SCIPE

Research code for Sustainable Census Independent Population Estimation (SCIPE) - a pipeline for predicting population maps using microcensus data and satellite imagery.

SCIPE is introduced in the paper 'Census-Independent Population Estimation using Representation Learning' [Scientific Reports (2022)].

Scripts

There are several scripts included with the project. They are intended to be run in the following order:

1. /scripts/rasterize_survey.py <path/to/config> - rasterize survey data to geoTiff. This script is dependent on survey format, and is designed for SpaceSUR population survey from Mozambique.
2. /scripts/estimate_footprints.py <path/to/config> - estimate probability maps using building footprint models defined in the config file.
3. /scripts/run_pipeline.py <path/to/config> - run preprocessing, build dataset, predict population, and output results of experiments according to the config file.
4. /scripts/split_imagery.py <path/to/imagery> <path/to/survey> <out/path> - split large raster files contained within <path/to/imagery> into grid defined by geoTiff at <path/to/survey>

Config

Each script takes a separate YAML config file as input, examples of each can be found in /scripts/config/.

Models

Pretrained population models are available from the authors upon request.

Documentation

Documentation is available at /docs/_build/html/index.html

Examples

Executing run_pipeline.py yields the following results for the example pipeline.yaml with outliers removed: