainties in hydrometeorological variables from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6) simulations and further reduce these biases using a Cycle-Consistent Generative Adversarial Network (CycleGAN). The bias-corrected CMIP6 data are then used to drive the SWAT model calibrated with both runoff and snow water equivalent (SWE) through a dual-objective approach for future projections. The results indicate that: (a) model uncertainty is the primary source of uncertainty in the original CMIP6 outputs. CycleGAN demonstrates substantial effectiveness in reducing model uncertainties; (b) most subbasins of the MRB will experience absolute SWE reduction in the future and the changes of SWE vary significantly across elevation bands; (c) The runoff in MRB has an increasing trend in future. As the ratio of rain to snow increases and snowmelt occurs earlier, low flows during the dry period will increase significantly, which will result in higher risk of spring floods. The findings will provide important guidance for projecting future snow dynamics and water resources management in the snow dominated watersheds.