Remote Sensing of Environment 183 (2016) 65–81

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Mapping spatial distribution and biomass of coastal wetland vegetation

in Indonesian Papua by combining active and passive remotely
sensed data
Aslan Aslan a, Abdullah F. Rahman b,⁎, Matthew W. Warren c, Scott M. Robeson d
a
School of Public and Environmental Affairs, Indiana University, Bloomington, IN 47405, USA
b
Coastal Studies Lab, University of Texas Rio Grande Valley, San Padre Island, TX 78597, USA
c
Northern Research Station, USDA Forest Service, Durham, NH 03824, USA
d
Department of Geography, Indiana University, Bloomington, IN 47405, USA

a r t i c l e i n f o a b s t r a c t

Article history: There is ongoing interest to develop remote sensing methods for mapping and monitoring the spatial distribution
Received 30 July 2015 and biomass of mangroves. In this study, we develop a suite of methods to evaluate the combination of Landsat-8,
Received in revised form 13 April 2016 ALOS PALSAR, and SRTM data for mapping spatial distribution of mangrove composition, canopy height, and
Accepted 30 April 2016
aboveground biomass in the wide intertidal zones and coastal plains of Mimika district, Papua, Indonesia.
Available online xxxx
Image segmentation followed by visual interpretation of composite PALSAR images was used to delineate man-
Keywords:
grove areas, whereas a flexible statistical rule based classification of spectral signatures from Landsat-8 images
Indonesia was used to classify mangrove associations. The overall accuracy of land cover classification was 94.38% with a
Mangroves kappa coefficient of 0.94 when validated with field inventory data and Google Earth images. Mangrove height
Classification and aboveground biomass were mapped using the SRTM DEM, which were calibrated with field-measured
Biomass data via quantile regression models. There was a strong correlation between the SRTM DEM and the 0.98 quantile
Landsat-8 of field canopy heights (H.98), which was used to represent the tallest trees in each of 196 10 m radius subplots
Radar (r = 0.84 and R2 = 0.804). Model performance was evaluated through 10,000 bootstrapped simulations, produc-
ing a mean absolute error (MAE) of 3.0 m for canopy height estimation over 30 m pixels of SRTM data. Quantile
regression revealed a relatively strong non-linear relationship between the SRTM derived canopy height model
and aboveground biomass measured in 0.5 ha mangrove inventory plots (n = 33, R2 = 0.46). The model results
produced estimates of mean standing biomass of 237.52 ± 98.2 Mg/ha in short canopy (Avicennia/Sonneratia)
stands to 353.52 ± 98.43 Mg/ha in mature tall canopy (Rhizophora) dominated forest. The model estimates of
mangrove biomass were within 90% confidence intervals of area-weighted biomass derived from field measure-
ments. When validated at the landscape scale, the difference between modeled and measured aboveground
mangrove biomass was 3.48% with MAE of 105.75 Mg/ha. These results indicate that the approaches developed
here are reliable for mapping and monitoring mangrove composition, height, and biomass over large areas of
Indonesia.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction
Mangroves grow exclusively within the intertidal zones of coastal,
estuarine, and riverine landforms of the tropics and subtropics (Giri
et al., 2011). The ecological functions of mangroves are critical to the en-
vironmental health of near-shore marine environments and adjacent
upstream terrestrial systems, as they are transitional ecosystems
where ocean, land, and freshwater converge (Kathiresan & Bingham,
2001; Lugo & Snedaker, 1974; Suratman, 2008; Tomlinson, 1986).
They provide at least US $1.6 billion each year globally in ecosystem ser-
vices while supporting the livelihoods of coastal communities (Polidoro
⁎ Corresponding author.
et al., 2010). Mangrove diversity and occurrence are highest in South-
east Asia, where nearly 75% of the World's mangrove species occur.
Indonesia alone contains about 19% of the World's mangrove forest
area, a total of 3.2 million hectares, and 45 of about 70 “true” mangrove
species (species which meet physiological criteria and are adapted to
occupy saline, hypoxic intertidal habitats; see Bakosurtanal, 2009;
FAO, 2007; Tomlinson, 1986). Approximately half of Indonesia's man-
groves are located in the Papua and West Papua provinces and remain
largely intact, whereas mangroves in other parts of Indonesia have ex-
perienced very high deforestation rates in recent decades due to aqua-
culture and infrastructure development (Ilman, Wibisono, &
Suryadiputra, 2011; Rahman, Dragoni, Didan, Barreto-Munoz, &
Hutabarat, 2013). A recent assessment suggests Indonesia has lost up
to 40% of its original mangrove cover since 1980. Most of these losses

