General comment:

Skillful and reliable sub-seasonal precipitation forecasts are complicated as the sources of predictability are much fewer than short-medium range and seasonal predictions. This study proposes a Spatial-Temporal Projection based Calibration, Bridging, and Merging (STP-CBaM) method to improve sub-seasonal precipitation forecast skills. The manuscript presents many results demonstrating that the STP-CBaM method can provide skillful and reliable sub-seasonal precipitation forecasts using dynamic and statistical models. The strategy appears novel and therefore merits publication after minor revisions.

Major comment:

1. This manuscript uses the intraseasonal oscillation signals forecasted by the ECMWF model as predictors. However, the forecast skill of intraseasonal oscillation is also limited at long lead times. Thus, it is essential to know the potential prediction skill of the STP-CBaM method when observing intraseasonal signals used as predictors.

Minor comments:

1. Page 1, Line 9,".." is. ".

2. Page 13, The graphical aspect of Figure 4 and Figure 5 could be improved (e.g., colored bars, etc.).

3. Page 14, The CRPS skill scores shown in Figure 6 indicate that the STP-CBaM method can provide more skillful forecasts than the calibration and bridging models alone. However, I noticed that the STP-CBaM forecasts are of lower prediction skill than the calibration model in several regions. The authors should provide some explanations.

4. Page 15, The authors should explain the weights' results in Figure 7.

5. L. 359~361: "The values of α-index are mostly over 0.7 for all hydroclimatic regions and lead times, suggesting that the merged forecasts are of high reliability."

I am unsure if the value of the α-index exceeds 0.7, indicating high reliability. I suggest providing some figures to prove such conclusions.

6. L. 365: The authors should also discuss recent progress on the prediction of the East Asian Monsoon. The prediction skill of extreme events should also be addressed.

Liu, B., Zhu, C., Ma, S., Yan, Y., & Jiang, N. (2023). Subseasonal processes of triple extreme heatwaves over the Yangtze River Valley in 2022. Weather and Climate Extremes, 40, 100572.

Yan, Y., Zhu, C., & Liu, B. (2023). Subseasonal predictability of the July 2021 extreme rainfall event over Henan, China, in S2S operational models. Journal of Geophysical Research: Atmospheres, 128(4), e2022JD037879.

Zhu, C., Liu, B., Li, L., Ma, S., Jiang, N., & Yan, Y. (2022). Progress and Prospects of Research on Subseasonal to Seasonal Variability and Prediction of the East Asian Monsoon. Journal of Meteorological Research, 36(5), 677-690.

This paper investigates the application of calibration, bridging and merging to forecast subseasonal precipitation anomalies in China based on ECMWF model output of precipitation and atmospheric circulation patterns (zonal winds, geopotential height, OLR). Observations are taken from ERA5, MSWEP and NOAA. Individual models are constructed using BJP and then merged with BMA. Forecast performance is assessed using leave-one-year-out cross-validation and in terms of CRPS, reliability and model weights. It is found that calibration is dominant at short lead times (5-10 days) and U200 and OLRA have increasing relevance at longer lead times. It is concluded that the BMA forecasts have best overall skill and the forecasts are reliable.

Overall, the paper is structured well and the presentation of figures is appropriate. The methods can be followed sufficiently well. My concerns with the paper lie around the marginal performance differences and the strength of the conclusions, particularly around outperformance and reliability. I therefore have some moderate comments for the authors to address ahead of publication, mostly minor, but perhaps requiring some further analysis.

Abstract: First line, add a supporting statement about where subseasonal forecasts are of value or delete.

Abstract: I suggest adding details about the study area

L37: is it better to say the variability is too slow (rather than too short)?

Section 2.1: Suggest some commentary on the quality of the MSWEP rainfall dataset and underlying sources over the 17 regions.

Figures 4 & 5: suggest using different colors for the outline

L328-330: It is difficult to interpret the differences visually from blue shading on the maps. To my eye the difference between BMA skill and calibration skill is marginal and sometimes the merged skill is even lower. I suggest the authors find some way to highlight that indeed “the BMA CRPS skill scores are higher compared to both calibration and bridging”. Perhaps include some statistics.

Figure 7: The weights don’t seem to match the skill patterns. At 15-day lead time the weights of OLRA are higher than calibration and U200 in Region 1, but the CRPS of the OLRA is lower than U200 and calibration. I suggest the authors discuss the discrepancy.

L339-341:  Related to the above, it is stated that the OLRA and U200 models are more useful at longer lead times based on the higher weights in Figure 7. However, it is difficult to discern from Figure 6 that the bridging models are skilful beyond about 15 days. I suggest revising this sentence discuss the value in terms of skill rather than the weights alone.

L350-355: I wouldn’t give too much credit to weakly positive skill scores at longer lead times, they may not be significantly different from zero. I suggest the paragraph be rewritten to focus on the stronger patterns of skill and not worry too much, e.g., about the differences between May and June at longer lead times.

Figure 9: Some of the reliability index values are quite low, around 0.7, which would indicate some problems with the reliability. I suggest further investigation is required to unpack what is the difference in reliability between 0.7 and 0.9. Perhaps the merging is causing some problems with uncertainty representation.

Conclusions: I suggest the first and last paragraphs are not really necessary.

L396: It’s not certain the skill will be improved, could just say it will be investigated.