3. MEASUREMENT

The Measurement module starts from the registration of the rural property or possession in the Moss Forest system through the link https://bit.ly/Moss-Forest and then takes the following flow:

To this end, Moss Forest uses ten sources of data obtained from geographic databases developed over the years by members of different sectors of the economy and based on robust methodologies, whether for the establishment of public policies or for project development (Table 2). It is important to highlight that currently the Moss Forest system currently only applies to Brazil and REDD+ projects, but will be expanded to other countries in the future.

The use of these data from official sources ensures compatibility with national and global efforts for land, environmental and forest monitoring regularization. Also, the data providers are widely recognized by the scientific community and the international community related to the AFOLU sector, providing the robustness, confidence and transparency required for Moss Forest.

The centralization of all this geographic information within a single digital tool, through the automation of the activities of querying, obtaining, treatment, processing and systematization of data, improves the performance of teams of experts in the development of REDD+ projects; as well as facilitating data observation by investors and auditors.

For projects in the AFOLU sector, in particular, the ex-ante², Carbon measurement and the consolidation of this data can be one of the most costly and lengthy steps. The IPCC Good Practices Guide outlines three levels of complexity for obtaining data and estimating ex-ante reductions: tier 1, which uses default values proposed by the IPCC; tier 2, which uses national values with respective estimates; and tier 3, which uses more elaborate methods such as modeling and which must be compatible with the other tiers (Figure 2).

In this way, the Moss Forest system for Brazil:

• Centralizes Tier 2 and Tier 3 geographic and official data;

• Enables the evaluation of the additionality of the REDD+ intervention;

• Adapts the analytical scale of the territory of interest;

• Applies the ex-ante phase to the implementation.

3.1. Land Evaluation

After receiving the rural owner's registration information and the property's official documents, the total area of the rural property is calculated, and the area's level of land regularity is evaluated using the (Brazilian) SICAR and SIGEF databases.

3.1.1. National Rural Environmental Registry System (Sicar)

Brazil’s “National Rural Environmental Registry System” (Sicar - https://www.car.gov.br/publico/imoveis/index) is an electronic public registry with national scope and mandatory for all rural properties in Brazil. It is extremely important when it comes to understanding the Brazilian territorial, land and environmental dynamics. For registration, identification is required by means of a plan and descriptive memorial containing the geographic coordinates with at least one mooring point on the perimeter of the property and the location of areas of relevance, such as remnants of native vegetation, of Permanent Preservation Areas, the Restricted Use Areas, the consolidated areas and the location of the Legal Reserve (Art. 29 of the Forest Code).

By routinely querying the Sicar database, Moss Forest integrates environmental information from rural properties and possessions, establishing a database for control, monitoring, environmental planning and combating deforestation. In this way, the query of the rural property's polygon area and all its sets of relevant information on the area of interest for the REDD+ project is automated, reducing the human effort of operationalizing Sicar, extracting geographic data and systematizing registration information (Figure 3).

3.1.2. Land Management System - SIGEF

Using the CPF (personal tax ID, or Brazilian social security number) of the owner of the rural property, based on the information contained in the property's registration number, it is possible to consult the system developed by the “National Institute of Colonization and Agrarian Reform” (INCRA) and the “Ministry of Agrarian Development” (MDA) for managing land information of the Brazilian rural environment, SIGEF (sigef.incra.gov.br/ - Figure 4). This system carries out the reception, validation, organization, regularization and provision of georeferenced information on rural property boundaries, assisting in land governance throughout the national territory. Over 889,000 plots of rural properties have been certified since the system was launched in November 2013 - gov.br/agricultura/pt-br/assuntos/noticias-2022/nova-funcionalidade-do-sigef-per mite-desmembramento-automatizado-de-parcelas.

When crossing the data of the boundary of the rural property registered in Moss Forest with the SIGEF certified information, Moss Forest automatically identifies whether there is any spatial overlap with other properties or protected areas. This consultation is done on a daily basis and includes the following steps:

• evaluation of whether the property is on the private and non-public property database;

• evaluation of the georeferencing status and validation by the system;

• download of the documents of the georeferenced area (plan / map and descriptive memorandum);

• status of applications for certification, registration, splitting, reassembly, rectification and cancellation.

3.2. Estimation of Carbon Stock

This module allows for an ex-ante estimation of carbon stocks in trees above and below ground and non-tree woody biomass for the baseline scenario, thus estimating pre- and post-deforestation carbon stocks. Estimating the carbon stock plays a vital role in the implementation of AFOLU projects. This phase is essential for forecasting carbon credit generation through the implementation of REDD+ project actions and the analysis of its financial viability.

The carbon storage potential varies according to abiotic factors, such as the local hydrology pattern, soil (minerals and nutrients), climate (temperature, light and water), geology and other factors. These attributes, in turn, have already been mapped and studied over the years by academia and correlated to the carbon stock of the vegetation. However, some studies have identified three important weaknesses in the carbon maps for the Amazon Forest: (1) overestimation of carbon values in general; (2) coarse resolution and limited spatial variability, and (3) high degree of patchiness, with carbon values in a single area changing significantly from one map to another (ENGLUND et. al, 2017).

Moss Forest uses the scientific study prepared by Englund et. al (2017)11 as its base carbon map. This study consolidates six of the main carbon maps of the Brazilian Amazon, weighting all these carbon stock estimates. This consolidated database is used to estimate the average carbon stock and standard deviation of the area of interest. Seeking a conservative estimate, the system uses the lower limit of the established confidence interval.

Subsequently, using the carbon map, the system carries out the conversion of aboveground biomass to belowground biomass. This is accomplished through the proportion of the root with the shoot appropriate to the biome type, according to the default values determined by the AFOLU Guidelines (IPCC 2019) shown in the table below. To determine the domain and ecological zone of the analyzed territory, the system uses FAO global ecological zones.

Furthermore, the system captures and uses databases for carbon stock data in pastures to determine the net changes in the carbon stock at the baseline, being represented by the pre-deforestation forest stock minus the carbon stock of the probable land use after deforestation, usually to make pasture areas. It is considered that the other carbon reservoirs of plant biomass do not have stored carbon, which is a conservative estimate.

3.3. Land cover and land use

Understanding land use and occupation from satellite images underpins the understanding of historical anthropic pressure, the status and trends of the studied territory, as well as facilitating the socioeconomic and ecological management of the region. For REDD+ projects, it is necessary to understand the dynamics of land use and occupation in the target territory region, mainly in relation to the deforestation vectors and agents.

To this end, the Moss Forest code, in its application for Brazil, uses the mapping carried out by the Project for the Annual Mapping of Land Use and Coverage in Brazil (MapBiomas), which is a multi­ institutional initiative formed by NGOs, universities and technology companies to generate annual land use and land cover maps from automatic classification processes applied to satellite images. This Brazilian system (MapBiomas) is extremely robust to verify and attest land use and occupation maps from native to non-forest conversion, in a significantly cheaper, faster and updated way, compared to other current methods and practices.

3.3.1 Raster vectorization for query optimization and macro mappings

The refined methodology used by MOSS.Earth, based on the MapBiomas platform and other carbon maps, for the vectorization of raster data sets, is specifically aimed at carbon mapping and the creation of Land Use and Cover and Land Use Change maps (LULUC). This procedure begins with the transition of data from its original raster format, in which each pixel is imbued with a value alluding to a singular category of land use and cover, to a vectorized format suitable for loading on platforms such as BigQuery. It is noteworthy that raster formats, being incompatible with BigQuery, require conversion to shapefiles, structures that efficiently represent the different classes of land cover, as well as their respective geometries in a tabular organization, with each line corresponding to a category distinct from land use.

The conversion operation of this dataset takes place through the use of open source Python libraries, which include rasterio, shapely, geopandas, ogr, osr, gdal, numpy and pyproj. Throughout the process, a predetermined shapefile serves as a mask, segmenting the raster and enabling the necessary transformations to be carried out. We chose the administrative limits of Brazilian municipalities as masking units for effective scalability, given the computational infeasibility of converting a full raster dataset for a nation or state on a single occasion.

