Rebuttal by keralaguest

VIEWS: 66 PAGES: 10

									Ref.: RSE-D-11-00332
Understory trees in airborne discrete-return LiDAR data


FIRST REVISION - Rebuttal

The reviewer comments were carefully examined, and major changes to the manuscript were
made accordingly. These included reordering of the structure, and re-doing and writing
everything concerning the sub-topic of transmission loss compensation. All calculations and
modeling efforts were re-done in this sub-topic, as we learned (by following the reviewer
advice to include several parameter optimization criteria) that the feasibility of the loss
compensation is less obvious. Section 2.7. now has this tpic.

We even changed the title – it now indicates the contents better, even reflects the results and
their cause.

An English language check was carried out by a native speaker to the first version and small
changes were also due to this. This revised version was sent to language review in parallel to
the submission to RSE.

In nearly all cases we agreed with the reviewers.

We have added a symbol table at the end of Section 1 to assist the reader. To save space we
started use the symbol I for intensity that is a common term in the text.

The length was 800 lines, and is 745 in this revised version. There was a Figure-numbering
error in the original manuscript with Fig-#s 8 and 9 competely missing. Now there is one more
Figure, and less one Table. The text was shortened across the manuscript, but we also had to
extend the description of the transmission loss modeling, Section 2.7. Original Fig. 5 was
removed. Original Tables 3 and 4 were merged. Original Tables 5 and 6 were removed. New
Table 1 (Symbols and abbreviations) was added, as was new Table 8. New figures are #9
and #10, they were needed in support of the new approach to transmission-loss modeling.

Our response to reviewer comments is written in indented text.



HELSINKI. OCT 5, 2011. ILKKA KORPELA, DOCENT, Academy research-fellow.
Comments from the Reviewers:

Reviewer #1:

General

This article concentrates on the usefulness of the intensity information from airborne LiDAR
sensors for mapping of understory in Boreal Pine stands. The aims are rather generally
stated. In the Abstract it is written "We examined the potential of LiDAR in mapping
understory trees in Scots pine stands". In the introduction L107, it is stated ". explore the
potential of LiDAR in mapping understory trees for forestry applications".

       We have now RQs 1-4, listed at the end of Introduction. Each has its own subsection in
       Section 2, The Experiment. In the Discussion (Section 3), the subsections link to the
       RQs.

A very detailed field survey has been made in two large plots, where the position of each
understory tree has been measured in field with dm accuracy. The position of over story trees
has been measured with photogrammetric techniques, and the canopies has been modeled in
order to allow ray tracing techniques when investigating the effects of laser beams
penetrating the over story before reaching the understory. LiDAR data from four different
acquisitions are available

Among the numerous results that are presented are:
- The probability for an eco from the understory is tree species dependent, but the intensity
signal is not very different between the species.
- The intensity is higher from the ground than from the understory.
- Compensation for the losses in the canopy improves classification accuracies, but only
marginally.
One of the conclusions is also that the different returns should not be added into a single
intensity variable.

The approach is exploratory and a wealth of data is presented, this utilizes the value of the
study, but this approach makes it also difficult to follow which findings that have been
made. Furthermore, the manuscript contains several minor typographical errors. I
recommend that the results should be published but that the manuscript should be entirely re-
written with a maximum of structure in mind. When doing this re-structuring, it is
recommended that the authors consider which the hypothesis are, how the hypothesis are
tested, and which results could be omitted or saved to another paper etc.

       We entirely restructured the manuscript and omitted some results, which are not
       relevant to the four RQs.


Details:

L11-12 and L44-45. It is of course correct that LiDAR penetrates a canopy much better than
passive optical imagery, but since 3D digital photogrammetry now quickly is becoming
operational is the wording that the signal from all layers is mixed in optical data a bit
unnecessary, since it is not true when multi view angle matching has been made on the
optical data (even if the penetration rate not is sufficient). I would suggest a slightly different
wording which describes the advantages of LiDAR without supposing that optical image data
always is 2-dimensional.

       We now have LiDAR as “promising addition/alternative”. Later in the text we highlight
       the nature of volumetric backscattering also for LiDAR.

