Comments on “Development  of East Asia Regional Reanalysis based on advanced hybrid gain data assimilation method and evaluation with E3DVAR, ERA-5, and ERA-Interim reanalysis” by Yang et al.

The authors generated the East Asia Regional Reanalysis (EARR) using Weather Research and Forecasting (WRF, v3.7.1) during the ten-year period 2010-2019, based on the advanced hybrid gain data assimilation method (AdvHG). The new advanced hybrid gain (AdvHG) data assimilation method combining E3DVAR and ERA5 based on WRF model is newly proposed and investigated in this study. The manuscript verified the EARR for two-year period 2017-2018 by comparing EARR against ERA-Interim, ERA5 and observations.

I think a lot of work behind this manuscript is worth publishing. However, there are some details and the underlying physics processes were not well discussed.

The detailed comments are listed as below:

Major comments:

1. It is not easy to understand the regional reanalysis method used in this paper. I’d suggest the authors use schematic diagram to clearly describe the regional reanalysis method. What does “forecast fields are integrated up to 36h” mean? Is reinitialization used? What does “6h forecast of ERA5 reanalysis based on WRF model” mean? Does it mean WRF simulations using ERA5 as LBC forcing? How many experiments are used in this manuscript? What experiments do “E3DVAR”, “AdvHG”, “WRF-ERA5” and “WRF-ERAIN” refer to respectively? These details should be more clarified.

2. The newly advanced hybrid gain (AdvHG) data assimilation method uses 6h forecast ERA5 instead of deterministic analysis, and uses ERA5 instead of producing their own analysis fields from a variational DA method. The method is expected to save time and computing cost compared to traditional data assimilation framework. Why does the author use the 6h forecast of ERA5 reanalysis based on WRF instead of the deterministic analysis in AdvHG? Will the regional reanalysis be more accurate if using ERA5 deterministic analysis in AdvHG?

3. EARR is developed during the ten-year period 2010-2019, while in the manuscript, the authors only verified the product for two-year period 2017-2018. Why is this two-year period instead of the entire ten-year period chosen to be verified? I would like to see how the EARR product performs compared to the other two ECMWF global reanalysis in a longer time period, and whether EARR is suitable for application in long-term climatology research. If so, the influence of this product will be greatly increased.

3. The precipitation of AdvHG in winter (Juanuary) is the most accurate among other results, while E3DVAR (ERA5) performs better for weak (strong) thresholds in summer (July). The assimilation method tends to have positive effects on the simulated winter rainfall while brings limited improvement on summer rainfall. The possible underlying physics have not been well understood. Which physical process could be mostly possible affected leading to such positive effects?

4. Figure 2, it is clear that E3DVAR has the lowest RMSE for up-level variables, so what are the advantages of AdvHG? Does ERA5 in Fig. 2 mean WRF results forcing by ERA5?

Minor comments:

1. The meaning of variables in Eq. (3)-(6) should be clarified.

1, Title. Suggest the authors change it. If it is the development of EARR, the information of input observations data quality or evaluations of more variables/indexes during the whole 10 years 2010-2019 should be included in the figures, but not only 201701 and 201707 mainly in Fig 2-4, 7-13 and only 2017-2018 in Fig 4-5. If it is the development of AdvHG, the innovation contents from your own group should be included in 2.2.2 (page5-8), otherwise, they are all the approaches you could adopt, but not develop. The main contents of the paper is evaluation in 2017-2018, including the method, results and usage in EARR, so maybe it is more suitable to call the title  like “Evaluation of EARR based on AdvHG”, for your reference. If more figures of longer time series results could be replaced here, it is better. Anyway, the results are not enough, the period is short. The representativeness of the result is limited, comparing with the ERRA (2010-2019).

2, horizontal resolution, 12km. It is suggested to mention the raw description of model/DA like other reanalysis, for it is not the same resolution anywhere in the global. Add the information only once in 2.1, like in line 84 (540*432 grid points), it is suggested.

3, 1. introduction. The motivation is described well enough here, like a full story, while the scientific background introduction is not enough, not like an excellent scientific introduction in a paper.

4, 2. system. Line 83-85, the sentence is not right, Fig 1 is the domain.

5, Line 134-140, what is alpha in EQ 7?

6, The authors make great effort in the DA approach, while what is the characteristics in East Asia of the EARR, comparing with other regional reanalysis, considering of the terrain, climate state like monsoon. In this scope, 50% is cover by the ocean, how is the ocean-land-atmospheric coupled here or just simply depends on all in WRF?

7, Line 160, it is wrong here to mention QuikSCAT which is 199907-200911, it is not in 2010-2019. For your reference: (1) Coriolis/WindSAT (20070813-20120804) from CFSR prepbufr, (2) Oceansat-2/OSCAT (20091215-20140220) KNMI reprocessed but not in CFSR,  (3) MetOp-A/ASCAT (20070101-20140331 KNMI reprocessed , GTS data till present in GDAS), (4) MetOp-B/ASCAT (20140408-present in GDAS). You may not use reprocessed ASCAT wind, but it is used in ERA5.

8, 4. Result, it is suggest to shorten the results to 60%. The emphasis is how good is EARR but not ERA5. There are many sentences/paragraphs with the subject of ERA5 but not your reanalysis. Like line 235-256, 260-262, 290-291, 328-331, 410-412, 449-452, 520-521. And the difference between ERA5 and ERA-I could be shorten like line 265-270. Pay more attention in how good EARR but not how is ERA5 like line 357-363. The order of the results from different reanalysis is also important. Line 312-314 is good in expression.

9, Line 324, except for strong thresholds, how is strong? >4 mm/6h? how is week?

10, Fig 7, Line 346-348, results of AdvHG in Jan is better than in Jul, FBI closer to 1. Different FBI results in Jul are larger than 1 (over-forecast) more than in Jan, more difficult to improve for summer than winter. Index ETS and FBI are more difficult to handle and analysis than POD and FAR which is better when it is larger and smaller, separately.

11, Line 354, (Figs. 8a and b), is it right? 8b is FAR.

12, 6. Summary, it is not concise in this paragraph.

13, Reference, line 608-614, it is repeated, please delete. Line 621-622 seems with larger font size.

14, Fig 2, is it the result of all EARR domain? Jan and Jul shown in Fig 2-6 while YYYYMM or YYYYMMDD shown in Fig 7-11, it is suggested to use unified expression. Fig 6, Temp revised to T like Fig 2.