Please see my detailed comments in the attachment.

This manuscript presents a three-step stochastic downscaling framework designed to downscale daily temperature and precipitation from the ERA5 reanalysis dataset. The manuscript is well-written overall, and the results demonstrate a solid and comprehensive analysis. However, I have two substantive concerns that warrant further consideration. First, regarding the precipitation occurrence modeling, the authors employ a Bernoulli distribution while acknowledging that temporal dependence structures cannot be adequately incorporated into the precipitation occurrence process due to the infeasibility of transforming binary precipitation occurrence into Gaussian random variables. This represents a notable limitation, as the temporal structure in precipitation occurrence may not be sufficiently preserved under this approach. Given this acknowledged technical constraint, I would encourage the authors to discuss why traditional Markov chain-based methods, which are well-established for modeling temporal dependencies in precipitation occurrence, were not adopted as an alternative. Second, while the authors acknowledge that their proposed downscaling method treats precipitation and temperature independently and thus cannot capture multivariate relationships between these variables, it remains unclear whether the framework is capable of preserving spatial or inter-site dependence structures within the downscaled precipitation and temperature fields themselves. The ability to maintain realistic spatial coherence is critical for many hydrological applications, and I recommend that the authors provide additional results and discussion addressing the performance of their method in reproducing spatial dependence patterns across the downscaled domain.