L15: Mention also the Latin names in the abstract, since abstracts tends to live their own
lives.

       Added (Pinus sylvestris L.)


L30: Write also the full English names.

       We left out the names of the understory tree species in Abstract. We would like to save
       space by using pine, spruce and birch, the short names, after the introduction of the
       scientific names, which are unambiguous.

L33: Specify that the angle count sampling is used in the Finnish NFI, (in many other
countries is callipering used in the NFI).

       Added “Finnish”

L39: You could also mention that data about understory trees provides information of
relevance for habitats for certain species, for example some birds.

       We could not find (from Finland) any studies to refer to in this aspect. However,
       Kuuluvainen (1998) mentions vertical structure of forests and the (implicit) link to birds
       and other Fauna should be evident to the reader. To save space we would not expand
       the “motivating part” of the introduction.

L50: "10 years" or "decade" would be time statements better than 5-8 years. The area based
method has been used operationally in Norway since 2002, and there were successful
research results about 5 years before that.

       Changed to “10 years” to include the earliest works.


L65: The statement "intensity observations are not physical entities" sounds a bit peculiar,
maybe it would be better to write something like: "intensity observations are not scaled to
physical entities"?

       Corrected: “Intensity observations are instrument values,...”.
L70: You might add "and nature conservation" after forestry.

       Our motivation is mainly from forestry, where we can ourselves see the benefits.


L73, 81, 92 etc: Please explain "h", "h-metrics" , "h-histograms", "h-distributions" etc first time
they occurs in the text,

       Height (h) appears now twice in this paragraph, to clarify that that h stands for height.
       h is also in the symbol Table #1.


L81, 82. Incomplete sentence: "Maltamo ."

       Something had happened when editing the text, it now reads:

       “Maltamo et al. (2005) studied seminatural boreal forests. If the h-histogram was bi-
       modal, the canopy was classified as two-layered. Jaskierniak et al. (2011) fitted
       theoretical distribution functions to LiDAR h-histograms to characterize the vertical
       structure of eucalyptus forests.”

L99: Check language.

       The placement of verb, and word “also” changed: “The echo-triggering probability,
       intensity, and transmission losses are also dependent on the geometry of the pulse-
       target intersection, due to the uneven energy distribution within footprints.”

Table 1: Don't include "m" in the first row if units not are included for the other rows. Explain
the site types letters used in row 2, or skip that information in the table.

       “m” removed from 1st row.

       The site type is actually explained by the characterization of the bottom flora,
       this row was removed.
Fig 1. Why separate figures for below and above 14 m?

       It is there for illustrative purpose, to enhance the readability of the h-distributions.

L182-183: The context of the sentence: "We used for illustrations ." is unclear.

       Sentence removed, Caption of Fig. 11 explains the 2004 and 2010 extra data sets.


L 424: Check language: " . an trees ."

       Rewritten “understrory trees”

Fig 2: Align the bars in the upper right panel to the panel below.

       This is Fig 11, now Fig. 7, and they are now aligned.

L605: Consider omitting the statement about a blind zone that precedes an eco.

       Omitted!

L656: Check headline " .. In the understory tree".

       Added “layer”

Reference list: Correct the details, such as brackets around years (L707, 774), ending with
point (L741, 746) etc.

       missing “()” and missing ending points corrected.
========================

Reviewer #2:

This is a useful manuscript with original ideas. It is using a high-positional-accuracy survey of
tree location, size, and crown extent to assign laser echoes to individual trees and then to
explore the potential of echo intensity and position for understorey species discrimination and
understorey/ground echo separation. It quantifies the benefits of modelling echo intensity to
account for energy reduction along the path of a pulse due to noise and transmission losses.
It shows that despite improvements in intensity consistency introduced in the modelling
phase, the overall utility of echo intensity alone as understorey species discriminant is, in the
best case, marginal.

The manuscript contains a complete and current literature review and has no language or
expression issues. There is room for organizational improvement though, given that
identical information is replicated in the text. Information that appears in the results
section would be better, and more naturally, fit in the methods. The discussion
contains a comparison to an earlier study by one of the co-authors that should be
placed much earlier.