http://dx.doi.org/10.1016/j.rse.2016.04.026
0034-4257/© 2016 Elsevier Inc. All rights reserved.
66 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
have occurred in Sumatra, Kalimantan, Sulawesi and Java (Ilman et al.,
2011; Murdiyarso et al., 2015).
The southern coast of Indonesian Papua contains some of the largest
contiguous mangrove forests on Earth, including large tracts on Bintuni
Bay and on the wide coastal plains of Mimika and Asmat districts
(Fig. 1). An estimated 193,226 ha of mangrove forest cover the coastal
lowlands of Mimika, a southern district of Papua Province (MoF,
Ministry of Forestry of Indonesia., 2008). These mangroves are mostly
pristine and of high ecological value as they support productive near-
shore fisheries, harbor unique biodiversity, and are critical sources of
food and other natural products to local communities (ARD, 2014).
However, despite the importance of Mimika's mangrove forests and
coastal wetlands in providing ecosystem services (Barbier, 2012;
Clough & Abdullah, 1993; Costanza et al., 1997; Mumby et al., 2004;
Vo, Kuenzer, Vo, Moder, & Oppelt, 2012), they are vulnerable to the
rapid large-scale conversion and mismanaged deforestation observed
elsewhere in Indonesia (Margono, Potapov, Turubanova, Stolle, &
Hansen, 2014; Miettinen, Shi, &Liew, 2011; Rahman et al., 2013). As
coastal resources of other Indonesian Islands become depleted, it is
likely that Papuan mangroves will become more attractive to devel-
opers intending to expand aquaculture production and other industries.
In addition to the well-known ecological benefits they provide, man-
groves are gaining increasing attention for their capacity to sequester
carbon (they are a key component of “Blue Carbon”) and potential in-
volvement in climate change mitigation efforts (Alongi, 2012; Chmura,
Anisfeld, Cahoon, & Lynch, 2003; Donato et al., 2011; Murdiyarso,
Fig. 1. Map of the study area in the Mimika district of Papua, Indonesia, showing the location of field sampling plots and random validation points bounded between 0 and 150 m ASL
elevation.
Kauffman, & Verchot, 2013). Mangroves in the Indo-Pacific region are
reported to contain carbon stocks exceeding 1000 MgC/ha, mostly
stored in buried sediments (Donato et al., 2011; Fujimoto et al., 1999).
Per unit area, this capacity to store carbon is several times higher than
that of tropical forests on dry land (Donato et al., 2011). Therefore, con-
servation and improved management of the remaining mangroves pro-
vides an avenue to avoid large-scale CO2 emissions and loss of
sequestered C from mangrove conversion. Such Blue Carbon mitigation
strategies could contribute substantially to Indonesia's voluntary pledge
to reduce greenhouse gas emissions 26% by 2020 (GOI, Government of
Indonesia, 2011). However, involving mangrove conservation in climate
mitigation programs such as Reducing Emissions from Deforestation
and Forest Degradation (REDD +) and Blue Carbon (McLeod et al.,
2011; Pendleton et al., 2012) requires intensive mapping and monitor-
ing to create baseline C stock data and to validate conservation efforts.
Such mapping is also needed for improved spatial planning and inte-
grated coastal management.
The goal of this study was to develop a systematic approach for map-
ping the distribution and biomass of different species and associations
of mangroves at a consistent pixel scale by combined use of multispec-
tral and Synthetic Aperture Radar (SAR) remote sensing data. Specific
objectives include:
1) To investigate whether improved mapping of mangroves at genus
or even species level may be achieved by using images from the
recently launched (February 2013) Landsat-8, which has 16-bit
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
radiometric resolution (65,536 grey levels). To address this objec-
tive, we develop a Flexible Statistical Expert Based (FSEB) system
and apply an image segmentation technique, rather than using tra-
ditional unsupervised or supervised classification methods; and
2) To evaluate the effectiveness of using a combination of SAR and
Landsat-8 data for retrieving a reliable estimation of mangrove can-
opy heights and its standing biomass.
Following this introduction section the paper is structured as fol-
lows. Section 2 describes a short literature review of past research.
The materials and methods are presented in Section 3, which provides
a brief description of study area, field and remote sensing data (SAR
and Landsat-8) used in this study, as well as approaches used to process
the field and remote sensing data. Section 4 demonstrates the imple-
mentation of the analysis methods developed in Section 3 and its corre-
sponding discussion, whereas Section 5 presents the conclusions
derived from this study.
2. Literature review
Remote sensing provides a fast, cost-effective, and efficient method
of mapping the spatial distribution and biomass of forests, and monitor-
ing their disturbance dynamics (Held, Ticehurst, Lymburner, &
Williams, 2003). It is particularly useful for mangrove monitoring
since many mangrove forests are located in remote locations accessible
only by boat, where field measurements are difficult, time consuming,
and expensive. Producing accurate maps of mangrove distribution and
biomass requires image classification procedures that can discern varia-
tions in mangrove forest structure, which often depends on age, canopy
height, and species associations.
A few studies have demonstrated the potential of hyperspectral and
high spatial resolution optical remote sensing data for detailed mapping
of mangrove forest composition (Green, Mumby, et al., 1998a; Held
et al., 2003; Wang, Sousa, & Gong, 2004; Neukermans, Dahdouh-
Guebas, Kairo, & Koedam, 2008). Green, Mumby, et al. (1998a) have
successfully classified the coastal forests of the Turks and Caicos Islands
into six mangrove and three non-mangrove classes, with overall accu-
racy of 78.2%, using hyperspectral data from the Compact Airborne
Spectrographic Imager (CASI) sensor. Another study used high spatial
resolution (0.7 m) QuickBird pan-sharpened imagery to map man-
groves in Gazy Bay, Kenya, into four dominant species with overall accu-
racy of 72% (Neukermans et al., 2008). Yet, one major problem with the
use of hyperspectral or high spatial resolution imagery is that unlike
Landsat or SPOT data, these datasets are not commonly available for
the vast majority of mangrove areas, and when available, are cost pro-
hibitive for monitoring and mapping mangroves changes at regional
to national scales.
Several studies have classified mangroves into broad classes using
medium spatial resolution multispectral optical sensors such as SPOT
(Système Pour l'Observation de la Terre) (Gao, 1998; Jensen et al.,
1991; Rasolofoharinoro et al., 1998), and Landsat Thematic Mapper
(TM) (Brondízio, Moran, Mausel, & Wu, 1996; Green, Clark, et al.,
1998b; Long & Skewes, 1996). These studies have indicated that de-
tailed mangrove characterization is generally cumbersome at such
moderate spatial resolution. Other studies have shown that vegetation
indices from multispectral sensors can be linked to mangrove leaf area
index (LAI) and canopy heights (Ramsey & Jensen, 1996). According
to Green, Clark, et al. (1998b), two probable reasons for the limited suc-
cess in differentiating among many classes of mangroves using moder-
ate resolution multispectral satellite data may have been related to:
1) sensor's resolution, particularly the relatively low 8-bit (256 grey
levels) radiometric resolution of those satellite sensors (Al-Wassai &
Kalyankar, 2013; Masek et al., 2001); and 2) the ability of the image pro-
cessing method (i.e., classification procedure) to discriminate between
various mangrove types.
67
Use of radar data in mangrove mapping is challenging because of the
complexity of the backscatter signal received from mangroves (Lucas
et al., 2007). Different bands of radar backscatter are affected differently
by the interactions between the transmitted signal and biophysical
properties of mangrove plants, such as size, geometry, orientation of
leaves, trunks, branches, different types of roots, and the moisture con-
tent of both plants and underlying soil (Kuenzer, Bluemel, Gebhardt,
Quoc, & Dech, 2011; Lucas et al., 2007). For example, the longer wave-
length L-band (wavelengths of 15–30 cm) has a greater capacity of pen-
etrating the foliage and small branches of the upper canopies of forests
including woody trunk and the underlying ground surfaces (Lucas,
Moghaddam, & Cronin, 2004), whereas the shorter wavelength C-
band (wavelengths of 3.75–7.5 cm) not only has a limited penetration
through the canopy of forest, but is also heavily influenced by scattering
caused by the canopy layer (Lang & Kasischke, 2008). SAR data have
been used for mapping mangrove cover, structure, and biomass with
some success (Fatoyinbo & Simard, 2013; Held et al., 2003; Lucas
et al., 2009; Mougin et al., 1999; Proisy, Mougin, Fromard, & Karam,
2000; Proisy, Mougin, Fromard, Trichon, & Karam, 2002; Simard et al.,
2006, 2008). Several recent studies combined elevation data from Shut-
tle Radar Topographic Mission (SRTM) with spaceborne lidar data from
the ICESat/GLAS satellite to estimate canopy height and biomass of
mangroves, producing promising results (Fatoyinbo, Simard,
Washington-Allen, & Shugart, 2008; Fatoyinbo & Simard, 2013; Simard
et al., 2006, 2008).
A few studies have also explored the potential of combining SAR and
optical remotely sensed data for mapping mangrove ecosystems
(Aschbacher et al., 1995; Held et al., 2003; Kushwaha, Dwivedi, & Rao,
2000; Nascimento Jr. et al., 2013; Ramsey III et al., 1998). Lucas et al.
(2007) have shown the potential of using L-band SAR data for mapping
mangrove extent and zonation in Australia, New Guinea, and Malaysia
where differences in structure occurred between zones as a function
of species, growth stage, and biomass distributions. A recent study by
Flores De Santiago, Kovacs, and Lafrance (2013) has successfully sepa-
rated mangrove areas from saltpan and water/shallow zones with an ac-
curacy of 91.1% for single polarization and 92.3% for dual polarization of
PALSAR images, respectively. Cloud-free multispectral data offer the
ability to characterize canopy coverage including different shades of
leaf greenness and composition of mangrove species in a given area,
whereas SAR data provide images that are sensitive to mangrove forest
structure including canopy height.
3. Materials and methods
3.1. Study area
This research was carried out within the Mimika district of
Indonesia's Papua province, located between 4° and 5°20′ S and
135°40′ to 137°45′ E, in the south-central coast of Papua. As this study
focused on mangroves and associated coastal wetland vegetation, we
limited the study area to 0–150 m above sea level (ASL), extending
from the coastline landward. Based on global mangrove data, our as-
sumption is that mangroves do not occur at elevations greater than
the 150 m threshold (Fig.1). The Mimika coastal plains are characterized
by broad swampy areas extending from the coast in a northeast direc-
tion with mountains rising abruptly from the edge of the plains. The
swamps are intersected into narrow catchments by a number of
north-south parallel rivers flowing from the mountains to the sea.
Tide records from Puriri Island (Fig.1) indicate diurnal to semi-diurnal
patterns with a tidal range between 3.3 and 3.6 m (Ellison, 2005). The
wet and dry seasons are not clearly distinguishable, but the southeast
monsoons typically increase rainfall from June to mid-September
(Pouwer, 1970). Small settlements of the indigenous Kamoro and
semi-nomadic clans of Asmat ethnic groups inhabit the coastal wet-
lands; however, traditional lifestyles have little direct impact on the
coastal wetland forests (Marshall, 2007). The rich natural and cultural
68 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
resources of coastal Mimika are currently facing increasing threats due
to rapid population growth and in-migration, a booming mining indus-
try, and rapid forest conversion to oil palm plantations in a recently
granted 38,000 ha concession to the west of Timika (Fig. 1).
3.2. Data acquisition and analysis
3.2.1. Field data
During July–December of 2013, two major field campaigns were
completed, including forest inventory measurements of 41 transect
plots of 0.5 ha each and 25 field verification plots of 0.09 ha each
(Fig. 2). Each of the 41 transect plots contained six nested circular sub-
plots (10 m radius), resulting in 246 subplots. For these 41 plots, we
followed the sample plot design and procedure used in the Indo-
Pacific Forest Carbon Study and current Sustainable Wetlands Adaption
and Mitigation Program (SWAMP; www.cifor.org/swamp) described in
detail by Kauffman and Donato (2012). At each of the field plots, tree
height (H) was measured using hypsometer and estimated to the
nearest meter, and stem diameter (cm) was measured at breast height
(DBH) or relevant point of measurement. Trees N 50 cm in diameter
were sampled in 125 m by 40 m rectangular plots, whereas trees
N5 cm were sampled in six 10 m radius subplots arranged every 25 m
along the 125 m center line of the plot (Fig. 2). We identified trees to
genus or species level, measured height and diameter, and recorded
the coordinates of subplot centers using a handheld global positioning
system (GPS) receiver.
We employed Ward's method of hierarchical cluster analysis to clas-
sify dominant mangrove forest types (Ward Jr., 1963). Input data for
this analysis were species-level ‘Importance Values’ of mangrove and
non-mangrove species recorded in each 0.5 ha plot (Peet, 1974). Impor-
tance Value is an integrated measurement of density (number of trees
per plot), dominance (proportion of total basal area occupied by a
given species), and frequency (number of subplots the species was re-
corded within the larger 0.5 ha plot). A stepwise quantile regression
technique was used to obtain the best representation of field canopy
heights at the scale of the 20 m diameter subplots because the sizes of
Landsat-8 and SRTM pixels (30 m) and PALSAR pixels (12.5 m) are rel-
atively more comparable to the subplot size than to the sizes of individ-
ual canopies. Aboveground biomass of mangrove and non-mangrove
trees was calculated at subplot and plot levels using published
species-specific and general allometric equations (Chave et al., 2005;
Clough & Scott, 1989; Kauffman & Donato, 2012; Komiyama,
Poungparn, & Kato, 2005). Allometric equations describe the close rela-
tionship between tree biomass and a combination of diameter, wood
density, and height (Baker et al., 2004; Chambers, Santos, Ribeiro, &
Higuchi, 2001; Chave et al., 2005; Nascimento & Laurance, 2002;
Phillips et al., 1998). When species-specific values were not available
for general allometric equations, average wood density for the genus
or most closely related species was used (a complete list of allometric
equations used for this study can be found in Table S1).
Fig. 2. Mangroves in Papua, Indonesia, and plot design. A demonstration of measuring and inventorying mangrove trees (A); design of main transect plots (0.5 ha) and six subsequent
subplots in the left, with additional plots for land cover map verification in the right (B). Coordinates of each location were taken with a GPS receiver.
3.2.2. Multispectral data
Four Landsat-8 Level 1 terrain-corrected (L1T) scenes were used in
this study (path/row 102/64 captured on 30 March 2013, path/row
102/63 capture on 30 March 2013, path/row 103/63 capture on 1
April 2014, and path/row 104/63 capture on 23 May 2013). The images
were downloaded from the US Geological Survey National Center for
Earth Resources Observation and Science through the EarthExplorer
data portal (http://earthexplorer.usgs.gov/). These images have been
geocoded in the Universal Transverse Mercator (UTM) projection sys-
tem zone 53 S and WGS84 datum. The digital number (DN) of each
band was converted to reflectance value using the ATCOR tool in
ERDAS Imagine®. For image analysis, we used the bands 1 to 7 from
the operational land imager (OLI) sensor of Landsat-8. As mentioned
earlier, Landsat-8 data has 16-bit radiometric resolution, which may
have the potential to identify spectrally distinct species/genus of man-
groves. In addition, Landsat-8 also has a low geolocation uncertainty,
b12 m circular error (Irons, Dwyer, & Barsi, 2012).
3.2.3. SAR data
Two types of SAR data were used in this study: 26 scenes of Phased
Array L-band Synthetic Aperture Radar (PALSAR) images from the Ad-
vanced Land Observing Satellite (ALOS) platform and four scenes of dig-
ital elevation model (DEM) images from the C-band SAR onboard the
Shuttle Radar Topography Mission (SRTM) spacecraft. PALSAR data
were used for land cover classification to distinguish mangroves from
non-mangroves. We used the level 1.5 Fine Beam Dual (FBD) Polariza-
tion PALSAR images that have 12.5 m spatial resolution, obtained from
the Alaska Satellite Facility (ASF) data portal (University of Alaska Fair-
banks, https://ursa.asfdaac.alaska.edu/). The FBD PALSAR images were
composed of HH (horizontal transmitted and horizontal received) and
HV (horizontal transmitted and vertical received) polarizations and
their acquisition dates were during June and July of 2007. These images
were processed, calibrated, and geo-rectified to UTM projection (zone
53 S and WGS-84 datum) with ASF MapReady software version 3.2.1
in order to convert the DN values to radar backscatter coefficients (σ°,
unit of decibels dB) using the following formula (Shimada, Isoguchi,
Tadono, & Isono, 2009):


