Carbon stock estimation based on remote sensing in the northern coast of Java

Mangroves are one of the land covers on the earth’s surface that being the largest storage of carbon reserves compared to other land covers. On the other hand, quick and accurate monitoring of carbon stocks on the earth’s surface is needed. This research was conducted on the northern coast of Java and used a flexible and efficient cloud computing-based remote sensing approach by using satellite imagery data. We identify land cover classification, especially mangrove, uses the Support Vector Machine (SVM) algorithm through the GEE (Google Earth Engine) platform. The estimated value of mangrove carbon was obtained from the NDVI index (Normalized Difference Vegetation Index) analysis on sentinel-2 images. The results showed that the estimated carbon value was 1,232,311.496 tones. Strong relationship is found between NDVI and carbon stocks with R2 of 98%. The study, therefore, strongly suggests the further use of NDVI to assess and monitor carbon stocks from mangroves in the future.


Introduction
As the largest carbon sink, forests play an important role in the global carbon cycle.One forest ecosystem that has an influence on carbon sequestration is the mangrove forest.Mangrove ecosystems are ecosystems that grow along the coastline and are influenced by tides.This ecosystem is also the most carbon-rich ecosystem in the equatorial region because it has a potential carbon stock of 1,023 Mg of carbon per hectare and is stored in soil containing a lot of organic matter at a depth of 0.5 m to 3 m [1].In research conducted by Murdiyarso et al. [2] stated that mangroves can store carbon ten times greater than other common ecosystem plants.The amount of carbon storage in mangroves is influenced by differences in vegetation shape, density, type, and tides [3].
The rapid development of technology makes people have many alternatives to do work quickly and solutively [4].One of these technological developments is mangrove mapping using remote sensing in the form of Google Earth Engine (GEE).Mangrove mapping collects data to observe the actual and efficient condition of mangrove forests [5].Meanwhile, according to Farizki and Anurogo [6], remote sensing is a technology that can provide information in the form of fast and accurate data sets.Mapping the extent of mangrove forests by utilizing remote sensing technology can provide information or an overview in the form of a map of the distribution of mangrove forest areas in an area [7].Google Earth Engine (GEE) is a web-based platform that is accessed online based on cloud computing and can run GIS IOP Publishing doi:10.1088/1755-1315/1315/1/012042 2 processes in the form of scripts that are made and can produce analysis according to the desired search [8].The use of this technology makes it possible to temporally analyze location data through image capture.One of the widely used images is Sentinel-2.GEE provides access to Sentinel-2 imagery that can be used for various applications, including vegetation monitoring with good enough resolution.
The North Coast of Java (Pantura) has beautiful beaches as well as being an ecosystem for many types of mangroves in Indonesia and a high store of blue carbon reserves.Mangrove rehabilitation efforts in the North Coast of Java are often carried out every year to support the needs of urban forests, making it necessary to have spatial-based distribution data at a detailed scale.Moreover, mapping involving the GEE platform is still quite small in the field of research, especially in Indonesia.This research, therefore, certainly provides an in-depth understanding of the application of the platform.Rsearch on mangrove distribution mapping is important to do to get an accurate distribution area with the latest cloud computing-based methods through the GEE platform, as well as accuracy assessment based on Support Vector Machine (SVM) values.

Time and Place of Research
This study was conducted in the northern coastal area of Banten Province covering three regencies or cities namely Serang City, Serang Regency, and Tangerang Regency.These three locations were chosen by considering the distribution of mangroves and damage due to human intervention such as deforestation or other land use conversion.

Data Sources
In this mangrove carbon stock research, estimation involves remote sensing data.The remote sensing data used is Sentinel-2 Multispectral instrument (MSI) satellite image data and is obtained on a computing platform.Sentinel-2 MSI imagery is a spectral-based earth observation satellite image utilizing wavelength emission from the earth's surface to space.This image has advantages when compared to various other spectral-based satellite images.One of them is with a spatial resolution of 10 meters to 60 meters, thus increasing the detail of the classification results.

Analysis of Mangrove Detection
Machine learning (ML) is a computer system capable of learning and adapting without relying on explicit instructions have been advancing rapidly.These systems employ algorithms and statistical models to analyze data patterns and draw inferences, enabling them to evolve and improve their performance autonomously.Machine learning models are used to estimate the aboveground biomass of different species of mangroves in the targeted area.The models that were used in these methods are Google Earth Engine (GEE) cloud computing platform and RStudio (R Core Team).Support vector machine (SVM) algorithm is also used in the method to analyze data for classification.

.4 Accuracy Test for Machine Learning Performance
Land classification analysis generally produces information that does not match field conditions.This occurs due to the capabilities of the classification algorithms involved and the availability of remote sensing data.One of them is the presence of cloud cover which affects the index value.This research uses accuracy analysis methods, namely Overall Accuracy (OA) and Kappa Statistics (KS).In addition, OA and KS calculations also involve the PA (Producer Accuracy) and UA (User Accuracy) formula data approaches.This accuracy analysis involves validation data obtained randomly using high-resolution image data in Google Earth pro software [20].This validation data is separate from the training data in the classification process.This considers the intervention of validation data in the classification process.The accuracy analysis formula involved is presented in the following.

