Building Land Cover Objects Following the IPCC Guidelines for Carbon Emission Estimation. Case Study in the Central Highlands of Vietnam

We estimated the carbon emissions in the field of Land Use, Land Use Change, and Forestry (LULUCF) by using advanced technology to build the input data. Remote sensing, including satellite remote sensing and Unmanned Aerial Vehicles (UAV) with transparency, multi-time, and wide coverage characteristics, is useful in this area. The article focuses on the proposed regulations and building process of subjects in the land cover of the Vietnamese mainland following the guidance of the Intergovernmental Panel on Climate Change as applied to carbon emission estimation. We propose a process to estimate carbon emissions using the Agriculture and Land Use Greenhouse Gas Inventory software with input data extracted from the remote sensing images. An experiment on land cover change was carried out over ten years between 2006 and 2016 in the Central Highlands of Vietnam. The results obtained with remote sensing data classification for land cover categories achieved a reliability of 69% for the year 2006 and 66% for the year 2016. The carbon emission estimation data were checked and used in Vietnam’s biennial update report to the United Nations Framework Convention on Climate Change, including content and updated information on the greenhouse gas inventory.


Introduction
In the past, Greenhouse Gas (GHG) emission estimation in Vietnam often used local statistics, calculated with Excel software. Currently, the United Nations Framework Convention on Climate Change (UNFCCC, 2012) recommended that countries should apply advanced technology to the built land use, land use change, and forestry dataset for carbon emission estimation; specifically, remote sensing technology was encouraged for use with data transparency continuity to make a basis for comparison between calculation years (IPCC, 2003;IPCC, 2006). In addition, the Agriculture and Land Use Greenhouse Gas Inventory Software (ALU), which was provided by Intergovernmental Panel on Climate Change (IPCC) is the software that has been applied in Vietnam. The ALU software is a dedicated software with the function of Quality Control (QC) and Quality Assurance (QA) (ALU, 2014) in regulation with the IPCC guidelines. IPCC (2003IPCC ( , 2006 provided the guidance for the method's use with three "Tiers", which could provide the tally results from the minimum to maximum degree of the uncertainty. The specific contents were as follows: TIER 1: Calculated data and emission coefficients are taken from globally announced data sources, such as the Food and Agriculture Organization of the United Nations (FAO) reports and websites.
TIER 2: This tier uses the same methods and formulas as TIER 1, but the calculation data and emission coefficients are taken from a national data source. TIER 3: The higher grade method used in TIER 3 includes improved inventory models and systems to focus on specific cases in each country repeated over time and is supported by high-resolution spatial data that is detailed at a provincial grade or ecological region. This tier gives an estimated result with higher certainty compared to TIER 1 and TIER 2.
Recently, Vietnam has been carrying out monitoring of the GHG emissions and absorption in the process of land use, land use change, and forestry planning using remote sensing materials. In the process of applying remote sensing technology, it is necessary to refer to the IPCC technical regulations for the input dataset in the field of LULUCF to the GHG emission estimations under Vietnamese natural conditions. The implementation of national GHG inventory must comply with the IPCC guidelines based on the Guidelines for National Greenhouse Gas Inventories, 1996Revised (IPCC, 1996 and Good Practice Guidance for Managing Uncertainty in GHG inventories (GPG, 2000). Depending on the availability of the input data, each country can choose a different approach, closely related to the level of increased complexity in the data accuracy requirements.
The emission coefficients used during the inventory process were the coefficients proposed by the IPCC and could be applied to many territories with the same climate zones. Countries often use national data sources for spatial data, default emission coefficients, and GHG emissions following the IPCC guidelines or FAO database (IPCC, 2003). The coefficients of stock change and GHG emissions were applied to specific regional data. We chose to use spatial and time data corresponding to the national coefficients with higher resolutions and greater detail. These data were identified for each specific region and spatial land cover system.
According to the IPCC guidelines, the total area data of land use should be equal to the country area.
Depending on the availability of the input data, each country determines the subjects in the land cover based on their natural conditions. In Vietnam, the dataset in the field of LULUCF is a data system collected though land statistical progress, such as forestry, agriculture, total land inventory, published research results, national statistical data, or extracted from remote sensing data. These data were combined into the LULUFC dataset by area classification, including forest land, cropland, grassland, wetlands, residential land (residential and infrastructure), and other land. A classification system of land cover objects was applied to the whole country as a land use status map classification system. Land cover was established To build the input dataset in the field of LULUCF for GHG emission estimations based on the classification system of land use objects in MONRE (2018), the combination of classes necessary to determine the six types (categories) of land cover objects (forestland, grassland, cropland, wetland, residential land, and other land) needed for the IPCC guidelines is proposed in Table 3. For example, for investigation, collecting, and classification, forestland includes evergreen broad-leaved forests, deciduous forests, planted forests, mangroves, and other forests. Cropland includes annual cropland, perennial cropland, and wet rice.
In the process of building a land cover object dataset, the technical requirements for those objects must comply with the current legal documents on the technical regulations for the production of optical sensing images with high and super high resolution, using the reference system and national coordinate system of Vietnam (VNPM, 2000), as well as the converted parameter system between the international coordinate system WGS-84 and the national coordinate system VN-2000(MONRE, 2007.
The article proposes regulations regarding the process for subjects in land cover for the Vietnamese mainland following IPCC guidelines; experimented carbon emission estimation in the Central Highlands region by using the ALU software with the input data extracted from the remote sensing images.

