close

Вход

Забыли?

вход по аккаунту

?

07038992.2017.1394181

код для вставкиСкачать
Canadian Journal of Remote Sensing
Journal canadien de télédétection
ISSN: 0703-8992 (Print) 1712-7971 (Online) Journal homepage: http://www.tandfonline.com/loi/ujrs20
Can Polarimetric Radarsat-2 images provide
a solution to quantify non-photosynthetic
vegetation biomass in semiarid mixed grassland?
Zhaoqin Li & Xulin Guo
To cite this article: Zhaoqin Li & Xulin Guo (2017): Can Polarimetric Radarsat-2 images provide a
solution to quantify non-photosynthetic vegetation biomass in semiarid mixed grassland?, Canadian
Journal of Remote Sensing, DOI: 10.1080/07038992.2017.1394181
To link to this article: http://dx.doi.org/10.1080/07038992.2017.1394181
Accepted author version posted online: 25
Oct 2017.
Submit your article to this journal
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=ujrs20
Download by: [UNSW Library]
Date: 26 October 2017, At: 06:36
ACCEPTED MANUSCRIPT
Can Polarimetric Radarsat-2 images provide a solution to quantify non-photosynthetic vegetation
biomass in semiarid mixed grassland?
Zhaoqin Li and Xulin Guo*
Department of Geography and Planning, University of Saskatchewan, 117 Science Place,
Saskatoon, SK, S7 N 5C8, Canada
Downloaded by [UNSW Library] at 06:36 26 October 2017
*Corresponding author: [email protected]
La quantification de la biomasse de la végétation non photosynthétique (VNP) dans les prairies mixtes semi-arides
par moyen de télédétection optique présente un défi. Cette difficulté est due à l’effet combiné de la végétation
photosynthétique (VP), la croûte de terre biologique, et le sol nu sur les spectres de la canopée. Radarsat-2 offre
un nouveau moyen de quantifier la biomasse de la VNP. Cette étude examine la capacité des images Radarsat-2 en
mode fin et quad-pol à quantifier la biomasse de la VNP et la biomasse aérienne dans les prairies mixtes semiarides. Les paramètres suivants ont été utilisés : l’indice de végétation radar (IVR), le rapport de polarisation croisé
(HH/VV), le rapport de dépolarisation, le composant de décomposition de Cloude et Pottier (entropie et l’angle
alpha) et les composants de décomposition Freeman-Durden (volume, surface, et diffusion multiple). Les
meilleures estimations de biomasse de la VNP (r2 = 0,70 et REQM relative = 9%) et de biomasse aérienne (r 2 = 0,51
et REQM relative = 8,4%) s’obtiennent en utilisant un rapport de polarisation croisé de l’image FQ23 (41.9°-43.3°)
au milieu de la saison de croissance. Avec un rapport de dépolarisation de l’image FQ4 (20.9°-22.9°) à la hauteur de
la saison de croissance, des valeurs de r2 = 0,65 et REQM relative = 12,6% ont été obtenus pour la biomasse de la
VNP, et r2 = 0,70 et REQM relative = 8,4% pour la biomasse aérienne.
ABSTRACT
Quantifying non-photosynthetic vegetation (NPV) biomass using optical remote sensing in
semiarid mixed grassland is challenging. This is due to the combined effects of photosynthetic
vegetation (PV), biological soil crust (BSC), and bare soil on the canopy spectra. Radarsat-2
provides a new way to quantify NPV biomass. This study investigated the potential of fine quadpol Radarsat-2 images for quantifying NPV biomass and total aboveground biomass in semiarid
mixed grasslands. The parameters used were Radar Vegetation Index (RVI), co-polarization ratio
(HH/VV), cross-polarization ratios (VH/HH and VH/VV), de-Polarization ratio, the Cloude and
Pottier decomposition component (Entropy and Alpha angle) and the Freeman-Durden
decomposition components (volume, surface, and multiple scattering). The best NPV and total
aboveground biomass estimations are achieved with an r 2 of 0.70 and 0.51 and relative Root
Mean Square Error (rRMSE) of 9% and 8.4%, respectively, using the VH/VV cross-polarization
ratio of the FQ23 (41.9°-43.3°) image in the middle growing season. The r 2 values are 0.65 and
1
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
0.70 and the rRMSE are 12.6% and 8.4%, respectively, for NPV and total biomass estimation
Downloaded by [UNSW Library] at 06:36 26 October 2017
using the depolarization ratio of the FQ3 (20.9°-22.9°) image in the peak growing season.
2
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
INTRODUCTION
Non-photosynthetic vegetation (NPV) includes aboveground standing dead vegetation and plant
litter on the surface (Guerschman et al. 2009). There is a substantial amount of NPV in grassland
ecosystems where grazing intensity is light to moderate and fires are infrequent (Li and Guo
2016). Globally, NPV, as a carbon pool and wildfire fuel, influences carbon sequestration,
greenhouse gas and black carbon emissions, and thus affects climate and altering ecosystem
structure and function (Li and Guo 2016). Locally, NPV adjusts light, heat, and water transfer
between the land surface and the atmosphere, while decomposed NPV supplies soil nutrients,
Downloaded by [UNSW Library] at 06:36 26 October 2017
affecting the structure, diversity, and productivity of ecosystems (Li and Guo 2016). NPV also
serves as a wildlife habitat (Davis 2005, Huang et al. 2009, Fisher and Davis 2010). Therefore, it
can be an indicator of grassland ecosystem health.
Research on quantifying NPV using remote sensing data has been motivated by its ecological
importance. However, most research has focused on the fractional cover of NPV in savannah
(Guerschman et al. 2009, Jackson and Prince 2016), cropland (Daughtry et al. 2006), shrubland
(Asner and Heidebrecht 2003), grassland (Xu et al. 2014, Smith et al. 2015), and forest (Roberts
et al. 1993). Although the fractional cover of NPV is a measurement of abundance, NPV biomass
is a more ideal measurement (MacDonald et al. 2012). A few studies have been carried out to
quantify NPV biomass in grasslands using optical hyperspectral data, including the studies
conducted in Amazon pasturelands (Numata et al. 2008) and inner Mongolian steppe (Ren and
Zhou 2012). These studies have been successful to an extent. However, optical remote sensing of
NPV biomass becomes difficult when canopy structure is complex (Numata et al. 2008) or the
fraction of green vegetation is higher than 30% (Daughtry et al. 2004, Ren and Zhou 2012).
Jacques et al. (2014) had shown the potential of MODIS data for quantifying dry biomass in
pastoral Sahel in the Gourma region of East Africa. Contrary to this, a study in the Canadian
mixed prairie (Cihlar 2012) has shown that multispectral data, including Landsat TM and
MODIS, cannot provide a solution for NPV biomass estimation. Our recent research suggests
that advanced multispectral images, including images from Landsat 8 OLI and Sentinel-2, can be
used for quantifying NPV biomass in semiarid mixed grasslands (Li and Guo submitted).
However, photosynthetic vegetation (PV), biological soil crust (BSC), and bare soil on canopy
3
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
spectra still provide challenges when measuring NPV biomass (Smith et al. 2015, Li and Guo
2016). On other hand, the acquisition of optical images for retrieving NPV biomass is hindered
by the presence of clouds, haze, and precipitation etc. In this regard, the Synthetic Aperture
Radar (SAR) data provide an opportunity.
Unlike optical images, SAR images can be acquired under all weather conditions, although
factors such as clouds, precipitation, and wind may exert an influence on the interpretation of
SAR data focusing on the land and sea surface (Danklmayer et al. 2009, Alpers et al. 2016).