The algorithm used initially divides the raster into smaller units, in line with the dimensions of the municipalities, promoting the export of each individual segment to a specific directory, until all municipalities are adequately processed. Next, the code retrieves each fraction of the raster, converting it to a shapefile using GDAL's Polygonize function, and then grouping the data by class. Completing the cycle, the shapefile is loaded into BigQuery as a table, comprising the geometry belonging to each class, and structured in columns: CLASS | geometry.

It is crucial to emphasize that segmentation at the municipal level is an intermediate step, simplifying vectorization, but not influencing the subsequent integration of data into BigQuery, where data at the state or national level is considered in its entirety, without divisions. Later in the database, any additional mask, that is, another shapefile, can be used at the intersection with the existing data set, making it possible to obtain information about overlaps and other spatial interactions.

Additionally, the vectorization process is not limited to the municipal iteration, it also includes an annual iteration, applying the same methodology to the slope and elevation attributes, vectorizing the continuous values for altitude (in meters) and slope (in percentage) for each pixel.

It is imperative to include that MapBiomas is a consortium that includes a large collaborative network responsible for creating annual maps of land cover and use in Brazil. These are generated from pixel-by-pixel classification of Landsat satellite images in Google Earth Engine using machine learning algorithms. The beta serieslaunched by MapBiomas Solo, which details soil organic carbon stocks up to 30 centimeters deep, was produced using data from SoilData and relevant environmental variables.

These are generated from pixel-by-pixel classification of Landsat satellite images in Google Earth Engine using machine learning algorithms. The beta series launched by MapBiomas Solo, which details soil organic carbon stocks up to 30 centimeters deep, was produced using data from SoilData and relevant environmental variables.

Finally, it is essential to mention that the data generated by MapBiomas is available to the public, under the CC-BY-SA license, and must be referenced appropriately when used, following the format: "MapBiomas Project – Collection 8 of the Annual Series of Coverage Maps and Land Use of Brazil (1985-2022), accessed in October 2023 through the link: https://plataforma.brasil.mapbiomas.org/". Furthermore, it should be noted that the described methodology and procedures for Moss Forest applies exclusively to LULUC, slope and elevation data, and there is no record of its application to other data sets.

Eventually Moss plans to expand to other countries, using the Mapbiomas database for LULUC monitoring and national carbon maps for the rest of Latin America (Bolivia, Chaco/Paraguay, Argentina, Peru) and Indonesia:

https://platform.indonesia.mapbiomas.org/

https://peru.mapbiomas.org/en/ https://chaco.mapbiomas.org/en/

https://bolivia.mapbiomas.org/en/

Brazil: MapBiomas Brasil: https://brasil.mapbiomas.org/en/

Amazon countries: Bolivia, Brazil, Colombia, Ecuador, Peru and Venezuela (through the MapBiomas Amazonía initiative) MapBiomas Amazonía: https://amazonia.mapbiomas.org/en/

Trinational Atlantic Forest: Argentina, Brazil and Paraguay (through the MapBiomas Bosque Atlántico initiative) MapBiomas Bosque Atlántico: https://bosqueatlantico.mapbiomas.org/en/

3.3.2 Remote sensing

Moss Forest also uses remote sensing data from historical collections of Landsat satellite images produced by NASA (National Aeronautics and Space Administration) and USGS (United States Geological Survey) with a resolution of 30m and formed by the RGB pattern (red, green, blue). Subsequently, indices are calculated to facilitate the interpretation of these images, such as NDFI, NDVI and NPV, aiming at automating the process of discrete separation of land use classes. Finally, the system performs cloud processing by applying a trained algorithm that classifies each spectral response into land use categories, such as pasture, agriculture, urban infrastructure, hydrology and native vegetation. Also, MapBiomas embeds into its system some local samples to calibrate the algorithm in relation to reality (on the ground truth data). This process guarantees the quality of image classification and the reduction of errors. The stages of the MapBiomas land use and land cover classification methodology are presented below.

As such, a mosaic is obtained with the classes of final land use and vegetation for each year in matrix format (30x30m pixel), used to calculate the deforestation observed in the territories studied. The result of land cover and use in 1985 and 2017 in the Brazilian biomes can be seen in the comparative map below.

