MULTI-FREQUENCY SAR CAPABILITIES FOR FOREST BIOMASS AND CARBON INVENTORY FOR REDD MONITORING

Full text

MULTI-FREQUENCY SAR CAPABILITIES FOR FOREST BIOMASS AND CARBON
INVENTORY FOR REDD MONITORING
Suman Sinha
1*
, A. Santra
2
, S. S. Mitra
3
, C. Jeganathan
4
, L. K. Sharma
5
, M. S. Nathawat
6
, A. K. Das
7
and S. Mohan
8
1
Haldia Institute of Technology, Haldia 725657, India,
Email: sumanrumpa.sinha@gmail.com
2
Haldia Institute of Technology, Haldia 725657, India,
Email: avisek.santra@gmail.com
3
Haldia Institute of Technology, Haldia 725657, India,
Email: shreyashi.mitra@gmail.com
4
Birla Institute of Technology, Mesra 835215, India,
Email: jeganathanc@bitmesra.ac.in
5
Central University of Rajasthan, Ajmer 305817, India,
Email: laxmikant1000@yahoo.com
6
Indira Gandhi National Open University, New Delhi 110068, India,
Email: msnathawat@ignou.ac.in
7
Space Application Centre (ISRO), Ahmedabad 380015, India,
Email: anup@sac.isro.gov.in
8
Physical Research Laboratory, Ahmedabad, 380059, India,
Email: shivmohan.isro@gmail.com
KEY WORDS : ALOS PALSAR, Radarsat-2, COSMO-Skymed, Forest, Biomass.
ABSTRACT
Greenhouse gas inventories and emissions reduction programs require robust methods to quantify carbon
sequestration in forests. Proper inventory of forest aboveground biomass (AGB) is required for accounting carbon
emissions that forms the most vital part of the carbon cycle modeling and climate change mitigation programs in
context to Reducing Emissions from Deforestation and Forest Degradation (REDD). Remote Sensing (RS)
technology provides cost and time effective means for accurate temporal monitoring over large synoptic extents at
local to global levels, and hence, is beneficial over conventional methods. The study presents a suitable approach
for estimating AGB through the synergic use of multi-frequency X-, C- and L-band Synthetic Aperture Radar
(SAR) data over tropical deciduous mixed forests of Munger (Bihar, India). Backscatter values generated from the
raw SAR images were correlated with field-based AGB values and then regressed to generate best-fit models for
AGB estimates with single and combined frequencies of COSMO-Skymed (X-band), Radarsat-2 (C-band) and
ALOS PALSAR (L-band). Among all the models for AGB estimation, the integrated model involving X, C- and L-
bands showed the best results with r
2
=0.95, RMSE=14.81 t/ha and Willmott’s index of agreement of 0.95. Resulting
modeled AGB were converted to carbon (C) and carbon dioxide (CO
2
) equivalents using conversion factors. Hence,
the study proposed L-band for single frequency analysis and the combination of X-, C- and L-bands for multi-
frequency analysis for tropical forest AGB and C estimation. The study revealed information regarding the spatial
distribution and quantification of forest AGB and C required for REDD monitoring.
1. INTRODUCTION
‘400 ppm World’; an universal alarming situation where the global carbon dioxide (CO
2
) concentration surpassed
the 400 ppm threshold level during 2016; a phenomenon which is permanent not likely to revert back in future
(Betts et al., 2016). India being a mega-biodiversity country harbors forests that account for more than one fifth of
the geographical area. Indian tropical forests are likely to experience extreme hasty and significant climate and
vegetation changes over the next decades (Ravindranath et al., 2006). Globally this is a serious issue as tropical
forests sequester one fifth of the global carbon (C) stock, and almost one half of the above-ground C, stored in
vegetation of all biomes (Hunter et al., 2013). Biophysical indicators of forest C storage, viz. the above-ground
biomass (AGB) are important for realizing the terrestrial C equilibrium (Sinha et al., 2017). Sequestration of C in
the vegetation is the only possible viable strategy in milieu of Reducing Emissions from Deforestation and forest
Degradation (REDD) to maintain the atmospheric C balance and account for the CO
2
released from forests
(Waikhom et al., 2017). Henceforth, accurate estimates of biomass are a prerequisite for forest C accounting under
REDD framework (Sinha et al., 2016).

