spatial_pop_projections

This is the methodology used to project spatialized school-age populations until 2030, based on UN SALB boundaries and WorldPop gridded population estimates. The projections are adjusted to national population projections, and then transformed into school-age groups with this plugin based on this methodology

The first use case is Togo.

Developing and running the analysis

The analysis is containerized. The only requirement for running it is a container engine such as Podman or Docker. You can of course run everything in an ordinary python virtual environment too, if you prefer it, but the following instructions assume you are running the container.

To start, spin up a container and attach to it:

docker compose up -d

docker attach population-analysis

After attaching you can use python to run code in the container as you normally would.

If you want to use jupyter lab from the container, run

invoke jupyter

and look for the jupyter url in the output.

Automated tests can be run with:

invoke test

and formatting with:

invoke format

Analysis workflow and structure

All of the analysis components are scripted and found in the scripts/ directory. The files meant to be run as scripts are at the top-level of the directory, while more general-purpose functions used by the scripts are stored in their respective packages. In addition, a minimal example of the projection methodology is available as an interactive notebook in notebooks/ example.ipynb.

Below are brief descriptions of the main analysis steps, and how to run them.

Step 1 - download all input data

Download historical population grids for 2000, 2005, 2010, 2015, 2020 for Togo, Ghana, Burkina Faso, and Benin from Worldpop
Result: Input data stored to their respective directories in the local repository
Download all input data by running:

python scripts/download_all.py

Step 2 - preprocess data

Using the SALB official boundaries for Togo, create the official Togo dataset for 2000, 2005, 2010, 2015, 2020.
- This is done by merging all of the input rasters (all different age, sex and year combinations), and then clipping the mosaics with the correct boundaries.
Result: The full Togo historical population grids for 2000, 2005, 2010, 2015, 2020 by 5-year age groups, with official boundaries.
Preprocess everything by running:

python scripts/preprocess_all.py

If analysing a country where the worldpop raster boundaries match with the interest area the preprocessing step is unecessary.

Step 3 - project population

Apply a linear model for each pixel for their population values in 2000-2020.
Based on the linear change estimated from the 2000-2020 data, project the population values for the period 2022-2030.
Adjust the projections to official population projections. Currently this is done on prefecture level, so every prefecture's total predicted population for a given year matches the official projection for that year.
Result: Adjusted population estimations for each pixel in each prefecture until 2030.
Create the prefecture-level projections by running:

python scripts/predict_all.py

Find an interactive axample of how creating one prediction works in notebooks/example.ipynb

Step 3 - merge projections

Merge all prefecture-level projections to get the projections for the entire Togo (all different age, sex and year combinations).
Result: Finished projections for Togo until 2030.
Merge predictions by running:

python scripts/merge_predictions.py