|The authors have generally improved the manuscript, but in my view, there are still some issues that require their attention.|
1) The manuscript still needs a thorough copy editing. I have included suggestions on grammatical corrections below, but I have stopped at about page 6, since the text contained errors almost every other line. It can be cumbersome for foreign authors, like myself, but nowadays, several software packages assist the authors. In any case, a manuscript should appear written in proper English. Furthermore, it should be the task of a reviewer to correct the grammar of a manuscript, even less so when this is the second version.
2) As with the first version, the data assimilation's real purpose in the snow depth context remains unclear. Usually, a model assimilates a set of limited observations but provides a much more complete output regarding variable and spatial coverage. For instance, a weather prediction model assimilates available station observations of some variables but produces a complete field of future predictions.
Here, the hydrological model assimilates snow depth observations but estimates snow depth at the same location, and it does not even produce predictions of snow depth. So the utility of the whole set-up is unclear. Shouldn't the model be validated by comparing prediction versus observations? The conclusion of the study that the results of assimilation are closer to observations is to ma a tautology. Of course., the model results are corrected towards the observations. It is no surprise that those corrected values are closer to observations. A natural progress would be if the model *predictions* for a specific time step t are closer to observations when the observations prior to t are assimilated.
3) line 78 'and AMSR-E SWE into a hydrologic model to improve modeled SWE ..'
4) line '81 assimilating ground-based snowfall and snowmelt rates, simultaneous assimilation of D-InSAR
The acronym has not been defined yet.
5) line 88 dynamic system usually has strong nonlinearity
6) line 95 The
greatest strength of PF technique is free from the constraints of model linearity and error following
Gaussian distribution, this makes the PF technique succeed applied in nonlinear and non-Gaussian
the greatest strength of PF technique is to be free from the constraints of model linearity and error following
a Gaussian distribution. This allows the successful application of the PF technique to nonlinear dynamical systems with non-Gaussian errors
7) line 97 dynamic systems. Additionally, PF technique give weights
dynamical systems. Additionally, the PH technique gives weights.
8) line 108 snowpack runoff simulations (Magnusson et al., 2017). Above studies demonstrated that either
assimilated the snow-related in-situ measurements or remotely sensed observation data through PF
technique can successfully update the predictions of snowpack dynamics,
The studies indicated above demonstrated that the assimilated snow-related in-situ measurement or the remotely sensed observation data through PF technique can successfully update the predictions of snowpack dynamics.
9) line 112 Nevertheless, particle degeneracy is still one potential limitation for PF technique, it occurs when
most of particles have negligible weight and only few particles have significant weights, which makes
the state probability distribution cannot be represented by the particles
Nevertheless, particle degeneracy is still one potential limitation of PF technique. It occurs when
most particles have negligible weight, and only a few particles carry significant weights, which hinders a realistic sampling of the underlying probability distribution of the state
10) line 116 an efficient approach which can effectively mitigate the problem of particle degeneracy, however,may lead to the resulting sample will contain many repeated points a
An efficient approach that can effectively mitigate the problem of particle degeneracy. However, it may lead to the resulting sample containing many repeated points a
11) line 147 is distributed at different latitudes in the northern hemisphere
12) line 148 located beside the Kitinen River in Finland and has a 2 m depths frost
located beside the Kitinen River in Finland. The upper 2 meters are frozen.
13) line 164 It is noteworthy that the spatial variance on the performance of the model is negligible
the spatial variance of the performance of the model
14) line 168 detailed information of snow climates, and
dataset process introduction of the eight sites can be also referenced in You et al. (2020a).
What is 'data process'. ....can also be found in You et al. (2020a)
15) line 170 The snow partial within Noah-MP model
This is not proper English. It is unclear what snow partial is
16) line 213 function p ( zt xt i ) , which measures the likelihood of a given model state concerning the
observation z t
The notation could be clearer. Usually, I would interpret p (z_t) | x_ti) as the probability of z_t conditional on x_ti
17) line 215 In general, a Gaussian distribution was assumed to perturb the observations and the
likelihood function was defined to represent the errors.
The observation errors are generally assumed to follow a Gaussian distribution, and the chosen likelihood function represents this assumption.
line 225 if the effective sample size
what is the effective sample size? Is it just the number of samples? The text does not mention autocorrelation at all, so the word 'effective' is unclear
242 The role of the survival rate is unclear. It seems that the survival rate is just a measure of the distance between the particle and the observations. This distance is already considered when assimilating the observations with the weighted average over particles. So is this a double counting?
line 268 All particles are disturbed with a gaussian error. Isnt is just the same as the mutation , only with a different type of error distribution ? what is the role of the mutation ?
line 339 ' Since the meteorological perturbations are unbiased, the nonlinearity of physical
processes within the model is supposed to be the main reason for the uncertainty'
I may not understand this sentence. The magnitude of uncertainties is not related to the linear or nonlinear character of the model. A linear model would just rescale the spread in the forcing and nonlinear model would expand or shrink disproportionally the forcing uncertainties. So the nonlinear character itself cannot be the reason per se of
Section 3.1 Open-loop ensemble simulations
The expression open loop-ensemble is used only once in the title of this section. What is its meaning? It is nowhere defined nor used again.
This section is also tough to read. It contains just one very long paragraph without clear structure.