Potential of SAR data for quantifying NPV has been demonstrated in croplands using field
Downloaded by [UNSW Library] at 06:36 26 October 2017
measurements (McNairn et al. 2001) and TerraSAR-X images (Pacheco et al. 2010), and the
forests of West Africa using ALOS-1 PALSAR data (Carreiras et al. 2012). The application of
C-band dual-pol Radarsat-2 imagery for NPV biomass estimation in Canadian mixed prairie
yielded an r2 of 0.30 (Finnigan 2013). However, the advantage of quad-pol Radarsat-2 images
was not investigated. The effects of incidence angle and polarization of Radarsat-2 images on
NPV biomass estimation were also not explored. As Radarsat Constellation Mission (RCM) will
be fulfilled in 2018, high temporal and spatial resolution of Radarsat-2 images will be a valuable
asset for ecosystem monitoring (http://www.asc-csa.gc.ca/eng/satellites/radarsat/).
Fully polarimetric SAR data have demonstrated potential in differentiating crops, croplands and
grasslands, and macrophyte species. For example, the Polarimetric L-band ALOS-1 PALSAR
data identified diverse macrophyte species in the Amazon floodplain wetlands (Sartori et al.
2011). TerraSAR-X images have demonstrated advantages over C-band Radsarsat-2 images for
identifying crops (McNairn et al. 2009). The research of Li et al. (2012) showed the superiority
of polarimetric decomposition over the linear polarization for rice mapping using C-band
Radarsat-2 images. The study of McNairn et al. (2009) also concluded that polarimetric
decomposition is superior to linear polarization for identifying crop types. The Freeman-Durden
classification of Radarsat-2 images could identify native grasslands from croplands, but had
difficulty in separating native grasslands from improved grasslands (Smith and Buckley 2011).
These studies contributed greatly to agriculture management and environment conservation.
Nevertheless, limited research was conducted to investigate the potential of fully polarimetric Cband SAR images to quantify NPV biomass in grasslands.
4
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
In this study, we investigated the potential of quad-pol Radarsat-2 images for quantifying NPV
biomass in semiarid mixed grasslands characterized by large amounts of NPV, PV, BSC, and
bare soil. Specifically, we determined the most suitable quad-pol Radarsat-2 image for
quantifying NPV biomass and explored the optimum SAR parameter (s) for NPV biomass
estimation. The potential of Radarsat-2 images for quantifying total aboveground biomass was
also investigated. This study was conducted based on the hypothesis that SAR parameters (e.g.,
volume scattering) that are sensitive to change in canopy vegetation are useful for estimating
standing dead vegetation biomass, while SAR parameters sensitive to change in ground surface
Downloaded by [UNSW Library] at 06:36 26 October 2017
are useful for quantifying plant litter on the surface. SAR parameters sensitive to canopy
vegetation were retrieved and applied to NPV and total biomass estimation, as were the
parameters sensitive to ground surface.
STUDY AREA
This study was conducted in the west block of Grasslands National Park (GNP, 49.10°N, 106.89
°W) (Figure 1). GNP, as a fragment of northern mixed grass prairie, has been extensively studied
because it is a refuge for native and endangered species, and is at the northern edge of
continental C4 grass species (Li and Guo 2012). GNP was established in 1984, and between
establishment and 2006 there was no large herbivore grazing in the west block. In 2006, for
conservation purposes, bison were introduced into the west block. Despite this, a substantial
amount of NPV is still visible due to low grazing intensity and low frequency of natural and
prescribed fire.
GNP is in a region of continental, semiarid climate with hot summers and cold winters. The
mean annual temperature is 3.8 ºC and the average of the annual total precipitation is 347.7 mm,
based on the climate records of Environment Canada from 1971 to 2000. Most of the annual
precipitation is from evening storms in May and June. It is worth noting that low soil moisture
content is a typical climatic feature of GNP (Wang and Davidson 2007). Precipitation is the
dominant factor limiting vegetation growth (Li and Guo 2012). GNP has rolling topography with
elevations ranging between 750 m and 905 m. Chernozemic and solonetzic soils are the two
major soil types in the park (Fargey et al. 2000).
5
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Vegetation in GNP was classified into upland, valley, and slope communities based on
topography, with disturbed (invasive species) communities identified independently. The
dominant upland vegetation communities are speargrass – blue grama (Stipa comata – Bouteloua
gracilis) and western wheatgrass – sedge (Agropyron smithii – Carex sp.) (Li and Guo 2014).
Valley vegetation communities mainly consist of western wheatgrass and silver sagebrush
(Agropyron smithi – Artemesia cana) with shrubs and occasional trees along the Frenchmen
River (Li and Guo 2014). The main disturbed communities are occupied by crested wheatgrass
(Agropyron cristatum) and smooth brome (Bromus inermis). Typical vegetation communities are
Downloaded by [UNSW Library] at 06:36 26 October 2017
shown in Figure 2.
Vegetation phenology in GNP is highly responsive to climate change. Extracted from AVHRR
NDVI images acquired from 1985 to 2007, green-up in GNP generally starts in mid-April to
mid-May), middle growing season in middle May to mid-June, peak growing season in late June
to mid-July and senescence begins in late July to early August (Li and Guo 2011).
DATA
Biophysical Data
Biophysical data collected for this study are biomass data and fractional cover of green grass,
forb, shrub, standing dead, plant litter on the surface, moss, lichen, rock, and bare soil. Biomass
data collection was carried out from June 20 to July 02, 2014. Fourteen sites were identified
using a stratified random sampling approach, including four upland, three valley, five sloped,
and two disturbed communities (Figure 1 a). At each site, the aboveground fresh biomass was
harvested within a 20 × 50 cm quadrat frame at 20 m intervals over two 100 m long transects
crossing at right angles (Figure 1b), yielding eight biomass quadrat sampling. Within each
quadrat, all the aboveground plant mass including green vegetation, NPV (standing dead
vegetation and plant litter on the surface), and moss and lichen were collected. The collected
samples were referred to as fresh biomass samples. The fresh biomass from each quadrat frame
was taken back to the laboratory and sorted into green grasses, forbs, shrubs, and NPV including
dry moss and lichen, then dried at 60 °C for 48 h in an oven. The dry biomass at each site was
averaged to avoid spatial autocorrelation for data analysis.
6
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Fractional cover of green grass, forb, shrub, standing dead, plant litter on the surface, moss,
lichen, rock and bare soil was visually estimated at a sum of 100%. Each estimation was within a
50 cm × 50 cm quadrat that was placed at 10 m interval over two 100 m long transects crossing
at right angles, yielding 20 quadrat measurements within one site. The general rule used for
cover estimation is that cover was estimated at a 5% increment when the fractional cover is
larger than 5% and smaller than 95%, or the cover was evaluated at a 1% increment.
Table 1 summarizes the descriptive information of PV biomass including dry green grass
biomass, green forb biomass, shrub biomass, NPV biomass, and total dry biomass at upland,
Downloaded by [UNSW Library] at 06:36 26 October 2017
slope, valley, and disturbed sites. The difference in vegetation present in these sampling sites is
shown in Figure 2. The different vegetation in upland, slope, valley, and disturbed sampling sites
resulted in high biomass variations (Table 1), indicating a good representative sample of sites for
the study area. The percentage of NPV biomass ranges from 59% of total dry biomass in valley
sites to 81% of total dry biomass in disturbed communities. The average NPV biomass in upland
is 386.5 g/m2, accounting for 70% of total aboveground biomass.