Paper presented in the 38th Asian Conference on Remote Sensing, October 23-27, New Delhi (INDIA)
Sinha et al. (2015) have outlined detailed information regarding the biomass estimation methods, wherein, the
remote sensing (RS)-based approaches have clearly outshined other methods. Optical sensors are recurrently used
for AGB estimation (Kumar et al., 2013; Sharma et al., 2013), however, saturates early owing to poor sensitivity to
forest parameters, unlike Synthetic Aperture Radar (SAR) and Light Detecting and Ranging (LiDAR) sensors
(Sinha et al., 2015). Though both the systems are sensitive to forest spatial structure and standing biomass, SAR is
favoured due to its wall-to-wall coverage which is absent in all LiDAR systems (Su et al., 2016). Currently SAR is
extensively used in retrieving forest biomass (Sinha et al., 2015).
Usually SAR data are acquired in X, C, L bands and sometimes in S and P bands as well. SAR backscatter from
longer wavelengths, like L- and P-bands relate more to the forest biophysical parameters due to their greater
penetration capabilities through the vegetation surfaces and are scattered/attenuated by trunk and main branches
(Sinha et al., 2015). Synergic use of radar and optical sensor data has the potential to improve the estimation of
forest AGB (Sinha et al., 2016). However, optical sensors suffers several drawbacks, like occurrence of frequent
clouds in the tropics hampering the acquisition of the high quality satellite data, lack of volumetric estimations due
to absence of penetrability of visible bands and low saturation levels of the spectral bands for biomass estimation
(Sinha et al., 2015). Hence, the use of integrated multi-frequency SAR reveals greater potential for AGB estimation
that can possibly overcome the limitations of optical sensors and single SAR sensors. Alappat et al. (2011) applied
synergic model integrating SAR C- and L-bands and Englhart et al. (2011) used SAR X- and L-bands for biomass
assessment. This study in addition integrates SAR backscatter data from X-, C- and L-bands to investigate
relationships with field AGB to derive a synergic model to estimate above-ground bole biomass, from which C and
CO
2
has been enumerated.
2. MATERIALS AND METHODS
2.1 Site under investigation
Munger forests in the state of Bihar (India) comprising the Bhimbandh Wildlife Sanctuary with geographic
coordinates of 25º19'30''N-24º56'50''N latitude and 86º33'33''E-86º11'51''E longitude, covering an area of approx
672.5 km
2
has been considered as the test site for investigation (Figure 1). Summer temperature reaches 45
o
C, while
winter experiences nearly 3-9
o
C. The average annual rainfall is around 1079 mm.
Figure 1: Location of the study site
2.2 Land use land cover details
The backdrop of the site is a moist-deciduous mixed forest with over 89% of the area under forests. The study site
mainly comprises of open and degraded mixed forests, with Shorea robusta, Acacia catechu, Madhuca longifolia,
Dendrocalamus strictus, Diospyros melanoxylon and Terminalia tomentosa as the dominant floral species (Sinha et
al., 2013). Located south of River Ganges, the entire area is drained by several small rivers, like Kiul, Man,
Narokol, Morwe, Dudhpanian, Kandani, etc. The forest is a virgin patch with limited disturbances in terms of
deforestation mainly due to dispersed anthropogenic activities like settlements, agricultural development and
plantations and to some extent mining activities. Conservation of the forest is of high relevance for the conservation
of biodiversity and also in context to REDD/REDD+.





 



Paper presented in the 38th Asian Conference on Remote Sensing, October 23-27, New Delhi (INDIA)
2.3 Data input
Multi-frequency SAR datasets of COSMO-Skymed, Radarsat-2 and ALOS PALSAR were procured to model for
AGB prediction. The COSMO-Skymed data is HH/VV dual polarized X-band obtained through PINGPONG
imaging mode with 15m (resampled to 25m) spatial resolution and 30km swath width. The Radarsat-2 data is
Standard Quad-pol C-band with 25m spatial resolution and 25km swath width. The ALOS PALSAR data is L-band
acquired in Fine Beam Single (HH) and Dual (HH,HV), and Quad Polarimetric (HH,HV,VH,VV) mode with 25m
spatial resolution, and off-nadir angles of 34.3° and 21.5°, and swath width of 70 and 30 km respectively.
Simultaneously, primary data was generated in terms of field inventory that included in-situ information generated
from field sample plots, like forest types, canopy density, species composition, stand height, and girth at breast
height (GBH).
2.4 In-situ field inventory for field biomass estimation
A total of 45 square sample plots of 0.1 ha (Hectare) were randomly selected, 36 of which were considered for
developing models, while the remaining for validating the models. The three most important data collected from
field within each plot i.e., tree species, stand height and GBH; from which the diameter at breast height (DBH) were
used to calculate the volume using the species-specific regional volumetric equations of FSI (1996). AGB is then
estimated by multiplying the resulting volume with the tree-specific specific gravity of FRI (1996). Global
Positioning System (GPS) was used to collect the latitude-longitude information of the plots for importing the
information in GIS framework. Volumetric equations and specific gravity of the tree species for the study site are
mentioned in Sinha (2016).
2.5 SAR processing
Raw SAR datasets were preprocessed, rectified, geocoded and calibrated using a series of standard steps in
SARscape software to generate the backscatter image following the equation (Sinha et al., 2016):
0
)(10log10
0ADNa +×=
σ
(1)
where, 
0
= backscatter coefficient or sigma nought values in decibels (dB),DN is the power (or intensity) image, A
0
is the calibration factor that vary with sensor type. A
0
= -115 dB for ALOS PALSAR, A
0
= -59.62 dB for HH
polarized COSMO-Skymed, A
0
= -58.88 dB for VV polarized COSMO-Skymed. Radarsat-2 has different values of
A
0
for each line and is processed in Geomatica.
2.6 Model development
SAR backscatter values were regressed to the 36 field-based AGB values to find the best fit model for calculating
AGB. The field-based estimated AGB was correlated with modeled AGB for 9 additional random points for
validation of the best-fit model. The model performance was evaluated based on certain statistical measures (Sinha
et al., 2016). Resulting modeled AGB (in t/ha) were converted to carbon (C) and carbon dioxide (CO
2
) equivalents
using conversion factors of 0.5 and 3.67 respectively (Mushtaq & Malik, 2014, Rashid et al., 2016).
3. RESULTS
3.1 SAR interactions with AGB
X-, C- and L-bands were used to develop the AGB model. Figure 2 shows the relationship between the SAR
backscatter values and the in-situ plot AGB values, which showed the existence of a logarithmic relation. The figure
also revealed that X-band saturated earliest, then the C-band, and lastly the L-band. The saturation level depended
on the radar frequency or wavelength, radar wave polarization and vegetation types.
Figure 2: Relationship between multi-frequency SAR backscatter with plot AGB

Paper presented in the 38th Asian Conference on Remote Sensing, October 23-27, New Delhi (INDIA)
X-band COSMO-Skymed VV polarized data interacted with the upper part of the canopy (leaves, small branches,
etc.); as revealed by the increased r
2
value in comparison to HH. With minimum penetration capability, X-band has
the minimum interaction with the trunk and the main branches, so has the poorest relationship with the bole AGB.
C-band having relatively greater penetration within the canopy layer interacts with the secondary branches and this
is evident from higher r
2
values. L-band ALOS PALSAR showed the best correlation with the bole AGB due to its
ability to penetrate the canopy layer and interact with the trunk. HH polarized data of L-band showed the greatest
interaction with the trunk due to its vertical structure. Hence, HH polarization was most sensitive towards the bole
AGB. It was observed that like-polarizations had high backscatter values than cross-polarizations and increase in
the wavelength leads to increase in the penetration capability of the radar signals, thus providing more accurate
information related to the bole above-ground biomass. Table 1 documents the correlation values. In this study, X-
band saturates earlier at about 40-50 t/ha, while C-band saturates next nearly at 100-120 t/ha, while, saturation is
highest for L-band among the three, at about 160-180 t/ha (Figure 2).
Table 1: Coefficient of determination (r
2
) between SAR backscatter and plot AGB
SAR datasets Polarization r
2
X-band COSMO-Skymed HH 0.01
X-band COSMO-Skymed VV 0.02
C-band Radarsat-2 HH 0.46
C-band Radarsat-2 HV 0.34
C-band Radarsat-2 VV 0.43
L-band ALOS PALSAR HH 0.85
L-band ALOS PALSAR HV 0.56
3.2 Integrated AGB model
Three best-fit models (Equations 2, 3 and 4) were generated from this logarithmic relationship between backscatter
values from each of the three SAR datasets (X-, C- and L-band respectively) and the plot AGB based on the
information in Table 1 from 36 plots used for model development and calibration. Synergic modeling was
developed using only those plots among the total 36 that were found in all the SAR datasets. An integrated AGB
model was developed using Multiple Linear Regression (MLR) of Equations 2, 3 and 4 where the data were used in
combination of all three datasets, expressed as Equation 5. The models are enlisted in Table 2. The models were
evaluated based on certain statistical calculation, also documented in Table 2. The table indicated that the synergic
model integrating SAR X-, C- and L-bands (Equation 5) showed the best results among all; while Equation 4
involving just the L-band showed better results among all the single band models.
Table 2: AGB model evaluation
Eq. Model SAR data used r
2
RMSE
(t/ha)
2
(
)
σ
o
e
VVX _
*0442.0
*984.94
X (VV) 0.01 46.18
3
(
)
σ
o
e
HHC _
*1874.0
*58.380
C (HH) 0.28 37.24
4
(
)
σ
o
e
HHL _
*2765.0
*3.1067
L (HH) 0.87 16.06
5
(
)
(
)
(
)
8778.58
*2765.0
*984.1028
*1874.0
*3375.44
*0442.0
*504.103
___
−+−
σσσ
o
e
o
e
o
e
HHLHHCVVX
L (HH), C (HH), X (VV) 0.90 15.29
3.3 Validation
The AGB models were statistically validated with nine additional plot AGB data and the corresponding statistical
measures, like r
2
, RMSE, slope, average absolute accuracy () and Willmott's Index of agreement (d) were executed.
The results are summarized in Table 3. The table confirmed that the integrated AGB model (Equation 5) showed the
best results with the highest r
2
value of 0.954, least RMSE of 14.813 t/ha, good model accuracy of about 79% and
greatest d value of 0.95. Significantly high r
2
, model accuracy and d values nearing unity and low RMSE value
reveal the acceptability of the model. However, Equation 4 with L-band information also showed promising results
amongst the single sensor models with a fairly high r
2
value of 0.713, moderate RMSE of 22.34 t/ha, good model
accuracy of 61.7% and d value of 0.88.

Paper presented in the 38th Asian Conference on Remote Sensing, October 23-27, New Delhi (INDIA)
Table 3: AGB model validation
Eq. r
2
RMSE (t/ha) Slope  d
2 0.094 32.151 0.105 39.478 0.306
3 0.002 50.083 -0.050 7.961 -0.095
4 0.713 22.340 0.871 61.768 0.883
5
0.954
14.813
0.964
78.894
0.950
3.4 AGB and C maps
The AGB map represented as Figure 3 was prepared from Equation 5 in GIS and reclassified in ten classes
according to biomass levels from very low (<25t/ha and 25-50 t/ha), low (50-75t/ha and 75-100t/ha), moderate
(100-125t/ha and 125-150t/ha), high (150-175t/ha and 175-200t/ha) to very high (>250t/ha). Resulting modeled
AGB were converted to carbon (C) and carbon dioxide (CO
2
) equivalents using conversion factors and reclassified
in five classes. Figure 4 illustrates both C and CO
2
spatial distribution map. Both the figures 3 and 4 have white or
empty regions within the boundary of the study area that lack the input SAR data owing to data restrictions. North
portion of the of study site in the figures show erroneous results due to presence of non-overlapping portions of
multi-sensor satellite data, demarcated in blue dotted line.