The Subjects in Land Cover in the Field of LULUCF in Vietnam Natural Condition
Based on IPCC guidelines (IPCC, 2003;IPCC, 2006) and MONRE (2018), the land cover in the field of LULUCF in Vietnam natural condition as follows: Input remote sensing image data must be preliminary assessed for cloud cover. Cloud cover assessment is divided into levels and denoted by letters: -Level A: Remote sensing image with cloud coverage under 10%; -Level B: Remote sensing image with cloud coverage from 10-25%; -Level C: Remote sensing image with cloud coverage of 25% or more.
In this study, only remote sensing images with level A or B were used.
The map must be established following the Vietnam regulation of MONRE (2000) The content of the land cover status or change map is divided into seven layers, including the math base layer, terrain layer, traffic layer, hydro system layer, and administrative boundary; each class is divided into subjects. Each layer can consist of one or several objects with the same properties, each object is marked with a unique and uniform code on the map.

The Process to Estimate Carbon Emissions Based on Land Cover Information Using Remote
Sensing Data Based on the the above basises, the process to estimate carbon emission based on land cover information using remote sensing data is expressed in Figure 1 http://www.scholink.org/ojs/index.php/se  b) Imagery processing and classification: First, we processed the remote sensing image data to create a homogeneous image, eliminating the clouded effect on data quality and creating color composite images for the study area, while using a random classification method to classify the images (Li et al., 2014) with a classification key that offers a suitable classification sample set. Then, we evaluated the accuracy and reliability of the classification results. The data layers were classified by type, with the categories following IPCC guidelines and natural object classification regulations for Vietnam (MONRE, 2018).
We assessed the accuracy of the object classification using the Kappa Khat error matrix method (Congalton, 1991). The first step in the accreditation process was to identify high-resolution image areas on Google Earth twice. Checkpoints were randomly generated in the ArcGIS software and then loaded into buffering areas with a size of 2 ha. These areas were converted into the KML file format and loaded into Google Earth. Through visual image analysis, the land cover properties were assigned to random checkpoints. The number of these checkpoints was then verified to again attribute them to topographic mapping data at the same time. The checking process was conducted for the whole study area out with the number of sampling points for each result. d) Land cover status/change: After inputting the classification results, an overlay dataset was built for two points at the same place to assess the area accuracy and consolidation according to the required ratio. In the process of building a land cover objects' dataset, the technical requirements for those objects must comply with the current legal documents on the technical regulations for the production of optical sensing images with high and super high resolution, using the reference system and national coordinate system of Vietnam (MONRE, 2015), as well as the converted parameter system between the international coordinate system WGS-84 (VNPM, 2000) and the national coordinate system VN-2000(MONRE, 2007. The next step was to import and overlay data for the two periods and calculate and determine the specific changes. g) Integrating, processing and synthesizing data/ export data for the ALU software. Land cover data serving for carbon emission estimation using the ALU software, combining the soil and climate/ecological zoning data. The work included converting data to the same data format and correlating the spatial relationships between geographic subjects.
h) Carbon emission estimation using the ALU software.
We performed emission estimation in the ALU software to check the data accuracy in the input data table related to the data entry. The next step was to run the software and export the emission results.

a) Study area
The Central Highlands of Vietnam is not a single plateau but a series of adjacent and stratified plateaus.
The Central Highlands is located in the range from 11 o 17' to 15º26' North latitude, from 107 o 19' to 108º54' East longitude. The Central Highlands has a special geographical position with an altitude from 250 to 2,500 m. It is upstream of four large river systems. This plateau is located near the middle of Vietnam with a radius of equal distance from Southeast Asian countries not exceeding 2,000 km.  The map material was a topographic map at the scale of 1: 100,000 or 1: 50,000 in the VN-2000 coordinate system. This was used as the base map serving for establishing the land cover status map from the remote sensing images.