      We have now Introduction with RQ1-RQ4. Section 2 is Experiment, and has
      subsections 2.4-2.7 for these RQs. They describe the methods and results, with
      very limited discussion style. Section 3 is Discussion. We moved text from earlier
      Conclusions to Discussion. While doing the restructuring, redundant text could be
      removed. Some redundancy still remains between paragraph 5 of Introduction (9 lines,
      top level) and Derivation of the transmission loss compensation models, lines 375-405
      (detailed) in Section 2.7.1.


Line 12 [abstract]. '... of each hot-spot reflection ...'. Consider excluding the hot-spot
designation. In a previous manuscript you have described the laser systems as 'monoscopic'.
Although this is an absolutely correct description, it may be confused with the equivalent term
in spectral imagery parlance.

       hot-spot view-illumination geometry is equivalent to “backscatter geometry”, which is
       obtained by operating the view-illumination system in monostatic configuration. It is
       also monoscopic, but not in sensu photogrammetry.

       The sentence was removed from Abstract to leave room for more relevant text.

Lines 365 - 370. Paragraph is ambiguous. I am wondering how it is compatible with the claim
of a homogeneous overstory made in line 307. It is later mentioned that the crown foliage
density increased with distance from the stem, and decreased from top of the crown to
bottom. This makes the crown anything but homogeneous. Unless of course you are referring
to a homogeneous overstory at the stand level. In any case a clarification is needed.
       Line 307 says that we assume scatterer reflectance to vary modestly in the overstory.
       Thus, the intensity variation up there would mostly be contributed to variation of
       illumination area and intersection geometry. Modeling of the losses after first echo in
       the crowns was expected to improve by considering horizontal and vertical trends in
       within-crown needle density. We do not perceive contradiction, weak assumptions
       maybe that relate to the ill-posed nature of the problem.

       However, we omitted ray-tracing the trees with needle-density functions. So this
       comment is now obsolete.


Line 378. Need to state the range of i in eq 10. In addition, there are many parameters
embedded into the computation of eq 10, including df, b0, b1, and c0 to c3. This is a large
number of parameters to optimize against via Monte Carlo. Were they and limits to the values
of those parameters within which the optimization of CV was attempted?

       This is now entirely re-done.

I have a couple of concerns related to the science part of the manuscript included below. I am
under the impression that certain parts of the manuscript are very detailed, perhaps
excessively occasionally, while a few others are mentioned very briefly. For those reasons I
suggested that the manuscript be revised to address these concerns.

General comments:

There is often ambiguity as of whether the authors refer to overstory or understorey tree
crowns. These ambiguities need to be clarified. There are too many to mention separately.

       All instances checked!

Background information in lines 69 - 90 should be condenses. The paper is already quite
long.

       Was condensed about 50%

Line 257. The essentials of the Java program that performs the pulse tree crown model
intersection should be described, including the tree model type (raster, vector), and spatial
support.

       More details are now given about the tool logics.

Lines 326 - 335. The entire paragraph needs to be rewritten, perhaps a graphical
representation be given, and an explanation on the rationale for the level scaling (0.5 - 1.5) be
provided. The meaning of penetration class is obvious but the practical part of it is not clear.
The caption of figure 7 should be modified accordingly. The choice of CVpooled minimization
as criterion for the echo intensity consistency optimization is questionable, especially when it
contains parameters used as exponents (c1 and c3 in eq 6). CVpooled minimization will
assign large weights to echoes with extreme intensity values. Given that second and third
echoes contain a lot of noise mostly because of the transmission losses you have
documented well, the Monte Carlo process will be driven primarily by extreme intensity values
which contain a lot of noise, not the signal (species-specific) component. Under the fairly safe
assumption that the mean noise in intensity is stationary within the study
area, the pooled CV will standardize the noise and not the signal. The immediate implication
is that you are not going to see in the model output difference among the various version of
the equations tried. The results presented in table 11 confirm this limitation. It is therefore
advisable to try alternative optimization measures.

It is probably worth stating in the text that, owing to the age of the two stands used in the
study, in the 3 to 5 years between field measurements and the laser data acquisition, crown
width did not change by an appreciable amount.

       This criticism was appropriate and we reformulated the models and their optimization
       criteria. While doing that we have realized the facts (higher degree of ill-posedness
       than originally considered) that are now seen in Sections 2.7.1. and 2.7.2.

       The parameter optimization criteria were extended from CV minimization to constrains
       on the remaining trends and the use of pulses was restricted to a valid range.

       The „selection process‟ is now highlighted. Selection of targets in the presence of
       transmission losses.


Figure 11. You could inform the reader that missing trunk and species categories in the figure
are due to inadequate number of echoes in that category for a particular sensor.

       Figure caption actually implies this already “(> 30 observations”).

Section 4.2. Does the text apply to understorey tree or to all? Modify the caption of figure 11
accordingly.

       Caption changed, added “understory”

Line 482. 'Significantly' implies a statistical test. If you have indeed performed a test please
name the process. Otherwise, you may wish to substitute 'significantly' with 'substantially'.

       Corrected: “Substantially”


Line 517. Which variables are you referring to?

       Table 8 added.

Line 640. I do not agree with this statement in the last sentence of the paragraph. A scatterer
that generates a third echo can be structurally, and in terms of reflectance identical, to one
that generates a first echo or a single echo. As you have stated in the test, it is the relative
position of the scatterer along the pulse's path that makes the difference in recorded intensity.
Hence, you should consider removing the last sentence.

       Our results show that scatterers that produce 2nd (or 3rd) echoes are not similar to
       those producing first-of-many or single echoes. Transmission losses filter the targets:
       the more losses, the higher the cross-section (silhouette area), orientation and/or
       reflectance needs to be to have the signal with proper S/N to trigger an echo.

       Photons pass thru gaps to the understory. If the gaps are large, e.g. due to large
       opaque overstory objects (crowns), the losses occur only at the edges of these objects.
       The probability for the pulse to hit the edge (and the target partially) is higher for larger
       footprints. This results in more pulses subject to losses for the sensor that has larger
       footprint. More pulses in need of compensation, intensity data affected by target cross-
       section variation more than reflectance variation.

       The discussion about radiometry is entirely rewritten.

Section 4.3.2. Which are the parameter values in eq 6 and table 6 that produced the results
shown on table 11?

       Values for the power terms (linearity of the intensity scale) appear now at the end of
       Section 2.7.2.

Typos and grammar suggestions:

Line 70. ... stands shows (that) the accuracy [missing word]
Line 82. eliminate the first 'by'

       Rewritten.

Lines 98 - 99. ... described in intensity data. Complexity is increased by the fact that intensity
does not depend only on target reflectance ... [rewording]

       The placement of verb, and word “also” changed: “The echo-triggering probability,
       intensity, and transmission losses are also dependent on the geometry of the
       pulsetarget intersection, due to the uneven energy distribution within footprints.”

Line 154. ... For each tree the relative height was determined ... [rewording]

       Removed.

Line 176. ... to assure accurate ranging ... 'ranging' may be confused with a distance
correction. Perhaps it should be preceded by 'value'.

       “ranging” seems to be unambiguous (in Wikipedia at least), and we use “range” (R)
       later so it is the best choice here.
Line 233, caption of figure 3. Equation should be #3, not #1.

       Corrected!

Figure 4. You may consider removing the dot marking the center of each circle. It can be
erroneously interpreted as an echo.

       The dots were removed!

Line 311. Eliminate 'by default'

       Removed, redundant.

Line 312. ... implies pulse with high likelihood to penetrate the canopy [rewording]

       The passage was removed.

Line 436. ... while the lowest first of two returns were at h of ~3.8 m. [rewording]

       “. In the ALS50, the low-intensity single-return data reached as low as 2 m in height,
       while the lowest first (-of-two) were at ~3.8 m height”

Line 456. ... corre(c)tion [missing character]

       “c”

Line 482. ... (Table 12). [closing parenthesis]

       Corrected.


Line 591. ... pattern(s) [plural]

       Corrected.

Line 614. ... far the most [add word]

       Corrected “...are by the far most useful”

								
To top