Table 2 presents the percentage of grasses, forbs, shrubs, standing dead vegetation, and plant
litter on the surface, and lichen, moss, rock, and bare soil sampled in peak growing season of
2014. Percentage of green grasses, standing dead vegetation, and plant litter on the surface vary
from 36% to 49%, 15% to 26%, 9% to 34%, respectively, in disturbed, slope, upland, and valley
vegetation communities. The small percentage of plant litter in valley is attributed to the natural
fire in April 2013. The fractional cover of moss in slope and upland is high as 15% and 10%,
respectively. Fractional cover of forbs and shrubs is smaller than 10%. Percentage of bare soil is
small in disturbed, slope, and upland, while it is high as 15% in valley vegetation communities.
Our previous study indicated that from early growing season to peak growing season, as
factional cover of green vegetation increases, the proportion of NPV, BSC, rock, and bare soil
correspondingly decreases (Li and Guo 2011). The change in fractional covers implies that
influence of green vegetation and backgrounds on quantifying NPV biomass using SAR data is
different at different growing stages. It is worth noting that NPV biomass remains unchanged
from the early growing season to peak growing season.
7
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Synthetic Aperture Radar (SAR) Data and Preprocessing
In total, this study analyzed 12 Radarsat-2 fine quad-pol single look complex (SLC) images
acquired from June 2 through July 09, 2014 (Table 3). The incident angles of the images ranged
from 18.54° to 46.5°, and the spatial resolution was 5 m. Temperature, dewpoint temperature,
wind speed, and precipitation around the acquisition time, as well as precipitation within three
days before acquisition at Val Marie, Saskatchewan (Table 3) were downloaded from the
Environment Canada website to check the effects of wind, dew, and rain on the data quality.
Based on the environmental data (Table 3), the quality of those images was not directly
Downloaded by [UNSW Library] at 06:36 26 October 2017
compromised by dew or rainfall. However, the wetness of the canopy, surface litter, and soil
moisture caused by heavy precipitation may exert influence on the June 15 FQ23 image, the June
18 FQ12 image, the June 19 FQ5 image, and the June 28 FQ3 image.
The Radarsat-2 images were orthorectified based on the Radar Specific Model in Radar Ortho
Suite, an add-on in PCI Geomatica 2015. The Digital Elevation Model (DEM) used for
orthorectification was ASTER Global Digital Elevation Model Version 2 (GDEM V2). The
digital number (DN) of Radarsat-2 images was converted to backscatter coefficient through
sigma-nought   ) (cibration. A 5 × 5 boxcar filter was applied to reduce speckle noise of the
orthorectified Radarsat-2 SLC images. The boxcar filter can be applied to both detected and SLC
data to reduce speckles through averaging the covariance (or coherency) metrics of neighboring
pixels (Lee et al. 2015). In this study, boxcar filtering was applied to the orthrectified Radarsat-2
SLC data using PSBOXCAR algorithm with 5 × 5 pixel size in PCI Geomatica. This step
increased the estimated number of looks from a single look to multiple looks, meeting the
requirements of algorithms for performing the Cloude and Pottier decomposition (Cloude and
Pottier 1997) and Freeman-Durden decomposition (Freeman and Durden 1998). The boxcar filter
is the most commonly used algorithm when there are no distinct features on the image and when
there is no concern on preservation of spatial resolution (Lee et al. 2015). It is effective in
reducing speckle for forest and cropland biomass estimation (Wiseman et al. 2014, Lee et al.
2015). The boxcar filter was used in this study because a comparison using our data indicated
that the images filtered with the boxcar approach had a better estimation on biomass than those
8
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
filtered by the Lee adaptive filter approach. A flow chart (Figure 3) was created to demonstrate
procedures of image pre-processing, data retrieval, and data analysis.
RESEARCH METHODS
Scattering Mechanism
The Cloude and Pottier decomposition (Cloude and Pottier 1997) was applied to the Boxcar
filtered Radarsat-2 SLC images to explore the scattering mechanism of the sampling sites at
different imaging incidence angles. The Cloude and Pottier decomposition calculated the
Downloaded by [UNSW Library] at 06:36 26 October 2017
eigenvalues of the covariance or coherence matrix of the image to obtain the entropy (H,
between 0 and 1) and the anisotropy (A, between 0 and 1), and parameterized each eigenvector
in terms of four angles, including the alpha angle (0°-90°) (Cloude and Pottier 1997). The
entropy (H) indicates the degree of mixing between surface, volume, and double bounce
scattering and the anisotropy (A) is dependent on the ratio between probabilities based on second
and third eigenvalues (Cloude and Pottier 1997). The alpha angle depicts the scattering
mechanism of the eigenvector. The alpha angle (0°, 45°, and 90°) indicates a trihedral scatter (a
smooth surface), a dipole scatter, and a dihedral scatter respectively. An entropy-alpha plane was
used to demonstrate the scattering mechanism of the sampling sites.
NPV Biomass Estimation from SAR Polarimetric Data
The asymmetric matrix of each filtered Radarsat-2 SLC image was converted to a symmetric
matrix prior to the Freeman-Durden decomposition. The Freeman-Durden classification
decomposes the total backscatter into the contribution of volume scattering (dipole scattering),
double bounce (dihedral scattering), and surface scattering (Bragg scattering) (Freeman and
Durden 1998). Co-polarization ratio (HH/VV) and cross-polarization ratios (VH/HH and
VH/VV) were generated. HV/HH, HV/VV were not analyzed as HV and VH backscatter are
similar (Moran et al. 2012 a). Depolarization ratio  xv  hich is sensitive to soil surface roughness
(Ulaby et al. 1986, Gherboudj et al. 2011) was calculated from e.q. (1):
xv   vh  dB    vv  dB  (1)
9
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Where  vh nd  vv re the VH cross-polarization and VV co-polarization backscatter coefficients in
decibel (dB) respectively. Radar vegetation index (RVI) (Kim and van Zyl 2009) was
characterized by the ratio of cross-polarization backscatter to the total scattering (e.q. (2)).
RVI 
 HH
8 HV
(2)
  vv  2 HV
Where σ HV s the cross-polarization backscattering and σ HH and σ vv re the co-polarization
backscattering in power unit. The RVI is sensitive to biomass variation, but less affected by
environmental conditions such as soil moisture (Kim and van Zyl 2009). The value of RVI is not
Downloaded by [UNSW Library] at 06:36 26 October 2017
only determined by vegetation condition, but also controlled by the incidence angle of the radar
images, because the increase of incidence angle will increase the path length of the radar pulse
through the vegetation canopy (Kim et al. 2012).
SAR parameters, including RVI, co-polarization ratio, cross-polarization ratios, depolarization
ratio, the Freeman-Durden decomposition components, and Entropy and Alpha angle of the
Cloude and Pottier decomposition, were retrieved within a 19 × 19 pixel window size to match
the 100 × 100 m sampling site. The retrieved SAR parameters were individually averaged within
each site to correlate with the biomass data within the site. Before the analysis, outliers of the
biomass data were checked at the quadrat level using SPSS. Outliers were detected based on
upper threshold value (e.q.(3)) and lower threshold value (e.q. (4)) (Hoaglin and Lglewicz 1987).
