Evaluating methods of counterfactual estimation with placebos

The repository contains the project that evaluates various ex-ante and ex-post methods of counterfactual estimation for REDD+ projects (Reducing Emissions from Deforestation and forest Degradation in Developing countries), using a set of placebo projects randomly selected across the wet tropics with comparable characteristics to existing REDD+ projects. Because of the absence of project activities, project and counterfactual deforestation should follow the identical trend in these placebo projects, allowing us to evaluate the predictive performance of a method by comparing predictions against observed deforestation rates in the placebo projects.

Input

The code ingests the parquet files of sampled pixels in the project area and parquet files of matched pixels. The matching procedures were conducted using the Tropical Moist Forest Accreditation Methodology Implementation code[1], which implements the Canopy PACT 2.0 methodology[2][3]. The matched pixels are used to constrcut counterfactuals using one of the three ex-ante forecasting methods as well as an ex-post estimation method. The code then estimates annual compound deforestation rates (%) of the placebo projects, and compares it against observed deforestation rates.

Structure

The project contains the following scripts, all written in Python:

1. ``generate_placebos.py`: randomly generates a set of placebo project areas with Google Earth Engine API
2. rates.py: estimates annual counterfactual deforestation rates (%) using the four methods and comparing against observed deforestation rates in placebo projects
3. graphs.py: plots results from script 2
4. yearly_rates.py: calculates the predictive performance of the four methods for near-future time intervals of different durations
5. yearly_graphs.py: plots results from script 4

System requirements

This project is developed under Python 3.12.2, and uses the libraries panda, numpy, scipy, matplotlib, and seaborn. Each script can be executed by simply running:

python3 -m [script name]

In addition, the Sherwood cluster (Dept of Comp Sci, University of Cambridge) currently hosts a Python environment wrapper tmfpython3, which contains all the libraries needed for the code (with the notable exception of seaborn). There, the scripts can be run by:

tmfpython3 -m [script name]

References

[1] Dales M, Ferris P, Message R, Holland J, and Williams A (2023). GitHub Repository: Tropical Moist Forest Accreditation Methodology Implementation, https://github.com/quantifyearth/tmf-implementation. commit:7f15246

[2] Balmford, A et al. (2023). PACT Tropical Moist Forest Accreditation Methodology v2.0. Cambridge Open Engage, https://www.cambridge.org/engage/coe/article-details/657c8b819138d23161bb055f.

[3] Swinfield, T and Balmford, A (2023). Cambridge Carbon Impact: Evaluating carbon credit claims and co-benefits. Cambridge Open Engage, https://www.cambridge.org/engage/coe/article-details/6409c345cc600523a3e778ae.