Regional Studies in Marine Science 44 (2021) 101730

Contents lists available at ScienceDirect

Regional Studies in Marine Science

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

Comparison of vegetation indices for estimating above-ground

mangrove carbon stocks using PlanetScope image
∗
Eva Purnamasari a , Muhammad Kamal b , , Pramaditya Wicaksono b
a
Master Program of Remote Sensing, Faculty of Geography, Universitas Gadjah Mada, Yogyakarta, Indonesia
b
Department of Geographic Information Science, Faculty of Geography, Universitas Gadjah Mada, Yogyakarta, Indonesia

article info a b s t r a c t

Article history: Mangroves play a pivotal role in providing ecological benefits and services to reduce and adapt to
Received 24 November 2020 climate change impact on coastal ecosystem. They are capable of absorbing carbon, which is crucial
Received in revised form 4 March 2021 in controlling CO2 levels in the atmosphere. This research aims to assess the accuracy of selected
Accepted 7 March 2021
vegetation indices for estimating above-ground carbon (AGC) stocks of mangroves using PlanetScope
Available online 10 March 2021
images in Bedul, Banyuwangi, East Java Province, Indonesia. A semi-empirical approach was used to
Keywords: assess and map mangrove AGC, starting with applying the allometric equation to calculate field-
Above-ground carbon stock measured species-specific AGC stocks. Regression analyses were applied to develop a relationship
Mangroves between field AGC and vegetation indices derived from PlanetScope Image, including Normalized
PlanetScope image Difference Vegetation Index (NDVI), Difference Vegetation Index (DVI), and Enhanced Vegetation Index
Vegetation index
(EVI). The Standard Errors of Estimates (SE) were 31.41, 32.93, and 31.63 tons/ha for DVI, EVI, and NDVI,
respectively. Thus, carbon stocks estimation, including DVI as an independent variable, is considered
more accurate than other vegetation indices tested in this research
© 2021 Elsevier B.V. All rights reserved.

1. Introduction Considering that mangrove forests often grow in hardly acces-

sible terrain, it is necessary that constraints related to time, cost,
Global warming is one of the most feared problems, and and energy in mangrove studies to be minimized. Remote sensing
carbon emissions are considered its strongest influencing factors satellites are now commonly used for estimating extension and
(Bindu et al., 2018). Global climate change has been attributed dynamics, like detecting changes in mangrove areas (Sulaiman
to the loss of forest area due to advances in crop and livestock et al., 2013). In 2016, the latest remote sensing data product was
agriculture, forest fires, and resource overuse. Coupled with the released, namely PlanetScope. This high-resolution satellite im-
release of greenhouse gases into the atmosphere by industrial agery offers several advantages in data acquisition, including the
activities and transportation development, it creates a daunting ability to sense and record 200 million km2 of the ground surface
challenge for humans to survive (Smith et al., 2017; Grant et al.,
area daily with 130 complete satellite constellations. According
2017; Pant et al., 2018). To overcome this circumstance, many
to Planet Labs (2020), each PlanetScope satellite is 3U in size
have suggested enhancing the role of forests as CO2 absorbers by
(CubeSat form factor = 10 cm × 10 cm × 30 cm). It has a 98◦
devising natural and planted forest management systems (Brown,
inclination angle, with an orbital altitude of 475 km in a sun-
1997). Coastal ecosystems such as mangroves and seagrass play a
crucial part in providing benefits and services related to climate synchronous orbit, allowing object sensing in the morning and
change impact reduction and adaptation (Howard et al., 2014). during the day and, especially along the equator, at 09:30−11:30
Mangroves are a group of vegetation species that can absorb am. PlanetScope has a 12-bit radiometric resolution carrying four
carbon, an element controlling CO2 concentrations in the atmo- spectral bands (blue, green, red, and near-infrared/NIR), a 3 m
sphere (Sidik et al., 2019). Mangrove forests in Indonesia have spatial resolution, and a high temporal resolution (one day) at
the most extensive area worldwide (Kusmana, 1996), making it nadir (Planet Labs, 2020).
tremendous potential for carbon absorption. The global mangrove Mangrove ecosystems are one of the objects detectable in
area is about 16,530,000 hectares, with 42% of it found in Asia, remote sensing data because they are geographically situated at
20% in Africa, and 15% in North and Central America (Giri et al., the interface between land and sea, resulting in unique sensing
2011). products compared to other vegetation objects (Faizal and Amran,
2005). Because remote sensing acquires information quickly and
∗ Corresponding author. with a wide area of coverage, it can provide data required in
E-mail address: [email protected] (M. Kamal). mangrove carbon stock estimation (Hastuti et al., 2017). This

https://doi.org/10.1016/j.rsma.2021.101730
2352-4855/© 2021 Elsevier B.V. All rights reserved.
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

method generally uses image transformation like vegetation in- creates PlanetScope images that are ready to use and geometri-
dices to measure vegetation density levels, which are related to cally, radiometrically, and atmospherically correct. Image mask-
biomass, by applying a specific algorithm to a satellite image. ing is aimed to separate the research area from the non-observed
The vegetation index transforms the spectral values of several re- area located in the same image scene. The research area was
mote sensing bands to accentuate those of vegetation objects. In the mangrove forest in the Bedul Mangrove Area, Banyuwangi.
practical terms, it is a mathematical transformation that involves Masking was also applied to separate between water bodies and
several bands at once and produces a new, more representative land cover types other than mangroves using visual interpretation
image of vegetation objects (Danoedoro, 2012). with the help of the image color composite. The RGB composite
In this research, three vegetation indices (VIs) were selected: used was 432 (NIR, Red, and Green).
Normalized Difference Vegetation Index (NDVI), Difference Veg- In this study, an image transformation, PCA, was conducted
etation Index (DVI), and Enhanced Vegetation Index (EVI). These automatically by utilizing tools and features in the image pro-
VIs were selected after factoring in numerous literature studies cessing software. The PlanetScope image bands, namely Blue,
and research reporting their strengths and weaknesses in dif- Green, Red, and Infrared, were inputted to the PCA. This image
ferent circumstances (Wicaksono et al., 2011; Wicaksono, 2017; processing produced new pixel values that allowed differences in
Kamal et al., 2016; Nguyen et al., 2019; Xia et al., 2020; Zhu et al., mangrove species to be visually observable from the PCA results.
2020; Aljahdali et al., 2021). Thus, it was intended to assess their Spectral transformation based on PCA accentuates masked or
accuracy in estimating above-ground carbon stocks from Plan- hidden information that ordinary spectral bands cannot display.
etScope images in the Bedul Mangrove Area, Banyuwangi, East PCA rotates data to produce new bands in a small amount, but
Java Province, Indonesia. The research differs from previous stud- each of these bands contains specific and unique information
ies in remote sensing data used and area observed. Accordingly, (Murti and Wicaksono, 2014).
it was also designed to investigate whether the three vegetation Table 1 shows the algorithms of the three vegetation indices
indices produced accurate estimates with such differences. analyzed in this study: NDVI, EVI, and DVI. The researchers chose
to use these spectral transformation tools because they had a
2. Data and methods high sensitivity to vegetation biomass, especially NDVI and EVI
(Jensen, 2004). Previous studies have also confirmed that they
2.1. Research site can estimate carbon stocks accurately in respective research lo-
cations. Therefore, this research was designed to prove which of
This research was conducted in the Bedul Mangrove Block, a the three indices could produce estimates with the best accuracy
mangrove forest part of the Alas Purwo National Park in Bloksolo in the area observed.
Sub-village, Sumberasri Village, Purwoharjo District, Banyuwangi
Regency. According to the minutes of the areal measurement on 2.3. Field sampling collection
May 27, 1983, the national park covers 43,420 hectares of land.
Lowland tropical rainforest ecosystem composes most of the park, The sampling point map was prepared based on the PCA
while the mangrove forest area, situated at the Grajagan Bay, is results. Samples were collected from several plots, the plot was
currently about 1200 hectares. The Grajagan Bay is geographically square and 20 m × 20 m in size, which was later subdivided
located between 114◦ 13’20.203’’ - 114◦ 20’459.79’’ E and 8◦ into three parts: 20 m × 20 m, 10 m × 10 m, and 5 m × 5
35’52.79’’ - 8◦ 37’28.697’’ S (Fig. 1). m, according to the spatial resolution of the image used. Each
The Alas Purwo National Park management identified 27 plot size was used to measure different tree diameters: the 20
species of true mangroves in several locations. About 24 of them m × 20 m plots for trees with a diameter of >20 cm, the 10
are distributed along the Segara Anak River, which is dominated m × 10 m plots for those with a diameter of 10−20 cm, and
by Rhizophora mucronata, Ceriops tagal, and Bruguiera gymnor- the 5 m × 5 m plots for those with a diameter of <10 cm
rhiza. The river influences the mangrove forest development at (SNI 7724: 2011). The fieldwork comprised observations, mea-
Grajagan Bay. Most mangrove species thrive in muddy soils, surement of diameter at breast height (DBH), mangrove species
particularly where silt accumulates. Along ±15 km of the Segara identification, and canopy density measurement. DBH was mea-
Anak River, the mud substrate has varying depths: from 0.14 m sured from mangroves in the size of sapling, pole, and stand,
to 2.1 m, and is composed mainly of mud, with some sandy mud and the mangrove species were identified using an ecological ap-
substrates found at a few sites. proach considering mangrove characteristics and growth zones.
An allometric equation was then applied to these data to produce
2.2. Image dataset biomass estimates for mangrove saplings, poles, and stands. This
research used the species-specific allometric equation formulated
This research used the level-3B PlanetScope image recorded on by Komiyama et al. (2008) because it has been designed for use
December 3, 2019. This image is an Ortho Scene Product, i.e., an in Asian mangroves and is, therefore, highly relevant for appli-
image product that has been orthorectified and projected to a cations in Indonesian regions. Above-ground carbon stocks (AGC)
map projection. It has also been geometrically corrected using were computed from above-ground biomass based on the rules
GCP (Ground Control Points) obtained from the Digital Elevation published in the Indonesian National Standards (SNI) 7724:2011,
Model (DEM) data with 30 m to 90 m resolution and position in which 0.47 or 47% of biomass is carbon.
accuracy of less than 10 m. The cartographic projection system
was UTM with the horizontal datum WGS 84, and the resampling 2.4. AGC estimation and mapping
method used was cubic convolution.
The image processing did not include radiometric correction The mangrove AGC mapping began with building a corre-
because the PlanetScope image product was already corrected lation between two variables: the vegetation index values and
radiometrically and corrected to Surface Reflectance. This radio- field-measured carbon data using correlation and regression anal-
metric correction involves conversion to absolute radiometric yses. However, before conducting these analyses, the normality
values based on calibration coefficients that are always updated Kolmogorov–Smirnov test was applied to both data to fulfill the
with on-orbit calibration techniques and correction of radiomet- requirement of data normality in parametric statistical modeling.
ric values to minimize calculation errors (Planet Labs, 2020). It Correlation analysis is a statistical method used to identify the
2
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

Fig. 1. The research location laid on the PlanetScope image, with areal boundaries defined based on the Principal Component Analysis (PCA) results.

Table 1
Vegetation Index algorithms.
No Index Algorithms References
1 Difference Vegetation NIR-Red (1) Tucker (1980)
Index (DVI)
2 Enhanced Vegetation G*((NIR- (2) Huete et al. (2002)
Index (EVI) Red)/(NIR+C1*Red-
C2*Blue+L))
3 Normalized Difference (NIR- (3) Rouse et al. (1973)
Vegetation Index (NDVI) Red)/(NIR+Red)

EVI coefficients: G=2.5, C1=6, C2=7.5, L=1.

Table 2
Field-collected data.
Sample Coordinates Average Average tree Frac- Dominant
points DBH (cm) height (m) tional species
canopy
cover
X Y
1 −8.59648 114.25760 27.8 10.8 0.40 R. mucronata
2 −8.59587 114.26702 29.3 11.7 0.34 E. agallocha
3 −8.59882 114.27336 40.9 14.5 0.33 R. mucronata
4 −8.59813 114.27344 54.8 23.1 0.14 R. mucronata
5 −8.59744 114.27371 39.2 17.6 0.18 R. mucronata
6 −8.59903 114.27770 45.0 13.6 0.22 A. officinalis
7 −8.59988 114.27762 40.9 18.9 0.21 R. apiculata
8 −8.60047 114.27749 50.6 18.8 0.09 R. apiculata
9 −8.60043 114.28008 47.9 11.9 0.05 R. apiculata
10 −8.60120 114.27996 47.1 23.3 0.21 R. apiculata
11 −8.59279 114.27028 38.5 15.9 0.31 R. mucronata
12 −8.59202 114.27024 28.2 8.2 0.18 C. decandra
13 −8.59250 114.26817 36.3 14.1 0.29 S. alba
14 −8.59010 114.26584 22.6 5.7 0.09 C. tagal
15 −8.59852 114.28223 48.8 19.5 0.20 R. apiculata
16 −8.59796 114.28265 48.2 16.7 0.17 R. apiculata
17 −8.59710 114.27887 41.3 19.1 0.15 S. alba
18 −8.59524 114.27602 34.2 16.1 0.27 S. alba
19 −8.59398 114.27277 35.3 10.8 0.15 S. alba
20 −8.59297 114.27310 22.8 5.9 0.18 C. decandra

relationship between two or more quantitative independent and that, according to previous studies, results in a strong correlation.
dependent variables and produce a coefficient (r) signifying their Linear regression analysis changed image pixel values into abso-
correlation degree. For this purpose, this research used the Pear- lute mangrove AGC and, thus, evaluated the best mangrove AGC
son product-moment correlation with a 95% significance level. estimation with vegetation index as the input.
Vegetation index values that exceeded the r-value for sample size All carbon stock models were tested for accuracy to obtain the
n and had a significant relationship with field-measured carbon best empirical modeling for each vegetation index. However, the
were inputted to the empirical modeling using regression analy- coefficient of determination resulting from the regression analysis
sis. Here, the regression model refers to simple linear regression (R2 ) cannot be used to compute the accuracy of the empirical
3
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

carbon stock modeling. Despite the widely used measures for Table 3
assessing predictive models for numerical data, Alexander et al. Biomass and carbon stock calculation results.

(2015) and Li (2017) suggested that R2 is not an appropriate Sample points Biomass (ton/ha) Carbon stock (ton/ha) Dominant species

measure to evaluate modeling results’ accuracy. R2 is only valid 1 193.53 90.96 R. mucronata
for developing models from training datasets and not suitable for 2 114.18 53.66 E. agallocha
3 210.71 99.03 R. mucronata
model accuracy tests using validation sample datasets. It does not 4 242.75 114.09 R. mucronata
evaluate the absolute value of the predicted models. Therefore, 5 121.23 56.98 R. mucronata
the Standard Error of Estimate (SE) was used to assess the quality 6 123.41 58.00 A. officinalis
and quantity of the estimated carbon stocks (i.e., AGC obtained 7 109.09 51.27 R. apiculata
8 133.77 62.87 R. apiculata
from the linear regression analysis) by comparing them with the
9 113.95 53.56 R. apiculata
observed carbon stocks. The lower the SE value, the higher the 10 137.76 64.75 R. apiculata
accuracy. However, SE only shows the value of the prediction 11 108.16 50.84 R. mucronata
of accuracy, but not the distribution pattern of the predicted 12 45.35 21.32 C. decandra
values. In addition to SE, we also applied the goodness-of-fit 13 63.94 30.05 S. alba
14 27.36 12.86 C. tagal
using the 1:1 line for accuracy testing, particularly to determine 15 234.01 109.98 R. apiculata
whether the estimated data and accuracy test samples showed an 16 154.00 72.38 R. apiculata
over-or under-estimated pattern. In this test, a red line marked 17 69.58 32.70 S. alba
the ideal 1:1 value between the field-measured and estimated 18 53.01 24.92 S. alba
19 70.73 33.24 S. alba
carbon stocks (Kamal et al., 2016). Over-estimation occurred if
20 38.10 17.91 C. decandra
the linear plot distribution skewed toward the image-estimated
carbon stocks, whereas under-estimation occurred if it skewed
toward the field-measured carbon stocks.
The relative rank proposed by Poudel and Cao (2013) was used affect the biomass and carbon content. According to Dharmawan
to compare the performance of each vegetation indices in AGB and Siregar (2008), 98.7% of biomass is influenced by the number
estimation and mapping. The relative rank shows the relative of trees and DBH because the amount of biomass calculated using
position of each vegetation indices used in the mangrove AGB the allometric equation strongly depends on the number and
estimation. The relative rank is defined as: diameter of mangrove trees.