Values larger than the upper threshold or smaller than the lower threshold were identified as
outliers. Values identified as outliers based on the criteria were doubled checked with the photos
taken at the quadrats. After checking, one disturbed plot with over 90% alfalfa and one upland
plot with apparently low biomass measurement were excluded from analysis. The sample
number for analysis of most images is 12, expect for the July 5 image that only covers 10
sampling sites (with one upland and one disturbed community exclusive). Since plant surface
litter and moss and lichen are a significant amount of NPV biomass, all the derived radar
parameters, including surface scattering, were investigated. Accuracy of NPV biomass and total
aboveground biomass estimations was quantitatively measured using leave-one-out crossvalidation. Relative Root Mean Square Error (rRMSE) was used to measure the accuracy.
10
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Utvalue  value @75% perc  2.20   value @75% perc  value @25% perc  (3)
Ltvalue  value @25% perc  2.20   value @75% perc  value @25% perc  (4)
Where Utvalue s upper threshold value, Ltvalue s lower threshold value, and perc s abbreviation of
percentile.
RESULTS
Scattering Mechanism
Figure 4 demonstrates the scattering mechanism of the entropy-alpha planes of the June 2 FQ1
Downloaded by [UNSW Library] at 06:36 26 October 2017
and the June 8 FQ27 Radarsat-2 images. The environmental effects on the quality of the two
images are negligible (Table 3). Therefore, variations in the scattering mechanism can be
considered a result of different incidence angles. Shown by the scattering mechanism plot, the
FQ1 image on June 2 is dominated by smooth surface backscattering, while the FQ27 image on
June 8 is characterized by rough surface scattering and volume scattering as a result of low
penetration capability through the canopy. Although vegetation growth from June 2 to June 8
may contribute to the volume scattering of the FQ27 image, the dramatic decrease in the surface
scattering of the FQ27 images on June 8 was more likely a result of a shallow incidence angle.
The scattering mechanism explored the capability of the images with steep incidence angles to
quantify the surface plant litter portion of biomass, and the ability of the images with shallow
incidence angles to quantify standing dead vegetation biomass.
Relationship between Radarsat-2 Response and NPV and Total Aboveground Biomass
The relationships between NPV biomass and various Radarsat-2 parameters are presented in
Table 4. The largest r2 value for quantifying NPV biomass is 0.70, achieved by using the crosspolarization ratio (VH/VV) calculated from the June 15 FQ23 image. The volume scattering of
the June 08 FQ27 image and the depolarization ratio of the June 15 FQ23 image have an r2 value
of 0.69 for NPV biomass estimation.
The cross-polarization ratio (VH/VV) and the de-polarization ratio calculated from the VH and
VV bands of all the images, except for the FQ10 images on June 12 and July 06 and the July 09
FQ23 image, yield better NPV biomass estimations than RVI, co-polarization ratio (HH/VV) and
cross-polarization ratio (VH/HH). However, performance of VH/VV and de-polarization ratio
11
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
changes as incidence angles of the images and vegetation phenology change. From middle
growing season (mid-May to mid-June) to peak growing season (late June to mid-July), NPV
biomass remains unchanged. From Table 4, in the middle growing season, the VH/VV ratio and
de-polarization ratios derived from the shallow incidence angle FQ23 image outperform those
calculated from the steepest incidence angle FQ1 image and the shallowest incidence angle
FQ27 image. The medium incidence angle FQ10 and FQ12 images have the worst performances
for NPV biomass estimation. In the peak growing season, the VH/VV and de-polarization ratio
of the June 28 FQ3 image are superior for quantifying NPV biomass, compared to those
Downloaded by [UNSW Library] at 06:36 26 October 2017
calculated from incidence angle images, including the July 02 FQ27, July 05 FQ7, July 06 FQ10
and July 09 FQ23 images.
The decomposition components of the Cloude and Pottier decomposition are more promising
than the Freeman-Durden classification for quantifying NPV biomass (Table 4), although the
largest r2 value achieved is not beyond those obtained by the VH/VV and de-polarization ratio.
Incidence angle of the images and vegetation phenology also have an influence on performance
of the decomposition components. In the middle growing season, when incidence angle is
steeper than that of the FQ10 image, both entropy (H) and Alpha angle have a significant
relationship with NPV biomass with an r 2 than 0.60, but smaller than 0.69. When incidence angle
is shallower than that of FQ10, there is no significant relationship between H/Alpha angle and
NPV biomass. Nevertheless, volume scattering of the Freeman-Durden decomposition extracted
from the shallowest incidence angle FQ27 image yields an r 2 value of 0.69 for quantifying NPV
biomass, followed by the steepest incidence angle FQ1 image. In the peak growing season, the H
and Alpha angle of the FQ3 images on June 28th have similar performance with the VH/VV and
de-polarization ratios. Volume scattering extracted from the July 9 FQ23 images has a significant
r2 value (0.38) for NPV biomass estimation. Surface scattering is not a significant contributor for
quantifying NPV biomass.
The performance of Radarsat-2 on total aboveground biomass estimation is summarized in Table
5. To consider the increase in green vegetation during the growing season until it reaches the
peak in late June or early July, only the images acquired near the field days were used for
aboveground biomass estimation. The June 28 FQ3 images demonstrated the greatest ability with
12
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
the r2 value of 0.70 achieved by the de-polarization ratio, followed by the VH/VV and RVI with
an r2 of 0.65 and 0.57, respectively.
Entropy and Alpha angle extracted from the June 28 FQ3 image have an r 2 value similar to that
of VH/VV and de-polarization ratio for quantifying total aboveground biomass. The entropy
extracted from the FQ5 image acquired on June 19th has an r2 value of 0.62, which is comparable
to that of the June 28 FQ3 image. Entropy and Alpha angle derived from images with an
incidence angle shallower than that of the FQ7 image have no significant relationship with total
aboveground biomass. Volume scattering of the FQ23 image acquired on July 09 2014 has a
Downloaded by [UNSW Library] at 06:36 26 October 2017
significant relationship with total aboveground biomass, while surface scattering of the June 28
FQ3 image can significantly account for the variations in total aboveground biomass.
The best relationships between Radarsat-2 response and NPV and total aboveground biomass
identified in Table 4 and 5 are plotted in Figure 5. The main purpose of Figure 5 is to address
whether NPV biomass is measured directly or indirectly as a component of total aboveground
biomass. From the middle to peak growing season, NPV biomass is unchanged without cure of
green vegetation and removal of NPV by ground overflow and wind etc., while total
aboveground biomass increases as green vegetation increases. So NPV biomass sampled in the
field season (June 20 to July 2) generally equal to that in middle growing season; however total
aboveground biomass sampled in the field season is larger than that in middle growing season.
Therefore, SAR parameters from June 15 and June 19 yield better estimation of NPV biomass
than total aboveground biomass. In the peak growing season, using the June 28 image, total
aboveground biomass estimate is slightly better than the NPV biomass estimate. The very similar
r2 values for measuring total aboveground and NPV biomass in the peak growing season suggest
that NPV, as a merely part of the total vegetation, may be indirectly measured as a part of the
total aboveground biomass.
Accuracy Assessment
Accuracy of NPV biomass and total aboveground biomass estimation using the VH/VV ratio
extracted from the June 15 FQ23 image and the depolarization ratio of the June 28 FQ3 image
was assessed using a leave-one-out cross-validation approach (Figure 6). Relative RMSE
(rRMSE) for NPV biomass and total aboveground biomass estimation using the VH/VV ratio of
13
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
the FQ23 image acquired on June 15, 2014 are 9% and 8.4%, respectively. The rRMSE (12.6%)
for NPV biomass estimation is higher, while it (6.9%) is lower for total aboveground biomass
estimation using the depolarization ratio of the June 28 FQ3 image.
