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)].
There are several scripts included with the project. They are intended to be run in the following order:
/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./scripts/estimate_footprints.py<path/to/config>- estimate probability maps using building footprint models defined in the config file./scripts/run_pipeline.py<path/to/config>- run preprocessing, build dataset, predict population, and output results of experiments according to the config file./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>
Each script takes a separate YAML config file as input, examples of each can be found in /scripts/config/.
Pretrained population models are available from the authors upon request.
Documentation is available at /docs/_build/html/index.html
Executing run_pipeline.py yields the following results for the example pipeline.yaml with outliers removed:
