TerraVista Methods

Ex-ante Carbon Stock Forecast

Native Restoration

TerraVista calculates ex-ante GHG removals based on state-of-the-art science, by model forest growth through time using on-the-ground-observations, globally distributed datasets, and satellite remote sensing, integrated with machine learning.

Stand-level carbon ($tCO_2e/ha$) is modeled with the Chapman-Richards (CR) growth equation, which is typical for this use case¹. This curve accounts for initial slow development as saplings establish a root system, followed by a period of rapid growth until canopy closure when competition among trees increases and growth is limited by light, water, and nutrients. At this point, biomass approaches the maximum mature biomass, which is determined by the climate and edaphic factors of the site.

A form of the CR curve was used that has three parameters: maximum biomass in tons dry matter per hectare ($MAX$), a dimensionless growth rate ($k$), and a dimensionless shape parameter ($m$):

$Biomass=MAX*[1-exp(-k*Age)]^{(11-m)}$

Where $Age$ is tree age in years and $Biomass$ is aboveground and belowground biomass.

Potential maximum biomass is a key aspect of the carbon curve because it characterizes the upper limit of the carbon projection. The potential biomass ($MAX$) is the greatest biomass density (t dm/ha converted to $tCO_2e/ha$) the site can support once the forest stand is fully developed. Although tree planting and interventions such as weeding or irrigation can speed up the restoration process and enable regeneration in places it would not occur otherwise², the local environment determines the mature forest biomass the project area can support. For stand-level modeling, it is assumed that the potential stand-level biomass is independent of the species mix, planting density, or other project interventions, given that a native forest establishes on the site. While these variables don’t affect the potential biomass, they do affect the growth rate. In other words, factors such as species mix don’t affect how much is sequestered but when.

For stand-level modeling a global satellite-based machine learning model of potential woody aboveground and belowground biomass, based on climate, soil, and topography³, was used to determine the potential biomass of the project site. This potential biomass is localized with a resolution of 500 m (compared to previous efforts at 900-m ⁴ at 900-m and 9.3-km ⁵) and reflects the potential in the specific project area.

Specialists at Earthshot assembled a growth database from peer-reviewed scientific literature, agency and partner project data, and an internal database of project field data for observations of stand-level biomass accumulation over time to estimate parameters in the CR growth equation. Observations were algorithmically selected that approximate the project's climatic, edaphic, and biological conditions.

Biomass values were converted to tonnes of $CO_2$ equivalent prior to modeling. Woody tree biomass is about 47% carbon, and this factor was used to convert biomass to carbon. Carbon was transformed into a $CO_2$ equivalent by multiplying the carbon weight by 44 and dividing by 12, following the molecular transformation of carbon into carbon dioxide.

To depict the range of variability in the GHG forecast we calculated uncertainty due to CR parameter uncertainty and variability in potential maximum biomass across the project area using a Monte Carlo simulation method. This method uses curve fit parameters and covariance between the parameters to generate 1,000 random parameter combinations. To effectively capture the spatial variability in maximum potential biomass in the project area, these 1,000 parameter combinations were applied to deciles of maximum potential biomass found within the project area and a final curve calculated from the 50th percentile, with a 95% confidence interval, was generated.

The realized planting density will depend on local site conditions and the mixture of included tree species. It will be essential to select species adapted to the project site’s growing conditions. Native species often meet these requirements. Even so, it is recommended to enlist a local restoration expert to determine the appropriate species mixtures and planting densities to be used across the potential project area.

The following map outlines areas that currently have sufficient data to make growth projections (red), areas where we are still collecting data and so do not automatically output growth projections (green), and ecozones where reforestation is not expected (grey).

INSERT IMAGE

GHG Emission Deduction Scenarios

Deduction scenarios were generated using assumptions about pre-existing biomass, uncertainty, buffer pool, performance benchmark, and leakage. Two leakage scenarios were provided: (1) following Verified Carbon Standard VMD0054 and (2) assuming the project would mitigate all leakage impact down to 0. Leakage was calculated according to the VCS VMD0054 which evaluates historic production in the project area and deducts carbon equivalent to the deforestation caused by shifting production. Historical production was defined using global livestock distribution maps and global crop production maps. These maps were overlaid on the project polygon(s) boundaries and total production was extracted. Livestock production includes buffalo, cattle, sheep, and goats. Crop production includes all commodity crops. Carbon stock on new lands brought into production was defined as the average observed current carbon stock on forested lands in the ecoregion. Other parameters were held at the default values provided in the methodology.

Table 1. Native reforestation deduction scenarios

Land Cover Classification Methodology and Land Eligibility Assessment

Following Verra’s Standard (v4.7) and VM0047 methodology, land eligibility for ARR includes the following requirements:

1. Evidence shall be provided that the native ecosystem to restore was degraded at least 10 years prior to the proposed project start date.
2. Project activities cannot occur on organic soils or in wetlands and should not convert native ecosystems (e.g. native grasslands or shrublands).
3. Evidence shall be provided that the project activity restores a native ecosystem type represented in the same ecoregion as the project.
4. Where the ecosystem was degraded within 10 years of the project start date, evidence shall be provided that the ecosystem was not degraded due to the project activity (e.g. that the degradation occurred in the pre-project land use due to natural disasters such as hurricanes or floods). Such evidence is not required where the dominant land cover is an invasive species that is threatening ecosystem health.

Depending on the project geography and data available in the region, TerraVista uses two different approaches to classifying the land cover.

Approach 1

When available, the most up to date MapBiomas Collection dataset for the assessment area has been selected as the source of LULC maps for the Land Eligibility Assessment of the present study. Selection of this dataset was possible as it has sufficient historical coverage with yearly LULC maps from 1985 to 2023 for a feasibility as well as high global accuracy in its classification varying across biomes from 81.8% to 96.6% in the level 1 classification.

Two LULC maps were selected for the assessment, one as the start year, approximately ten years before the expected project start date (2013) and another at the end of a designated period, typically prior to the project start date (2023). Classes of the MapBiomas LULC maps were reclassified into forest, non-eligible non-forest, wetland, and eligible non-forest for ARR project activities. Eligible areas at this stage are considered as all non-forest areas that have been converted from forest into productive land use (eg. cropland & pastureland), and/or into degraded forest (eg. non-regenerating areas due to logging or wildfires, such as non-natural occurring grasslands or sparsely forested areas), where native forest regeneration could occur. A final step is then implemented by comparing the ARR eligible non-forest areas for both years to identify all eligible areas that have remained unchanged throughout the specified period (10 years). This process provides the required evidence to showcase that the native ecosystem to restore was degraded at least 10 years prior to the proposed project start date.

All calculations are conducted using Google Earth Engine. See below for additional details regarding the LULC class reclassification.

Table 2. LULC class reclassification

Approach 2

Outside regions covered by the MapBiomas dataset, two supervised classifications are performed using Copernicus Sentinel‑2 MSI Level‑2A imagery and a digital elevation model (SRTM v4) covering a 10‑year period in accordance with the standard requirement. Each classification was performed with a surface‑reflectance composite at 10‑meter spatial resolution using a Random Forest classifier (100 classifier trees). For the supervised classification the following spectral bands were included: B2, B3, B4, B5, B6, B7, B8A, B11 and B12 — with clouds and cirrus masked by the Sentinel‑2 QA60/SCL bit masks — and several spectral indices (NDVI, NDII, NDWI and EVI). Ancillary data, such as canopy height models, global LULC datasets, and ground truth points (when available) are used to support and guide the labelling process.

Two land cover classifications are conducted, one at the beginning (approximately ten years before the project start date) and another at the end of a designated period (typically prior to the project start date). Eligible areas at this stage are considered as all non-forest areas that have been converted from forest into a productive land use (eg. cropland & pastureland), and/or into degraded forest (e.g. non-regenerating areas due to logging or wildfires, such as non-natural occurring grasslands or sparsely forested areas), where native forest regeneration could occur. A final step is then implemented by comparing the ARR eligible non-forest areas for both years to identify all eligible areas that have remained unchanged throughout the specified period (10 years). This process provides the required evidence to showcase that the native ecosystem to restore was degraded at least 10 years prior to the proposed project start date.

Footnotes

1. 2001. Modeling and Prediction of Forest Growth Variables Based on Multilevel Nonlinear Mixed Models. Hall & Bailey. Available at https://academic.oup.com/forestscience/article/47/3/311/4617390
   ↩
2. 2017. A global review of past land use, climate, and active vs. passive restoration effects on forest recovery. Meli et al. Available at https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0171368
   ↩
3. 2022. The global potential for increased storage of carbon on land. Walker et al. Available at https://www.pnas.org/doi/10.1073/pnas.2111312119
   ↩
4. 2019. The global tree reforestation potential. Bastin et al. Available at https://www.science.org/doi/10.1126/science.aax0848
   ↩
5. 2018. Unexpectedly large impact of forest management and grazing on global vegetation biomass. Erb et al. Available at https://www.nature.com/articles/nature25138
   ↩