3.2. Above-ground carbon stocks

Ri = 1 + (m − 1)(Si − Smin )/(Smax − Smin ) (4)
where Ri is the relative rank of the vegetation index i (i = 1, 2, . Table 3 shows the calculation results of the field-measured
. . , m), m is the number of vegetation indices evaluated, Si is the biomass and carbon at each sampling point. Based on the statis-
evaluation statistic value of vegetation index i, and Smin and Smax tical analysis results, a higher amount of biomass corresponds to
are respectively the minimum and maximum value of Si . Ri is a a larger carbon stock. Point 4 had the largest carbon stock among
value between 1 (best) and m (worst). the sampling stations, while point 14 had the smallest one, only
12.86 tons/ha. The carbon stock estimation model was built on
3. Result and discussion the assumption that biomass was closely related to mangrove tree
dimensions (diameter and height). It is consistent with Hairiah
3.1. Identified mangrove species characteristics and Rahayu (2007), which stated that the potential carbon stocks
could be measured from existing tree stand biomass. Biomass
Table 2 presents the list of mangrove species found at several affects the amount of carbon stored in every part of the tree.
locations during the fieldwork. Data were collected from only Therefore, an increase in biomass will be followed by an increase
20 of the planned 40 points due to the mangrove forest’s harsh in carbon content.
condition, making it difficult to access all sample sites. These 20 Based on the results presented in Table 3, it is known that the
samples data were split up equally into two groups: ten sam- species dominating all sampling points are Rhizophora mucronata
ples were used for modeling carbon stocks, while the other ten and Rhizophora apiculata. Also, the largest carbon stocks (114
were for accuracy testing. To obtain an optimum prediction from tons/ha) and biomass content (242.75 tons/ha) were found at the
limited samples, we selected these sampling sets carefully by sampling point where Rhizophora mucronata was predominant.
considering the variation of physical characteristics (Table 2), AGC The same case applied to the carbon stocks (109.98 tons/ha)
value variation, and location distribution of sample plots. There and biomass content (234.01 tons/ha) of a sampling point where
were 14 mangrove species identified along the Segara Anak River Rhizophora apiculata was predominant. Similarly, Suryono et al.
in the Alas Purwo National Park. These species were Avicennia (2018) stated that Rhizophora sp. had higher biomass and carbon
alba, Avicennia marina, Avicennia officinalis, Bruguiera gymnorhiza, contents than other species. It is because the average DBH of
Ceriops tagal, Ceriops decandra, Excoecaria agallocha, Lumnitzera Rhizophora sp. is much larger than that of other species. DBH,
racemosa, Rhizophora apiculata, Rhizophora mucronata, Sonneratia biomass density, and canopy cover are thought to be the influ-
alba, Sonneratia caseolaris, Xylocarpus granatum, and Xylocarpus encing factors of carbon stock in mangrove stands. Trees with
moluccensis, with Rhizophora sp. as the dominant species. Rhi- a diameter of more than 30 cm significantly contribute to the
zophora sp. is the most well-known species in tropical coastal carbon stocks of tree biomass. The higher the number of trees
mangrove ecosystems (Cerón-Souza et al., 2010). with a diameter of larger than 30 cm covering the land, the higher
Table 2 also shows that these mangroves had varying diam- the carbon stock of this land (Hanafi and Bernardianto, 2012).
eters at breast height (DBH), from 22.6 cm (Ceriops tagal) to
54.8 cm (Rhizophora mucronata). The dominant R. mucronata and 3.3. Image-estimated Aboveground Carbon Stock (AGC)
R. apiculata had the highest average DBH: 54.8 cm and 50.6 cm. R.
apiculata is known to have larger DBH and higher wood density A high vegetation index value illustrates vegetation with a
than R. mucronata (Komiyama et al., 2008). In this research, tree high level of greenness in the area observed. On the other hand,
height and mangrove fractional canopy cover did not significantly a low vegetation index value indicates that the monitored land
4
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

Table 4 Table 5
The Kolmogorov–Smirnov normality test results. Comparison of the vegetation indices used in this research.
Input Statistics Input Data normality r R2 SE
Samples Mean StDev Dn KStable DVI Normal 0.820 0.673 31.41
Carbon observed 20 55.57 29.94 0.13 0.30 EVI Normal 0.779 0.607 32.93
DVI 20 0.18 0.03 0.10 0.30 NDVI Normal 0.716 0.513 31.63
EVI 20 0.49 0.09 0.07 0.30
NDVI 20 0.61 0.06 0.11 0.30

this research produced overestimated values, potentially because

the satellite sensors recorded the canopy density of the objects
has vegetation with a low level of greenness, or it is most likely around the mangrove forest area more than the tree diameter
that the object observed is not vegetation. The vegetation indices measured during the fieldwork.
produced different range of values: DVI = 0.1−0.2, EVI = 0.2−0.6,
and NDVI = 0.4−0.6. A normality test was conducted to ensure 3.5. Comparison of vegetation indices
well-modeled data and determine their feasibility for use in fur-
ther analyses: correlation and regression. Table 4 presents the Table 5 compares the coefficients of correlation (r), coefficients
normality test results, with 20 data input (according to the num- of determination (R2 ), and standard errors of estimate (SE) of
ber of collected samples). The results showed that Dn < KStable , the three vegetation indices. It shows that DVI had the highest r
meaning that these 20 data meet the assumptions of parametric and R2 values in all statistical analyses, from the correlation test
statistical research and can be used for further analysis. to the accuracy test. However, the test results did not correlate
Based on the normality test results presented in Table 4, it with each other. There was no relationship between R2 and SE,
can be concluded that the estimated data were suitable for use in meaning that vegetation index with high R2 does not necessarily
statistical analysis. Also, the correlation test results showed that represent high accuracy and vice versa. NDVI had a higher accu-
DVI had the highest value (0.820), as compared to EVI (0.779) and racy than EVI, but the former had lower r and R2 values than the
NDVI (0.716). These figures mean that, in this research, DVI has latter. The regression analysis was used to model above-ground
the strongest relationship with field-measured carbon stocks. carbon stocks in mangrove forests, resulting in a distribution map
The vegetation index values, and the field-measured carbon of carbon stocks for each vegetation index (Fig. 3). Besides, the
stocks were inputted to the estimation model using regression relative ranking calculation based on SE (Table 5) shows that
analysis. This statistical analysis produced coefficients of deter- DVI, NDVI, and EVI were ranked 1, 1.29, and 3.01, respectively.
mination (R2 ) and regression functions for each of the model- It means that DVI has a relatively higher rank than NDVI and EVI
ing inputs, and the resulting linear regression model showed a in estimating mangrove AGB in the study site.
straight line because it was a linear function. Fig. 2 shows that DVI is effective for use because the formula is relatively simple,
DVI yielded the highest R2 value, 0.673, for each input, compared using only the red and infrared bands. This is consistent with
to EVI (0.607) and NDVI (0.513). It indicates that the pixel values Wicaksono et al. (2015), which consider DVI as the most effective
transformed by DVI can explain 67.3% of the variation in field- because: (1) the formula is simple; (2) it requires only red and NIR
measured carbon stocks and that the two variables have a strong bands—thus applicable to most available passive multispectral
relationship. The regression equation obtained was y = 1513x images; and (3) the performance is consistent across radiometric
– 229.5, where y is field-measured carbon stocks, and x is DVI levels and improves with the application of radiometric correc-
values. Despite the different R2 values, the three indices show a tion. DVI is a spectral transformation used to enhance vegetation
strong relationship between the x and y variables. The regression aspects and is developed to avoid negative values. Among the
analysis results were then used to produce estimated carbon three indices, DVI had the highest coefficient of correlation (r).
stocks from image processing for accuracy testing. Besides, prior scholars have reported that it has relatively better
The basic function of the best model per data was applied to performance than other indices (Barati et al., 2011; Naji, 2018).
the vegetation index. In total, there are three maps of the spatial The absorption of electromagnetic waves in the red band by
distribution of carbon stocks above the mangrove surface in this chlorophyll and their high reflection in the NIR band are the
study. The carbon stock information obtained is the carbon value reasons for the high coefficient of correlation in each class of DVI
derived from the results of the regression equation so that the (Tucker, 1980).
pixel value in the image is the carbon value information itself. NDVI is the most frequently used vegetation index in research,
Fig. 3 shows the spatial distribution pattern of the estimated especially to monitor the ground surface (Robinson et al., 2017;
carbon stocks. Bhandari et al., 2012; Barati et al., 2011), and it uses the red and
NIR bands. In some literature and previous studies, NDVI shows
3.4. AGC maps accuracy assessment varying accuracy (Hamdan et al., 2013; Bindu et al., 2018; Anand
et al., 2020). In Anand et al. (2020), the NDVI-estimated carbon is
This study used the Standard Error of Estimate (SE) for ac- higher than the EVI-estimated one. Also, in Muhsoni et al. (2018),
curacy testing: the lower the SE value, the higher the accuracy. NDVI produces the best results among the 24 vegetation indices
The SE calculation results showed that the standard errors were analyzed. These results are possible because NDVI is influenced
31.41 tons/ha for DVI, 32.93 tons/ha for EVI, and 31.63 tons/ha for by the atmosphere, topography, soil, and biomass (Anand et al.,
NDVI. DVI produced the lowest standard error, meaning that the 2020). In the current research, NDVI yielded good scores in the
DVI-estimated carbon stocks have the highest accuracy. Mean- correlation and regression analyses but did not produce the low-
while, EVI yielded the highest standard error, indicating the low- est standard error of estimates in the accuracy test. NDVI is not
est accuracy among the three indices tested in this research. the best index for estimating carbon but has relatively consistent
However, T-test results suggest that the SE results were not accuracy at various levels of radiometric correction (Wicaksono
statistically significant at 95% confidence level. The estimated and et al., 2011).
field-measured carbon stocks were inputted into the SE calcula- EVI had the lowest accuracy, as compared to DVI and NDVI.
tion. Based on the distribution pattern of the estimated mangrove It is possible because, unlike other vegetation indices that only
carbon stocks in Fig. 4, all models built for AGC mapping in use the red and infrared bands, the EVI formula also uses an
5
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

Fig. 2. Regression function between field-measured AGC and (a) DVI, (b) EVI, and (c) NDVI.

Fig. 3. Spatial distribution of AGC derived from (a) DVI, (b) EVI, and (c) NDVI (the AGC value are classified for visualization and comparison purposes only).

additional blue band. Besides, it has a constant value in its cal- Candra et al. (2016) conducted a study in the mangrove forest
culation, affecting the resulting index value. However, Frananda area of the Ngurah Rai Grand Forest Park, Bali, and found that
the mRE-SR index had higher accuracy than DVI and EVI. This
et al. (2015) found that EVI could produce a high R2 value (0.71) area has silty sand-textured and muddy substrates, creating a
but not the best value in the accuracy test. suitable environment for Rhizophora mucronata that thrive in mud
6
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

Fig. 4. The 1:1 goodness-of-fit plot of field-measured AGC against (a) DVI-estimated AGC, (b) EVI-estimated AGC, and (c) NDVI-estimated AGC.

to gravel substrates (Hermawan et al., 2014; Wiyanto and Faiqoh, Declaration of competing interest
2015). Meanwhile, Hamdan et al. (2013) found that NDVI had
the highest coefficient of correlation when used in the Matang The authors declare that they have no known competing finan-
Mangrove area in the Northwest of Peninsular Malaysia. Here, cial interests or personal relationships that could have appeared
mangroves grow under the influence of a tropical climate (warm to influence the work reported in this paper.
and humid), with an average annual temperature of 23.7−33.4◦
Celsius, a relative humidity of 76.5−83.5% (Goessens et al., 2014), Acknowledgments
and year-round rainfall varying from 2000 to 3000 mm, and
semi-diurnal tides with an amplitude of about 3.3 m (Ashton This research was funded by the 2020 Penelitian Dasar (Fun-
et al., 1999). This current research has different results from damental Research) Grant scheme provided by the Ministry of Re-
the above studies in that it observes mangrove forests with dif- search and Technology/ National Research and Innovation Agency
ferent characteristics, compositions, and tide-sensitive waterlog- of the Republic of Indonesia (contract number
ging conditions. Moreover, the distribution and dynamics of soil 2979/UN1.DITLIT/DIT-LIT/PT/2020). The authors would like to
organic carbon content may differ from other research loca- thank (1) the Department of Geographic Information Science at
tions mentioned above as these variables are influenced by tides, the Faculty of Geography, Universitas Gadjah Mada, for providing
vegetation biomass and productivity, species composition, and research facilities and equipment, (2) the Alas Purwo National
sedimentation (Sherman et al., 2003). Park management for granting fieldwork permits, (3) R.F. Lestari,
S.A. Hafid, S.M. Ridha, A.D. Rahmandhana, M.F. Hidayatullah, M.A.
Purnomo, and A. Widodo for their invaluable assistance during
4. Conclusion
the fieldwork.

There are 14 mangrove species found along the Segara Anak References
River in the Alas Purwo National Park. Among those species,
Rhizophora sp. were found to be dominant in this area and con- Alexander, D.L.J., Tropsha, A., Winkler, D.A., 2015. Beware of R(2): Simple,
tributed to the high values of AGC. Overall, PlanetScope image unambiguous assessment of the prediction accuracy of QSAR and QSPR
models. J. Chem. Inf. Model 55 (7), 1316–1322. http://dx.doi.org/10.1021/
has a potential for mangrove AGC estimation and mapping. All
acs.jcim.5b00206.
the three vegetation indices resulted almost similar pattern of Aljahdali, M.O., Munawar, S., Khan, W.R., 2021. Monitoring mangrove forest
estimated AGC value distribution. However, based on the accu- degradation and regeneration: Landsat time series analysis of moisture and
racy testing results, DVI has the lowest standard error of estimate vegetation indices at Rabigh Lagoon, Red Sea. Forests 12 (1), 1–19. http:
//dx.doi.org/10.3390/f12010052.
(31.41 tons/ha) among the three vegetation indices used. It means
Anand, A., Pandey, P.C., Petropoulos, G.P., Pavlides, A., Srivastava, P.K.,
that DVI has the highest AGC estimation accuracy as compared to Sharma, J.K., Malhi, R.K.M., 2020. Use of hyperion for mangrove forest
EVI and NDVI. Nevertheless, all mapping models produced over- carbon stock assessment in Bhitarkanika Forest Reserve: A contribution
estimated mangrove AGC. This could be attributed to the satellite towards blue carbon initiative. Remote Sens. 12, 597. http://dx.doi.org/10.
3390/rs12040597.
sensor that records not only the targeted mangrove objects, but
Ashton, E.C., Hogart, P.J., Ormond, R., 1999. Breakdown of mangrove leaf litter
also the objects around the mangrove area, such as mangrove in a managed mangrove forest in peninsular Malaysia. Hydrobiologia 413,
seedling, low growth mangroves, and organic litters. In addition, 77–88. http://dx.doi.org/10.1023/A:1003842910811.
the variation of mangrove canopy density depicted by an image Barati, S., Rayegani, B., Saati, M., Sharifi, A., Nasri, M., 2011. Comparison the
accuracies of different spectral indices for estimation of vegetation cover
pixel could also play a significant role in determining the AGC
fraction in sparse vegetated areas. Egypt J. Remote Sens. Space Sci. 14, 49–56.
estimation results. Furthermore, this research shows different http://dx.doi.org/10.1016/j.ejrs.2011.06.001.
results from previous studies owing to differences in mangrove Bhandari, A.K., Kumar, A., Singh, G.K., 2012. Feature extraction using normalized
forest characteristics and composition and tide-induced water- difference vegetation index (NDVI): A case study of Jabalpur city. Procedia
Tech. 6, 612–621. http://dx.doi.org/10.1016/j.protcy.2012.10.074.
logging conditions. Future research is required to investigate the
Bindu, G., Rajan, P., Jishnu, E.S., Joseph, K.A., 2018. Carbon stock assessment of
contributing factors affecting the accuracy of an image-based AGC mangroves using remote sensing and geographic information system. Egypt J.
estimation based on the use of tree diameter. Remote Sens. Space Sci. 23, 1–9. http://dx.doi.org/10.1016/j.ejrs.2018.04.006.
Brown, S., 1997. Estimating Biomass and Biomass Change of Tropical Forest: A
Primer. FAO Forestry Paper 134, FAO, USA.
CRediT authorship contribution statement Candra, E.D., Hartono, Wicaksono, P., 2016. Above ground carbon stock estimates
of mangrove forest using worldview-2 imagery in Teluk Benoa, Bali. IOP Conf.
Eva Purnamasari: Conceptualization, Methodology, Formal Ser. Earth Environ. Sci. 47, http://dx.doi.org/10.1088/1755-1315/47/1/012014.
Cerón-Souza, I., Rivera-Ocasio, E., Medina, E., Jimenez, J.A., McMillan, W.O.,
analysis, Writing - original draft. Muhammad Kamal: Validation,
Bermingham, E., 2010. Hybridization and introgression in new world red
Interpreting, Writing - review & editing. Pramaditya Wicaksono: mangroves, Rhizophora (Rhizophoraceae). Am. J. Bot. 97 (6), http://dx.doi.
Methodology, Data curation, Writing - review & editing. org/10.3732/ajb.0900172.

