Spatial structure in urban water consumption: A Bayesian space-time analysis of Bogotá (2019–2024)

Abstract

In this study, we apply fully Bayesian spatio-temporal models to locality-level water consumption in Bogotá, Colombia, from 2019 through 2024. An additive model identifies spatial differences after adjusting for common temporal dynamics and city-wide climatic and socioeconomic time series. Because these covariates are shared across all localities, they provide temporal adjustment but cannot explain between-locality spatial variation. A formal sensitivity analysis adding a structured space–time interaction substantially improves model fit (WAIC: 604.0 versus 2502.7; DIC: 333.9 versus 2500.5), showing that locality deviations are not adequately represented as constant throughout the study period. Student- t likelihoods indicate heavy residual tails (posterior mean degrees of freedom: 3.02 in the additive model and 2.60 in the interaction model), but the Gaussian interaction model retains substantially better information criteria than its Student- t counterpart. The principal spatial main effects remain highly correlated across likelihoods ( r  = 0.975). The results therefore support spatially heterogeneous temporal dynamics and show that the main spatial ordering is robust to heavy-tailed errors.