This study applied the Markov Chain Monte Carolo algorithm to calibrate a simple water balance model using GRACE TWS observations. The posterior model parameters, model states and simulated TWS for one watershed were shown in this study. The results suggested the potential of using GRACE to constrain model parameters. The topic is relevant for reader of HESS. However, I believe some critical points need to be clarified and supported by additional results.

1. It was mentioned that the ET in the model was derived from the satellite observations of precipitation and TWS and ground-based river runoff (L138). I assumed that it must be GRACE TWS here. If so, the GRACE data has been used in the modelling and resued in the model calibration through MCMC. The GRACE data was not independent to the model simulations. Please clarify. 

2. The results for one watershed is not convincing. In particular, it was mentioned that additional testing over other watershed has been done in L369. Results for few more watersheds can help supporting this study. Since the inputs for the model are satellite rainfall, TWS and in-situ river runoff, it would be surprise there is no other watershed has enough river runoff data over Amazon.

3. Results of model simulations without MCMC should be compared with the posterior TWS to demonstrate the improved performance.

4. L76: Is the model proposed by the authors for the first time? Or any reference for the model?

5. L138: Was the satellite precipitation from TRMM here as well?

6. L173: Spherical harmonic solutions here?

7. L260: The information about each variable for each subfigure has been included in the figure caption. More discussion about the results instead would be helpful.

8. L285-290: could you include the r2 for the model simulated TWS and de-seasonalized TWS without MCMC as a comparison? Also plot the time-series together in Fig6?

9. The sensitivity results from figure 8 would make more sense to mentioned in the beginning of the section since the results were summarised in Table1.  

10. Changing the x-axis to years for Figure 4 and 8 like the other plots would be more reader-friendly.

The article is well written and all the concepts are explained in required details. However, the inference drawn from the study need more clarifications and probably a major revision. After writing this review, I happened to check the review comments from reviewer 1 and I had exactly the same comments as their comment no 1, 2, 3 and 10. I am not repeating them here. My additional concerns are:  

1. Authors have stated, in the title and in the introduction (line 65 to 69) that they are able to identify and estimate the processes responsible for TWS variability. Explicitly they have stated their aims as: a)“inform and reduce uncertainties of terrestrial hydrologic processes regulating the seasonal and inter-annual variability of TWS in the western Amazon, the Gavião watershed” b)Develop a model to represent the first order controls on seasonal-to-decadal soil moisture dynamics, and c) employ a Bayesian model-data fusion approach to constrain model parameters (namely initial states and time-invariant process variables), such that differences between GRACE and simulated TWS anomalies are statistically minimized

Therefore the first impression is that the study will help us understand the cause and drivers of variability in the water storage. However, the results and summary lack discussion on the parameters and physical processes that drive the dry season or wet season, and why?

2. The calculation of basin average TWS from GRACE data is not clearly explained. Did the authors use mascon data or spherical harmonic data? How do they combine these three solutions? How do they tackle the limitations of the coarse spatial resolution of GRACE data and the different spatial resolution of the three data products.
3. Can authors comment or compute their model for an arid or semi-arid region or a river catchment with trends (such as the Great Basin in the USA)? What do they expect to obtain from such a model at global scale?
4. Please mention the catchment name in the title instead of ‘Amazon’ or compute the model for whole Amazon.
5. Line 361: Authors state “GRACE-informed model parameters can be used for predicting seasonal and inter-annual soil water hydrology in the absence of concurrent GRACE measurements. We showed that using a 5-year data record of TWS allows the parameter inference to still be applicable to the remaining 5-year data record, which is simulated without the use of information from GRACE”. If the evapotranspiration product was obtained using GRACE, then the simulated part is not completely free from GRACE.