We start with a project area in the tropical moist forest (TMF) biome.
We can think of the project area as a grid of small pixels, with tens of thousands in a typical project, each of which has a land use class (forested, deforested, etc.).
We can extract the land use classification - derived by third parties using satellite-based earth observation - and look at how this changes over time to measure the deforestation that has occurred in the project. If the project developer is truly protecting the forest, we expect there to be less deforestation within the project compared to the wider landscape. In other words, there should be less deforestation than would have occurred in the business-as-usual scenario.
We can determine the business-as-usual scenario by measuring deforestation in places that are similar to the project (i.e., subject to the same forces of deforestation) but are not protected. These places form our counterfactual. One way of identifying the counterfactual is through pixel-matching, which allows us to ensure similarity across a set of variables that have been found to be important predictors of deforestation rate.
By comparing the deforestation in the project with that in the counterfactual, we can measure additionality: deforestation avoided as a direct result of the project.
The reduction in deforestation corresponds to a reduction in carbon emissions, which can be quantified and used to issue carbon credits.
There are four phases to the pipeline: preparing the input layers, creating project-specific inputs, doing the matching, and calculating additionality.
Divide the TMF project area into a grid of 30 metre by 30 metre pixels, since this matches the pixel size used by the JRC TMF dataset. Using the JRC TMF data, make a transition map containing the per-pixel land use changes in the 10 years prior to the project start (between $t_{-10}$ and $t_0$). Possible transitions include undisturbed → undisturbed (what we want), as well as undisturbed → degraded, undisturbed → deforested and deforested → regrown forest.
This transition map serves a dual purpose, as both an input for pixel-matching, and a way of measuring the land use change and carbon impact of the project.
Next we prepare the layers for the other matching variables. From the land use transition map, we create a Coarsened Proportional Cover (CPC) map. CPC is a measure of the proportion of pixels in either the undisturbed or the deforested class within a 1km radius of a particular pixel, and thus acts as a summary of the deforestation environment. CPC maps are created at three time points: the project start ($t_0$), 5 years prior to project start ($t_{-5}$) and 10 years prior to project start ($t_{-10}$).
We extract elevation and slope data from NASA SRTM. The higher the elevation and slope of a particular place, the lower its risk of deforestation.
We extract accessibility data from the Malaria Atlas Project, measured as motorized travel time to healthcare (as of 2019). This acts as an indicator of remoteness (and hence deforestation risk) for a particular place.
Finally, we prepare maps of ecoregion and country borders, our final matching variables.
Start with a shapefile, which is a way to represent the boundaries of the project. This project has a start year of $t_0$.
Create a land use transition map for the project area.
Extract NASA GEDI data for the project area. This provides mean carbon density values for each land use class which allow us to convert the land use transitions into carbon stock changes.
Take a dense sample of points from within the project area, known as the project sample (or Set K). The density of the project sample is determined by the size of the project: those under 250,000 hectares are sampled at a density of 0.25 pixels per hectare; projects over 250,000 hectares are sampled at a density of 0.05 pixels per hectare. This lets us get a good, but not too large, sample of project pixels.
Extract the following characteristics from each point in the project sample from the prepared layers:
Generate the set of potential matches based on the characteristics of the project sample. First, create the potential matching space: the intersection of the countries and ecoregions that lie within the project plus a 2000 km buffer around the project, but excluding: