General comments

Wang et al. presents an interesting study about comparison between two downscaling approaches i.e. statistical downscaling (ISIMIP) and dynamical downscaling (WRF) combined with bias correction of the future projection of ESM scenarios (i.e. RCP4.5, RCP8.5, G4) over Beijing provincial region. The idea of comparing statistical with dynamical downscaling is a novel path of research. The authors focus on mean temperature, humidity, and wind speed, which are all relevant for climate impacts. The manuscript is generally clearly written and well structured, and the analysis is done in an organized way. However, some parts need to be clarified to easily understand some contents of the manuscript. I think the manuscript is scientifically sound and merits publication after some minor revisions based on my specific comments below.

Specific comments:

1. In table 1, assessment on the performance of ISIMIP and WRF with bias correction is done only through RCP4.5 scenario. Is there any reason why RCP8.5 is excluded in this part?
2. This research used ERA5 reanalysis data as the proxy of observation. The authors need strong justification of choosing ERA5 reanalysis in this research as reanalysis data itself might contain some degree of bias.
3. Step1 of ISIMIP statistical downscaling and bias correction is simplified and make it unclear. Please provide a more detailed description of it.
4. The statement “QDM is similar to QM but is non-stationary”. In what sense it is non-stationary?
5. From figure 3i we see that the WRF simulations have spatial patterns very different from Moreover, figure 4 also shows that the pdf of WRF deviates from ERA5 significantly. Are there any reasons behind these findings?
6. Authors use three different terms for ensemble i.e., “ensemble-mean”, “multi-ensemble mean”, “multi-model ensemble mean”. Are they referring to the same metric?
7. In Figure 5, the caption states “labelled as ERA5”, but in the figure we cannot find the label “ERA5”. Do you mean the “observed” label? Figure 8 seems correct. Please make it consistent.
8. It is stated that “wind speed of all four ESMs outputs have correlation coefficients <0.1 with ERA5”. I would suggest that authors describe in the discussion section the consequence of this low value to the results of the analysis.
9. The statement “Humidity and windspeed anomalies from ISIMIP appear somewhat spatially anti-correlated, while for WRF there are no particular patterns” need to be clarified. What does the term “anti-correlated” mean? Is it a negative correlation?
10. In Table 3, I would suggest using Asterisk sign to indicate that the differences are significant, instead of writing the number in bold.

General Comments:

This paper looks at statistical and dynamical downscaling for the Beijing region under three future climate scenarios. It also appears to be the first time that dynamical downscaling has been used with a geoengineering scenario. The paper is fairly straightforward but seems to be a good starting point for future work on geoengineering modeling. Probably the biggest question I have is regarding the choice of analyzed variables. The authors compare temperature, humidity, and wind speeds under the different scenarios; temperature is of course important, but, at least to me, the latter two are not so much. I wonder if the authors thought about analyzing precipitation instead.

Another general question I have is: how is the ISIMIP spatial downscaling done? The two steps in Section 2.3 address bias correction and temporal downscaling, but how do you go from the coarse ESM grids to the ERA5 grid?

Finally, why is quantile delta mapping not done for the historical WRF simulations (in which case I note that it would simplify to the regular quantile mapping method)? Wouldn’t it be fairer to compare ERA5 with bias-corrected ISIMIP/statistical downscaling and bias-corrected WRF/dynamical downscaling?

Specific Comments:

Abstract, line 20: It is not clear to me what “larger spatial ranges” means. I also do not see further reference to this in the main text. While I can see from Figure 9 that the range (maximum minus minimum) of temperature and of humidity is larger in WRF than ISIMIP, it’s not obvious that this is the case for wind since although WRF has lower minima, ISIMIP has higher maxima.

pg 3, line 60: Give the full name of ISIMIP at least once (perhaps in the abstract as well).

pg 12 / Figure 4: Are the authors concerned about RH values exceeding 100%?

pg 14, line 277: ISIMIP seems to increase error and variance in wind speed in the case of HadGEM2-ES.

pg 16, line 309-310: Assuming I’m looking correctly, it doesn’t look like WRF-QDM-MIROC and WRF-QDM-MIROC-CHEM are wetter than ERA5 from June to October. If anything, they look drier, especially the former.

pg 18, line 335-336: Wind speeds in ISIMIP look highest (as opposed to lowest) in the southeast of the domain.

pg 20, line 360: To me, only the G4-RCP8.5 anomalies for ISIMIP look anti-correlated. I might suggest removing this sentence.

pg 24, line 417: I think the authors meant to say that WRF relative humidity exhibits a dry bias in winter.

pg 25, line 429-430: This sentence is confusing as written, and the first part is not obvious since WRF-QDM-ESM shows opposite trends in roughly equal parts of the domain when comparing G4 and the 2010s. I might suggest removing this sentence.