 
σ ðdBÞ ¼ 10 x log10 DN2 −83:4: ð1Þ
These backscatter images were then mosaicked to cover the entire
area of study. Additional georectification adjustment of the mosaicked
PALSAR image was done using the Landsat-8 images as reference,
resulting in Root Mean Squared Error (RMSE) of 7.28 m. A 7 by 7 Lee fil-
ter was applied to the georectified and stacked PALSAR mosaic to re-
move speckle noise (Lee, 1980). Using the HH and HV data, we
calculated the Radar Forest Degradation Index (RFDI), which is a ratio
of the power difference between HH and HV polarizations and the
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
total power of HH and HV polarizations (Mitchard, 2012).
HH−HV
RFDI ¼
HH þ HV
The RFDI is potentially a useful input layer for wetlands vegetation
mapping because HH polarized data has been found to be sensitive for
detecting water beneath canopy (Hess, Melack, Filoso, & Wang, 1995),
whereas HV polarized data has been known for its sensitivity to volume
scattering and biomass (Bourgeau-Chavez, Riordan, Powell, & Miller,
2009; Mitchard, 2012). Finally, the HH, HV and RFDI layers were stacked
to make one multi-layered image for further analyses.
In addition, we also used the SRTM DEM data with 30 m spatial res-
olution to identify potential mangrove areas based on elevation. The
DEM data were provided by the US Forest Service. Although the primary
mission of SRTM was to deliver high-resolution topographic data, man-
grove trees normally thrive in intertidal zones; hence, a ground eleva-
tion close to the mean sea level can be used to identify potential
mangrove areas (Treuhaft, Law, & Asner, 2004). The SRTM DEM derived
from the Interferometric Synthetic Aperture Radar (InSAR) technique
has been shown to correspond with mangrove canopy height (Simard
et al., 2006). Although there is a 13 + year difference between the
SRTM data collected in February of 2000 and field data collected in
late 2013, this discrepancy does not affect our results because these
are intact ecosystems and presumably in a steady state. Any error at
the pixel level because of treefall gaps or natural forest dynamics should
be inherent in the random variability. There may be some dynamics on
the edge of the coastlines but that is such a small proportion of the total
area it should not have any impacts on the results. We georectified the
SRTM images with Landsat-8 images resulting in an RMSE of 4.82 m.
Overall, the image processing procedure of this study is schematically
shown in Fig. 3.
Fig. 3. A flow diagram showing the steps involved in the image analysis for producing land cover map and estimating mangrove canopy height and standing biomass.
69
3.2.4. Mangrove delineation
We separated mangrove areas from non-mangroves by using image
segmentation on the composite PALSAR image containing HH, HV, and
ð2Þ RFDI (Fig. 4). Image segmentation refers to the process in which raster
images are partitioned into spatially continuous, disjointed and homo-
geneous regions (segments) based on pixel values and locations
(Jensen, 2005). This method is suitable for initial separation of man-
grove types based upon statistical similarity of regions (Held et al.,
2003), and focuses on separating groups of similar pixels (textural and
spectral), rather than individual pixels (Blaschke, 2010). The resulting
segmented mangrove delineation image was then exported to a vector
map to be processed further through refining the mangrove boundary
with visual interpretation using on-screen digitization in ArcGIS (ESRI,
2014, version 10.2.1). In the process of refining the mangrove boundary,
we used atmospherically corrected Landsat-8 images as a background
(Green, Clark, et al., 1998b).
3.2.5. Land cover classification using flexible statistical expert-based
method
We introduce a novel technique in this study for classifying multispec-
tral images, called Flexible Statistical Expert Based (FSEB) classification.
The FSEB is a “pixel-based” sequential classification technique developed
based on flexible determination of standard deviation thresholds of spec-
tral profiles generated for “surface objects” that have been derived from
the training area of 16-bit multispectral Landsat-8 data.
With the FSEB method, the land cover classification process starts by
extracting the most distinct surface features on the input Landsat-8 im-
agery (water and bare soil) by applying water and soil masks built from
normalized difference water index (NDWI) (Gao, 1996) and soil ad-
justed vegetation index (SAVI) (Huete, 1988) respectively. Both of
these classes are excluded from the rest of classification procedure.
Then the Landsat-8 images were masked into two major areas: ‘man-
groves’ and ‘non-mangroves,’ based on input from the mangrove
70 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
Fig. 4. An RGB composite image (R: HH polarization, G: HV polarization, and B: RFDI) of mosaicked ALOS PALSAR images showing delineation of mangrove forests in Mimika district (green
line). Mangrove extent was then verified and refined using atmospherically corrected Landsat-8 images. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)
delineation file, as described earlier. Then we identified several training
sites of targeted classes (e.g., Rhizophora dominated mangroves) from
the high-resolution QuickBird images available on Google Earth. Google
Earth was utilized as a visualization tool in locating samples of the train-
ing area of each targeted class, and the spectral signature used for clas-
sification were derived from the same areas of the Landsat-8 images.
Other studies have demonstrated the potential of using Google Earth
as visualization tool or reference data to collect training area samples
for assisting land use/cover classification (Cha & Park, 2007; Hu et al.,
2013; Yu & Gong, 2012). Based on the spectral information (Landsat-8
bands 1–7 reflectance) from these training sites, we identified all pixels
in the ‘mangrove’ areas of the Landsat-8 image using the following clas-
sification algorithm:
if μ i −σ ij ≤bandi ≤μ i þ σ ij then bandi ¼ 1; otherwise bandi ¼ 0 ð3Þ
where μ and σ respectively represent mean and standard deviation of
grouped pixel samples from the training sites, i represents band num-
bers of the OLI sensor of Landsat-8, and j represents the positive real
number of selected standard deviation. The output is a binary image
consisting of the ‘targeted’ class and ‘all other mangroves.’
After one mangrove class was successfully extracted, we masked out
that extracted class from the mangrove areas of the Landsat-8 scene. On
the rest of the pixels we then repeated the procedure for other targeted
classes of mangrove types until all pixels within the mangrove areas
were properly assigned a class. Once the mangrove areas were classi-
fied, we repeated the same procedure for several classes of coastal veg-
etation in non-mangrove areas including marshlands, sago palm, and
riparian forest. Finally, the mosaic tool in ERDAS Imagine® was used
to combine all extracted classes and produce a coastal land cover map
of the entire area of study. The sequential nature of FSEB is an important
aspect of this study, which we found to be an improved classifier of
mangroves.
3.2.6. Estimation of mangrove canopy height and biomass
The SRTM based elevation data were used to estimate canopy height
and aboveground biomass of the mangroves. Canopy height was
modeled by regressing the SRTM data onto the field canopy height mea-
surements at the 246 subplots within the 41 forest plots. The 25 m tran-
sect interval of linearly arranged subplots inside each main plot was
comparable to the size of a 30 m SRTM pixel. Ideally, when overlaid
with SRTM imagery, the coordinate of each of these 246 subplots should
closely match a single pixel. However, in some cases multiple subplot
coordinates were located within a single SRTM pixel because of the var-
iable accuracy of handheld GPS devices used in the field. In cases where
this occurred, duplicates were averaged to achieve one value of canopy
height per pixel. This procedure resulted in 196 out of 246 subplots
being used to develop the canopy height model for the 30 m SRTM
DEM pixels. The best model for mangrove canopy height was deter-
mined from the highest coefficient of determination (R2). In addition,
a nonlinear quantile regression (NLQR) technique (Koenker & Bassett
Jr., 1978) was used to model standing biomass of mangrove via
regressing the mean of estimated SRTM canopy height resulting from
the previous process and field measured biomass of 33 0.5 ha mangrove
forest inventory plots (Fig. 5). Analysis of variance (ANOVA) was used to
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81 71

4. Results and discussion

4.1. Spatial and structural heterogeneity of Mimika's coastal wetlands

Forest

A total of 8353 stems from 20 mangrove species and 38 non-

Fig. 5. Different intervals of nonlinear quantile regression (NLQR) were examined using
ANOVA to determine the best predictor of biomass model. The nonlinear least squares
(NLS) relationship is shown for reference.
evaluate and select the best model for estimating aboveground biomass
of mangroves in the study area from the series of NLQR.
3.2.7. Accuracy assessment and model performance evaluation
Due to the limited number of ground validation points (66 points,
the total number of field plots), we collected additional random valida-
tion points from the available high resolution DigitalGlobe images on
Google Earth to examine the accuracy of the land cover map that was
produced using the FSEB method (Cha & Park, 2007; Hu et al., 2013;
Yu & Gong, 2012). Due to the limited high resolution image coverage
of the study site, we were able to collect 94 random validation points.
Thus, our total validation points were 160 (Fig. 1). Kappa statistic was
used to evaluate classification accuracy of produced land cover map
and presented in the form of an error matrix. The error matrix is a sim-
ple cross-tabulation of the mapped class label against the observed in
the ground or reference data (Congalton & Green, 2008). Model perfor-
mance of canopy height and aboveground biomass estimation derived
from SRTM DEM were evaluated from 10,000 bootstrap analysis runs
(Efron, 1979; Efron & Tibshirani, 1994) of field measurements of canopy
height and standing biomass with replacement to yield mean absolute
error (MAE) and root mean squared error (RMSE) (Willmott, 1982).
mangrove species were measured from the 41 0.5 ha sample plots.
The 20 mangrove species documented in this study represent approxi-
mately half of the total 43 mangrove species reported in Papua and
New Guinea (Duke, Ball, & Ellison, 1998). However, our study was not
designed to be an exhaustive survey of biodiversity and many trees
were identified only to the genus level. Therefore, our data do not rep-
resent total mangrove species diversity of the region.
Cluster analysis based on forest community composition led to eight
major forest associations in the inventory plots (Appendix Fig. S1);
however, many species remained unidentified especially in the highly
diverse transitional forests which included numerous non-mangrove
lowland swamp species. A summary of structural characteristics and
biomass of the field plots are presented in Table 1 and their percent con-
tribution are shown in Fig. 6. The forest types are dominated by typical
mangrove species in the Rhizophoraceae family (Rhizophora, Bruguiera,
and Ceriops), with co-dominant Camptostemon and Lumnitzera being re-
markably similar in forest structure and aboveground biomass
(Table 1).
Substantial variation remains within forest types (Fig. 6), attribut-
able to plot-level differences in the abundance of large trees (N50 cm di-
ameter) among plots. Based on field data, Bruguiera dominated forests,
composed of mixed Bruguiera, Rhizophora, and Lumnitzera species, were
more dominant than other forest types in Mimika, illustrated by the
largest share of basal area (28.78%), stem density (18.94%), and above-
ground biomass (37.43%). Their dominance was also observed by the
largest average biomass (416.79 ± 175.41 Mg/ha) and the number of
occupied plots assigned to this forest type (10 plots). In fact, two plots
from this forest type had biomass in excess of 600 Mg/ha. The tallest
tree within Bruguiera dominated forests varied between 16 m and
30 m. These findings coincide with those of Ellison (2005), who sug-
gested that species of Bruguiera were dominant in the Ajkwa and
Tipoeka estuaries of Mimika, Papua.
Spatial analysis of remote sensing data, however, shows that
Bruguiera-dominated forests contain less overall area than Rhizophora-
dominated forests with a total occupancy of 23,623 ha compared to
77,496 ha, respectively, mostly covering the central part of coastal
area extending to the northwest (Fig. 7). As noted in other publications,
the Bruguiera dominated forests tend to occupy landward mangrove
zones (Ellison, 2005), but thrive in areas with sufficient drainage
(Tomlinson, 1986), and occasionally mix with transitional forests. Tran-
sitional forests refer to areas where a diverse mix of freshwater swamp
species dominate and mangrove associates and true mangroves are
present but not abundant. During our field campaigns we encountered
six Bruguiera species but only five of them were positively identified.
The five species include Bruguiera exaristata, Bruguiera gymnorrhiza,

Table 1
Eight major mangrove associations were identified from the 41 0.5 ha forest inventory plots in Mimika district, Papua Indonesia. For landscape scale biomass inventory, mangroves dom-
inated by Rhizophoraceae (Rhizophora, Bruguiera, and Ceriops) were combined in subsequent analyses as these forests were similar in structure and aboveground biomass. Values are
mean ± standard deviation.

Forest type Basal area (m2/ha) Stem density (#/ha) Aboveground biomass (Mg/ha) Range of max height (m) # of occupied plot

Avicennia, Sonneratia 15.6 ± 3.5 1926 ± 247 106.3 ± 38.6 8–10 4

Rhizophora 27.1 ± 4.8 1045 ± 438 333.3 ± 122.9 15–32 7
Camptostemon, Rhizophora, Bruguiera 34.8 ± 1.5 733 ± 197 300.9 ± 96.7 25–31 2
Bruguiera, Rhizophora, Lumnitzera 33.3 ± 7.4 786 ± 215 416.8 ± 175.4 16–30 10
Ceriops 34.4 ± 6.4 827 ± 491 387.1 ± 18.4 28–29 2
Transitional (coastal) 35.3 ± 9.8 1132 ± 250 286.7 ± 68.2 30–35 3
Transitional (landward) 26.4 ± 11.6 1108 ± 313 182.0 ± 97.9 25–27 6
Sago 26.4 ± 5.1 960 ± 289 137.7 ± 24.6 25–31 7
72 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
Fig. 6. Percentage of basal area, stem density, and biomass across eight major groups of forest types within the areas of Mimika derived from cluster analysis. AVI = Avicennia, BRU =
Bruguiera, CAM = Camptostemon, CER = Ceriops, RHI = Rhizophora, SON = Sonneratia, TRANS = Transitional. Percentages were calculated from 41 0.5 ha forest inventory plots.
Bruguiera hainesii, Bruguiera parviflora, and Bruguiera sexangula. Based
on Ellison's (2005) study, the sixth unidentified species is likely
Bruguiera cylindrica. Bruguiera hainesii is listed by the International
Union for Conservation of Nature (IUCN) as a critically endangered
mangrove species (Duke et al., 2010), and was recorded in the transi-
tional coastal zones of the study area where freshwater swamp species
mix with mangroves on elevated beach ridges.
The second dominant group of mangroves was composed of
Rhizophora species that according to field data, occupied 16.37% basal
area, 17.62% stem density, and 20.96% of the aboveground biomass of
the delineated mangrove area (Fig. 6). Rhizophora inhabit seaward mar-
gins with lower elevations (Ellison, 2005), and were comprised of
Rhizophora mucronata, Rhizophora apiculata, and Rhizophora stylosa spe-
cies, with R. apiculata the most abundant. Their mean biomass was rel-
atively high, (333.31 ± 122.95 Mg/ha) and the tallest trees varied from
14 to 32 m. Rhizophora-dominated forests spread from the central part
of the mangrove zone to the southeast, although Rhizophora was also
abundant in the western part of the study area (Fig. 7). In addition,
the co-dominant Ceriops- and Camptostemon-dominated forests were
similar in their percentage of basal area occupied (5.93% and 6%, respec-
tively) and stem density (3.98% and 3.53%, respectively), with slight var-
iations in standing biomass (6.95% and 5.41%, respectively) (Fig. 6).
They also possessed relatively high mean biomass, (387.14 ±
18.40 Mg/ha, Ceriops-dominated forests; and 300.89 ± 96.75 Mg/ha,
Camptostemon-dominated forests), with maximum canopy height vary-
ing between 28 and 29 m for Ceriops and from 25 to 31 m for
Camptostemon. Bruguiera, Ceriops and Camptostemon species inhabit
the landward portion of the mangrove zone (Ellison, 2005) with higher
drainage (Tomlinson, 1986). Avicennia/Sonneratia forests colonize
newly exposed mudflats as pioneering stands. The land cover map indi-
cates that the Avicennia/Sonneratia forests are confined to coastal zones
covering 5472 ha of the study area. They are characterized by a lower
canopy height (between 8 and 10 m), the lowest basal area (5.38%)
and aboveground biomass (3.82%), but the highest density of smaller
stems (18.55%) compared to other mangrove types. Their stem density
surpassed Rhizophora-dominated forest and was similar to Bruguiera-
dominated forest (Fig. 6). Avicennia/Sonneratia stands contained the
lowest mean aboveground biomass, (106.26 ± 38.55 Mg/ha), which is
approximately 300 Mg/ha lower than the biomass of Bruguiera-
dominated forest.
Based on land cover classification and field inventory data from non-
mangrove plots, Sago palms tend to be more predominant in transi-
tional landward and coastal forests within the study area. Sago palms
represented 15% of basal area, 16.18% of stem density, and 8.66% of bio-
mass in the study area. Transitional mixed (non-mangrove) landward
forests occupied 12.45% basal area, 13.02% stem density, 9.05% biomass,
whereas transitional mixed coastal forests contained 9.14% basal area,
81.18% stem density, and 7.72% biomass of the study area respectively.
The mean biomass of Sago was 137.72 ± 24.62 Mg/ha with maximum
height varying from 25 to 31 m among plots. However, the biomass es-
timation of Sago was calculated from the best information available, as
Sago specific allometric equations have not been developed. Equations
used for this study were derived from data collected for a different pur-
pose (Table. S1; Gusmayanti, unpublished data). Since palm fibers are
much less dense than wood, the biomass of Sago palms in adult life
stages is much lower than trees of similar diameter. This was reflected
in the low biomass estimates for Sago plots, where the majority of
standing biomass occurs in the interspersed hardwoods. Sago palm is
distributed throughout the eastern part of study area, intermixed with
swamp forest and/or swamp shrublands, pandanus, and grasslands cov-
ering a total area of 107,129 ha (Fig. 6). Transitional forests represent an
ecotone between mangroves and freshwater swamp forests. They can
be further subdivided into “coastal” and “landward” types in which
both forests have unique characteristics and occupy distinct landforms
(Marshall, 2007). These forests are where the endangered Bruguiera
hainesii were documented. The mean stem density of the coastal tran-
sitional forests was almost similar to that of landward transitional
forests; however, their average biomass was about 100 Mg/ha higher
than landward forests. The difference could be attributed to signifi-
cant variation in canopy height between both types of transitional
forest (Table 1). For example, the tallest trees in the transitional
coastal forests plots were between 30 and 35 m; conversely, the
maximum canopy heights recorded for landward forest plots were
between 25 and 27 m.
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
4.2. Evaluation of land cover classification methods
The area and statistical accuracy of the 17 classes included in the
land cover map derived from a combination of image segmentation
and FSEB classification methods are shown in Table 2. The total area of
mangroves in the study was 186,291 ha with Rhizophora forest account-
ing for the largest coverage (5.78%) and Avicennia/Sonneratia forest with
the smallest coverage (0.41%).
Moreover, the total classified forest area, including swamp forest,
dryland/riparian forest, and all types of mangrove forest, was
941,891 ha. It is about 3% of the 35.2 million hectares of primary forest
in Papua Island (Margono et al., 2014). Our results indicate that al-
though the settlements, roads and palm oil concessions accounted for
only 0.67% of the total area of the study site, rapid land clearing for
palm oil plantation to the west of the city of Timika is contributing to
an increase in deforestation of Mimika's coastal wetlands. The classifica-
tion methods presented in this study produced an overall classification
accuracy of 94.38% with kappa statistics of 0.94, surpassing many typical
recommended targets of 85% of overall accuracy (Foody, 2002). The
combination of image segmentation and FSEB methods were effective
for mapping tropical wetland vegetation indicated by high percentages
of producer's and user's accuracy (N81.82% and N 85.71%), respectively.
Our methods have been successful in classifying 10 groups of wetland
vegetation including several genera of mangroves (Fig. 7). Within man-
grove genera category, for example, all groups of mangrove have
producer's accuracy N 85% whereas for user's accuracy, only mixed man-
grove class had b 85% user's accuracy. Wetland cover classes included:
swamp grassland, swamp shrubland, swamp forest, Sago forest,
Fig. 7. Coastal land cover map of study area shown 17 distinct land cover classes derived from Landsat-8 imagery using FSEB method.
73
Rhizophora forest, Avicennia/Sonneratia forest, Bruguiera dominated for-
est, other (mixed) mangrove forest, transitional landward and coastal
forests, and Nypa stands.
The rule-based FSEB classification substantially improved mangrove
classification accuracy compared to supervised classification methods.
These results are in agreement with those of Gao, Chen, Zhang, and
Zha (2004), who reported overall accuracy of 46.7%and 68.3% for super-
vised parametric classification used to map stunted and lush mangroves
in the western portion of the Waitemata Harbor, Auckland, New
Zealand, whereas after the spatial knowledge (spatial situation and
spectral properties) was sequentially incorporated, these accuracies im-
proved substantially, to 83.3% and 96.7%, respectively. Another study
that mapped Alpine vegetation cover in southwest Yukon Territory,
Canada demonstrated improved classification accuracy from 74%
when using satellite data alone to 85% accuracy when incorporating
geomorphometric variables extracted from DEM data (Wilson &
Franklin, 1992).
Three factors were important for our classification technique's over-
all accuracy: proper band selection, proper determination of standard
deviation value (σ), and sequential classification procedure. Proper
band selection was done by analyzing the spectral profiles of seven
multi-spectral bands of Landsat-8 and finding the appropriate bands
for characterizing certain genus/species of mangrove/non-mangrove
forest (Fig. 8A). Proper determination of σ was introduced to include di-
versity in a class. For instance, 1σ may or may not be enough to classify
particular classes of land cover, and a factor of 2σ or larger may be
needed to include the variance when a relatively high dispersion of
spectral reflectance occurs within a species in an area (Figs. 8B and C).
74 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81

Table 2
Accuracy matrix of land cover map showed producer's and user's accuracy including area for each class cover types.

Class name Area (ha) Area (%) Reference Classified # Correct Producer's (%) User's (%) Kappa

Dryland/riparian forest 278,136.45 20.75 17 16 15 88.24 93.75 0.93

Swamp grassland 13,254.12 0.99 6 6 6 100.00 100.00 1.00
Swamp scrubland 54,036.00 4.03 3 3 3 100.00 100.00 1.00
Sago palm dominated forest 107,129.07 7.99 10 9 9 90.00 100.00 1.00
Mixed swamp forest 301,918.23 22.52 15 16 15 100.00 93.75 0.93
Grassland/scrubland 18,416.07 1.37 7 5 5 71.43 100.00 1.00
Settlements/roads 3663.81 0.27 4 6 4 100.00 66.67 0.66
Palm oil concession 5328.00 0.40 7 7 7 100.00 100.00 1.00
Open water 277,186.41 20.68 15 15 15 100.00 100.00 1.00
Cloud 2059.20 0.15 7 7 7 100.00 100.00 1.00
Barren land – soil 24,583.86 1.83 8 8 8 100.00 100.00 1.00
Rhizopora dominated forest 77,496.39 5.78 22 20 20 90.91 100.00 1.00
Avicennia/Sonneratia forest 5471.19 0.41 7 6 6 85.71 100.00 1.00
Bruguiera dominated forest 23,622.66 1.76 7 8 7 100.00 87.50 0.87
Mixed mangrove forest 68,696.73 5.13 10 11 9 90.00 81.82 0.81
Transitional landward and coastal forest 68,416.47 5.10 9 10 9 100.00 90.00 0.89
Nypa palm forest 11,004.12 0.82 6 7 6 100.00 85.71 0.85
Totals 1,340,418.78 100.00 160 160 151

Fig. 8. A demonstration of the spectral response profiles used for the FSEB classification method. The difference in mean spectral profile response from Rhizophora and Metroxylon Sago on
multispectral band of Landsat-8 (A); the spectral profile of Rhizophora associations species with ±2 σ used for FSEB classification (B); and the spectral profile of Sago palm (Metroxylon
Sago) with ±2.575 σ used for FSEB classification (C).
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81 75

Fig. 9. Optimal quantile for the mangrove canopy height model (between SRTM DEM and field canopy heights) was determined using a stepwise quantile regression technique at subplot
scale (A). The scatterplot (B) shows the agreement between SRTM DEM and the 0.98 quantile of field canopy height, H.98.

We found that the incremental use of σ is important for implemen- image, and the remaining image to be used for classifying each subse-
tation of the FSEB classification method, instead of depending solely on quent land cover class until all surface features within Landsat-8 images
a fixed σ as a default. The sequential classification procedure allowed an are properly classified. This procedure helped to avoid overlapping re-
easily identifiable land cover to be classified and removed from the gions between classes.

Fig. 10. Map of mangrove canopy height derived from SRTM elevation data. The map showed cluster patterns in canopy heights corresponding to different forest types.
76 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
Fig. 11. The 0.5 NLQR was the best predictor of relationship between SRTM-derived
canopy height model (based on Eq. 3) and measured biomass from 33 0.5 ha mangroves
inventory plots (R2 = 0.46). Two outliers from the atypical sandy soil plots are shown
inside a rectangle.
Fig. 12. A map of mangrove standing biomass (Mg/ha) derived from the model based on SRTM elevation data. The inbox shows clustered patterns of biomass that correspond to different
forest types.
This study shows for the first time that the Landsat-8 data of higher
radiometric resolution can be used to map mangroves up to genus levels
at broad spatial scales. Nonetheless, although the FSEB method was suc-
cessful in classifying mangrove forests of Papua, it also showed some
limitations where it failed to separate some genera of mangroves such
as stands of Lumnitzera, Ceriops, and Camptostemon. A possible cause
of this limitation is that we could not select any “pure” spectra of
these mangrove types for training because they are intermixed with
other mangrove species/genera, including Rhizophora and Bruguiera.
The ability of the image segmentation method applied to PALSAR
composite images (i.e., HH, HV, and RFDI) to separate mangroves from
non-mangroves indicates that the methodology can be applied to refine
current boundaries of Landsat-based global mangrove distribution
maps produced by Giri et al. (2011). Several studies have emphasized
the effectiveness of using L-band SAR data for mapping mangrove
(Flores De Santiago et al., 2013; Kovacs et al., 2013; Lucas et al., 2007;
Lucas et al., 2014; Rocha de Souza Pereira et al., 2012). Flores De
Santiago et al. (2013) have also shown that image segmentation of
ALOS PALSAR data can be used for mapping mangrove areas in tropical
areas with persistent cloud cover. Our study demonstrates that in addi-
tion to separating mangroves from non-mangroves, the combined use
of ALOS PALSAR and Landsat-8 imagery can be used to map mangrove
composition to the genus level at regional scales. The methods pre-
sented here to produce detailed coastal wetland maps has the potential
to provide decision makers with a valuable tool for assessing mangrove
resources, as well as their use in climate adaptation and mitigation
efforts.
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81 77

Table 3
A comparison of biomass obtained from the modeled, measured, and extrapolated to an entire class from the field measurements, for each land cover types. Values are mean ± standard
deviation.

Forest types Area (Ha) Modeled Measured Modeled total (Tg) Measured (extrapolated) % Difference
estimation (Tg)

Mean ± STD (Mg/ha) Mean ± STD (Mg/ha) Lower 90% CI Total Upper 90% CI

Avicennia/Sonneratia forest 5472 237.52 ± 98.20 106.26 ± 38.55 1.24 0.32 0.58 0.85 72.53
Bruguiera dominated forest 23,623 295.09 ± 86.17 416.79 ± 175.41 6.95 4.64 9.85 15.05 34.52
Other mangroves forest 79,701 284.93 ± 103.89 344.02 ± 57.56 22.29 21.59 27.42 33.25 20.64
Rhizophora dominated forest 77,496 353.30 ± 98.43 333.31 ± 122.95 27.32 13.85 25.83 37.81 5.61
All mangroves 186,291 313.58 ± 105.28 299.56 ± 98.62 57.79 32.71 55.81 78.9 3.48
Transitional forest 68,417 308.20 ± 96.51 234.34 ± 83.05 21.02 8.89 16.03 23.17 26.94

4.3. Evaluation of canopy height model
We found a strong linear correlation between the SRTM DEM and
the 0.98 quantile of field canopy height (H.98; the tallest trees in each
10 m radius subplot) indicated by high correlation coefficient (r =
0.84) and predicted coefficient of determination (Pred-R2 = 0.804)
(Fig. 9).
The high correlation was expected because the SRTM DEM was de-
rived from single-pass interferometry of the C-band radar signal from
which heights are measured according to scattering phase center, or
the weighted mean of all contributing backscatter throughout the
depth of canopy (Madsen et al.,1993). Since mangrove forest canopies
are typically dense and SRTM C-band contains short wavelengths, the
InSAR scattering phase center would be located close to the top of can-
opy, corresponding closely with field-based canopy-height estimation
(Kellndorfer et al., 2004; Madsen, Zebker, & Martin, 1993). The resulting
optimal model fit between SRTM DEM and H.98 in our study is
expressed by the following equation:
CHMSRTM ðmÞ ¼ 2:7191 ðSRTM DEMÞ0:676
where CHMSRTM represents a canopy-height model derived from SRTM
DEM data. A bootstrapping error analysis (10,000 resamples with re-
placement) produced average errors of 3.0 m for MAE and 3.7 m for
RMSE for the 30 m SRTM pixels. The RMSE and predicted R2-
coefficient of the canopy heights of our model indicated less uncertainty
than those previously reported by Kellndorfer et al. (2004). Another
study by Simard, Rivera-Monroy, Mancera-Pineda, Castañeda-Moya,
and Twilley (2008) calibrated SRTM DEM with field measurement of
mangrove canopy height, resulting in an RMSE of 2.7 m. However,
their error estimation is more limited as it was estimated from 30
plots with only 166 trees in total, whereas our study is derived from
41 field plots with N 8000 trees measured. Similar research by
Fatoyinbo and Simard (2013) estimated mangrove canopy height across
Africa and obtained the overall RMSE of 3.6 m, whereas another study
by Simard et al. (2006) in Everglades National Park revealed the RMSE
of 2.0 m. Although the uncertainty estimation of our canopy height
model is comparable to Fatoyinbo and Simard (2013) and Simard
et al. (2006); their error estimations were not based on ground-truth
measurements, rather inferred from the fusion of SRTM DEM data and
Light Detection and Ranging (LiDAR) data. Several studies have shown
a consistent underestimation of canopy height using LiDAR in
temperate forest and tropical rainforests (Clark, Clark, & Roberts,
2004; Gaveau & Hill, 2003; Rönnholm et al., 2004; Wang & Glenn, 2008).
Using Eq. (4), we produced a canopy height map at 30 m spatial res-
olution of mangrove areas in Mimika district (Fig. 10). The modeled
mangrove canopy heights ranged from 4.3 to 36.2 m with an estimated
mean of 20.8 m. Mangrove trees in Papua and New Guinea can reach up
to 40 m in height (Alongi, 2007); however, in general, their canopy
heights are between 25 and 30 m, typically due to the mangrove species
in the Rhizophoraceae family (Ellison, 2005).
For comparison, Simard et al. (2008) reported that 15 m was the
maximum height of mangroves grown in Ciénaga Grande de Santa
Marta, Colombia, while Lucas et al. (2002) demonstrated that tall man-
groves of northern Australia have maximum heights varying from 16 to
20 m. A study by Pool, Snedaker, and Lugo (1977) reported that the
tallest mangrove trees varied from 9 m in Florida to 15 m in Puerto
Rico, 16 m in Costa Rica, and 17 m in Mexico. Based on our findings,
mangrove trees in our study area are consistently among the tallest
reported.
Superimposing the land-cover map (Fig. 7) on the canopy height
map (Fig. 10) showed a clustered pattern (zonation) among mangrove
ð4Þ
forest types in Mimika. For example, the majority of tall mangroves
(from the Rhizophoraceae family) occupy the central part of the study
area downstream of larger rivers. In contrast, the shortest mangrove
trees, mainly Nypa, were concentrated in the southeastern portions of
our study area which receive less freshwater inflow from rivers. Short
canopy Avicennia- and Sonneratia-dominated mangroves were also
present in the coastal fringe and along aggrading coastlines. Such clus-
tered patterns are attributable to periodic changes in the intertidal envi-
ronment and hydrogeological settings. Zonation of mangrove species is
influenced by the frequency of inundation, drainage, and/or elevation,
as estuarine shorelines are largely determined by the interaction be-
tween sediment supply and variation in sea level (Ellison, 2000, 2005).
4.4. Evaluation of biomass model
Based on an ANOVA of the series of non-linear quantile regressions,
the 0.5 quantile was the best model to calculate the spatial distribution
of aboveground mangrove biomass at the landscape scale in the study
area (Fig. 11). The two outliers highlighted in Fig. 11 are plots domi-
nated by stunted Lumnitzera littoralis and Rhizophora apiculata found
in coarse sands rather than fine silt/clay substrates that are typical to
this region. The architecture, and therefore allometry, of trees growing
in the coarse sand plots is atypical, dominated by secondary growth

Table 4
Quantitative error measures and paired mean t-test of mangrove standing biomass model performance. Values are mean ± standard deviation.

Forest types N Predicted mean ± STD (Mg/ha) Observed mean ± STD (Mg/ha) MAE (Mg/ha) RMSE (Mg/ha) r

All mangroves 145 292.72 ± 121.89 314.37 ± 172.62 105.75 147.98 0.55
Avicennia/Sonneratia, Bruguiera, Rhizophora 104 303.96 ± 128.02 313.43 ± 176.20 91.30 129.51 0.67
Other mangroves 41 264.22 ± 100.59 316.97 ± 165.29 142.39 186.84 0.14
Transitional forest 45 346.10 ± 105.29 254.20 ± 134.79 123.57 143.27 0.60
Mangroves and non-mangroves 190 305.36 ± 120.07 300.12 ± 166.11 109.97 146.88 0.50
78 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81

Fig. 13. Comparison between the observed and predicted standing biomass of mangroves: all mangroves combined (A), Avicennia + Sonneratia + Bruguiera + Rhizopora combined (B),
mixed mangroves (C), transitional forests (D) and mangroves and non-mangroves combined (E).
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
that produce short, stunted trees that grow large in diameter but not
height. Forest biomass in these areas is underestimated, since the archi-
tecture of this mangrove type is not represented in species-specific allo-
metric equations.
Unlike the previous studies that used linear regression to model
mangrove biomass from canopy heights (Fatoyinbo & Simard, 2013;
Fatoyinbo et al., 2008; Simard et al., 2006, 2008), we applied a non-
linear regression (exponential function) to model aboveground biomass
of mangroves because theoretically is a more justifiable approach and
can be explained as follows. Normally, at an early stage (seedlings and
saplings) different mangrove species will grow and have heights that
linearly correspond with their biomass. However, in nature, each tree
species will compete and respond differently to the sunlight. Therefore,
some trees from the same species or from different species may have
higher rate of growth and biomass than others. As a result, these partic-
ular trees will be taller than others and wider in diameter and biomass
as well, since they received more sunlight. Naturally, at one point the
height and diameter of trunks of these taller trees will slow their growth
rate due to their maturity, but their biomass will still increase as a result
of secondary growth from branches. Several other studies have used
non-linear regression approach, particularly an exponential function,
to model aboveground biomass in tropical rainforests based on field
data collection (Chave et al., 2005; Kauffman & Donato, 2012). In this
study, the relationship between CHMSRTM (m) and measured above-
ground biomass (Mg/ha) at the 0.5 quantile was found to be:
BIOMASSSRTM ðMg=haÞ ¼ expð3:9042 þ 0:0858 ðCHMSRTM Þ ðmÞÞ ð5Þ
The map of spatial distribution of mangrove biomass derived from
Eq. 5 is provided in Fig. 12. To compare biomass from the model and
that from simple extrapolation of field data (Table 3), total mangrove
biomass was scaled to the landscape by applying the canopy height-
biomass model to all forest types within the study area (modeled),
and multiplying the mean observed biomass value of each cover class
by its coverage (extrapolated) (Fig.7). There was considerable variation
in aboveground biomass among forest types within mangroves both
from field calculation and modeled biomass (Table 3).
For instance, average aboveground biomass varies from 237.52 ±
98.2 Mg/ha, in a low canopy young Avicennia/Sonneratia stands to
353.52 ± 98.43 Mg/ha in a mature tall canopy Rhizophora-dominated
stands. Modeled biomass estimation of Avicennia/Sonneratia forests
was higher than the biomass estimates resulting from simple extrapola-
tion from field plots by 131.26 Mg/ha. The difference could be partly at-
tributed to taller statured Avicennia/Sonneratia forests not well
represented in the field dataset. Nonetheless, at the landscape scale,
the canopy height-biomass model indicated only a 3.48% difference be-
tween the modeled and the measured estimates of total biomass of all
mangroves combined, meaning that the height and biomass models de-
veloped in this study are reliable and can be applied broadly for map-
ping and monitoring mangrove biomass over large areas of Indonesia.
Furthermore, modeled standing biomass estimations were within the
90% confidence intervals for Rhizophora-dominated forest, Bruguiera-
dominated forests, and all combined mangroves forests when com-
pared with measured area weighted biomass. The average biomass
value of all mangrove forest types except for the field-measured
Avicennia/Sonneratia forest greatly exceeded the Intergovernmental
Panel on Climate Change (IPCC) default value for aboveground biomass
of wet tropical mangroves (192 Mg/ha) (Hiraishi et al., 2014). The mean
values of field-measured biomass for Bruguiera-dominated forest were
double the IPCC default value (416.79 versus 192 Mg/ha, respectively),
indicating that mangrove biomass in southern Papua is much higher
than the global average, and among the highest on Earth.
A summary of error analysis between predicted and observed man-
grove standing biomass at the subplot level is given in Table 4.
At the subplot level, the predicted values of aboveground biomass
were well within the range of observed values for all mangroves
79
(Fig. 13A). The difference between means is nearly 7% of the observed
mean biomass with MAE = 105.75 Mg/ha and RMSE = 147.98 Mg/ha.
Mangroves that are shorter statured, such as those dominated by
Ceriops, Lumnitzera, and stunted Rhizophora (included in the “other
mangroves” cover class) generally have lower predicted biomass than
observed in the field, which is an artifact of the allometric equations
used rather than any error in the model. When the “other mangroves”
cover class is excluded from analysis, model fit improves substantially
and the difference between predicted and observed means is 3% of the
observed mean (Fig. 13B). Mangroves dominated by Avicennia/
Sonneratia, Bruguiera and Rhizophora stands account for 57% of the
total mangrove area in Mimika. Nevertheless, when applied to other
mangroves, the canopy height-biomass model does not fit observed bio-
mass data very well even though predicted biomass values are within
the range of observed values (Fig. 13C). The difference between
means is about 18% of the observed values. Again, the lack of fit could
be partly due to the underestimation of biomass in the field using allo-
metric equations that are not representative of shorter statured, large
diameter Rhizophora and Lumnitzera mangroves growing on beach
ridges. SRTM data may also not capture the few larger trees present in
low canopy mangroves. These larger diameter trees contribute most
to the estimates of plot level (0.5 ha) aboveground biomass. Although
the Pearson's correlation coefficient between predicted and observed
is relatively high (r = 0.60) for transitional forests growing on elevated
beach ridges, their standing biomass were overestimated by the canopy
height-biomass model (Fig. 13D). Biomass observed in the field may be
lower than predicted because of different height-biomass relationships
of non-mangrove species, lower wood density, and smaller diameter
trees dominating the canopy. Nevertheless, when all forest inventory
data are combined, the differences between predicted and observed
mean biomass values are within 2% of the observed value (r = 0.5;
Fig. 13E).
5. Conclusions
In the development of new approaches for applying remote sensing
data to derive ecosystem properties, accuracy and simplicity are critical
factors. Furthermore, the methods must be efficient in terms of effort
(time and cost) to implement and should be capable of broad applica-
tion with minimum supervision. In this study, a remote sensing ap-
proach was developed by combining multispectral and SAR data for
mapping the spatial distribution and standing biomass of coastal wet-
land forests of southern Papua, Indonesia. A new classification method,
FSEB, was introduced for mapping spatial distribution of coastal wet-
land forests, and nonlinear quantile regression was employed to
model aboveground biomass of mangrove. The results showed that
the incremental use of variable standard deviations is a realistic ap-
proach for implementing the FSEB classification method, instead of de-
pending solely on a fixed standard deviation classification approach.
This study shows for the first time that Landsat-8 data can be used to
map mangroves at the genus level rather than classifying mangroves as
a single general class. In addition, contrary to the previously published
mangrove biomass studies, the use of a nonlinear regression (exponen-
tial function) to model standing biomass of mangroves was used be-
cause it follows the growth pattern of trees regardless of their species.
The reliability of our approach is demonstrated by high overall accuracy
and kappa coefficient (94.38% and 0.94, respectively) for the land cover
classification (surpassing many typical recommended targets of overall
accuracy; e.g., Foody, 2002). A point of caution is that given the rela-
tively limited number of validation data points (160) that we had avail-
able for this vast area of study (N 1.3 million ha), the high accuracy of
classification achieved by our method may require further validation
in other areas. Additionally, our method also produced a high percent-
age of the predicted R2 with low margins of error (MAE and RMSE) for
the canopy height model, and predicted estimates within 90% confi-
dence intervals between modeled and measured values for the
80 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
landscape-scale mangrove standing biomass. Our results show that the
methods proposed in this study can be used can potentially provide de-
cision makers with a valuable tool for assessing mangrove resources,
and can be used for climate change adaptation and mitigation efforts.
Acknowledgements
This work was supported by the USAID Indonesia Forest and Climate
Support (IFACS) (contract # AG-3187-C-13-0010) through Kamoro
Collaborative Wetlands Management Program. The authors would like
to thank to the USAID Indonesia Forest and Climate Support (IFACS)
Program and their Timika field office for support in collecting field
data. Field data collection was accomplished by the tireless efforts of
Taryono Darusman, Rudi, Hendri Saleh, Irvansyah, and Leman Noerman
of Puter Foundation, Indonesia. The ALOS PALSAR imagery used in this
study was provided by the Alaska Satellite Facility (ASF) under ALOS
