Spatial Application of Southern U.S. Pine Water Yield for Prioritizing Forest Management Activities

Round 1

Reviewer 1 Report

This paper extends a point-scale model of pine forest water yield developed for the SE US to the landscape scale tto guide forest management decision making.  It couples together a variety of climatic, geological, and forest structural data sets to implement the model, and uses the results to help prioritize regions of the Apalachicola National Forest/Tate’s Hell for which pine thinning may yield hydrological benefits in the form of increased total stream flow.  I found the execution and presentation of the work to be excellent, and while I have some critiques and questions below, I believe this will be a valuable contribution to the literature and to the managers of SE forests.  I want to especially applaud the authors for the commitment to transparency and open-source science.

The core idea of the work is that water yield from pine-dominated forests can, using a recently published model, be effectively predicted with three attributes: LAI, water table depth, and climate (aridity).  Several assumptions were necessary to create wall-to-wall estimates, but the authors make these clear, and the workflows are easy to follow, and yield some important and granular management insights.  I have four primary issues which I’ll pose first, and then several secondary questions that I pose in line-number order thereafter.

1.       The original model used leaf area index as the primary predictor of water yield.  The authors chose instead to use basal area (BA), which is modestly correlated with LAI, but was an inferior predictor in the model.  If the association between LAI and BA is linear, then this decision weakens the accuracy of the resulting predictions, but doesn’t alter the core response.  However, the association is generally NOT linear.  In McLaughlin et al. 2013 (cited in the manuscript), the association is clearly not linear (following Michaelis-Menten growth), and many other papers (Gholz and Fisher 1982, Vose et al. 1994, Gonzalez-Benecke et al. 2016) document that pine stand LAI saturates early in stand development, while BA continues to increase.  It’s hard to say how much this matters, but it sounds like the “dense pine” stands scrutinized as WY potential are quite old (i.e., high basal area, but no longer accruing more LAI), which may make the linear extrapolation of LAI from BA problematic.  The authors assert that BA from lidar is easier to measure, and that BA is a more common field measurement in forest measurements.  While the latter is clear, the former is not, particularly since remote detection of LAI is an operational tool.  If I were to speculate on the consequences of this, it would be that the LAI imputed for the dense stands is too high (compared to what’s there), and that the water yield estimates are (slightly) exaggerated.

2.       The original model was developed for upland pine lands.  The spatial extrapolation appears to be for all lands, including those areas too wet to be well predicted by the model.  Since the model predicts variation in water yield due to LAI, there must be some transient water limitation of ET, and wetlands are not venues where this occurs.  That is, wetlands typically achieve PET except under the extremely rare circumstances that the water table drops below the root zone.  The consequence of including wetlands in the spatial domain of WY management is that the model will dramatically overestimate the impacts.  In Fig. 9 (compartment 68) the predictions appear to be confined to regions determined to the “dense pine” which precludes more sparse uplands and non-pine wetlands, but a look at the National Wetlands Inventory for that region indicates that many of the locations that are listed as “dense pine” are palustrine forested wetlands, albeit with the code of seasonal inundation (i.e., very short hydroperiod).  They appear to be pine lands, but the shallow water table may preclude water limitation.  One thing the authors ought to do is mask the landscapes they are working in for wetlands, and at least discuss why these justify inclusion for the water yield predictions.  Since I don’t know that they should not, I cannot insist they be masked, but this should at least be a discussion point.

3.       A sizable portion of the paper is devoted to the temporal variation in aridity.  It’s clear that rainfall variation (and to a far lesser extent PET variation) impacts water yield, but I am not sure that this is relevant to the predictions of forest management decisions, which affect the marginal water yield. This shows up later L490) when the authors assert that water yield increases are biggest in the western compartments; this is not the case because the model used has no interactions, so the marginal effect of LAI reduction is the same regardless of the aridity of groundwater setting.  The effects of aridity variation between wet and dry years becomes relevant if, and in my view only if, the aridity is non-stationary (e.g., with climate change).  If the model is short term, over which aridity can be conceived of as variation around a stationary mean, then the fact that wet years have high WY and dry years have low is not germane to how forest management should change.  For me, this simplifies Fig. 4, Table 2, and eliminates Fig. 8.  I also have questions about the content of Table 2. The difference between “wy_18_xxxx” and “wy_11_xxxx” (where “xxxx” refers to mean, max or min) is always the same change in LAI.  As such, the water yield difference in cm should be the same.  It’s not…it’s close, but it’s not identical.  This may be because some existing stands are below BA = 18, but above BA = 11, but this is not clear.  Also, the precision of the WY predictions does not justify 4 significant figures. Probably everything after the decimal points is irresolvable uncertainty.