3.4. Eligibility of forests

For forest carbon project certification purposes, it is necessary to assess the eligibility of the forest area in relation to the methodology adopted in terms of defining the forest. Therefore, based on the classification of land use, presented in the previous item, Moss Forest performs some adjustments via code written on Google Earth Engine and other GIS software to align with the eligibility criteria of forests for avoided deforestation projects.

Thus, a geospatial analysis algorithm is used to determine the eligible forest area by excluding the following areas:

• non-forest natural formations (flooded fields and wetlands, grassland formations, apicum, rocky outcrops and herbaceous restinga) and savanna formations, mangroves and wooded “restinga.” The term "restinga" in Brazilian Portuguese refers to a specific type of coastal ecosystem found in Brazil, characterized by sandy soils and a variety of plant species. In English, the closest equivalent term would be "coastal woodland" or "coastal scrubland." However, it's worth noting that these terms may not capture the full ecological and botanical characteristics of a restinga.

• areas identified as forests less than 10 years old (we establish 10 years as the eligible forest parameter due to the lack of longer historical data and since it is the eligibility requirement by most carbon credit certification methodologies). This criterion is generally used to prevent areas of secondary forest vegetation falling fallow between cycles of agricultural production from being considered as forests.

This geospatial processing is performed using code in Google Earth Engine for the number of years determined by a methodology for analysis of an eligible forest (10 years in most cases, therefore, from the start of the assessment of the area, going back 10 years). The algorithm calculates the deforestation rate in the jurisdiction (municipality), according to the adopted REDD+ methodology for generating carbon credits. Thus, the historical behavior of land and forest use is identified for determining the baseline of deforestation and the forecast of the generation of carbon credits for avoided deforestation.

3.5. Environmental Co-benefits

Bearing in mind the importance of generating co-benefits through REDD+ projects, especially regarding aspects of biodiversity conservation linked to the forest carbon project, Moss Forest also consults data sources related to the topic and automates the assessment of impacts on biodiversity for Brazilian areas. In this way, valuable information is provided regarding the potential for change and impact of a carbon project in the evaluated territory, the potential for obtaining co-benefit certifications, which commonly add value to REDD+ projects, and, therefore, for optimized decision-making.

To this end, the system consults automatically three globally recognized databases on the characteristics of biodiversity conservation in a territory, namely:

• IUCN Red List of Threatened Species: catalog of the conservation status of plant, animal, fungal and protozoan species from around the planet from the most up-to-date list;

• World Database on Protected Areas: Global database on protected land and marine areas collected from international convention secretariats, governments and collaborating NGOs. The IUCN definition of a protected area is used as the main criterion for entries to be included in the database;

• World Database of Key Biodiversity Areas (KBA): Database of sites that contribute significantly to global biodiversity, covering terrestrial, freshwater and marine ecosystems. A KBA must meet one or more of eleven criteria, grouped into five categories, as follows: threatened biodiversity; geographically restricted biodiversity; ecological integrity; biological processes; and irreplaceability.

Such databases are periodically updated and maintained by the IBAT (Integrated Biodiversity Assessment Tool), an online data subscription platform for accessing global biodiversity datasets, which is included in the Moss Forest code, identifying any overlaps of such attributes to the project area and the 5 km, 20 km and 50 km buffers (Figure 8).

In this way, Moss Forest assesses the potential impact on the territory's biodiversity by the REDD+ project, expediting the assessment of the feasibility of implementing a REDD+ project; in the obtaining of primary data in loco by the technical team specialized in faunal groups; optimizing the field assessment of endangered species; and also during the preparation of reports to obtain co-benefit certifications, such as the CCB (Climate, Community and Biodiversity) and SD Vista (Sustainable Development Verified Impact Standard).

3.6. Risk of deforestation

To forecast the risk of deforestation and identify areas of future deforestation in the avoided unplanned deforestation modality, the spatial statistical modeling of “Forest at Risk”, recommended by the World Bank, is used as the spatial model for forecasting deforestation in 92 countries, covering all of the world's tropical rainforests. Due to the territorial extension of Brazil and the complexity of the analysis linked to this continental characteristic, the analysis was carried out for each Brazilian state. The interactive map resulting from the modeling is available at https://forestatrisk.cirad.fr/maps.html and its respective scientific article cited as Vieilledent et. al 2022.

The model is derived from high-resolution images and built based on factors interconnected with the existence of deforestation in tropical forests, such as: topography (altitude and slope), accessibility (distances to the nearest road, city and river), forest landscape (distance from the forest edge), historical deforestation (distance to past deforestation) and land conservation status (presence of protected area). The source and attainment of such data can be better understood in the supplementary material in Vieilledent et. al (2022). Such spatial variables, observed in the 2010s, are used as predictors to train the model to predict deforestation risk data and future forest cover under a “business as usual” scenario.

In this way, the model estimates the future change in forest cover by determining the baseline, a crucial step for avoided deforestation carbon projects and programs. Keeping the observed deforestation and considering the remaining forest in 2020, the BAU scenario of Brazil estimates a potential loss of 40% of forest cover during the 21st century (Figure 9) and, in the year 2204, the country would have lost 75% of its remnant forest cover.

What follows is a map of the probability (risk) of deforestation in the Brazilian Amazon, that is, where the flow of deforestation intersects with the carbon stock. Forest areas in dark red are at higher risk of deforestation than forest areas in green. One may observe that the probability of deforestation is lower within protected areas (black polygons) and increases when the forest is located close to roads (dark gray lines). Therefore, Moss Forest selects pixels with a high risk of deforestation to determine the area of future deforestation in the eligible forest of the analyzed territory (Figure 10).

3.7. Carbon credit generation potential

The carbon credit potential generation code used by Moss Forest considers all previous processes to identify the forest area eligible for the generation of carbon credits arising from planned and unplanned avoided deforestation.

For planned deforestation, when the owner has the legal right to deforest but does a REDD project to issue carbon credits and remunerate the land via the sale of these credits, the area that is not legally authorized for deforestation is identified as a Legal Reserve. The percentage of rural property registered as a Legal Reserve varies according to the biome, as follows: 80% in properties located in forest areas in the Brazilian “Legal Amazon”; and 35% in properties located in “Cerrado” areas in the Legal Amazon. The term "cerrado" in Brazilian Portuguese refers to a vast tropical savanna biome found mainly in Brazil, but also in parts of Bolivia and Paraguay. When translating "cerrado" into English, the most appropriate term would be "Brazilian savanna" or simply "cerrado." The cerrado is characterized by a unique combination of grasslands, shrubs, and small, twisted trees, and is considered one of the richest and most diverse tropical savanna ecosystems in the world.

Based on these and other guidelines contemplated in the Brazilian Forest Code, the rural landowner determines the location of their legal reserve on their property, which must be registered in the CAR (Rural Environmental Registry) and approved by the environmental agency with jurisdiction over the matter. It is common for the legal reserve not to be approved by the environmental agency. In order to help minimize this situation, Moss Forest assists in determining the possible legal reserve area based on the defined percentage of overlap with the eligible forest area (process presented in item 7). This value is finally divided by ten years of avoided planned deforestation credit generation.

As for the avoided unplanned deforestation modality, the methodology for calculating the potential for generating carbon credits is different. Based on the “risk of deforestation” processes presented, they are superimposed on the eligible forest map, which will result in an estimate of the absolute area deforested for the year 2050. The assumption is that the generation of these credits will be constant over the years and thus an estimate of the generation of carbon credits disaggregated by year is obtained.

Moss Forest also indicates situations in which the area of interest has the potential to generate carbon credits in both modalities (Figure 11).

For the purposes of analyzing the risk of non-permanence of carbon pools, whether due to illegal deforestation, fires or natural disasters, the Moss Forest system applies an automatic non permanence risk buffer of 20%.

For the increase in GHG emissions or decrease in removal outside the project boundaries that occurred because of project actions, referred to as leakage, Moss Forest applies the 10% discount.

For the estimation of the gross revenue resulting from the implementation of the project, economic factors such as the current value of the credit, the exchange rates to Brazilian Real and the percentage of participation of the owner in the distribution of revenues are defined.