7
E. Purnamasari, M. Kamal and P. Wicaksono Regional Studies in Marine Science 44 (2021) 101730

Danoedoro, P., 2012. Pengantar Penginderaan Jauh Digital. Penerbit ANDI, Naji, T.A.H., 2018. Study of vegetation cover distribution using DVI, PVI, WDVI
Yogyakarta. indices with 2d-space plot. J. Phys.: Conf. Ser. 1003. http://dx.doi.org/10.
Dharmawan, I.W.S., Siregar, C.A., 2008. Karbon Tanah dan Pendugaan Karbon 1088/1742-6596/1003/1/012083.
Tegakan Avicennia marina (Forsk.) Vierh. di Ciasem, Purwakarta. J. Penelitian Nguyen, L.D., Nguyen, C.T., Le, H.S., Tran, B.Q., 2019. Mangrove mapping and
Hutan Konservasi Alam V, 317–328. above-ground biomass change detection using satellite images in coastal
Faizal, A., Amran, M.A., 2005. Effective vegetation index transformation model areas of Thai Binh Province. Vietnam. For. Soc. 3 (2), 248–261. http://dx.
for the prediction of mangrove density Rhizophora mucronata. PIT MAPIN XIV doi.org/10.24259/fs.v3i2.7326.
ITS. Pant, G., Kumar, P.P., Revadekar, J.V., Singh, N., 2018. Climate change and Ut-
Frananda, H., Hartono, Jatmiko, R.H., 2015. Komparasi Indeks Vegetasi Untuk tarakhand: Policy perspective. In: Climate Change in the Himalayas. Springer,
Estimasi Stok Karbon Hutan Mangrove Kawasan Segoro Anak Pada Kawasan Berlin, Heidelberg, Germany, pp. 135–145. http://dx.doi.org/10.1007/978-3-
Taman Nasional Alas Purwo Banyuwangi, Jawa Timur. Majalah Ilmiah Globe 319-61654-4_8.
17 (2), 113–123. Planet Labs, 2020. Planet Imagery Product Specification. Planet Labs Inc.
Giri, C., Ochieng, E., Tieszen, L.L., Zhu, Z., Singh, A., Loveland, T., Masek, J., Poudel, K.P., Cao, Q.V., 2013. Evaluation of methods to predict Weibull param-
Duke, N., 2011. Status and distribution of mangrove forests of the world eters for characterizing diameter distributions. For. Sci. 59 (2), 243–252.
using earth observation satellite data. Global Ecol. Biogeogr. 20, 154–159. http://dx.doi.org/10.5849/forsci.12-001.
http://dx.doi.org/10.1111/j.1466-8238.2010.00584.x. Robinson, N.P., Allerd, B.W., Jones, M.O., Moreno, A., Kimball, J.S., Naugle, D.E.,
Goessens, A., Satyanarayana, B., Stocken, T. Van der, Zuniga, M.Q., Mohd- Erickson, T.A., Richardson, A.D., 2017. A dynamic landsat derived normalized
Lokman, H., Sulong, I., Dahdouh-Guebas, F., 2014. Is Matang Mangrove difference vegetation index (NDVI) product for the conterminous United
Forest in Malaysia sustainably rejuvenating after more than a century of States. Remote Sens. 9, 863. http://dx.doi.org/10.3390/rs9080863.
conservation and harvesting management? PLoS One 9, http://dx.doi.org/10. Rouse, J.W., Hass, R.H., Scell, J.A., Deering, D.W., 1973. Monitoring vegetation
1371/journal.pone.0105069. systems in the great plains with ERTS. In: Third ERTS Symposium. NASA,
Grant, D.B., Trautrims, A., Wong, C.Y., 2017. Sustainable Logistics and Supply pp. 309–317.
Chain Management: Principals and Practices for Sustainable Operations and Sherman, R.E., Fahey, T.J., Martinez, P., 2003. Spatial patterns of biomass and
Management. Kogan Page Publishers, London, UK. aboveground net primary productivity in a mangrove ecosystem in the Do-
Hairiah, K., Rahayu, S., 2007. Pengukuran Karbon Tersimpan di Berbagai Macam minican Republic. Ecosystems 6, 384–398. http://dx.doi.org/10.1007/s10021-
Penggunaan Lahan. World Agroforestry Center. 002-0191-8.
Hamdan, O., Khairunnisa, M.R., Ammar, A.A., Hasmadi, I.M., Aziz, H.K., 2013. Sidik, F., Adame, M.F., Lovelock, C.E., 2019. Carbon sequestration and fluxes of
Mangrove carbon stock assessment by optical satellite imagery. J. Trop. For. restored mangroves in abandoned aquaculture ponds. J. Indian Ocean Reg.
Sci. 25 (4), 554–565, JSTOR: https://www.jstor.org/stable/23616997. 15, 177–192. http://dx.doi.org/10.1080/19480881.2019.1605659.
Hanafi, N., Bernardianto, R.B., 2012. Pendugaan Cadangan Karbon Pada Sistem Smith, P., Howden, M., Krug, T., Masson-Delmotte, V., Mbow, C., Portner, H.-
Penggunaan Lahan di Areal PT. Sikatan Wana Raya. Media SainS 4. O., Reisinger, A., Canadell, J., O’Brien, P., 2017. Special Report on Climate
Hastuti, A.W., Suniada, K.I., Islamy, F., 2017. Carbon stock estimation of man- Change, Desertification, Land Degradation, Sustainable Land Management,
grove vegetation using remote sensing in Perancak estuary. Int. J. Remote Food Security, and Greenhouse Gas Fluxes in Terrestrial Ecosystems (SR2).
Sens. Earth Sci. 14 (2), 137–150. http://dx.doi.org/10.30536/j.ijreses.2017.v14. IPCC, Geneva, Switzerland.
a2841. Sulaiman, N.A., Ruslan, F.A., Tarmizi, N.Md., 2013. Mangrove forest change
Hermawan, A.R., Pribadi, R., Ario, R., 2014. Struktur dan Komposisi Vegetasi analysis along klang coastal using remote sensing technique. In: IEEE 3rd
Mangrove Alami di Kawasan Ekowisata Taman Hutan Raya Ngurah Rai. Bali. International Conference on System Engineering and Technology. 19–20 Aug.,
J. Mar. Res. 3, 405–414. http://dx.doi.org/10.14710/jmr.v3i4.8361. http://dx.doi.org/10.1109/ICSEngT.2013.6650190.
Howard, J., Hoyt, S., Isensee, K., Telszewski, M., Pidgeon, E., 2014. Coastal Blue Suryono, Soenardjo, N., Wibowo, E., Ario, R., Rozy, E.F., 2018. Estimasi Kandungan
Carbon: Methods for Assesing Carbon Stocks and Emissions Factors in Man- Biomassa dan Karbon di Hutan Mangrove Perancak Kabupaten Jembaran
groves, Tidal Salt Marshes, and Seagrasses. Intergovermental Oceanographic Provinsi Bali. Bul. Oseanografi Marina 7, 1–8. http://dx.doi.org/10.14710/
Commissions of UNESCO, Arlinton, USA, p. 180. buloma.v7i1.19036.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Tucker, C.J., 1980. A spectral method for determining the percentage of green
Overview of the radiometric and biophysical performance of the MODIS herbage material in clipped sample. Remote Sens. Environ. 9, 175–181.
vegetation indices. Remote Sens. Environ. 83, 195–213. http://dx.doi.org/10. http://dx.doi.org/10.1016/0034-4257(80)90007-3.
1016/S0034-4257(02)00096-2. Wicaksono, P., 2017. Mangrove above-ground carbon stock mapping of multi-
Indonesian National Standard (SNI), 2011. Pengukuran dan Perhitungan Cadan- resolution passive remote-sensing systems. Int. J. Remote Sens. 38 (6),
gan Karbon – Pengukuran Lapangan untuk Penaksiran Cadangan Karbon 1551–1578. http://dx.doi.org/10.1080/01431161.2017.1283072.
Hutan (Ground Based Forest Carbon Accounting). Badan Standardisasi Wicaksono, P., Danoedoro, P., Hartono, Nehren, U., 2015. Mangrove biomass
Nasional. carbon stock mapping of the Karimunjawa Islands using multispectral
Jensen, J.R., 2004. Introductory Digital Image Processing – A Remote Sensing remote sensing. Int. J. Remote Sens. 37 (1), 26–52. http://dx.doi.org/10.1080/
Perspective. Pearson Prentice Hall, New Jersey. 01431161.2015.1117679.
Kamal, M., Phinn, S., Johansen, K., 2016. Assessment of multi-resolution image Wicaksono, P., Danoedoro, P., Hartono, Nehren, U., Ribbe, L., 2011. Preliminary
data for mangrove leaf area index mapping. Remote Sens. Environ. 176, work of mangrove ecosystem carbon stock mapping in small island using
242–254. http://dx.doi.org/10.1016/j.rse.2016.02.013. remote sensing: Above and below ground carbon stock mapping on medium
Komiyama, A., Poungparn, S., Ong, J.E., 2008. Allometry, biomass and productivity resolution satellite image. In: Neale, C.M.U., Maltese, A. (Eds.), Proceedings
of mangrove forest: A review. Aquat. Bot. 89, 128–137. http://dx.doi.org/10. of SPIE 8174, Remote Sensing for Agriculture, Ecosystem, and Hydrology XIII,
1016/j.aquabot.2007.12.006. 81741B. SPIE Remote Sensing, Prague, http://dx.doi.org/10.1117/12.897926.
Kusmana, C., 1996. Nilai Ekonomis Hutan Mangrove. Media Konservasi V (1), Wiyanto, D.B., Faiqoh, E., 2015. Analisis Vegetasi dan Struktur Komunitas Man-
17–24. grove di Teluk Benoa, Bali. J. Mar. Aquat. Sci. 1, 1–7. http://dx.doi.org/10.
Li, J., 2017. Assessing the accuracy of predictive models for numerical data: 24843/jmas.2015.v1.i01.1-7.
Not r nor r2, why not? Then what? PLoS One 12 (8), e0183250. http: Xia, Q., Qin, C.-Z., Li, H., Huang, C., Su, F.-Z., Jia, M.-M., 2020. Evaluation of sub-
//dx.doi.org/10.1371/journal.pone.0183250. merged mangrove recognition index using multi-tidal remote sensing data.
Muhsoni, F.F., Sambah, A.B., Mahmudi, M., Wiadnya, D.G.R., 2018. Comparison of Ecol. Indic. 113, 106196. http://dx.doi.org/10.1016/j.ecolind.2020.106196.
different vegetation indices for assessing mangrove density using sentinel-2 Zhu, Y., Liu, K., Liu, L., Myint, S.W., Wang, S., Cao, J., Wu, Z., 2020. Estimating
imagery. Int. J. Geomate 14, 42–51. http://dx.doi.org/10.21660/2018.45.7177. and mapping mangrove biomass dynamic change using worldview-2 images
Murti, H.S., Wicaksono, P., 2014. Analisis Saluran Spektral Yang Paling Berpen- and digital surface models. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
garuh Dalam Identifikasi Kesehatan Terumbu Karang: Studi Kasus Pulau 13, 2123–2134. http://dx.doi.org/10.1109/JSTARS.2020.2989500.
Menjangan Besar dan Menjangan Kecil, Kepulauan Karimunjawa. Majalah
Ilmiah Globe 16, 117–124.