Annex 1 — Parameters and reproduction

Inputs

Table 7: Inputs to the calculation.

Input

Specification

Surface reflectance

Landsat Collection 2, Level 2. LANDSAT/LE07/C02/T1_L2 to and including 2014; LANDSAT/LC08/C02/T1_L2 from 2015.

Spectral endmembers

CODED, five fixed endmembers (Bullock et al., 2020, Table 1); Table 1.

Aboveground biomass

ESA CCI Biomass, 2017 to 2020.

Land cover

Copernicus Global Land Service Land Cover 100 m, collection 3, epoch 2018 (V3.0.1); MapBiomas Colombia, Collection 3.0, 2018, 30 m (Fundación Gaia Amazonas, 2025).

Project area

The project activity instance boundary, 2,371.35 ha.

Screening applied

Years. Years in which Landsat data were missing for a large number of project plots were excluded, on the counts of Table 3 above. The years retained are 2008, 2011 and 2014 to 2023. The effective historical series is therefore seven years of the eleven in the t = −10 to t = 0 window.

Plots. A plot with a missing value in any retained band was removed in full, not only for the year affected.

Land cover. Donor plots whose land cover class is not represented among the project plots were removed. The classes present among project plots are 30, 40, 122 and 126 of the Copernicus Global Land Service land cover (epoch 2018). Donor plots were further restricted to the MapBiomas Colombia 2018 classes held by at least 5% of the project-plot candidates, 21 and 12 (Select donor pool area).

Availability windows. A plot was retained only if it has a valid value in at least one year between t = −10 and t = −8, at least one year between t = −7 and t = −1, at t = 0, and in every year of the monitoring period.

Valid range. A plot with an observation outside the valid range of the surface reflectance product (DN 7,273 to 43,636) in any band, or with an undefined NDFI, was removed in full (Stocking index calculation).

Of the 150 project-plot candidates, 105 passed the screens above and 102 the valid-range screen; these were carried into the matching, and of those the 45 with the closest match to their control plots over the historical period were retained. Of the 15,000 donor candidates, 2,885 passed the Copernicus land-cover screen and the criteria of Table A1, 2,264 the MapBiomas screen and 2,228 the valid-range screen, and were carried into the matching.

Parameters applied

Table 8: Every parameter applied in the calculation.

Parameter

Value

Plot size and configuration

30 x 30 m, one Landsat pixel, non-overlapping

Project-plot candidates generated / carried / retained

150 / 102 / 45

Donor candidates generated / carried into matching

15,000 / 2,228

Donor search radius

50 km from the project centroid, excluding the project area (within the 100 km maximum of VM0047, Table A1)

Donor candidates per project-plot candidate

100

Control plots per project plot

4, without replacement

Years of imagery

2008 to 2023, 1 May to 31 August, annual median

Reflectance scaling / normalisation

x 2.75e-5 − 0.2 (surface reflectance); for the band covariates, then unit L2 norm per observation

Valid surface reflectance range

DN 7,273 to 43,636 in every band (USGS Collection 2, Level 2); a plot with an observation outside it, or with an undefined NDFI, is removed

Years retained

2008, 2011, 2014-2023

t = 0

2018, the earliest planting year and the instance start date

Historical window

t = −10 to t = 0, that is 2008 to 2018

Land cover classes of project plots

30, 40, 122, 126 (Copernicus, epoch 2018); MapBiomas 2018: 21 and 12, the classes held by at least 5% of the project-plot candidates

Donor land-cover screen (second)

MapBiomas Colombia, Collection 3.0, 2018: classes held by at least 5% of the project-plot candidates

Spectral unmixing

fully constrained linear unmixing (fractions non-negative, summing to one), five fixed endmembers of CODED (Bullock et al., 2020, Table 1)

AGB sample trim for the biomass-to-index regression

2.5th to 97.5th percentile

Stocking index

NDFI (Souza et al., 2005), Eq. 1

Covariate matrix for the matching

the stocking index of each retained historical year (2008, 2011, 2014 to 2018) and the principal components of the six 2018 bands that explain at least 80% of their variance (two), standardised over project and donor plots: 9 covariates in all (Table 5); no topographic or land-cover covariate

Distance metric

Manhattan

Assignment

Hungarian algorithm (linear sum assignment), four rounds, each excluding the donor plots already assigned

Match quality threshold

standardised difference of means below 0.25 for every covariate

Regression of the stocking index on year

ordinary least squares over the individual plot-by-year observations, cumulative from t = 0

Significance thresholds

abs(z) > 1.96; control trend p < 0.05

Ex-ante rolling window

10 years

How to reproduce the delivered figures

The ex-post results can be reproduced from the delivered data alone, without the code. The project proponent has verified the following and the validation/verification body can repeat it:

  • From annex G, regressing the column SI, wp,t and the column SI control,t on Year over all plot-by-year observations from 2018 to each monitoring year, every column of annex B is obtained: both slopes, both standard errors, z, the control p value and PBt, for each of the five monitoring years.

  • Restricting the same regression of SI control,t to the years up to and including 2018 returns the historic control slope and p value of annex A.

  • The control stocking index of each project plot is the weighted average of its four control plots, using the weight columns of annex G.

The matching can be reproduced from Matching covariates at t=0.kml, which holds, for the 2,330 plots carried into the matching, the band values at t = 0 (2018) and the stocking index of every retained historical year, from which the covariates of Step 2 are formed, by applying Step 2 of the procedure with the parameters above.