PALSAR and Student Proposal Program.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.rse.2016.04.026.
References
Alongi, D. M. (2007). 5.4. Mangrove forests of Papua. Ecology of Indonesian Papua part two
(pp. 824).
Alongi, D. M. (2012). Carbon sequestration in mangrove forests. Carbon Management, 3,
313–322.
Al-Wassai, F. A., & Kalyankar, N. (2013). Major limitations of satellite images. arXiv preprint
arXiv:1307.2434.
ARD, T. T. (2014). USAID IFACS quarterly report year 4. Second quarter of year 4 workplan:
Indonesia forest and climate support.
Aschbacher, J., Tiangco, P., Giri, C., Ofren, R., Paudyal, D., & Ang, Y. (1995). Comparison of
different sensors and analysis techniques for tropical mangrove forest mapping. Geo-
science and remote sensing symposium IGARSS'95.'. Quantitative remote sensing for
science and applications', international. (pp. 2109–2111) IEEE.
Baker, T. R., Phillips, O. L., Malhi, Y., Almeida, S., Arroyo, L., Di Fiore, A., ... Laurance, W. F.
(2004). Variation in wood density determines spatial patterns in Amazonian forest
biomass. Global Change Biology, 10, 545–562.
Bakosurtanal (2009). Peta mangroves Indonesia. Cibinong: Pusat Survey Sumberdaya Alam
Laut. Badan Koordinasi Survey dan Pemetaan Nasional.
Barbier, E. B. (2012). Progress and challenges in valuing coastal and marine ecosystem
services. Review of Environmental Economics and Policy, 6, 1–19.
Blaschke, T. (2010). Object based image analysis for remote sensing. ISPRS Journal of
Photogrammetry and Remote Sensing, 65, 2–16.
Nowels, M. (2009). Improving wetland characterization with multi-sensor, multi-
temporal SAR and optical/infrared data fusion. In L. L. Bourgeau-Chavez, K. Riordan,
R. B. Powell, N. Miller, & G. Jedlovec (Eds.), Advances in geoscience and remote sensing
(pp. 679–708). India: InTech Publishers.
Brondízio, E., Moran, E., Mausel, P., & Wu, Y. (1996). Land cover in the Amazon estuary:
Linking of the Thematic Mapper with botanical and historical data. Photogrammetric
Engineering and Remote Sensing, 62, 921–930.
Chambers, J. Q., Santos, J. d., Ribeiro, R. J., & Higuchi, N. (2001). Tree damage, allometric
relationships, and above-ground net primary production in central Amazon forest.
Forest Ecology and Management, 152, 73–84.
Cha, S., & Park, C. (2007). The utilization of Google Earth images as reference data for the
multitemporal land cover classification with MODIS data of North Korea. Korean
Journal of Remote Sensing, 23, 483–491.
Chave, J., Andalo, C., Brown, S., Cairns, M., Chambers, J., Eamus, D., ... Kira, T. (2005). Tree
allometry and improved estimation of carbon stocks and balance in tropical forests.
Oecologia, 145, 87–99.
Chmura, G. L., Anisfeld, S. C., Cahoon, D. R., & Lynch, J. C. (2003). Global carbon sequestra-
tion in tidal, saline wetland soils. Global Biogeochemical Cycles, 17.
Clark, M. L., Clark, D. B., & Roberts, D. A. (2004). Small-footprint lidar estimation of sub-
canopy elevation and tree height in a tropical rain forest landscape. Remote Sensing
of Environment, 91, 68–89.
Clough, B., & Scott, K. (1989). Allometric relationships for estimating above-ground bio-
mass in six mangrove species. Forest Ecology and Management, 27, 117–127.
Clough, B., & Abdullah, A. (1993). The economic and environmental values of mangrove for-
ests and their present state of conservation in the south-East Asia/Pacific region: Techni-
cal report of the project. (International Society for Mangrove Ecosystems).
Congalton, R. G., & Green, K. (2008). Assessing the accuracy of remotely sensed data: Princi-
ples and practices. CRC press.
Costanza, R., d'Arge, R., De Groot, R., Farber, S., Grasso, M., Hannon, B., ... Paruelo, J. (1997).
The value of the world's ecosystem services and natural capital. Nature, 387, 253–260.
Donato, D. C., Kauffman, J. B., Murdiyarso, D., Kurnianto, S., Stidham, M., & Kanninen, M.
(2011). Mangroves among the most carbon-rich forests in the tropics. Nature
Geoscience, 4, 293–297.
Duke, N., Ball, M., & Ellison, J. (1998). Factors influencing biodiversity and distributional
gradients in mangroves. Global Ecology and Biogeography Letters, 7, 27–47.
Duke, N., Kathiresan, K., Salmo, S. G., III, Fernando, E. S., Peras, J. R., Sukardjo, S., ... Ngoc
Nam, V. (2010). Bruguiera hainesii. The IUCN red list of threatened species Version
2014.3. bwww.iucnredlist.orgN. Downloaded on 05 April 2015.
Efron, B. (1979). Bootstrap methods: Another look at the jackknife. The Annals of Statistics,
1–26.
Efron, B., & Tibshirani, R. J. (1994). An introduction to the bootstrap. CRC press.
Ellison, J. C. (2005). Holocene palynology and sea-level change in two estuaries in
Southern Irian Jaya. Palaeogeography, Palaeoclimatology, Palaeoecology, 220,
291–309.
Ellison, J. C. (2000). How South Pacific mangroves may respond to predicted climate
change and sea-level rise. Climate change in the South Pacific: Impacts and responses
in Australia, New Zealand, and small island states (pp. 289–300). Springer.
ESRI (2014). ArcGIS desktop release 10.2.1. Redlands CA: Environmental Systems Research
Institute.
FAO (2007). The world's mangroves 1980–2005. FAO forestry paper 153. Italy: Rome.
Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote
Sensing of Environment, 80, 185–201.
Fatoyinbo, T. E., & Simard, M. (2013). Height and biomass of mangroves in Africa from
ICESat/GLAS and SRTM. International Journal of Remote Sensing, 34, 668–681.
Fatoyinbo, T. E., Simard, M., Washington-Allen, R. A., & Shugart, H. H. (2008). Landscape-
scale extent, height, biomass, and carbon estimation of Mozambique's mangrove for-
ests with Landsat ETM+ and Shuttle Radar Topography Mission elevation data.
Journal of Geophysical Research, 113 G02S06.
Flores De Santiago, F., Kovacs, J. M., & Lafrance, P. (2013). An object-oriented classification
method for mapping mangroves in Guinea, West Africa, using multipolarized ALOS
PALSAR L-band data. International Journal of Remote Sensing, 34, 563–586.
Fujimoto, K., Imaya, A., Tabuchi, R., Kuramoto, S., Utsugi, H., & Murofushi, T. (1999). Be-
lowground carbon storage of Micronesian mangrove forests. Ecological Research, 14,
409–413.
Gao, B. C. (1996). NDWI—A normalized difference water index for remote sensing of veg-
etation liquid water from space. Remote Sensing of Environment, 58, 257–266.
Gao, J. (1998). A hybrid method toward accurate mapping of mangroves in a marginal
habitat from SPOT multispectral data. International Journal of Remote Sensing, 19,
1887–1899.
Gao, J., Chen, H., Zhang, Y., & Zha, Y. (2004). Knowledge-based approaches to accurate
mapping of mangroves from satellite data. Photogrammetric Engineering & Remote
Sensing, 70, 1241–1248.
Gaveau, D. L., & Hill, R. A. (2003). Quantifying canopy height underestimation by laser
pulse penetration in small-footprint airborne laser scanning data. Canadian Journal
of Remote Sensing, 29, 650–657.
Giri, C., Ochieng, E., Tieszen, L., Zhu, Z., Singh, A., Loveland, T., ... Duke, N. (2011). Status
and distribution of mangrove forests of the world using earth observation satellite
data. Global Ecology and Biogeography, 20, 154–159.
GOI (Government of Indonesia) (2011). Presidential regulation no 61/2011 on national ac-
tion plan to reduce green house gasses (GHGs) emission. Jakarta, Indonesia: Office of the
President.
Green, E., Mumby, P., Edwards, A., Clark, C., & Ellis, A. (1998a). The assessment of man-
grove areas using high resolution multispectral airborne imagery. Journal of Coastal
Research, 433–443.
Green, E. P., Clark, C. D., Mumby, P. J., Edwards, A. J., & Ellis, A. (1998b). Remote sensing
techniques for mangrove mapping. International Journal of Remote Sensing, 19,
935–956.
Held, A., Ticehurst, C., Lymburner, L., & Williams, N. (2003). High resolution mapping of
tropical mangrove ecosystems using hyperspectral and radar remote sensing.
International Journal of Remote Sensing, 24, 2739–2759.
Hess, L. L., Melack, J. M., Filoso, S., & Wang, Y. (1995). Delineation of inundated area and
vegetation along the Amazon floodplain with the SIR-C synthetic aperture radar.
IEEE Transactions on Geoscience and Remote Sensing, 33, 896–904.
Hiraishi, T., Krug, T., Tanabe, K., Srivastava, N., Jamsranjav, B., Fukuda, M., & Troxler, T.
(2014). 2013 supplement to the 2006 IPCC guidelines for national greenhouse gas inven-
tories: Wetlands.
Hu, Q., Wu, W., Xia, T., Yu, Q., Yang, P., Li, Z., & Song, Q. (2013). Exploring the use of Google
Earth imagery and object-based methods in land use/cover mapping. Remote Sensing,
5, 6026–6042.
Huete, A. R. (1988). A soil-adjusted vegetation index (SAVI). Remote Sensing of
Environment, 25, 295–309.
Ilman, M., Wibisono, I. T. C., & Suryadiputra, I. N. N. (2011). State of the art information on
mangrove ecosystems in Indonesia. Bogor: Wetlands International–Indonesia
Programme.
Irons, J. R., Dwyer, J. L., & Barsi, J. A. (2012). The next Landsat satellite: The Landsat data
continuity mission. Remote Sensing of Environment, 122, 11–21.
Jensen, J. R. (2005). Introductory digital image processing 3rd edition. Upper saddle river:
Prentice hall.
Jensen, J. R., Lin, H., Yang, X., Ramsey, E., III, Davis, B. A., & Thoemke, C. W. (1991). The
measurement of mangrove characteristics in southwest Florida using SPOT multi-
spectral data. Geochemistry International, 6, 13–21.
Kathiresan, K., & Bingham, B. L. (2001). Biology of mangroves and mangrove ecosystems.
Advances in Marine Biology, 40, 81–251.
Kauffman, J. B., & Donato, D. (2012). Protocols for the measurement, monitoring and
reporting of structure, biomass and carbon stocks in mangrove forests. Center for In-
ternational Forestry Research Center (CIFOR) working paper (pp. 86).
 A. Aslan et al. / Remote Sensing of Environment 183 (2016) 65–81
Kellndorfer, J., Walker, W., Pierce, L., Dobson, C., Fites, J. A., Hunsaker, C., ... Clutter, M.
(2004). Vegetation height estimation from shuttle radar topography mission and na-
tional elevation datasets. Remote Sensing of Environment, 93, 339–358.
Koenker, R., & Bassett, G., Jr. (1978). Regression quantiles. Econometrica: journal of the
Econometric Society, 33–50.
Komiyama, A., Poungparn, S., & Kato, S. (2005). Common allometric equations for esti-
mating the tree weight of mangroves. Journal of Tropical Ecology, 21, 471–477.
Kovacs, J., Lu, X., Flores-Verdugo, F., Zhang, C., Flores de Santiago, F., & Jiao, X. (2013). Ap-
plications of ALOS PALSAR for monitoring biophysical parameters of a degraded black
mangrove (Avicennia germinans) forest. ISPRS Journal of Photogrammetry and Remote