Results
a) Image processing and land cover results in the Central Highlands.
From the imagery processing result at the time of 2006, the obtained land cover status with six criteria is presented in Figure 3.

b) Carbon emission estimation results.
Carbon emissions and absorption in this field included the carbon stock change in the five carbon tanks (aboveground biomass, belowground biomass, dead tree, falling object and soil, and soil), the above and below ground biomass were generally called "fresh biomass", and dead wood and falling subjects were generally called "litter".

Discussion
Different ground objects, including forestland, grassland, cropland, wetland, residential land, and other land, have different spectral characteristics (Yu, 2017). The total carbon emissions always followed the overall world trend (Olivier, 2017). In this study, we focused on the proposed categories of land cover based on the IPCC guidelines applied in the Vietnam natural condition to illustrate the results of Vietnam's land cover and carbon emission calculations in the Central Highlands of Vietnam, which was estimated with the ALU software using remote sensing data. As described in the Results Section, with the ground objects' characteristics of an afforestation effort in Vietnam, the land cover classification was conducted using remote sensing imagery and the carbon emissions/absorption following each specific period.
Using remote sensing data for estimating carbon emissions is useful (Lu et al. 2013), and satellite-based measurements of greenhouse gases have been facilitating an effective method of monitoring atmospheric constituents with the development of highly accurate sensors. This is also becoming the major data source to detect changes in the atmospheric CO 2 concentration at regional and global scales (Yoshida et al., 2011;Crisp et al., 2015;Lei et al., 2014;Buchwitz et al., 2015).
Statistical analysis showed that the results achieved a reliability of 69% for the year 2006 and 66% for the year 2016, which shows that this method obtained a relatively high accuracy. The classification accuracy of 2016 shows that the classification result achieved a lower accuracy compared to the year 2006. The evaluated results for the classification accuracy showed the percentage of errors in the classification process that were distributed over most layers; however, the largest error percentages fell into the groups of forestland and grassland. The cause of this confusion comes from the fact that these two objects have large spectral homogeneity for their subcategories, which created two categories; whereas residential land and vacant land belong to other land types (Yifan et al., 2018).
The accuracy assessment result showed the fact that grassland is interspersed with forestland was also a cause of confusion during classification by spectral similarity. The mixed population with vegetation is also a cause of the confusion when classifying, as the source of this error is difficult to avoid even with the traditional method (interpretation by eye). The homogeneity in the spectrum was the main cause of Many countries showed a decrease in CO 2 emissions in recent years, most notably the United States (-2.0%), the Russian Federation (-2.1%), Brazil (-6.1%), China (-0.3%), and within the European Union (-6.4%). In contrast, the largest absolute increases were seen in India (+4.7%) and Indonesia (+6.4%) with smaller increases in Malaysia, the Philippines, Turkey, and Ukraine. For many of the largest emitting countries, this is a continuation of a common trend. With an estimated 0.2% increase in CO 2 emissions, emissions in the European Union remained more or less the same in 2016 (Olivier and Peters, 2018). In contrast to most of the main emitters, the collective emissions from the rest of the world show a rising trend.

Conclusions
Remote sensing technology has become a useful tool in greenhouse gas emission estimation due to its transparency, accuracy, multi-time qualities, and wide coverage. To create an inventory of greenhouse gas and carbon stock changes over time, the results cannot be directly calculated from the remote sensing data but must be used in combination with software, such as the ALU software, to estimate the carbon emissions.
The results showed that the total amount of negative greenhouse gas emissions (absorption) in the field of LULUCF was 19.13 million tons of CO 2 . In particular, forestland, cropland, and grassland absorbed greenhouse gases with an uptake volume of 17,36 million tons of CO 2 , 2,31 million tons of CO 2 , and 0.77 thousand tons of CO 2 , respectively, whereas the greenhouse gas emissions for the remaining land types were as follows: wetland, with 65,79 thousand tons of CO 2 ; residential land, with 273,09 thousand tons of CO 2 ; and other land, with 201,25 thousand tons of CO 2 .
The development of technical regulation will help the State management units gradually improve the legal document system, implementing the state management function to be easier and more convenient.
The proposal and experimentation of the carbon emission calculation process in the field of LULUCF showed that land cover information extracted from remote sensing images reached high accuracy and effectively determined the object variation.
Remote sensing technology application is also of use for Vietnam to monitor the GHG emissions and absorption in the field of LULUCF. In the long term, highly accurate GHG inventory data and information sources help to determine the amount of carbon emissions and absorption in the field of LULUCF to help manage the quantitative emissions participating in the carbon market in the future.