The paper applies different already reasonably mature data-assimilation (DA) techniques to the estimation of states and parameters of two land surface models (LSM). The problem is important, the methods appropriate, and the paper operationally instructive.

The key finding -- the differences between the LSMs are larger than the differences between the DAs (abstract, lines 23-25) -- corroborates the maturity of the DA schemes, which is an important result. It raises the question, however, if such a study is required at all and if so, if its focus is correct. Given the comparable results of the DAs, my interest is directed towards computational cost, ease of use, and range of applicability beyond these two LSMs. While the authors decided not to address this issue in depth, a paragraph on it is certainly warranted. Looking at the LSMs, the situation is much more difficult since addressing the implications and relevance of their differences, and possible remedies, is clearly beyond the scope of this manuscript. Still, statements like the paragraph on time-varying parameters that emerge when heterogeneity is neglected (reply to an issue raise by reviewer #2) has me wonder what such parameters would be good for. Such approaches are fine for process control, in a chemical production plant for instance, but are they of any use in an environmental context where we eventually want to predict something? Some clarifying statements in the revision appear to be in order.

With the above, the focus of this paper can only be on the methodological aspects of DA, illustrated for the difficult case of LSMs. This must be clearly worked out in the revised version.

In addition, the numerous issues raised by the reviewers need to be addressed and discussed as confirmed in the authors' replies. Since the focus is on methods, the concepts must be more strict than in an application. What is for instance "fundamental" and what is not (reply to rev #1, pg 16 l 33) or how to measure distance in probability space (reply to rev #1, pg 16 l 1-2). [I do not want to become too philosophical at this point, just focus attention.]

(I should comment that the manuscript would have passed my desk with "minor revisions" if the authors had already included successful results/useful answers of/for their "will do in revision" replies.)


As a very minor point I should like to comment on the authors reply to reviewer #1 concerning the SPADE sensor:
(i) The numbers given for the permittivity refer to the relative value, which is dimensionless (not F/m).
(ii) The key question is: What are the sensor's center frequency and frequency range for the observed range of soil water contents? This is relevant because the relative dielectric permittivity depends on frequency, and temperature for that matter, which may turn this particular setup quite complicated to interpret. (I could not check this aspect myself because I could not get an electronic version of the cited paper. In the context of this paper, this is a minor issue, however.)

The manuscript improved significantly. This is also attested by the reviewers. Still, some important questions pointed out by the reviewers remain open and need to be addressed. Please go through them carefully. Since the raised issues are beyond the mere technical, I'll keep the reviewers involved also in the next, hopefully final iteration.

I currently judge the manuscript to be in a form that it can be accepted and to be a useful contribution to the data-assimilation community in hydrology. The fact that the two models are more uncertain than the various DA methods is unfortunate for the evaluation of the methods. On the other hand, this is one of the key operational issues in environmental sciences at large. It can be illustrated here, but apparently not resolved.


Suggestion: I would be more reluctant with the use of the concept of "effective parameters" for models and their assimilates that deviate so strongly and characteristically from the data. Maybe "apparent" would be more appropriate. Instances of this are "As the estimated parameters represent effective properties of the Rollesbroich site,…" in the response to reviewer 2, which also appears at least formally at odds with "The resulting parameter estimates might therefore not be representative for dry periods…" further down in the response.

Foreseeably, both reviewers were disturbed/confused by the late and extensive rewriting of the manuscript and not amused by the additional workload.

One reviewer found the manuscript good enough to publish with some comments at the discretion of the authors. These concerned: (i) dropping "joint" from the title and (ii) the use of the term "effective".

The other reviewer submitted a report and proposed major revision in order to be able to check the responses. In particular, the request is to mark all essential modifications that may go unnoticed in the typical late-stage reading of a manuscript. An example of such a modification, which I completely agree is noteworthy and which has escaped my notice, is given in the reviewer's report. I ask you to
1. go carefully through the manuscript in its transition to the latest version and mark all the key modifications
2. address carefully the reviewer's questions, also the unanswered ones from the previous iteration (you need not agree with the reviewer, but you must comment on the raise issues).

Given the real long time this manuscript has been in the making and the iterative process, I want this to be the last round with one of two possible outcomes: publish or submit as a new manuscript.

This manuscript took an unusually long path but now is an overall good contribution.

I am still concerned about the issue of a significant parameter dependence on the choice of the random generator's seed (hess-2016-42-author_response-version3.pdf, pg 3). Such a dependence is an alarming signal in my understanding of these methods and certainly warrants further exploration. Just having the proposed sentence

“This initial parameter ensemble is the same for all the assimilation methods.”

in the manuscript neither addresses the source of the problem nor gives any hint to the reader. I am not content with this approach! Still, as far as I can see it, this does not have an impact on the message of this paper. I thus

1. ask the authors to take the chance and set things straight by at least mentioning the issue in the manuscript

2. other than that accept the paper as is for publication.