In this paper, the authors investigate the difference in the impact of data assimilation for convergent and divergent systems.  They conclude that a different approach is needed for these two types of systems.

I have a dual opinion of the paper after having read it.  On the one hand, this is an issue that has been under-investigated in hydrology, and could use some more research.  On the other hand, there are a number of issues with the paper that really need to be addressed before the paper can be published.

Introduction Line 32: In ensemble data assimilation methods, ..., this leads to decreasing uncertainties over time, which favors filter degeneration.  If the filter degenerates, that means that the ensemble is not generated correctly.  An adequate generation of the ensemble should prevent this.

Line 37: You are forgetting uncertainties in the meteorological forcings!

Line 40: The Kalman filter has been developed to estimate states. In system theory, a parameter is defined as a variable that does not change. I know it is common practice to also update parameters, but it is worth noting that this violates the theoretical basis of the Kalman filter.

Line 43: A random walk is also often used as parameter model.

Line 47: Inflation methods are a complete violation of the theoretical foundation of the Kalman filter. The update equations are derived by minimizing the posterior state error. Adding this inflation will, by default, lead to suboptimal results. A correct ensemble generation will solve the need to do this.

Line 51: If the parameter uncertainty is kept constant, the Kalman filter is effectively reduced to optimal interpolation. This should be clarified.

Line 53: Same comment as for line 47.  Using this damping factor is a complete contradiction of the theoretical foundation of the Kalman filter. The author of that paper is Hendricks Franssen, by the way.

Line 107-111: This is very unclear. Please think about reformulating in a manner to make this easier to digest.

Figure 1: Does "forecast" mean the open loop (forward) run?

Line 150: Do you mean Fig. 1c?

Line 189: Usually, in soil moisture data assimilation studies, only the surface layer soil moisture is assimilated. Please justify this strategy.  Also, an observation error of 0.007 seems quite low.

Line 192: This is very important. It is stated: To generate the initial ensemble, the ensemble mean is perturbed by a correlated multivariate Gaussian distribution. This is very unclear.  Please provide a better explanation on how the ensemble is generated, as this is very important, and could help interpret the results later in the paper.

Immediately below figure 2: This case shows that the predictability for a perfect convergent system is infinite. I do not agree. It shows that the ensemble is not adequately generated.

Line 230: I would argue that disturbing the n values leads to an ensemble that is well generated, which explains the better results.

Line 268: A model is never perfect, so a better word than "perfect" should be used.

Line 283: Again, you are forgetting meteorological forcings.

Line 290, 291 and 296: Again, inflation, and using a constant ensemble spread, are tricks that are used to compensate for a poorly generated ensemble.

As stated above, I am in favor of the idea of the study, as this is important. But the issues I mentioned should be resolved before the paper can be published.

General comments: The technical note titled “Sequential ensemble data assimilation in convergent and divergent systems” written by Hannes Helmut Bauser et al. investigated the performance of data assimilation in convergent and divergent systems. Although the manuscript showed how a sequential data assimilation method (EnKF) works in two convergent and divergent systems, at the current version, manuscript only showed the results of different case studies without detailed explanation why we observed those results, which is crucial for technical studies. In addition, the manuscripts did not provide suggestions to improve the performance of data assimilation in convergent and divergent systems. In my opinion, it is better if the manuscript considers scenarios that use one or two techniques to maintain ensemble spread or to avoid convergence to wrong states in convergent systems as stated in the introduction section. Examples presented in the manuscript is ideal. Authors may consider examples that are nearer to the realistic cases. 

Specific comments:

- Introduction: Evidences on why we need to assess the performance of data assimilation in convergent and divergent system should be more clearly stated (why we need to do this study). How convergent/divergent system impacts on the effectiveness of data assimilation? What is the disadvantage of divergent systems (slowly converge to the truth state?) ? Provide more examples of convergent/divergent systems in hydrology and geosciences.

- Section 3 and 4: Provide more technical explanation of the obtained results and show the quantitative criteria (NSE, relative error, correlation coefficient, etc.) that compare results obtained with data assimilation and with forecast (open loop), especially explain the changes of ensemble variance with time. Why the initial condition for data assimilation was not set to be different from the true one? What is the observation values used for assimilation and their corresponding errors?

- Section 4: Which method was used for joint state and parameter estimation?

- Figure 1: Show the results corresponding with forecast in figures 1a,b,d for comparison. Explain what is and .

- Figure 2: results obtained with the forward run without data assimilation should present in figures b,c,d for comparison.