Figure 3: AGB map developed from integrated model
Figure 3 portrays less vegetative parts with low biomass levels of <50 t/ha at the vicinity of the boundary of the
study area, water bodies and the built-up regions that are concentrated in the periphery and at the central parts. Most
of the forested region was observed to lie within the biomass range of 25–100 t/ha, with an average value of 56.24
t/ha. Most of the high density vegetation was observed to have biomass ranging from 75–125 t/ha, mostly covering
the interior parts of the study area; however, some scattered areas with even higher biomass values that too are
generally restricted to the interior regions. Likewise, Figure 4 shows similar observation as C and CO
2
were
observed to be concentrated more in the interior parts of the area. Average C was calculated to 28.12 t/ha, while
CO
2
was 103.2 t/ha. It can be also noticed that the relative early saturation of biomass with the use of single SAR
frequency data alone can be counteracted with the integrated use of multi-frequency SAR data due to the better
relationship of AGB compartments to each of the SAR frequency bands. Hence, the integrated multi-frequency
SAR model provided the most accurate result for predicting AGB.

Paper presented in the 38th Asian Conference on Remote Sensing, October 23-27, New Delhi (INDIA)
Figure 4: C and CO
2
spatial map
4. CONCLUSIONS
In this ‘400 ppm World’ , global climate change is the most alarming situation that this world is experiencing. In this
context, REDD and associated concepts are gaining attention; henceforth, biomass/carbon assessment is becoming
crucial to address this issue. The current study targets in using multi-frequency SAR to assess above-ground bole
biomass and in-turn the forest carbon stock over a tropical deciduous heterogeneous virgin forest patch of Munger
in India. Synergic use of multi-frequency SAR has the potential to augment the AGB estimations in comparison to
any optical data as well as any single frequency spaceborne SAR data till date. Out of the X-, C- and L-bands, the
best model predicting AGB comprises of L-band information. Subsequently, the estimation improves on integrating
all the three SAR wavelength bands. The exponential model was observed as the best fit model for estimating
biomass on regressing SAR backscatter values to plot estimated AGB. The integrated model was validated and r
2
of
0.95, RMSE of 14.81 t/ha, model accuracy of 79% and Willmott’s index of agreement of 0.95 was calculated
without much over or under-estimation as denoted by the slope value of 0.96. The consequence of the approach in
resolving the relative early saturation of biomass with optical RS data or single SAR frequency data alone is evident
thus can be counteracted with the integrated use of multi-frequency SAR data due to the better relationship of AGB
compartments.
Hence, in this study, a synergy regression model for predicting AGB was developed with synergic use of SAR
multi-frequency X-, C- and L-band information that was further transformed to generate C stock and CO
2
emission
models in GIS. Quantifying multi-temporal changes in CO
2
via this approach can account for the climate change.
The integrated multi-frequency SAR approach adopted in the study gave valuable information of the spatial
distribution and quantification of the forest biomass and carbon; important for REDD monitoring.
5. ACKNOWLEDGEMENTS
The authors sincerely acknowledge Space Application Centre (ISRO, India) for providing technical help and JAXA
(Japan), CSA (Canada) and ASI (Italy) for providing SAR data through ALOS-RA Cosmo Skymed A.O. project.
The first author expresses sincere gratitude to Science and Engineering Research Board (SERB), Department of
Science and Technology (DST), Government of India for providing funds under SERB National Post-Doctoral
fellowship (SERB NPDF) scheme (File Number: PDF/2015/000043).

Paper presented in the 38th Asian Conference on Remote Sensing, October 23-27, New Delhi (INDIA)
6. REFERENCES
Alappat, V. O., Joshi, A. K. and Krishnamurthy, Y. V. N., 2011. Tropical dry deciduous forest stand variable
estimation using SAR data. Journal of the Indian Society of Remote Sensing, 39 (4), pp. 583-589.
