Bayesian INLA-SPDE Spatio-Temporal Small-Area Estimates of Under-Five Mortality in Ethiopia (2000–2015): 5 km Mortality Probability and Death Count Surfaces.

Overview This dataset provides annual high-resolution (5 km × 5 km) estimates of under-five mortality probability and expected under-five death counts across Ethiopia for the period 2000–2015. The estimates were generated using a Bayesian spatio-temporal small-area estimation framework incorporating: Integrated Nested Laplace Approximation (INLA); Stochastic Partial Differential Equation (SPDE) modeling for spatial random effects; Negative binomial likelihood with population offset; First-order random walk (RW1) temporal effects; Harmonized environmental, health, demographic, and socioeconomic covariates. The dataset provides gridded posterior mean estimates together with lower and upper bounds of the 95% credible interval for each year. Dataset Contents Under-Five Mortality Probability Rasters Mortality probability is expressed as the probability of dying before reaching five years of age and ranges from 0 to 1. File Pattern Description eth_u5_mortality_risk_mean_YYYY.tif Posterior mean mortality probability eth_u5_mortality_risk_lower_YYYY.tif Lower bound of the 95% credible interval eth_u5_mortality_risk_upper_YYYY.tif Upper bound of the 95% credible interval Expected Under-Five Death Count Rasters File Pattern Description eth_u5_deaths_number_mean_YYYY.tif Posterior mean expected number of under-five deaths eth_u5_deaths_number_lower_YYYY.tif Lower bound of the 95% credible interval eth_u5_deaths_number_upper_YYYY.tif Upper bound of the 95% credible interval Temporal Coverage Annual raster layers are provided for: 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, and 2015. Total dataset contents: 48 mortality probability rasters (16 years × 3 summaries); 48 death count rasters (16 years × 3 summaries); 96 GeoTIFF raster files in total. Spatial Information Property Value Geographic coverage Ethiopia Spatial resolution 5 km × 5 km Coordinate reference system EPSG:32637 (UTM Zone 37N) File format GeoTIFF (.tif) Compression LZW Interpretation of Mortality Estimates Mortality probability values range from 0 to 1. Examples: 0.05 = 5% probability of death before age five; 0.08 = 8% probability of death before age five; 0.10 = 10% probability of death before age five. For comparison with conventional under-five mortality rates reported per 1,000 live births: Under-Five Mortality Rate (per 1,000 live births) = Mortality Probability × 1,000 Example: 0.089 × 1,000 = 89 deaths per 1,000 live births. Model Development and Validation The mortality surfaces were generated using Bayesian spatial modeling informed by nationally representative survey data and geospatial covariates. Model development included: Harmonization of raster covariates to a common 5 km grid; Adjustment of child survival outcomes for informative censoring; Selection of uncorrelated candidate predictors; Bayesian INLA-SPDE spatial modeling; Annual prediction of mortality probability and expected death counts. Model performance was evaluated using spatial block cross-validation and out-of-sample testing. Validation metrics: Metric Value DIC 17,672.89 WAIC 17,665.10 Mean Test Coverage 0.91 Validation Approach Spatial block cross-validation Relationship to Previous Repository This dataset represents the final modeled outputs derived from the reproducible geospatial framework archived at: Bayuh Asmamaw H. (2026). Small Area Estimation using a Reproducible Geospatial Data Framework at 5 km Resolution: Harmonized Covariates, Censoring-Adjusted Child Survival Outcomes, Under-Five Population Surfaces, and Inhabited Area Masks. Zenodo. DOI: 10.5281/zenodo.20080209 The source repository contains: Harmonized covariate rasters; Under-five population surfaces; Inhabited area masks; Censoring-adjusted and child survival and covariate adjusted R Markdown code; Complete preprocessing and adjustment workflows; Reproducible R Markdown code; In addition, the source repository includes a complete worked example for the year 2000, demonstrating the full modeling workflow from covariate preparation through Bayesian spatial prediction. The accompanying code can be adapted for application in other countries where equivalent survey and geospatial data are available. Intended Use These raster datasets are intended for: Small-area estimation research; Child mortality surveillance; Spatial epidemiology; Public health planning; Health inequality assessment; Geospatial health modeling; Reproducible methodological research. Citation Please cite this dataset as: Bayuh Asmamaw, H. (2026). Small-Area Estimation of Under-Five Mortality in Ethiopia at 5 km Resolution (2000–2015): Bayesian INLA-SPDE Mortality Probability and Death Count Surfaces. Zenodo. https://doi.org/10.5281/zenodo.20524950 License Creative Commons Attribution 4.0 International (CC BY 4.0) Code Availability Source code and supporting materials are available through: https://github.com/Bayuh23 Author Information Bayuh Asmamaw HailuORCID: https://orcid.org/0000-0002-7810-2774 Email: [e-mail] Acknowledgements This work uses data obtained under authorization from The DHS Program. Geospatial covariates were derived from publicly available sources including ERA5, the Global Burden of Disease Study (GBD), MODIS, WorldPop, and the Malaria Atlas Project. The resulting mortality estimates were generated using Bayesian spatial modeling techniques implemented in R with INLA and SPDE methodologies.