4.       The model used is empirical, based on and designed for stand level inferences.  The authors acknowledge that they are extrapolating to landscapes, and I think that’s a justified (and very interesting) thing to do.  That said, the mean water yields in Table 3 are implausibly high for a flatwoods landscape (nearly double the water yield estimated using stream flow).  The authors also point out (L428) that this sort of water mass balance check on the water yield is not possible here because of data constraints.  But it’s still worth exploring whether these yields are plausible based on landscape-level water yields in similar landscapes.  There is a lot that can happen between the stand and the sea, and one explanation for the discordance between these numbers and the stream flow is that ET is augmented elsewhere in the landscape, which begs the question of whether the water yield predicted to occur from density reduction will translate into flow where it’s wanted (e.g., in a stream or aquifer).  This could be explored and discussed without undermining the conclusions of the paper. 

 

 

Additional questions:

- The introduction is trying to connect two related ideas, but could use some smoothing.  The first is that forests are important protectors of drinking water sources at national scales.  I don’t think it’s fair to imply (L52) that forests yield more water per unit area than other land cover types since most of that inference occurs because forests don’t occur where rainfall is low (and thus WY is low).  I also am not convinced that forests that do protect drinking water have relatively high water yields, just that they are in places with abundant rainfall (often at elevation), and that they dramatically protect water quality.  In nearly every study in every biome (with the exception of cloud forests and some snow-melt dominated alpine forests) the reduction in forest biomass increases water yield.  I’d argue that forests can both be integral to protecting US water supply (because they are located in wet places, and protect water quality), and be subject to management that increase the volume of water.  As I read the introduction (L49-68) I got the sense that the authors were trying to navigate between those two contentions and the prose got confusing.

- L80 While I like that paper, I think it’s not fair to say that it has any empirical insights about water yield from forests; their models may suggest that, but they have no new measurements of it.   

- L159 The idea of using the model to assess the hydrological impacts of Hurricane Michael is quite an excellent idea.

- L175 suggests that there were 6 forests.  I would argue that there were actually 30, with 6 plots (quite distinct forest attributes) in each of 5 sites.

- It’s not clear to how the algorithm discriminates “pine basal area” from “total basal area” (L195).  The efficacy of this parsing of biomass should be explored, and if there is basal area that is not pine in the stands, the effects of that biomass on water yield should be considered.  The model used has sites with hardwood encroachment, and there was no effort to excise that hardwood LAI from the total LAI for purposes of WY predictions.

- Fig. 2 should show the distribution of LAI as well as BA, since the former is what’s actually used in the model.  Fig 2 could also show the extent of NWI wetlands.

- Fig. 3 current retains shallow vs  deep as synonymous with confined vs. unconfined.  Since the workflow to get the water table depth was not exclusively hydrogeologic confinement, I recommend deleting the “confined vs. unconfined”

- The words “in situ” are used often throughout the paper.  I am not sure this is the correct use of this term.  The model was developed from in situ data, but the model itself is not applied in the original place (or in the field).  Rather it is an empirical association that is extrapolated from in situ observations.  Nit picking, I know.

-  L327 to 334 can be downplayed if the authors agree that temporal variation in aridity does not affect the estimates of marginal water yield change, only the absolute values of water yield.  As such, this prose and some of the content of table 2 can be summarized as ranges without sacrificing any meaningful content about the WY changes from forest management.  In the discussion where the authors consider climate change, the temporal trends in aridity are relevant.

- The implementation of the BA scenarios is useful, but also slightly confusing.  It would be helpful to report the proportion of each target basin that is over each BA threshold.  FWIW, the WY values in Table 3 are reported with what appears to excessive precision, so perhaps the information about the proportion of pine stands that exceed each BA threshold could be included parenthetically next to each WY estimate.