Figure 6 shows that using the June 15 FQ23 image, both NPV biomass and total aboveground
biomass in valley and disturbed vegetation communities are well estimated (stay close to the 1:1
line). Comparatively, accuracy of biomass estimation in slope and upland vegetation
communities is lower. Using the steep incidence angle FQ3 image, NPV biomass is generally
underestimated, except for one upland site and one slope site. Using the FQ3 image, total
aboveground biomass in valley and slope land is better estimated than that in upland and
disturbed communities. This indicates that vegetation biomass can be better estimated in valley
and disturbed communities where vegetation biomass is larger than other sampling sites in the
study area, as long as the saturation threshold of SAR images is not reached in dense vegetation
communities.
DISCUSSION
The largest r2 values for NPV biomass estimation were achieved by the VH/VV ratio of the
FQ23 image acquired on June 15, 2014. It indicated the potential of Radarsat-2 data for
quantifying NPV biomass in middle growing season when optical images, including Landsat 8
OLI and Sentinel-2 A, have limited capability (Li and Guo submitted). In the peak growing
season, depolarization ratio of the June 28 FQ3 image has the best performance on NPV and
total aboveground biomass estimation. However, the performance of June 28 image for biomass
estimation was affected by increased SAR backscattering as a result of increased moisture from
the 0.7 mm daily precipitation on June 28 th, and 20 mm total precipitation within 3 days (June
26th to June 28th). The performance of Radarsat-2 images for NPV and total aboveground
biomass estimation in the early growing season was not evaluated due to the lack of images.
Incidence Angle Effects
The performance of the Radarsat-2 images on quantifying NPV biomass was affected by
incidence angle, ground cover that was highly related to vegetation phenology, and environment
condition (e.g., wetness of canopy and soil surface, etc.). NPV biomass was nearly unchanged
from the early growing season until the peak growing season in GNP. The volume scattering of
the June 2 FQ1 image is not as good as that of the June 8 FQ27 image for estimating NPV
biomass, because the FQ1 image with steep incidence angle has more ability to penetrate the
canopy (Figure 4(c)), and therefore cannot quantify standing dead vegetation biomass very well.
14
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
The good performance of volume scattering of the June 8 FQ27 images on NPV biomass
estimation is attributed to its shallow incidence angle which enables them to capture the standing
dead biomass, a substantial component of NPV biomass. Besides the precipitation influence on
the FQ12 image, the moderate canopy penetration capability to plant surface litter and ability to
capture canopy volume scattering of the FQ10 and FQ12 images accounted for the lower r 2
values for quantifying NPV biomass.
The superiority of the June 15 FQ23 and the June 8 FQ27 images for NPV biomass estimation is
consistent with the finding that radar images with a shallow incidence angle and reduced
Downloaded by [UNSW Library] at 06:36 26 October 2017
penetration to soil surface are more sensitive to crop residue in harvested cropland (McNairn et
al. 1996). Also, SAR images with shallow incidence angles are more sensitive to surface
roughness (Baghdadi et al. 2002, Baghdadi et al. 2008).
Environmental Effects
Besides incidence angle effects, wind and precipitation play a role in NPV and total aboveground
biomass estimation using SAR data. Strong wind blows down standing dead vegetation and
green vegetation, alters surface roughness and changes exposure of plant litter on the surface,
and thus affects backscattering. Such change in backscattering affects NPV biomass estimation.
The poor performance of the July 02 FQ27 image for NPV biomass estimation can be attributed
to wind effects (Table 3). The difference in performance of the July 05 FQ7 and July 06 FQ10
images for biomass estimation is possibly partially attributed to the strong wind on July 5,
besides the small incidence angle difference. Also, the July 5 image covers only 10 sites, which
also possibly makes a difference.
Precipitation increases background and canopy moisture, which increases dielectricity and
further enhances backscattering. This enhanced backscattering from increased moisture reduced
the ability of SAR data to detect NPV that usually has very low water content. The reduced r 2
value for quantifying NPV and total aboveground biomass using the July 09 FQ23 image may be
accounted for by precipitation. Although incidence angle of the June 19 FQ5 and June 18 FQ12
makes a lot difference in biomass estimation, the large amount of precipitation (Table 3) is also a
possible reason of the bad performance of the June 18 image.
15
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Suitable SAR Parameters
Selecting a suitable SAR parameter is vital for quantifying NPV biomass and total aboveground
biomass. The cross-polarization ratio (VH/VV) and depolarization ratio outperformed the copolarization ratio (HH/VV) for NPV and total aboveground biomass estimation. This finding
agrees with the finding of Ferrazzoli et al. (1997) that the availability of cross-polarization was
important for biomass estimation in cropland and forest. It also explained the much smaller r 2
(0.30) on NPV biomass estimation achieved by the co-polarization Radarsat-2 image in the study
area (Finnigan 2013). The good performance of VH/VV is attributed to the sensitivity of crossDownloaded by [UNSW Library] at 06:36 26 October 2017
polarization backscatter coefficients (VH) to standing vegetation biomass and the sensitivity of
VV backscatter coefficient to the vertical structure of vegetation (Bartsch et al. 2016). The
inferority of the co-polarization ratio is because co-polarization backscattering is primarily from
surface scattering (Wiseman et al. 2014).
The VH/VV and depolarization ratio are also superior to other SAR parameters analyzed in this
study for NPV and total biomass estimation, including RVI and decomposition components of
the Freeman-Durden decomposition with an exception of volume scattering of the June 8 FQ27
image. Superiority of the VH/VV and depolarization is also demonstrated, in contrast with
entropy and Alpha angle. Entropy and Alpha angle of steep incidence angles (in this study, the
FQ1-FQ3 mode) have similar performance with the VH/VV and depolarization ratio for
quantifying NPV and total aboveground biomass. However, when using shallow incidence angle
images, such as the June 15 FQ23 and June 8 FQ27 images, entropy and Alpha angle are inferior
to the VH/VV and depolarization ratio.
Future Research
Further research is needed to investigate how the presence of green vegetation affects the
performance of Radarsat-2 images on NPV biomass estimation using a theoretically based
scattering model. Research is also essential to explore how incident angles influence the biomass
estimates, and the steep and higher thresholds of the incidence angle of the Radarsat-2 images for
NPV biomass estimation. The influence of surface plant litter on NPV biomass is also worthy of
investigation. Also, given the strong potential of the cross-polarization ratio VH/VV and
16
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
depolarization ratio for quantifying NPV, the potential of Sentinel-1 A and Sentinel-1B images is
also something that should be investigated in the future.
CONCLUSION
This study investigated the application of C-band, fine quad-pol, and Radarsat-2 data for
quantifying NPV biomass and total aboveground in a conserved semiarid mixed grassland,
characterized by a large amount of dead vegetation material and high percentage of biological
soil crust. The FQ3 Radarsat-2 image is most suitable for quantifying NPV and total
Downloaded by [UNSW Library] at 06:36 26 October 2017
aboveground biomass in the peak growing season. However, Radarsat-2 images with a shallow
incidence angle, such as FQ23, are recommended for NPV and total aboveground biomass
estimation in middle growing seasons.
The depolarization ratio and the cross-polarization ratios (VH/VV) are the best SAR parameters
for quantifying NPV and total aboveground biomass. Entropy and alpha angle decomposed using
Radarsat-2 images with steep incidence angles (FQ1-FQ3) also has certain capability to
quantitatively estimate NPV and total aboveground biomass.
This was the first study, to our knowledge, done to investigate the potential effectiveness of
multi-angular, multi-temporal fine Quad-pol Radarsat-2 images for quantifying NPV biomass in
grasslands. This has the potential to significantly contribute to grassland management that uses
NPV biomass and (or) total aboveground biomass as an indicator of ecosystem health, fire risk
assessment and herbivore carrying capacity estimation, among other things. It also contributes to
our understanding of grassland ecology, hydrology, and climatology that use biomass as a model
input.
ACKNOWLEDGEMENTS
We thank the Canadian Space Agency and the Canada Centre for Remote Sensing for providing
RADARSAT-2 data through the Science and Operational Applications Research—Education
Initiative (SOAR-E) program. This research was financially supported by a Discovery Grant
from the Natural Sciences and Engineering Research Council of Canada (NSERC) (Dr. X. Guo),
NSERC Alexander Graham Bell Canada Graduate Scholarship (Z. Li), and Qinghai Science and
Technology Department (2016-HZ-807). Thanks also go to the anonymous reviewers for their
valuable comments that help improve this manuscript. Logistic support was provided by
17
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Grasslands National Park, Parks Canada. Thanks also to Dr. Dandan Xu, Tengfei Cui, Meng Li,
Bin Lu, and James Liu for field data collection. RADARSAT-2 Data and Products ©
Downloaded by [UNSW Library] at 06:36 26 October 2017
MacDonald, Dettwiler and Associates Ltd. (2014)—All Rights Reserved.
18
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
REFERENCES
Alpers, W., Zhang, B., Mouche, A., Zeng, K., and Chan, P.W. 2016. ―Rain footprints on C-band synthetic
aperture radar images of the ocean-Revisited.‖ Remote Sensing of Environment 187:pp.169-185
Asner, G.P., and Heidebrecht, K.B. 2003. ―Imaging spectroscopy for desertification studies: comparing
aviris and eo-1 hyperion in argentina drylands.‖ IEEE Transactions on Geoscience and Remote
Sensing 41 (6):pp.1283-1296
Baghdadi, N., King, C., Bourguignon, A., and Remond, A. 2002. ―Potential of ERS and Radarsat data for
surface roughness monitoring over bare agricultural fields: Application to catchments in Northern
France.‖ International Journal of Remote Sensing 23 (17):pp.3427-3442
Baghdadi, N., Zribi, M., Loumagne, C., Ansart, P., and Anguela, T.P. 2008. ―Analysis of TerraSAR-X
data and their sensitivity to soil surface parameters over bare agricultural fields.‖ Remote Sensing
of Environment 112 (12):pp.4370-4379
Bartsch, A., Widhalm, B., Kuhry, P., Hugelius, G., Palmtag, J., and Siewert, M. 2016. ―Can C-Band SAR
be used to estimate soil organic carbon storage in tundra.‖ Biogeosciences 13:pp.5453-5470
Carreiras, J.M.B., Vasconcelos, M.J., and Lucas, R.M. 2012. ―Understanding the relationship between
aboveground biomass and ALOS PALSAR data in the forests of Guinea-Bissau (West Africa).‖
Remote Sensing of Environment 121:pp.426-442
Cihlar, J. 2012. Pilot study on quantification of non-photosynthetic vegetation biomass in Canadian
prairie grasslands. Technical report submitted to Environment Canada. Solicitation number:
K1A50-11-0036.
Cloude, S.R., and Pottier, E. 1997. ―An entropy based classification scheme for land applications of
polarimetric SAR.‖ IEEE Transactions on Geoscience and Remote Sensing 35 (1):pp.68-78
Danklmayer, A., Doring, B.J., Schwerdt, M., and Chandra, M. 2009. ―Assessment of atmospheric
propagation effects in SAR images.‖ IEEE Transactions on Geoscience and Remote Sensing 47
(10):pp.3507-3518
Daughtry, C.S.T., Doraiswamy, P.C., Hunt, E.R., Stern, A.J., McMurtrey, J.E., and Prueger, J.H. 2006.
―Remote sensing of crop residue cover and soil tillage intensity.‖ Soil and Tillage Research 91
(1-2):pp.101-108
Daughtry, C.S.T., Hunt, E.R., and McMurtrey, J.E. 2004. ―Assessing crop residue cover using shortwave
infrared reflectance.‖ Remote Sensing of Environment 90 (1):pp.126-134
Davis, S.K. 2005. ―Nest-site selection patterns and the influence of vegetation on nest survival of mixedgrass prairie passerines.‖ The Condor 107 (3):pp.605
Fargey, K.S., Larson, S.D., Grant, S.J., Fargey, P., and Schmidt, C. 2000. ―Grasslands National Park field
guide‖, Prairie Wind & Silver Sage – Friends of Grasslands Inc.
Ferrazzoli, P., Paloscia, S., Pampaloni, P., Schiavon, G., Sigismondi, S., and Solimini, D. 1997. ―The
potential of multifrequency polarimetric SAR in assessing agricultural and arboreous biomass.‖
IEEE Transactions on Geoscience and Remote Sensing 35 (1):pp.5-17
Ferrazzoli, P., Paloscia, S., Pampaloni, P., Schiavon, G., Sigismondi, S., Solimini, D. 1997. ―The
potential of multifrequency polarimetric SAR in assessing agricultural and arboreous biomass.‖
IEEE Transactions on Geoscience and Remote Sensing 35:pp.13
Finnigan, C. 2013. ―Developing a Grassland Biomass Monitoring Tool Using a Time Series of Dual
Polarimetric SAR and Optical Data.‖ MSc, University of Saskatchewan.
Fisher, R.J., and Davis, S.K. 2010. ―From Wiens to Robel: A Review of Grassland‐ Bird Habitat
Selection.‖ The Journal of Wildlife Management 74:pp.9
Freeman, A., and Durden, S.L. 1998. ―A three-component scattering model for polarimetric SAR data.‖
IEEE Transactions on Geoscience and Remote Sensing 36 (3):pp.963-973
19
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
Gherboudj, I., Magagi, R., Berg, A.A., and Toth, B. 2011. ―Soil moisture retrieval over agricultural fields
from multi-polarized and multi-angular RADARSAT-2 SAR data.‖ Remote Sensing of
Environment 115 (1):pp.33-43
Guerschman, J.P., Hill, M.J., Renzullo, L.J., Barrett, D.J., Marks, A.S., and Botha, E.J. 2009. ―Estimating
fractional cover of photosynthetic vegetation, non-photosynthetic vegetation and bare soil in the
Australian tropical savanna region upscaling the EO-1 Hyperion and MODIS sensors.‖ Remote
Sensing of Environment 113 (5):pp.928-945
Hoaglin, D. C., & Iglewicz, B. (1987). Fine-tuning some resistant rules for outlier labeling. Journal of the
American Statistical Association, 82(400), 1147--1149.
Huang, S., Crabtree, R.L., Potter, C., and Gross, P. 2009. ―Estimating the quantity and quality of coarse
woody debris in Yellowstone post-fire forest ecosystem from fusion of SAR and optical data.‖
Remote Sensing of Environment 113 (9):pp.1926-1938
Jackson, H., and Prince, S. 2016. ―Degradation of Non-Photosynthetic Vegetation in a Semi-Arid
Rangeland.‖ Remote Sensing 8 (9):pp.692
Jacques, D.C., Kergoat, L., Hiernaux, P., Mougin, E., and Defourny, P. 2014. ―Monitoring dry vegetation
masses in semi-arid areas with MODIS SWIR bands.‖ Remote Sensing of Environment
153:pp.40-49
Kim, Y., Jackson, T., Bindlish, R., Lee, H., and Hong, S. 2012. ―Radar vegetation index for estimating
the vegetation water content of rice and soybean.‖ IEEE Geoscience and Remote Sensing Letters
9 (4):pp.564-568
Kim, Y., and van Zyl, J.J. 2009. ―A time-series approach to estimate soil moisture using polarimetric
radar data.‖ IEEE Transactions on Geoscience and Remote Sensing 47 (8):pp.2519-2527
Lee, J.-S., Ainsworth, T.L., Wang, Y., and Chen, K.-S. 2015. ―Polarimetric SAR speckle filtering and the
extended sigma filter.‖ IEEE Transactions on Geoscience and Remote Sensing 53 (3):pp.11501160
Li, K., Brisco, B., Yun, S., and Touzi, R. 2012. ―Polarimetric decomposition with RADARSAT-2 for rice
mapping and monitoring.‖ Canadian Journal of Remote Sensing 38 (2):pp.169-179
Li, Z., and Guo, X. 2012. ―Detecting Climate Effects on Vegetation in Northern Mixed Prairie Using
NOAA AVHRR 1-km Time-Series NDVI Data.‖ Remote Sensing 4 (12):pp.120-134
Li, Z., and Guo, X. 2016. ―Remote sensing of terrestrial non-photosynthetic vegetation using
hyperspectral, multispectral, SAR, and LiDAR data.‖ Progress in Physical Geography 40
(2):pp.276-304
Li, Z., and Guo, X. 2017. ―Quantifying non-photosynthetic vegetation (NPV) biomass in semiarid mixed
grasslands using Landsat 8 OLI and Sentinel-2 images.‖ Internaltional Journal of Remote
Sensing (submitted)
MacDonald, R.L., Burke, J.M., Chen, H.Y.H., and Prepas, E.E. 2012. ―Relationship between
Aboveground Biomass and Percent Cover of Ground Vegetation in Canadian Boreal Plain
Riparian Forests.‖ Forest Science 58 (1):pp.47-53
McNairn, H., Boisvert, J.B., Major, D.J., Gwyn, Q.H.J., Brown, R.J., and Smith, A.M. 1996.
―Identification of Agricultural Tillage Practices from C-Band Radar Backscatter.‖ Canadian
Journal of Remote Sensing 22 (2):pp.154-162
McNairn, H., Duguay, C., Boisvert, J., Huffman, E., and Brisco, B. 2001. ―Defining the Sensitivity of
Multi-Frequency and Multi-Polarized Radar Backscatter to Post-Harvest Crop Residue.‖
Canadian Journal of Remote Sensing 27 (3):pp.247-263
McNairn, H., Shang, J., Champagne, C., and Jiao, X. 2009. ―TerraSAR-X and RADARSAT-2 for crop
classification and acreage estimation.‖ Geoscience and Remote Sensing Symposium, 2009 IEEE
International, IGARSS 2009.
20
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
Moran, M.S., Alonso, L., Moreno, J.F., Cendrero Mateo, M.P., de la Cruz, D.F., and Montoro, A. 2012 a.
―A RADARSAT-2 Quad-Polarized Time Series for Monitoring Crop and Soil Conditions in
Barrax, Spain.‖ IEEE Transactions on Geoscience and Remote Sensing 50 (4):pp.1057-1070
Moran, M.S., Alonso, L., Moreno, J.F., Mateo, M.P.C., De La Cruz, D.F., and Montoro, A. 2012b. ―A
RADARSAT-2 quad-polarized time series for monitoring crop and soil conditions in Barrax,
Spain.‖ IEEE Transactions on Geoscience and Remote Sensing 50 (4):pp.1057-1070
Numata, I., Roberts, D., Chadwick, O., Schimel, J., Galvao, L., and Soares, J. 2008. ―Evaluation of
hyperspectral data for pasture estimate in the Brazilian Amazon using field and imaging
spectrometers.‖ Remote Sensing of Environment 112 (4):pp.1569-1583
Pacheco, A.M., McNairn, H., and Merzouki, A. 2010. ―Evaluating TerraSAR-X for the identification of
tillage occurrence over an agricultural area in Canada.‖ Remote Sensing for Agriculture,
Ecosystems, and Hydrology XII, 2010/10/07.
Ren, H., and Zhou, G. 2012. ―Estimating senesced biomass of desert steppe in Inner Mongolia using field
spectrometric data.‖ Agricultural and Forest Meteorology 161:pp.66-71
Roberts, D.A., Smith, M.O., and Adams, J.B. 1993. ―Green vegetation, nonphotosynthetic vegetation, and
soils in AVIRIS data.‖ Remote Sensing of Environment 44 (2-3):pp.255-269
Sartori, L.R., Imai, N.N., Mura, J.C., de Moraes Novo, E.M.L., and Silva, T.S.F. 2011. ―Mapping
macrophyte species in the Amazon floodplain wetlands using fully polarimetric ALOS/PALSAR
data.‖ IEEE Transactions on Geoscience and Remote Sensing 49 (12):pp.4717-4728
Smith, A.M., and Buckley, J.R. 2011. ―Investigating RADARSAT-2 as a tool for monitoring grassland in
western Canada.‖ Canadian Journal of Remote Sensing 37 (1):pp.93-102
Smith, A.M., Hill, M.J., and Zhang, Y. 2015. ―Estimating Ground Cover in the Mixed Prairie Grassland
of Southern Alberta Using Vegetation Indices Related to Physiological Function.‖ Canadian
Journal of Remote Sensing 41 (1):pp.51-66
Ulaby, F., Kouyate, F., Brisco, B., and Williams, T.H. 1986. ―Textural Infornation in SAR Images.‖ IEEE
Transactions on Geoscience and Remote Sensing GE-24 (2):pp.235-245
Wiseman, G., McNairn, H., Homayouni, S., and Shang, J. 2014. ―RADARSAT-2 Polarimetric SAR
Response to Crop Biomass for Agricultural Production Monitoring.‖ IEEE Journal of Selected
Topics in Applied Earth Observations and Remote Sensing 7 (11):pp.4461-4471
Xu, D., Guo, X., Li, Z., Yang, X., and Yin, H. 2014. ―Measuring the dead component of mixed grassland
with Landsat imagery.‖ Remote Sensing of Environment 142:pp.33-43
21
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
TABLE 1. Descriptive analysis of aboveground biomass data sampled in summer 2014 (NPV
includes standing dead vegetation, plant litter on the surface, and moss and lichens)
Sites
Statistical
description
Average
Max
Downloaded by [UNSW Library] at 06:36 26 October 2017
Upland
Slope
Valley
Disturbed
Min
StdDev
Average
Max
Min
StdDev
Average
Max
Min
StdDev
Average
Max
Min
StdDev
Grass
(g/m2)
122.7
222.1
38.2
51.6
95.6
295.1
12.8
61.9
124.6
273.0
16.1
62.7
136.5
228.9
19.0
77.5
PV
Forb
(g/m2)
22.2
115.0
0.0
24.8
28.8
246.0
0.0
42.5
31.7
136.1
0.0
39.2
0.0
0.0
0.0
0.0
22
Shrub
(g/m2)
16.9
311.5
0.0
64.0
7.3
123.7
0.0
23.5
12.1
110.4
0.0
27.4
9.5
75.9
0.0
26.8
NPV
(g/m2)
386.5
791.0
65.7
236.9
258.6
909.0
12.7
216.8
247.8
972.2
20.9
211.9
624.3
950.0
480.4
147.0
Total aboveground
biomass(g/m2)
548.3
1196.5
172.1
265.8
390.3
983.2
46.6
244.4
416.3
1084.0
76.2
226.5
770.2
1175.3
603.2
185.2
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
TABLE 2. The fractional cover of green vegetation, including grasses, forbs and shrubs, nonphotosynthetic vegetation (standing dead vegetation (S.D) and plant litter on the surface),
biological soil crust (lichen and moss), rock, and bare soil (B. Soil) sampled in summer 2014
S.D%
15
18
16
26
Litter% Lichen% Moss% Rock% B.Soil%
34
0
0
0
0
21
1
15
3
2
28
0
10
0
0
9
0
2
0
15
Downloaded by [UNSW Library] at 06:36 26 October 2017
Sites
grass% forb% shrub%
disturbed
49
0
2
slope
30
7
3
upland
39
6
1
valley
36
6
6
23
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
TABLE 3. Radarsat-2 data description and environmental conditions
Date
Jun
02
Jun
08
Jun
12
Jun
15
Jun
18
Jun
19
Jun
28
Jul
02
Jul
05
Jul
06
Jul
09
Spatial
Wi
Daily
Be
Resoluti Wind nd Te Dewp Preci
am Incident on X
Dir
(K mp
oint
pitati
mo
angle
(m) × Y (10’s m/ (°C Temp
on
de range (°) (m)
deg) hr)
)
(°C)
(mm)
FQ
18.54- 4.73 ×
1
20.34
4.83
33
6 7.8
6.9
0
FQ
45.24.73 ×
27
46.5
4.85
35
5 2.5
1.1
0
FQ
29.24.73 ×
10
30.9
5.18
5
4 1.3
-0.5
0
FQ
41.94.73 ×
10.
