Seagrass Aboveground Carbon Stock Mapping using PlanetScope SuperDove Imagery in Nemberala, Rote Island, East Nusa Tenggara

Seagrass ecosystem is natural objects that can be used to adapt and mitigate climate change through blue carbon sequestration. There are 16 seagrass species in Indonesia and the high diversity of these species supports the high potential of carbon that can be absorbed and stored. This study aimed to develop a general equation to estimate seagrass aboveground carbon stock (AGC) from the percent cover (PC), and to map seagrass AGC using PlanetScope imagery in Nemberala, Rote Island. Nemberala has very diverse seagrass species such as Enhalus acoroides (Ea), Cymodocea rotundata (Cr), Halophila ovalis (Ho), Syringodium isoetifolium (Si), Thalassodendron ciliatum (Tc) and Thalassia hemprichii (Th). The results of laboratory analysis to obtain the value of carbon stocks using the Loss on Ignition method from each species were 0.016, 0.004, 0.001, 0.002, 0.001, and 0.0001 (gC/leaf) for Ea, Th, Cr, Si, Tc, and Ho, respectively. The general equation for predicting AGC from PC is SeagrassAGC = (0.051*SeagrassPC) – 0.635 with r of 0.61 and R2 of 0.36. This formula will be applied to convert field seagrass PC data to AGC, which will then be used to train and test the accuracy of seagrass AGC mapping based on PlanetScope SuperDove 8 bands image.


Introduction
Global warming causes instability in the wave and tides and increasing carbon dioxide also changing seawater acidification [1].Human disturbances such as sand dredging, reclamation, and aquaculture could also reduce the potential of carbon sinks to absorb and store carbon [2].The degrading coastal ecosystems has become an urgent matter.Protecting and restoring the stability of coastal habitats can reduce the impact of global warming and climate change as well as increase the potential of blue carbon.Blue carbon is carbon that is sequestered and stored by the carbon sink in the ocean or coastal ecosystems.There are seagrasses, mangroves, algae, and other coastal habitats that act as carbon sinks, which could store so much carbon [1].The seagrass ecosystem is natural objects that can be used to adapt and mitigate climate change through blue carbon sequestration [3][4][5].Seagrass ecosystems have a lot of potential blue carbon stock in Indonesia's coastal ecosystem.Based on [6], seagrass can absorb and store about 18 percent of blue carbon in the world.There are 16 seagrass species in Indonesia that spread along Indonesia's coast [7].The high diversity of these species supports the high potential of carbon that can be absorbed and stored.Carbon that is absorbed and stored in seagrass can be obtained by mapping the aboveground carbon stock (AGC) using remote sensing [5,8] and seagrass percent cover data [9][10][11].Biophysical identification of seagrass such as seagrass percent cover can be done using remote sensing images.Remote sensing has the ability to map vast areas with rapid time, also one of the non-destructive approaches to mapping biophysics objects [5,12].According to [13] seagrass percent 1291 (2024) 012013 IOP Publishing doi:10.1088/1755-1315/1291/1/012013 2 cover has a good correlation with carbon stock, with higher coverage will have higher carbon uptake.Combination of laboratory analysis and remote sensing-based mapping could be useful for developing an equation to measure the seagrass AGC.One of the relatively new remote sensing products from PlanetScope is PlanetScope SuperDove with 8-bands, covering spectral bands of coastal blue, blue, green I, green, yellow, red, red edge, and near infrared.This study aimed to develop a general equation to estimate seagrass aboveground carbon stock from the percent cover and to map seagrass AGC using PlanetScope imagery in Nemberala, Rote Island.

Study area
We carried out the seagrass aboveground carbon stock mapping in Nemberala, Rote, East Nusa Tenggara.Nemberala is located in an island in south-eastern Indonesia with a wide white-sand beach.The tides in Nemberala have a dynamic pattern because at low tide sometimes all the seagrasses come up and people can walk along the coast easily.Besides seagrass, Nemberala also has a lot of green algae farm that lives along with the seagrass.According to PlanetScope and other satellite images, seagrasses in Nemberala spread out approximately 5 km from north to south and 700-meter from the shoreline until the breaker zone.The seagrass meadow grows in clear water.Nemberala has very diverse seagrass species such as Enhalus acoroides (Ea), Cymodocea rotundata (Cr), Halophila ovalis (Ho), Syringodium isoetifolium (Si), Thalassodendron ciliatum (Tc), and Thalassia hemprichii (Th).(Th).Each species was measured by shoot density and percent cover using the photo-quadrant 1x1 meter method.Based on [14] the size of the photo-quadrant plot is close to the pixel size of the imagery or larger than that, for PlanetScope with a 3-meter resolution 1 m 2 plot is still effective.Calculation of AGC with gram carbon per leaf (gC/leaf) was calculated using Loss on Ignition (LOI) methods according to [4,15].
There were 78 samples obtained using random sampling considering the seagrass percent cover in the field, so that all of the range percent cover classes can be obtained.All the samples were calculated to extract the percent cover value and shoot density (by converting to leaf count).Every single leaf in each quadrant was multiplied by the result of laboratory AGC (gC/leaf) on each species to find the AGC value in each quadrant (gC/m 2 ).All AGC values on each species were added to calculate the total AGC in each quadrant to make the general equation.The general equation was calculated using regression based on the percent cover value and AGC value from a laboratory in each quadrant.This formula will be applied to convert seagrass PC data to AGC on PlanetScope SuperDove 8 bands image.