Sensing, 82, 102–111.
Kuenzer, C., Bluemel, A., Gebhardt, S., Quoc, T. V., & Dech, S. (2011). Remote sensing of
mangrove ecosystems: A review. Remote Sensing, 3, 878–928.
Kushwaha, S., Dwivedi, R., & Rao, B. (2000). Evaluation of various digital image processing
techniques for detection of coastal wetlands using ERS-1 SAR data. International
Journal of Remote Sensing, 21, 565–579.
Lang, M. W., & Kasischke, E. S. (2008). Using C-band synthetic aperture radar data to mon-
itor forested wetland hydrology in Maryland's coastal plain, USA. IEEE Transactions on
Geoscience and Remote Sensing, 46, 535–546.
Lee, J. S. (1980). Digital image enhancement and noise filtering by use of local statistics.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 165–168.
Long, B. G., & Skewes, T. D. (1996). A technique for mapping mangroves with Landsat TM
satellite data and geographic information system. Estuarine, Coastal and Shelf Science,
43, 373–381.
Lucas, R., Rebelo, L., Fatoyinbo, L., Rosenqvist, A., Itoh, T., Shimada, M., ... Trettin, C. (2014).
Contribution of L-band SAR to systematic global mangrove monitoring. Marine and
Freshwater Research.
Lucas, R. M., Bunting, P., Clewley, D., Proisy, C., Souza Filho, P. W. M., Viergever, K., ...
Rosenqvist, A. (2009). Characterisation and monitoring of mangroves using ALOS
PALSAR data.
Lucas, R. M., Mitchell, A. L., Rosenqvist, A., Proisy, C., Melius, A., & Ticehurst, C. (2007). The
potential of L-band SAR for quantifying mangrove characteristics and change: Case
studies from the tropics. Aquatic Conservation: Marine and Freshwater Ecosystems,
17, 245–264.
Lucas, R. M., Ellison, J., Mitchell, A., Donnelly, B., Finlayson, M., & Milne, A. (2002). Use of
stereo aerial photography for quantifying changes in the extent and height of man-
groves in tropical Australia. Wetlands Ecology and Management, 10, 159–173.
Lucas, R. M., Moghaddam, M., & Cronin, N. (2004). Microwave scattering from mixed-
species forests, Queensland, Australia. IEEE Transactions on Geoscience and Remote
Sensing, 42, 2142–2159.
Lugo, A. E., & Snedaker, S. C. (1974). The ecology of mangroves. Annual Review of Ecology
and Systematics, 39–64.
Madsen, S. N., Zebker, H. A., & Martin, J. (1993). Topographic mapping using radar inter-
ferometry: Processing techniques. IEEE Transactions on Geoscience and Remote
Sensing, 31, 246–256.
Masek, J. G., Honzak, M., Goward, S. N., Liu, P., & Pak, E. (2001). Landsat-7 ETM+ as an ob-
servatory for land cover: Initial radiometric and geometric comparisons with
Landsat-5 Thematic Mapper. Remote Sensing of Environment, 78, 118–130.
Margono, B. A., Potapov, P. V., Turubanova, S., Stolle, F., & Hansen, M. C. (2014). Primary
forest cover loss in Indonesia over 2000–2012. Nature Climate Change, 4, 730–735.
Marshall, A. J. (2007). Ecology of Indonesian Papua part two. Tuttle Publishing.
McLeod, E., Chmura, G. L., Bouillon, S., Salm, R., Björk, M., Duarte, C. M., ... Silliman, B. R.
(2011). A blueprint for blue carbon: Toward an improved understanding of the role
of vegetated coastal habitats in sequestering CO2. Frontiers in Ecology and the
Environment, 9, 552–560.
Miettinen, J., Shi, C., & Liew, S. C. (2011). Deforestation rates in insular Southeast Asia be-
tween 2000 and 2010. Global Change Biology, 17, 2261–2270.
MoF (Ministry of Forestry of Indonesia). (2008). Rekalkulasi Penutupan Lahan (Forest Re-
source Recalculation) Indonesia Tahun 2008. (Badan Planology Kehutanan
Departemen Kehutanan Indonesia).
Mitchard, E. T. A. (2012). Using satellite remote sensing to quantify woody cover and biomass
across Africa. University of Edinburgh.
Mougin, E., Proisy, C., Marty, G., Fromard, F., Puig, H., Betoulle, J., & Rudant, J. P. (1999).
Multifrequency and multipolarization radar backscattering from mangrove forests.
IEEE Transactions on Geoscience and Remote Sensing, 37, 94–102.
Mumby, P. J., Edwards, A. J., Arias-González, J. E., Lindeman, K. C., Blackwell, P. G., Gall, A., ...
Renken, H. (2004). Mangroves enhance the biomass of coral reef fish communities in
the Caribbean. Nature, 427, 533–536.
Murdiyarso, D., Purbopuspito, J., Kauffman, J. B., Warren, M. W., Sasmito, S. D., Donato, D.
C., ... Kurnianto, S. (2015). The potential of Indonesian mangrove forests for global cli-
mate change mitigation. Nature Climate Change, 5, 1089–1092.
Murdiyarso, D., Kauffman, J. B., & Verchot, L. V. (2013). Climate change mitigation strate-
gies should include tropical wetlands. Carbon Management, 4, 491–499.
Nascimento, H. E., & Laurance, W. F. (2002). Total aboveground biomass in central Ama-
zonian rainforests: A landscape-scale study. Forest Ecology and Management, 168,
311–321.
Nascimento, W. R., Jr., Souza-Filho, P. W. M., Proisy, C., Lucas, R. M., & Rosenqvist, A.
(2013). Mapping changes in the largest continuous Amazonian mangrove belt using
object-based classification of multisensor satellite imagery. Coastal and Shelf Science:
Estuarine.
81
Neukermans, G., Dahdouh-Guebas, F., Kairo, J. G., & Koedam, N. (2008). Mangrove species
and stand mapping in Gazi Bay (Kenya) using Quickbird satellite imagery. Journal of
Spatial Science, 53, 75–86.
Peet, R. K. (1974). The measurement of species diversity. Annual Review of Ecology and
Systematics, 285–307.
Pendleton, L., Donato, D. C., Murray, B. C., Crooks, S., Jenkins, W. A., Sifleet, S., ... Marbà, N.
(2012). Estimating global “blue carbon” emissions from conversion and degradation
of vegetated coastal ecosystems. PloS One, 7, e43542.
Phillips, O. L., Malhi, Y., Higuchi, N., Laurance, W. F., Núnez, P. V., Vásquez, R. M., ... Brown,
S. (1998). Changes in the carbon balance of tropical forests: Evidence from long-term
plots. Science, 282, 439–442.
Polidoro, B. A., Carpenter, K. E., Collins, L., Duke, N. C., Ellison, A. M., Ellison, J. C., ... Koedam,
N. E. (2010). The loss of species: Mangrove extinction risk and geographic areas of
global concern. PloS One, 5, e10095.
Pool, D. J., Snedaker, S. C., & Lugo, A. E. (1977). Structure of mangrove forests in Florida,
Puerto Rico, Mexico, and Costa Rica. Biotropica, 195–212.
Pouwer, J. (1970). Mimika land tenure. New Guinea research bulletin. 3. (pp. 24–33). Can-
berra: Australian National University.
Proisy, C., Mougin, E., Fromard, F., & Karam, M. (2000). Interpretation of polarimetric
radar signatures of mangrove forests. Remote Sensing of Environment, 71, 56–66.
Proisy, C., Mougin, E., Fromard, F., Trichon, V., & Karam, M. (2002). On the influence of
canopy structure on the radar backscattering of mangrove forests. International
Journal of Remote Sensing, 23, 4197–4210.
Rahman, A. F., Dragoni, D., Didan, K., Barreto-Munoz, A., & Hutabarat, J. A. (2013). Detect-
ing large scale conversion of mangroves to aquaculture with change point and
mixed-pixel analyses of high-fidelity MODIS data. Remote Sensing of Environment,
130, 96–107.
Ramsey, E. W., & Jensen, J. R. (1996). Remote sensing of mangrove wetlands: Relating can-
opy spectra to site-specific data. Photogrammetric Engineering and Remote Sensing, 62,
939–948.
Ramsey, E. W., III, Nelson, G. A., & Sapkota, S. K. (1998). Classifying coastal resources by
integrating optical and radar imagery and color infrared photography. Mangroves
and Salt Marshes, 2, 109–119.
Rasolofoharinoro, M., Blasco, F., Bellan, M., Aizpuru, M., Gauquelin, T., & Denis, J. (1998). A
remote sensing based methodology for mangrove studies in Madagascar.
International Journal of Remote Sensing, 19, 1873–1886.
de Souza Pereira F., R., Kampel, M., & Cunha-Lignon, M. (2012). Mapping of mangrove for-
ests on the southern coast of São Paulo, Brazil, using synthetic aperture radar data
from ALOS/PALSAR. IEEE Geoscience and Remote Sensing Letters, 3, 567–576.
Rönnholm, P., Hyyppä, J., Hyyppä, H., Haggrén, H., Yu, X., & Kaartinen, H. (2004). Calibra-
tion of laser-derived tree height estimates by means of photogrammetric techniques.
Scandinavian Journal of Forest Research, 19, 524–528.
Shimada, M., Isoguchi, O., Tadono, T., & Isono, K. (2009). PALSAR radiometric and geomet-
ric calibration. IEEE Transactions on Geoscience and Remote Sensing, 47, 3915–3932.
Simard, M., Rivera-Monroy, V. H., Mancera-Pineda, J. E., Castañeda-Moya, E., & Twilley, R.
R. (2008). A systematic method for 3D mapping of mangrove forests based on Shuttle
Radar Topography Mission elevation data, ICEsat/GLAS waveforms and field data: Ap-
plication to Ciénaga Grande de Santa Marta, Colombia. Remote Sensing of
Environment, 112, 2131–2144.
Simard, M., Zhang, K., Rivera-Monroy, V. H., Ross, M. S., Ruiz, P. L., Castañeda-Moya, E., ...
Rodriguez, E. (2006). Mapping height and biomass of mangrove forests in Everglades
National Park with SRTM elevation data. Photogrammetric Engineering and Remote
Sensing, 72, 299–311.
Suratman, M. N. (2008). Carbon sequestration potential of mangroves in southeast Asia.
Managing forest ecosystems: The challenge of climate change (pp. 297–315). Springer.
Tomlinson, P. (1986). The botany of mangroves. Cambridge tropical biology series. In (p.
419): Cambridge University Press, Cambridge.
Treuhaft, R. N., Law, B. E., & Asner, G. P. (2004). Forest attributes from radar interferomet-
ric structure and its fusion with optical remote sensing. Bioscience, 54, 561–571.
Vo, Q. T., Kuenzer, C., Vo, Q. M., Moder, F., & Oppelt, N. (2012). Review of valuation
methods for mangrove ecosystem services. Ecological Indicators, 23, 431–446.
Wang, L., Sousa, W., & Gong, P. (2004). Integration of object-based and pixel-based clas-
sification for mapping mangroves with IKONOS imagery. International Journal of
Remote Sensing, 25, 5655–5668.
Wang, C., & Glenn, N. F. (2008). A linear regression method for tree canopy height estima-
tion using airborne lidar data. Canadian Journal of Remote Sensing, 34, S217–S227.
Ward, J. H., Jr. (1963). Hierarchical grouping to optimize an objective function. Journal of
the American Statistical Association, 58, 236–244.
Willmott, C. J. (1982). Some comments on the evaluation of model performance. Bulletin
of the American Meteorological Society, 63, 1309–1313.
Wilson, B. A., & Franklin, S. E. (1992). Characterization of alpine vegetation cover using
satellite remote sensing in the Front Ranges, St. Elias Mountains, Yukon Territory.
Global Ecology and Biogeography Letters, 90–95.
Yu, L., & Gong, P. (2012). Google Earth as a virtual globe tool for Earth science applications
at the global scale: Progress and perspectives. International Journal of Remote Sensing,
33, 3966–3986.