23
43.3
4.94
34
15
1
8.8
0.2
FQ
31.54.73 ×
14.
10.3
12
32.9
4.96
22
11
5
24.9
FQ
23.44.73 ×
21
15 9.7
7.1
5
25.3
4.97
0.4
FQ
20.94.73 ×
15.
27
25
8
3
22.9
5.33
2
0.7
FQ
45.24.73 ×
11.
9
6
10.9
27
46.5
4.85
9
0
FQ
25.84.73 ×
27.
28
30
6.9
7
27.6
4.74
9
0
FQ
29.24.73 ×
13.
21
6
11.6
10
30.9
5.18
7
0
FQ
41.94.73 ×
12.
2
8
11
23
43.3
4.94
6
0
* All the images were collected at the right look direction.
24
3
days'
preci
pitati
on
(mm)
5.3
0
0.6
9.6
89
113.7
20
2.7
0
0
4.5
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
TABLE 4. The Relationship (r2 values) between various Radrsat-2 parameters and nonphotosynthetic vegetation (NPV) biomass (D-ratio is depolarization ratio; Entropy (H) and Alpha
angle were derived from the Cloude and Pottier decomposition; V and S present volume
scattering and surface scattering, respectively, which were derived from the Freeman-Durden
Downloaded by [UNSW Library] at 06:36 26 October 2017
decomposition)
Beam
HH/ VH/ VH/
Date
mode
RVI VV HH VV
D-ratio
H
02-Jun
FQ1
0.45 0.46 0.53
0.51
0.51
08-Jun
FQ27
0.34
0.48
0.43
12-Jun
FQ10
0.46 0.43 0.35 0.44
0.4
15-Jun
FQ23
0.57 0.58 0.36 0.70
0.69
18-Jun
FQ12
0.38
19-Jun
FQ5
0.45
0.34 0.47
0.58
0.65
28-Jun
FQ3
0.55
0.44 0.63
0.59
0.65
02-Jul
FQ27
05-Jul
FQ7
0.34
0.35
06-Jul
FQ10
0.36
09-Jul
FQ23
0.48
0.35
2
Note: only r values significant at the 0.05 level are demonstrated.
25
Alpha
0.57
V
0.52
0.69
S
0.41
0.60
0.63
0.44
0.38
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
Table 5. The relationship (r2 values) between various Radrsat-2 parameters and total
aboveground biomass (D-ratio is depolarization ratio; Entropy (H) and Alpha angle were derived
from the Cloude and Pottier decomposition; V and S present volume scattering and surface
scattering, respectively, which were derived from the Freeman-Durden decomposition)
Date
15-Jun
18-Jun
19-Jun
28-Jun
02-Jul
05-Jul
06-Jul
09-Jul
Beam mode RVI HH/VV VH/HH VH/VV D-ratio H Alpha V
S
FQ23
0.39
0.51
0.51
0.5
FQ12
0.38
FQ5
0.37
0.38
0.52 0.62
0.55
FQ3
0.57
0.42
0.45
0.65
0.70 0.60
0.65
FQ27
FQ7
0.36
0.48
0.52
0.52
FQ10
FQ23
0.41
* Note: only r2 values significant at the 0.05 level are demonstrated.
26
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
FIG.1. (a) The geographic location of the study area with the sampling sites in 2014 shown as
orange squares and (b) the sampling design at each sampling site (The background map is the
Radarsat-2 FQ1 image acquired on June 2, 2014; and RGB was assigned as HH, HV, and VV
backscatter coefficients, respectively)
27
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
FIG.2. Typical vegetation communities in (a) disturbed grassland, (b) slope grassland, (c) upland
grassland, and (d) valley grassland
28
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
FIG.3. The procedure of Radarsat-2 image processing and data retrieval and analysis
29
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
FIG.4. The Radarsat-2 images (Red: HH, Green: HV, and Blue: VV) with the field sampling sites (yellow
dots) on (a) the June 2 FQ 1 image and (b) the June 8 FQ 27 image; and the scattering mechanism of
Radarsat-2 images on: (c) the June 2 2014 FQ1image and (d) the June 8 2014 FQ27 image are
demonstrated using backscattering at one upland sampling site encompassed by the orange square in (a)
and (b).
30
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
FIG.5. The Radarsat-2 response and non-photosynthetic vegetation (NPV) and total aboveground
biomass: (a) the cross-polarization VH/VV ratio of the June 15 FQ 23 image, (b) the Entropy of the June
19 FQ5 image, (c) the cross-polarization ratio of the June 28 FQ 3 image, and (d) the Alpha angle of the
June 28 FQ3 image.
31
ACCEPTED MANUSCRIPT
Downloaded by [UNSW Library] at 06:36 26 October 2017
ACCEPTED MANUSCRIPT
FIG.6. Comparison of estimated and field measured non-photosynthetic vegetation (NPV) and total
aboveground biomass: (a) the cross-polarization VH/VV ratio of the June 15 FQ 23 image for NPV
biomass estimation (rRMSE = 9%), (b) the cross-polarization VH/VV ratio of the June 15 FQ 23 image
for quantifying total aboveground biomass (rRMSE = 8.4%), (c) the de-polarization ratio of the June 28
FQ 3 image for NPV biomass estimation (rRMSE = 12.6%), and (d) the de-polarization ratio of the June
28 FQ3 image for quantifying total aboveground biomass (rRMSE = 6.9%)
32
ACCEPTED MANUSCRIPT
Документ
Категория
Без категории
Просмотров
44
Размер файла
2 059 Кб
Теги
1394181, 2017, 07038992
1/--страниц
Пожаловаться на содержимое документа