- Delete Fig. 8.  It’s not helpful.  The patterns are the same as the mean, so what it conveys, to me, is that wet years mean more flow and dry years less.  I’d add that the assertion that we need water yield predictions over the entire range of climate conditions is relevant only if aridity has a trend.  Otherwise, it’s variation around a mean, and the absence of an interaction between LAI and aridity (which makes sense to occur, but is not in the model), there is no reason to expect variation in the marginal water yield.  (This is more or less point #3 above)

- Somewhere make clear (e.g., L400) that any management actions need to be sustained.  Thinnings that allow LAI regrowth will subsequently inhibit water yield, so to ensure persistent WY increases, the forest management decisions need to be thought of as a new regime.

- Throughout the document the authors use total volumes of water (m3 per year) to report impacts.  This is justified, and makes the totals sound more impressive, but the scaling of that number is only possible when it’s indexed to the area over which that increased yield is sourced.  For example, the compartment 68 site (Fig. 9) has an area, and the increased water yield over that entire area is a measure of the added flow that would accrue.  Unless there are some minimum flows and levels targets that specifically identify the augmented flow needed at a specific location, reporting only volumes (not WY depths) is a bit hard to really understand.

- L456-460 – this is a really cool utilization of the model!

- In Fig. 9 many of the “dense pine” are wetlands.  Can you overlay the NWI polygons and justify why the model predictions should include sites where the water table often goes above ground?  The three target sites on L472 all sound pretty wet.

- The authors posit (L507) that using LAI from lidar point clouds would require parsing between pine and non-pine.  I am not sure this is true.  The authors haven’t suggested that this is require for basal area estimates, nor was is the case that no hardwoods contributed to LAI in the original model (which included several mixed stands).  I think it’s fairer to say that the stands should be predominantly pine, not that the model needs to parse the biomass (which would then create the extremely odd situation of reporting pine water yield, but neglecting hardwood water yield from the same site).

 

 

Finally, I have a few operational questions:

a. How often is the BA data set updated, and thus how well can one look back in time to assess the changing forest mosaic?  If the answer is “never” and “not well”, then I wonder if that’s an argument to LAI inferences from satellite imagery rather than basal area.

b. How readily can the model be switched to monthly water yield?  This is relevant because there is strong evidence of annual variation in LAI, as well as known variation in rainfall and ET.  A more finely resolved representation of the input variables may be important to resolving the mismatch that seems apparent between point and landscape scale water yield estimates.

c. It would be helpful to make clear what a prospective user needs to run the model.  I didn’t actually try the code, but the data repository had the aridity values and the water yields, but not the BA file (nor the recurrence attributes of those data) or the water table depth. 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

 This paper created a practical workflow for creating high resolution, landscape-level spatial water yield data from an in-situ water yield model. I think that this manuscript is of great value with Forest management and should be published after substantial revision. Several aspects need more details and I would like to share with my specific comments and suggestions below:

 

1) It is mentioned in the introduction that LAI is the most directly related index of evaporation and transpiration. Is there any index related to the comparison of evaporation and transpiration? Please give a brief description in the introduction.

2) In 2.2 collected a number of data such as soil texture, soil moisture, etc. Are these data related to evaporation? The relationship between these data and WY could be listed to prove the rationality of choosing the LAI, DTW and ARID.

3) In Table 1, the time information of the data set could be given to prove the time consistency of the data selected in the experiment.

4) The water yield models of other researchers are mentioned in this paper, which can be compared with other models to reflect the advantages of this model.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

1. Title: Authors must improve title to capture study area (Florida Panhandle?)

2. Objectives: Poorly stated. It is better to list objectives explicitly. See line 129-142.

3. Lidar & ESRI are acronyms and must be so recognized throughout the text

4. Introduction and Discussion sections: Too long and laborious to read. Authors must demonstrate concise presentation for readers. They must cut several repetitions about the importance of the WY flow process for decision support.

5. Line 514-524 discusses weaknesses of methodology, but there is no mention of weakness in the methodology section itself. This is so important because application of water balance equation (dS = R-PET) is too crude and subject to serious hydrological errors over a large study area of 1.9m ha. 

6. Authors must find a way of addressing ground water flow and surface runoff in their WY model. If not, they must discuss key assumptions and feed that into error dynamics in the discussion section of the article.

7. Authors must explicitly specify the resolutions (temporal and spatial) of the remote sensing data (PRISM & MODIS) used in calculating aridity index.

8. Mixing of different datasets at different temporal and spatial resolutions makes harnessing a composite dataset difficult. Authors must show explicitly how that was achieved.

9. Poor citation of literature in the text: Acharya et al. [X]

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf