Environ Ecol Stat DOI 10.1007/s10651-013-0269-9 Predicting arsenic concentration in groundwater of Bangladesh using Bayesian geostatistical model Paritosh K. Roy · Syed S. Hossain Received: 3 July 2012 / Revised: 4 December 2013 © Springer Science+Business Media New York 2013 Abstract The pattern of the spatial variation in arsenic concentration in groundwa- ter of Bangladesh is usually needed for the planning of safe drinking water. Often a model-based prediction is required for this purpose. In this paper, we fit a Bayesian hierarchical geostatistical model by utilizing data from the project, ‘Groundwater studies of arsenic concentration in Bangladesh’ conducted by the British Geological Survey and the Department of Public Health Engineering of Bangladesh. We also develop a predictive model for arsenic concentration at different levels of well-depth using the same approach. The resulting predictive model has been cross-validated by appropriate statistical tools. Finally, we obtained reliable spatially continuous predic- tive maps and predictive probability maps showing the areas with high probability of arsenic concentration for different levels of well-depth. Results indicate that our model fits the data well and captures a substantial amount of spatial variation. More- over, well-depth is found to have a significant contribution in explaining the observed variation in arsenic concentration. The predictive maps that have been produced are observed to be different for various levels of well-depths and are expected to be helpful to the policy makers in preparing proper regional planning for safe drinking water. Keywords Arsenic concentration · Bayesian prediction · Geostatistical model · Predictive probability map · Spatially continuous map · Spatial process Handling Editor: Pierre Dutilleul. P. K. Roy (B) · S. S. Hossain Institute of Statistical Research and Training (ISRT), University of Dhaka, Dhaka 1000, Bangladesh e-mail: pkroy@isrt.ac.bd S. S. Hossain e-mail: shahadat@isrt.ac.bd 123