Betts, R. A., Jones, C. D., Knight, J. R., Keeling, R. F. and Kennedy, J. J., 2016. El Niño and a record CO
2
rise.
Nature Climate Change, 6, pp. 806–810.
Englhart, S., Keuck, V. and Siegert, F., 2011. Aboveground biomass retrieval in tropical forests — the potential of
combined X- and L-band SAR data use. Remote Sensing of Environment, 115, pp. 1260-1271.
FRI, 1996. Indian Woods. Forest Research Institute, Dehradun.
FSI, 1996. Volume equations for forests of India, Nepal and Bhutan. Forest Survey of India, Ministry of
Environment and Forests, Government of India, Dehradun.
Hunter, M. O., Keller, M., Victoria, D. and Morton, D. C., 2013. Tree height and tropical forest biomass estimation.
Biogeosciences, 10, pp. 8385–8399.
Kumar, P., Sharma, L. K., Pandey, P. C., Sinha, S. and Nathawat, M. S., 2013. Geospatial strategy for tropical
forest-wildlife reserve biomass estimation. IEEE Journal of Selected Topics in Applied Earth Observations and
Remote Sensing, 6 (2), pp. 917–923.
Mushtaq, H. and Malik, T., 2014. Accounting carbon dioxide emission and stratification of carbon stock in Western
Ghats, India. A geospatial approach. International Journal of Remote Sensing & Geoscience, 3 (1), pp. 1-5.
Rashid, I., Bhat, M. A. and Romshoo, S. A., 2016. Assessing changes in the above ground biomass and carbon
stocks of Lidder valley, Kashmir Himalaya, India. Geocarto International, doi: 10.1080/10106049.2016.1188164.
Ravindranath, N. H., Joshi, N. V., Sukumar, R. and Saxena, A., 2006. Impact of climate change on forests in India.
Current Science, 90 (3), pp. 354–361.
Sharma, L. K., Nathawat, M. S. and Sinha, S., 2013. Top-down and bottom-up inventory approach for above
ground forest biomass and carbon monitoring in REDD framework using multi-resolution satellite data.
Environmental Monitoring and Assessment, 185 (10), pp. 8621–8637.
Sinha, S., 2016. Polarimetric Scattering Parameter products of ALOS PALSAR for forest biomass assessment.
Research & Reviews: Journal of Space Science & Technology, 5 (1), pp. 1-9.
Sinha, S., Jeganathan, C., Sharma, L. K. and Nathawat, M. S., 2015. A review of radar remote sensing for biomass
estimation. International Journal of Environmental Science and Technology, 12 (5), pp. 1779-1792.
Sinha, S., Jeganathan, C., Sharma, L. K., Nathawat, M.S., Das, A.K. and Mohan, S., 2016. Developing synergy
regression models with space-borne ALOS PALSAR and Landsat TM sensors for retrieving tropical forest biomass.
Journal of Earth System Science, 125 (4), pp. 725-735.
Sinha, S., Santra, A., Sharma, L. K., Jeganathan, C., Nathawat, M.S., Das, A.K. and Mohan, S., 2017. Multi-
polarized Radarsat-2 satellite sensor in assessing forest vigor from above ground biomass. Journal of Forestry
Research, doi: 10.1007/s11676-017-0511-7.
Sinha, S., Sharma, L. K. and Nathawat, M. S., 2013. Integrated geospatial techniques for land-use/land-cover and
forest mapping of deciduous Munger forests (India). Universal Journal of Environmental Research and Technology,
3 (2), pp. 190-198.
Su, Y., Guo, Q., Xue, B., Hu, T., Alvarez, O., Tao, S. and Fang, J., 2016. Spatial distribution of forest aboveground
biomass in China: Estimation through combination of spaceborne lidar, optical imagery, and forest inventory data.
Remote Sensing of Environment, 173, pp. 187–199.
Waikhom, A. C., Nath, A. J. and Yadava, P. S., 2017. Aboveground biomass and carbon stock in the largest sacred
grove of Manipur, Northeast India. Journal of Forestry Research, doi: 10.1007/s11676-017-0439-y.