The manuscript has been quite significantly altered from its previous version, with many relevant and good aspects added to the model development, suitability map building, map evaluation and sensitivity assessment, in terms of methodology, results and discussion. It really has become an interesting paper to read, with very good use of references throughout. I maintain my opinion that the strongest part is the regression model, and that this model can be applied very well in the future for scenario analysis within the same study area. I continue to believe that the sustainability map is less interesting, because it basically uses the regression model that was already locally optimised using GWR to calculate a new map, but it is only applicable to the study area, precisely due to the local nature of the regression coefficients. Moreover, it does not fit the original vegetation (Kernel density) map as well as the authors claim based on their validation. Please allow me to start with these two main concerns I would like to see addressed in the discussion.
1. The calculation of R-squared of the GWR model provides very good results. Nothwithstanding, the resulting local coefficients vary largely, from highly positive to highly negative. Moreover, the variations sometimes occur on very small distances. This means that the effect of Ai or groundwater depth on groundwater dependent vegetation can vary from highly positive to highly negative throughout the area and even within very short distances. This seems purely a statistical exercise with apparently little physical meaning and needs to be addressed in the discussion. What does it mean? How then is this method valid and applicable elsewhere?
2. Given the high weight of the aridity index (Ai) in the regression map, the groundwater dependent vegetation (GDV) suitability map now closely follows the Ai categorised map (Fig. B1a), as also mentioned by the authors. The good agreement observed by the authors between the suitability map and the groundwater depth map, in my view is a coincidence, as the groundwater depth map in fact follows the aridity index map. In other words, in the more humid areas the groundwater level seems shallower, and vice-versa. In addition, the GDV suitability map does not show a good correspondence with the GDV occurrence map (Fig. 1), unlike the previous suitability map that was produced in the first version of the manuscript. In the former map (version 1 of the manuscript) soil type was the most important parameter, but that parameter was now taken out. As a direct consequence, the highest GDV density in the central north now occurs in an area of very poor to poor mapped GDV suitability, whereas in the southeastern area the GDV density is very low in an area of very good suitability. I acknowledge that the reality is always more complex and that the authors already refer to this in their discussion, but please also address the issues I have mentioned. The fact that the suitability maps fits well with the NDWI map, could be a logical consequence of the fact that the latter represents moisture content in vegetation. Why would the highest stress be indicative for groundwater dependency? Wouldn’t you expect stress to decrease if the trees have access to groundwater?
Some other comments are given below:
The abstract is well written.
The introduction provides a very good overview on the need of study, but could mention the other work/studies carried out so far in the field. That is currently limited to one sentence (ln 127-129), so that the paper does not show the added value of the implemented methodology as compared to existing studies, some of which are actually referred to later on in the manuscript (e.g. Barron et al., 2014; Condesso de Melo, 2015; Costa et al., 2008; Doody et al., 2017). Therefore, no new references are needed.
In material and methods, section 2.3.1, attributing a low GDV suitability score to soils of high clay content can be debated. Soils of a finer texture will have large extinction depths due to an increased capacity of capillary rise. I would expect coarser soils to have vegetation of lower groundwater dependency. Please briefly elucidate on this aspect.
In the model development (material and methods, section 2.5), how many data points are used (and what is the search radius) for the calculation of local model coefficients?
In section 2.7 of material and methods briefly explain for what purpose the NDWI anomaly map was calculated.
Please explain why you select slope (s) rather than soil thickness (S), if the latter has a higher correlation with principle component axis 2 (PC2).
What happens to R-squared when reducing the set to four or even three variables? Given the large weight of Ai and O4 I would expect the impact to be small. Have you considered using a reduced set? This would largely facilitate the application of the method in other areas.
Other minor comments and technical corrections:
Ln 17: delete the word “scenarios”
Ln 19: delete the words “the density of”
Ln 25: “closely followed”: this is not true. The other three parameters (groundwater depth, drainage density and slope) follow at a large distance, i.e. they are of much lower importance in the regression model.
Ln 28: “relative proportion”. Please briefly clarify what it means. Is it the local coefficient divided by sum of local coefficients? When negative, do you use absolute values (which would make sense)? This needs to be explained in detail in section 2.6 (pg 11 ln 329-341).
Ln 60: include
Ln 61: “subsurface groundwater” seems a pleonasm, although I understand what you mean, when comparing it to “surface groundwater”. Perhaps you could consider using the terms “emerging groundwater” vs. “resident groundwater”.
Ln 62: “a visible source”
Ln 64-65: place the references after GDE
Ln 74: “relying on”, perhaps use “entirely relying on”
Ln 76: “root system”
Ln 115: “rising temperature”
Ln 129-130: rephrase “coefficients proportions”, e.g. to “coefficients as proportion of total sum of absolute coefficients”.
Ln 184: “low drainage capacity”, “high clay fraction”
Ln 325-328: lower drainage density leads to higher suitability, which is correct, but the explanation is incorrect, as the explanation in fact suggests the opposite, or so it seems.
Ln 342-343: I suggest using “representing” instead of the word “referred”.
Ln 402: and in the south?
Ln 422: the maximum value on the map seems much higher than the value indicated in the text (0.714).
Ln 488: I suggest changing to: “poor suitability to GDV, corresponding to”
Ln 572: “did not only allow”
Figure 3: What are the units in this figure?
Figure 4: The reference to the different maps in the figure title is incorrect. Figure 4a is aridity index, not soil type, etc.
Figure 10: I would not use green to indicate highest stress.
Table 4: Values for slope and aridity index are incorrect in the table (the order of the scores 1-3 is inversed, as can be seen in the maps of Fig. B1, which are correct). |