Animal husbandry is widely practiced on the household scale in communities in low- and middle-income countries (LMICs) and, while having economic and health benefits, exposes household members to risk of zoonotic infections to an extent that is unclear. While demand for georeferenced information on infectious disease risk factors and drivers is growing, spatial variation in livestock ownership remains poorly characterized at high resolution. This study aimed to use geostatistical methods to model and map the prevalence of livestock husbandry in LMICs for three major animal taxa: poultry, swine, and ruminants. Microdata relating to ownership of livestock animal species were sourced from various population-based survey programs, which together cover the majority of LMICs, and categorized. These were georeferenced and spatially matched with a panel of time-fixed environmental and demographic spatial covariates. INLA models were fitted to the resulting database, and probabilities for ownership of each livestock taxon were predicted based on the model parameter estimates. The results indicated widespread poultry ownership across rural Central America, the Amazon basin, tropical Africa, and river basins and forests of East Asia. Swine husbandry is the least widely practiced among the three livestock taxa and concentrated in an undulating belt of higher prevalence extending from central China, through southeast Asia to Northeastern India. Rearing of ruminant livestock appears widespread across subequatorial Africa, Central Asia, the Gobi Desert, the Himalayas, Mongolia, and northern India. The models perform impressively by most standard evaluation metrics, and the patterns in their predictions align with external evidence. The distribution of this important risk factor for infectious disease transmission can be modeled using publicly available data sources to generate plausible and potentially actionable predictions over wide geographic areas and identify regions of high exposure to animal disease reservoirs. The resulting predicted prevalence estimates are made available as supplementary files in GIS-compatible format.

Changes after Oct 14, 2025:

These updated files are now derived from models fitted using R Template Model Builder (RTMB), rather than INLA, and fitted to the whole Plan-EO Dem database, rather than a subset of one sample per cluster as was the case in the previous iteration. Full methods are set out in our now published article. Households were georeferenced using an automated algorithmic cluster location assignment process. Three spatially explicit logistic models were fitted, one to each of the binary outcome variables with a set of spatial covariates using the RTMB package in which locations are projected onto a coarsened mesh of vertices that carry the spatial information and can be reprojected onto the observed data. All coordinates were transformed via the Mollweide projection and scaled into kilometers prior to analysis. The meshes used for modeling poultry and ruminants had 29,926 vertices, while the swine mesh had 29,580. Predicted probabilities for each livestock outcome were made for all locations in the domain of interest at 0.05 decimal degree resolution (~5.6 km at the equator) and exported in Georeferenced Tag Image File (GeoTIFF) format based on a polygon boundary shapefile for all LMICs obtained from the U.S. Department of State—Humanitarian Information Unit (available in the public domain with no restrictions, CC0). The spatial covariates along with a log-transformed time variable were used to generate predicted logistic distribution probabilities of a household reporting the presence of animals of that taxon from the corresponding model.