Small Area Estimation at 5 km Resolution Using a Reproducible Geospatial Data Framework: General-Purpose Code, Ethiopia-Context Raster (Covariates, Inhabited Areas, Under-Five Population), and Synthetic Data for Censoring-Adjusted Child Survival Analysis

This repository provides reproducible resources for geospatial small-area estimation at 5 km spatial resolution. The deposit includes:

• Generalizable analysis code for raster preprocessing, harmonization, rescaling to a common 5 km grid, population and inhabited-area data integration, and informative censoring adjustment;
• Ethiopia-specific raster datasets (covariates, under-five population surfaces, inhabited area masks) processed to 5 km resolution;
• Synthetic DHS dataset for reproducible testing without restricted DHS access;
• Complete R Markdown workflows demonstrating covariate/inhabited/population adjustment and censoring adjustment.

The code is designed to be reusable and can be applied to other countries and study settings with comparable input data.

What This Repository Contains

1. General-Purpose Code (Fully Reproducible)

01_covariate_adjustment.Rmd
Processes and harmonizes 25 raster covariates from multiple geospatial sources (ERA5, GBD, MODIS, WorldPop, Malaria Atlas Project). The workflow applies reprojection, resampling to a common 5 km grid, masking to Ethiopia, log1p transformation, z-score standardization, capping of extreme values, and pairwise correlation analysis to generate a reduced set of uncorrelated predictors.

02_censoring_adjustment.Rmd
Uses processed raster covariates together with Ethiopia DHS child-level data to construct survival datasets and perform informative censoring adjustment. Three survival modeling approaches are evaluated (CoxPH, Random Survival Forest, Gradient Boosting Machine), with the Random Survival Forest model selected as the final censoring-adjustment model based on predictive performance.

2. Ethiopia-Context Raster Datasets (5 km resolution)

• Harmonized covariate rasters (2000–2016, extended to 2023 where available);
• Under-five population surfaces;
• Inhabited area masks;
• Covariate summary statistics and multicollinearity assessment outputs;
• JSON metadata files accompanying all raster products.

3. Synthetic Data for Testing

• synthetic_dhs.rds – A synthetic DHS dataset provided to demonstrate, test, and validate the DHS survival-data preparation and censoring-adjustment workflows without requiring restricted DHS microdata access.

4. Worked Example: Individual Year Analysis (Year 2000)

Important clarification: This repository includes a worked example for the year 2000 only, demonstrating the complete modeling workflow from covariate preparation through Bayesian spatial prediction for a single year of analysis. The year 2000 raster outputs included here are generated from this individual-year workflow and serve as an illustrative example for researchers who wish to apply the framework to a single year of data in other settings.

This individual-year analysis is conceptually and methodologically distinct from the full spatio-temporal model described below.

What This Repository Does NOT Include (and Where to Find It)

Restricted DHS Microdata

This repository does not include restricted DHS microdata, household identifiers, or cluster-level geographic coordinates. All DHS-derived analyses were conducted under DHS Program data-use agreements.

Final Spatio-Temporal Under-Five Mortality Estimates (2000–2015)

This repository does NOT contain the final spatio-temporal under-five mortality estimates for Ethiopia.

The annual high-resolution (5 km × 5 km) under-five mortality probability and expected death count surfaces for the period 2000–2015 (all 16 years together) are available in a separate Zenodo deposit:

Bayuh Asmamaw, H. (2026). Bayesian INLA-SPDE Spatio-Temporal Small-Area Estimates of Under-Five Mortality in Ethiopia (2000–2015): 5 km Mortality Probability and Death Count Surfaces. Zenodo. https://doi.org/10.5281/zen…

That deposit contains:

• 48 mortality probability rasters (16 years × 3 summaries: mean, lower, upper 95% CrI);
• 48 expected death count rasters (16 years × 3 summaries);
• 96 GeoTIFF files total;

The separate spatio-temporal deposit represents the final modeled outputs generated using methods documented in this code repository, but applies a temporally structured model (2000–2015) rather than the individual-year worked example (2000 only) shown here.

Relationship Between the Two Deposits

Feature	This Repository (Zenodo 20080209)	U5M Spatio-Temporal Repository (Zenodo 20524950)
Purpose	General framework + code + input data + worked example	Final output surfaces
Temporal scope	Worked example: Year 2000 only (illustrative)	Full spatio-temporal: 2000–2015 (all years together)
Model structure	Individual-year cross-sectional (example)	Temporally structured (RW1 + SPDE)
Contents	Code, covariates, population, masks, synthetic data, example 2000 outputs	Final mortality probability + death count rasters
Intended use	Reproducibility, adaptation to other settings, method development	Direct use for Ethiopia U5M estimation

Researchers who wish to:

• Reproduce or adapt the methodology → Use this repository;
• Obtain final Ethiopia under-five mortality estimates → Use the separate U5M spatio-temporal deposit.

Intended Use

This repository is intended for:

• Reproducible geospatial small-area estimation research;
• Adaptation of the methodology to other countries or study settings;
• Understanding covariate preprocessing and censoring-adjustment workflows;
• Testing and validation using synthetic DHS data;
• Researchers who wish to run individual-year analyses (the year 2000 example provides a complete template).

For researchers seeking final annual under-five mortality estimates for Ethiopia (2000–2015) , please refer to the separate spatio-temporal deposit cited above.

Software and Code Availability

GitHub repository: https://github.com/Bayuh23/…

License: MIT License (code); raster datasets are CC BY 4.0

Citation

Please cite this repository as:

Bayuh Asmamaw, H. (2026). Small Area Estimation at 5 km Resolution Using a Reproducible Geospatial Data Framework: General-Purpose Code, Ethiopia-Context Raster (Covariates, Inhabited Areas, Under-Five Population), and Synthetic Data for Censoring-Adjusted Child Survival Analysis. Zenodo. https://doi.org/10.5281/zen…

If using the final spatio-temporal under-five mortality estimates for Ethiopia (2000–2015), please cite the separate deposit:

Bayuh Asmamaw, H. (2026). Bayesian INLA-SPDE Spatio-Temporal Small-Area Estimates of Under-Five Mortality in Ethiopia (2000–2015): 5 km Mortality Probability and Death Count Surfaces. Zenodo. https://doi.org/10.5281/zen…

License: CC BY 4.0 (rasters)

Dates: Submitted 2026-04-19