Deriving seasonal dynamics in ecosystem properties of semi-arid savanna grasslands from in situ-based hyperspectral reflectance

Tagesson, T.; Fensholt, R.; Huber, S.; Horion, S.; Guiro, I.; Ehammer, A.; Ardö, J.

doi:https://doi.org/10.5194/bg-12-4621-2015

Articles | Volume 12, issue 15

https://doi.org/10.5194/bg-12-4621-2015

© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

Special issue:

EUROSPEC – spectral sampling tools for vegetation biophysical...

https://doi.org/10.5194/bg-12-4621-2015

© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 12, issue 15

Research article

|

05 Aug 2015

Research article |

| 05 Aug 2015

Deriving seasonal dynamics in ecosystem properties of semi-arid savanna grasslands from in situ-based hyperspectral reflectance

T. Tagesson, R. Fensholt, S. Huber, S. Horion, I. Guiro, A. Ehammer, and J. Ardö

Download

Final revised paper (published on 05 Aug 2015)
Supplement to the final revised paper
Preprint (discussion started on 20 Feb 2015)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC C845: 'referee comment on paper bg-2014-619', Anonymous Referee #1, 26 Mar 2015
- AC C1699: 'Reply to Anonymous Referee #1', Torbern Tagesson, 24 Apr 2015
RC C1716: 'Referee comment on bg-2014-619', Anonymous Referee #2, 25 Apr 2015
- AC C1836: 'Response to Reviewer 2', Torbern Tagesson, 30 Apr 2015

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Torbern Tagesson on behalf of the Authors (06 May 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (06 May 2015) by Micol Rossini

RR by Anonymous Referee #2 (21 May 2015)

Suggestions for revision or reasons for rejection

>>The analysis of effects of varying sun / sensor geometry has been done over 15 days (of which 3 have
been removed) during the peak of the growing season. This misses the highest zenith angles and times
of different vegetation conditions. I suggest to repeat the analysis for other time periods as well to gain
a full picture of sun / sensor geometry effects. Furthermore, why have only NDSIs been investigated
and not the reflectances themselves? This information would help to understand the behaviour of the
NDSIs and would support the claim in the discussion that NDSIs reduce angular effects.

>Response: The reason for not doing the analysis of the varying sun/sensor conditions at the point
in time with the highest zenith angles, is that this occurs during the dry season (two months prior
to the onset of the growing season) where there are no vegetation (herbaceous) influencing the
reflectance spectrum in the measured area. The focus of the manuscript is to investigate how
NDSI is coupled with vegetation parameters, and we hence choose to use the point in time with
most vegetation on the ground.
We agree that it would make a very interesting study to investigate how sun/sensor geometry
influences NDSI differently across the year. However, this is not a minor task and this
manuscript is long as is. We therefore feel that this is beyond the scope of this manuscript. But it
is a very good idea for a future manuscript to investigate seasonal dynamics in anisotropy of both
the reflectance spectrum on its own and on NDSI estimates. This is something that will hopefully
be possible to do in a not too distant future.
The reason for focusing on NDSI, and not on the anisotropy on the reflectance values
themselves is that it has already been done (Huber et al., 2014; Tagesson et al., 2015). The focus
of the paper by Tagesson et al. (2015) is to present all research activities at the Dahra field site.
Among them, a section of the anisotropy of the reflectance spectrum is presented. The aim of the
paper by Huber et al. (2014) is to present the ASD set-up and investigate the quality of the
measurements. A second aim is to study the effects of varying sun/sensor geometry on the
reflectance spectrum. Therefore, in order not to present the same information two times, the
effects of varying sun/sensor geometry part of this paper focus on the effects on the NDSI.
However, the comment is relevant and in the revised manuscript we have included a discussion
regarding the behaviour of the NDSI in relation to the behaviour of the reflectance spectrum and
referred to figures in Huber et al. (2014) and in Tagesson et al. (2015).

Comment: The study uses data from 15
July 2011 until 31 December 2012
. Relationships between ecosystem variables and spectral indices have been investigated for the whole time period. This means phenology from no living vegetation to max living vegetation is included. Therefore it is NOT APPROPRIATE to restrict the analysis of effects of varying sun / sensor geometry to 15 (-3) days during the peak of the growing season.

>> Why has the analysis of the relationship between reflectance / NDSI and ecosystem variables been
restricted to a linear relationship? E.g. other studies found a non-linear relationship between reflectance
and biomass due to saturation effects. Also why have only daily median reflectances / NDSIs been used
when GPP, LUE and FAPAR were daily integrals? Averages would be more appropriate in these cases.
And why have the off-nadir views not been analysed?

> Response: In case the linear relationship is strong, it indicates limited issues with saturation. For
wavelength regions where there are issues with saturations, exponential and logarithmic
regressions could fit better. However, in case the aim is to find wavelength regions which are as
sensitive as possible for investigating seasonal dynamics in an ecosystem property, wavelength
regions with saturtion issues should be avoided. Therefore linear models are better to use than
non-linear models. This was the main reason for fitting linear rather than non-linear regressions.
There is also a practical aspect to it, fitting the reduced major axis linear relationships using the
bootstraping methodology required a full month of processing for these 4 variables (GPP, LUE,
FAPAR and biomass). In case we would try several other regression models, these would require
several months of processing.
Median values were used in order to minimise the influence of errors in the analysis. Median
provides the most common model output and it is thereby more robust against outliers than
average values. This info was provided in the manuscript, but it was not mentioned the first time
that median values were used. Thank you for pointing this out to us, it has been corrected in the
revised manuscript.
We have investigated the seasonal dynamics in the off-nadir views as well, but as seen in the
figure below, there was no difference in seasonal dynamics for the different viewing angles. We
thereby choose to only use the nadir one, as it would not make any difference in the analysis.

Comment: This is not a satisfactory response. Many studies have found non-linear relationships to work much better than linear ones. Therefore, restricting the analysis to linear relationships might not yield the best wavelength combinations.

>> page 3330, lines 11-14: This is not the reason for the saturation of the NDVI. The NDVI saturates at
high biomass because the NIR reflectance is much larger than the red reflectance. NDVI therefore
reduces to R_NIR / R_NIR which equals 1.

> Response: We agree with you, and we are talking about the same thing, we are just using
different phrasing, where you consider it from an equation point of view, we consider it from a
leaf optical property point of view.
All vegetation indices using red will suffer from saturation problems. The reason for this is
related to the fact that there are only so many photons striking a plant leaf and at a certain point,
the chlorophyll absorbs nearly all the red energy to the point where no matter how much
vegetation you add, more photons cannot be absorbed because they are already all absorbed. It is
normally the red band that saturates. So any index using the red energy will suffer from the same
limitation. For example, the Enhanced Vegetation Index (EVI) is not supposed to saturate as
badly because in the equation empirical constants have been added to put more weight in the
NIR spectrum that preserves sensitivity to higher loads of biomass (more layers of leafs) because
here much more radiation is transmitted and reflected from the leaves.

Comment: No, we are not talking about the same thing!!! I say the saturation stems from the specific equation applied (i.e. normalised difference). You say the saturation stems from the red band showing no changes. R_NIR << R_RED leads to NDVI=R_NIR/R_NIR=1. If you use a different index, e.g. the simple ratio R_RED/R_NIR there are not saturation issues if R_RED is small and changes little as long as R_NIR still changes.

Hide

RR by Anonymous Referee #1 (25 May 2015)

ED: Reconsider after major revisions (10 Jun 2015) by Micol Rossini

AR by Torbern Tagesson on behalf of the Authors (10 Jul 2015) Author's response Manuscript

ED: Publish as is (13 Jul 2015) by Micol Rossini

AR by Torbern Tagesson on behalf of the Authors (14 Jul 2015)

Download

Article (3536 KB)
Full-text XML

Short summary

Relationships between ecosystem properties of semi-arid savanna and reflected solar radiance between 35 and 1800nm were investigated. Normalised combinations of reflectance for the near infrared, shortwave infrared, and 600 to 700nm were strongly affected by solar and viewing angle effects. Ecosystem properties of savannas were strongly correlated with reflectance at 350-1800nm, and normalised combinations of reflectance were strong predictors of the savanna ecosystem properties.