|The authors have done a good job revising this MS, and should be congratulated, I hope my comments have proved constructive rather than obstructive. The statistical analysis is a lot more transparent, and I would urge the authors to refer to the model as a 'statistical model' throughout the paper so as to distinguish it from a model based on physics. I would also encourage the authors to provide the new information contained in their response at least as supplementary information. |
The increased information contained in the response (not available during the initial review) does raise more questions. Specifically, the large negative departure of the predicted time series of FA area when FA-1 is not included in the model (Figure 1b in the response). Regardless of the R2 value, this does raise some issues. The first is that if FA-1 acts as an autoregressive variable, then an autoregressive model is usually adopted to formally account for this. The second, and somewhat perplexing issue, is how the authors actually know what the variable FA-1 is if they are using it to predict FA for the following month? I understand that this is one of four variables used to predict dFA in the calibration period (i.e. when they are able to measure this from the remote sensing data), but the extrapolation of this series back to 1912 implies they have data for all these independent variables going back that far (such as it appears for total rain and rain days). From what I can see the authors have data for two of the variables going back to 1912, but use a model with 4 variables for prediction, but there doesn’t appear to be actual data for the latter 2 (FA-1 and Int). Is this the case, and if so, what numbers do they use in the model in their place? Or have I missed something here? If I’ve got this totally wrong I think the authors need to at least clarify this last step. Finally, the Cross-Validation Error should be plotted with the final predicted time series as error bounds to show the values are not a precisely known line, and give a visual sense of this error, not just one contained within the table (otherwise the figure is at risk of being misleading).
I also appreciate the authors have modified the interpretation of the negative FA values in the MS, and this seems reasonable. However, negative values are presented as a continuous series in the time series figure giving the impression that these are representative of an actual physical process. Positive values are obviously an accurate estimate of the actual water area on the marsh itself, but negative values are not an accurate estimate of anything (i.e. they certainly can't be expected to match the time series of the vadose zone water content or the groundwater elevation), so we are left with these precise looking wiggly lines below 0 whose values doesn't actually mean anything (other than they are less than 0 and an unsaturated zone is developing). One option is to highlight this in the figure, or at the very least provide this explanation in the text within the relevant section.
With these comments to be addressed, the paper looks good for publication.