The paper has been improved, thought there are still areas where further improvement is in my view needed (see comments below). I think the paper needs some further revision before being ready for publication.
1) Page 1, line 19 and elsewhere: While a hard space is included between a number and the unit, the exception is where the unit starts with a superscript. Percentages fall into this category, and the space between the number and the % symbol should be removed.
2) Figure 3: Why use Lake Mead in this figure, seeing as the same results are shown in Figure 4? Could a different lake (not shown in Figure 4) be used? Note that the results for Lake Mead suggest the relationship is slightly non-linear, which means that extrapolating beyond the range shown will result in larger errors, beyond the statistical error shown.
3) Figure 4: confidence bounds seem to be based on the scatter in the residuals (+- 2 std dev). The confidence bounds should be based on the uncertainty in the regressed coefficients (both intercept and slope), which will result is slightly higher single point prediction confidence bounds near the limits of the plot. The uncertainty in the fit near the centre of the plot will be much less than the single point confidence bounds, so this will have much greater variation in the width of the confidence bounds.
4) Figure 4d: Uncertainty results for Nasser are dominated by 2 observations. Any idea why these observations should be problematic? What happens if these are removed? Presumably, the confidence bounds will decrease by almost a factor of 2?
5) Page 11, line 10: In what way are the linear regressions accurate? How do you define "accurate". One possible definition is that an accurate regression is one that isn't biased. However, by definition, linear regressions are unbiased, so this doesn't help. An R^2>0.8 doesn't mean an "accurate" regression - just means that the residuals are sufficiently small to result in an R^2 of 0.8. Why 0.8? Why not 0.75, or 0.85? Better to just say that the linear regression was used to estimate the volumes in the subsequent analysis
6) Page 11, line 14: better to add here that the LvP group will not be used in the following analysis.
7) Figure 5: explain logic behind these thresholds? From what I can see given the tables in the appendix, the 0.008 threshold in CV is set based on the minimum CV for the LVG group. The 0.8 threshold in R^2 appears to beaimed at capturing in the LvG group, the point near (0.83,0.82). What is the uncertainty introduced in this classification? Would it be better to have x-axis on a log scale so that distinction between categories can be more easily seen? Also, looks like there is no sharp boundary between Lc and LvG/LvP. With slight change to the threshold, the numbers will change?
8) Page 11, line 20: mention that these are in group LvG?
9) Figure 6: The uncertainty bounds shown do not consider the time between values. They seem to be linear connections between the uncertainty in each point. Not an issue for panels a and c, but a significant issue for panels b and d. In the figures, the uncertainty in the gaps is more apparent than the uncertainty in the points. The issue is that the uncertainty in the gaps is at times grossly under-estimated. Either the plot should be modified to help emphasise the uncertainty in the points, or there should at least be some discussion about this in the text.
10) Figure 7: Might be good to comment on the spatial distribution of the different classes? Seem to be a high percentage of Lc in East USA and Canada for example, with only 2 lakes in the LvG category. West USA seems to be dominated by lakes in the LvG category. Similar in Scandinavian countries. Suggests a significant fraction of the Lc class lie above 45deg N, with from what I can see only 7 LvG lakes. The five southern most lakes (South America) are also either Lc or LvP, reinforcing a possible latitudinal influence? Also a high density of Lc lakes in Africa (south of 15deg N). Concerning the distribution of LvP lakes, how sensitive is this result to the choice of R^2 threshold? If decreased form 0.8 to 0.7 or 0.6 is there a significant change in the spatial distribution? Where are the really bad cases (e.g. R^2<0.3)?
11) Page 16, line 18: Might need to expand the discussion on the behaviour in the extrapolation for Lake Mead? Plot suggests that the magnitude of the variations is about right, but that there is a bias in the result. This potentially means that the pattern of short term variation may look okay, but the amplitude of the variation is likely to be less than the actual variation. In comparison, Lake Powell shows also shows a bias, but this is smaller, and an over-estimation rather than an under-estimation.
12) Figure 12: Cubic regression is not necessarily the best option. Why not try a hyperbolic form (asymptote for small A near 1271, and for high A increasing linearly)?