Seagrass photo-transect analysis.
The photo-transect method was conducted on August 2021 to obtain PC data samples to model the AGC map in Nemberala.These samples were divided into two categories, the ones for modeling (65%) and validation (35%) respectively.PC data from photo-transect was calculated using Coral Point Count with Excel extension (CPCe) to map the seagrass PC [16].Due to other classes existing in each photo-transect sample, only seagrasses with more than 25% coverage were classed as seagrass.

Image processing
We used PlanetScope Analytical Surface Reflectance with a SuperDove sensor which has 16-bit radiometric quality and eight bands, recorded on 27 August 2021.This imagery has already been orthorectified and radiometrically corrected, so that the pixel values show the real reflectance values of the objects [17].PlanetScope with SuperDove sensor was relatively new, which started producing images in mid-march 2020 [17], and there were not much studies about seagrass have used this imagery.AGC estimation was applied to all the single bands of coastal blue, blue, green I, green, yellow, red, red edge, and near infrared.The purpose of this study was to find which band has a correlation with AGC, especially in seagrass meadow like in Nemberala Beach.Land and deep water were masked due to the focus on optically shallow water because of the presence of seagrass.We applied multispectral classification to differentiate benthic habitats and mask out other classes.By using percent cover data from photo-transect, we used machine learning-based random forest classification with four classes (seagrass, coral, algae, and bare substrate).

AGC mapping
Estimation of AGC was carried out using statistical analysis and stepwise regression with the dependent variable (y) being field data AGC and the independent variable (x) being the pixel values of PlanetScope.Statistical analyses in terms of normality test and Pearson correlation test were used to prove the relationship between variables.The regression equation can be used to estimate the AGC by considering the correlation coefficient (r) and determinant coefficient (R 2 ).An accuracy test was carried out on the result of AGC mapping on the PlanetScope image based on the AGC value from the previous equation.The method used is a Standard Error of Estimate (SEE) to see the error value, and accuracy, also an accuracy test using a 1:1 plot to see the distribution of the AGC value from the model based on reference AGC.

AGC seagrass estimation from percent cover
Harvested seagrass from the field survey was dried to measure dry biomass.Loss on ignition method to measure aboveground carbon stock using wet biomass (seagrass weight before drying) and dry biomass.Drying each species was done using an oven at 60°C for approximately 48 hours [18].Wet biomass, dry biomass, and AGC were calculated per leaf based on the leaf average in the shoot density of each species.Based on Table 1, Ea has the highest AGC with 0.0159 gC/leaf and Ho has the lowest AGC with 0.0001 gC/leaf.Leaf structure and size cause a big difference between each species to store carbon [19,20].By developing a general equation for AGC from the percent cover, we conducted a normality test, correlation test, and regression analysis.The normality test using Kolmogorov-Smirnov shows that the AGC value is significant with 0.2 being more than 0.05.For the correlation test, we use Pearson correlation with an r value is 0.61 which shows a significant correlation between AGC and seagrass percent cover.Then, for the regression, we use linear stepwise regression to find the best equation to predict seagrass AGC using percent cover.The result shows SeagrassAGC = (0.051*SeagrassPC) -0.635 with R 2 of 0.36.This shows that 36% of AGC can be described using percent cover without considering the species and the other was because of other factors.

Mapping seagrass AGC based on PlanetScope image
Based on the developed equation, percent cover data from the photo-transect analysis was calculated to measure aboveground carbon stock value.The range of seagrass percent cover in the modeling samples is 27%-100% resulting in the range AGC value of 0.762 gC/m 2 to 4.465 gC/m 2 ; while the validation samples with 29%-100% percent cover have the range of AGC from 0.852 gC/m 2 to 4.465 gC/m 2 .All modeling samples were tested for normality and correlation.It was found that normality test using Kolmogorov-Smirnov resulting in the AGC samples for modeling is abnormal.Therefore, we perform bootstrapping for the correlation test due to abnormality data and the large data.Bootstrap correlation does not consider the distribution of the data, hence we do bootstrap to estimate the confidence for the correlation test.The correlation test between the AGC sample and PlanetScope single-band pixel value was conducted using Pearson-Correlation resulting in all the samples have significant correlation as shown in Table 2.The blue band on PlanetScope has a wavelength of 465 -515 nm.Based on the spectral reference of seagrass, due to the low tides on 27 August 2021, seagrass has low reflectance in the blue band.Thus, there is a huge difference in reflectance values between seagrass with water and bare substrate.This would probably make the blue band has the highest correlation.The blue band has a quite strong penetration through the water, so with the low tides, blue bands could differentiate the water from seagrass.Based on the results of the seagrass AGC map in Nemberala (as shown in Figure 2) shows the highest value, i.e. 4.116 gC/m 2 and the lowest value, i.e. 0.143 gC/m 2 with the average of 2.81 gC/m 2 .For the entire study area, the total aboveground carbon stock in Nemberala according to the AGC modeling is 0.98 tC, covering an area of 3.13 km 2 .The high AGC is mostly found in the northern part of Nemberala near the coastline.Tides and substrate give influence on the distribution of the seagrass.On the other hand, near the breaker zone, coral and dead coral algae are predominant.

Accuracy assessment
Based on the accuracy assessment using SEE for the AGC model and AGC reference, the SE value is 0.914 gC/m 2 , which means that there is approximately a 0.914 gC possibility error in every square meter.
The maximum accuracy that can be obtained is 71.93%.According to [19] also mapped AGC at the community level in Karimunjawa Islands using PlanetScope with an accuracy of 58.79%.Also, [20] have already mapped AGC in Parang Island with the highest accuracy of 66.9%.Beside that [15] also used PlanetScope to map seagrass AGC in Nirwana beach with the highest accuracy at 59.98%.The previous studies show similar accuracy results, where the difference is in the standard errors.Error at 0.914 gC is quite high because if we look at the distribution of the data, some of the validation samples have a huge difference as compared to the reference AGC data.From the 1:1 plot (figure 3) we can see the pattern of the value along the 1:1 line.The seagrass AGC model tends to be underestimated.The AGC model with a value of more than 3.5 gC/leaf has been saturated due to the high energy uptake by the high-density seagrass.The errors were also caused by the low correlation between AGC from the reference with the pixel value of PlanetScope.Calculated AGCs in the laboratory based on the correlation with seagrass percent cover show quite good results, with more seagrass in each quadrant more AGC will present.But when it is applied to PlanetScope imagery, some of the pixels do not represent the real percent cover of seagrass, thus the AGC would have an anomaly.This was caused by the instability of the number of sample points that represent PC in every pixel from the photo-transect analysis or errors in GPS coordinates.According to [15] PlanetScope has a 3-meter resolution, which differs from the sampling plot for photo-quadrant or photo transect.Consequently, in every pixel there will be a different value of samples, which can be more than one value for one pixel.Also, we used handheld GPS to obtain seagrass coordinates in the field.Today's GPS accuracy could affect the accuracy of the position of the samples, which then might have an impact on the pixel location.The handheld GPS has a relatively low accuracy up to 5 meters, as compared to that of PlanetScope (3 meters).This could be one of the reasons for the high error in this research.Some of the samples were also located near bare substrate or dead coral, which makes the reflectance of the pixels have been mixed up, causing the error of AGC is relatively high.According to [21], the mixed pixel values of seagrass and sand are quite difficult to be used as a basis for estimating the AGC values.Moreover, the general equation was developed by converting all the seagrass species into single equation, which might lead to error when it is applied to percent cover data.Each species has a different number of leaves, leaf size, and leaf structure which might also affect the AGC value [2].

Conclusion
Estimation and mapping of seagrass aboveground carbon stocks can use seagrass percent cover data.With the r value of 0.61 and R 2 = 0.36, the general equation is SeagrassAGC = (0.051*SeagrassPC) -0.635.This general equation could be applied to imagery with percent cover data to map aboveground carbon stock.However, this equation is still limited to the coastal area condition like in Nemberala with the variation of seagrass Cr, Ea, Ho, Si, Tc, and Th.The accuracy of the seagrass AGC map in PlanetScope SuperDove imagery has maximum accuracy of 71.93%.The error might occur due to the different structures of seagrass in each species, instability of samples in every pixel, as well as coordinate error from the GPS.With the 3.13 km 2 area of seagrass in Nemberala, the total AGC is 0.98 tC.For a better result, the mapping of AGC should consider the distribution of species, because the different structures of leaves have variations of AGC per leaf.In addition, the size of the plot sample should consider the pixel size of imagery, because it relates to the value that will represent by each pixel.The purity value in each pixel could give a better result of percent cover that is also applied in AGC.

3 )Figure 2 .
Figure 2. Map of the Seagrass AGC in the study area.

Table 1 .
The results of laboratory analysis of seagrass aboveground carbon stock per leaf.

Table 2 .
Correlation between AGC value and Planetscope Superdove single band.