Analysis of Carbon Stock
Carbon stock analysis was conducted in the study using an allometric approach.This was done to estimate the existing carbon stock potential in the three regions.The estimation variable in allometrics uses the vegetation index variable, NDVI.This index was chosen by considering the ability obtained in the process of estimating carbon stocks in mangrove forests.The NDVI value obtained was entered 788.434302 with a quadratic model with the equation Y = 574.05((NDVI) 2) -17.24 [21].According to IPCC [22], the carbon contained in organic matter is 47%, and therefore, the estimated amount of stored carbon can be known by multiplying biomass and a multiplying factor of 47% or 0.47.The formula is as follows: Cn (ton/ha) = Biomassa (ton/ha) x 0,47

Results and Discussion
The mangrove cover classification process in this study resulted in mangrove distribution covering the entire administrative area.The SVM algorithm successfully detected 788.434 ha of mangroves (Figure 2 and Table 3).This mangrove area has aquaculture ponds managed by residents as the main livelihood.
Support Vector Machine (SVM) classification is a classification system that works by separating an object from other objects using a kernel that works based on euclidian distance in a multi-dimensional feature space [23].According to the ENVI User's Guide in 2008 states that, SVM classification results are also able to present relatively satisfactory data, especially in complex remote sensing and there are many disturbances.Research conducted by Khaira et al. 2016 shows that SVM technology successfully classifies land cover with dense vegetation classes, sparse vegetation, to non-vegetation.SVM classification can show better accuracy test results compared to other arenas that can minimize errors in image interpretation [24].Similar method on the SVM usage was also used in Wikantika and Darmawan [25] research where identification of mangrove areas is done through a classification process using NDVI values as a reference.In selecting the best guided classification method, they used the 1993 Landsat 5 TM image of Subang Regency.Classification is done by taking training samples based on NDVI values.Classification is divided into 4 classes, namely mangrove, nonmangrove vegetation, settlements and waters NDVI has the capability to observe the well-being of expansive forested regions by identifying alterations in plant coverage, like shifts in the extent of tree canopies, the index of leaf area, and the overall biomass.It can also identify modifications in the composition of forests, such as variations in tree density, as well as pinpoint zones where new forest growth is taking place [26].Among the various vegetative indices, NDVI stands out as the optimal and widely employed choice for investigating carbon dynamics.The landscape-level approach (utilizing NDVI) is a technique for estimating carbon stocks through the utilization of satellite imagery.NDVI serves as an indicator of vegetation quantity and vitality on the earth's surface, and composite NDVI images are constructed to facilitate the differentiation between flourishing green vegetation and exposed soil.Scantier vegetation such as shrubs and grasslands, or maturing crops, can lead to moderately ranged NDVI values (approximately 0.2 to 0.5).Elevated NDVI values (approximately 0.6 to 0.9) correspond to dense vegetation, such as that found in temperate and tropical forests, or crops during their peak growth phase [27].
This study revealed the strong correlation between NDVI and carbon stock with R 2 of 98%.The R² value of 0.98 indicates that the equation can describe 98% of the relationship between the spectral value of the vegetation index and carbon.The greater the R² value, the better the correlation between NDVI index and carbon.The fitting lines for the simple linear regression can be seen in Figure 3.The results of Syahrial and Novita's research (2018) showed that mangrove species found in Serang Regency were 5 species namely Rhizophora apiculata, R. stylosa, R. mucronata, Bruguiera gymnorrhiza, Sonneratia caseolaris and Lumnitzera racemosa.Rhizophora sp. has a high carbon sequestration rate of 398.6 tons CO 2 /ha.Rhizophora sp. has a very important role in carbon sequestration potential when compared to other mangrove species due to the magnitude of the potential uptake and carbon content produced.Other mangrove species, namely B. cylindrica, A. officinalis, Xylocarpus sp. have carbon sequestration rates of 212.24; 34.83; and 12.70 tons of CO 2 /ha, respectively [28].Mangrove forests are capable of storing carbon three times the average carbon storage per hectare of land tropical forests [29].The optimal function of carbon absorption by mangroves is up to 77.9%, the carbon absorbed is stored in mangrove biomass such as stems, leaves and sediment [1].The average global carbon stock in mangrove ecosystems is estimated at 956 Mg C ha−1, which is much higher than tropical rainforests, peat swamps, salt marshes and seagrass beds (Bachmid et al. 2018), strongly show the importance of mangrove preservation to maintain this high global carbon stock.

Conclusions
This study revealed that estimated carbon stock from mangroves in the northern coast of Java obtained from the NDVI index on sentinel-2 images was 1,232,311.496tones.High correlation with R 2 of 98% is found between NDVI and carbon stock, strongly suggest the further use of NDVI for quick and accurate monitoring of carbon stocks from mangroves in the future.

Figure 3 .
Figure 3. Correlation between NDVI value and carbon stock

Table 1
List of the indexes used in this study

Table 3 .
Carbon distribution analysis