The widespread arsenic concentration in groundwater in Bangladesh has been recognized as posing a serious health problem to millions of people. So the reliable map of showing the level of arsenic concentrations are required to help plan for safe drinking water. This study extend a part of the analysis of the BGS’1999 report by demonstrating how the model-based geostatistical framework (Diggle et al., 1998) can be adapted to develop a predictive model to obtain a reliable map showing level of arsenic concentrations in groundwater of Bangladesh. Data used in this study are obtained from the BGS’1999 which are publicly available at http:www.bgs.ac.uk/arsenic. This study consider the conditionally defined transformed linear Gaussian geostatistical model with the Box-Cox transformation parameter (i.e., log-transformation) to fit the data based on the likelihood approach and the Bayesian approach.

The evidence of this study reveals that the spatially varying mean, which is a trend on the longitude, latitude and tubewells depth, is able to explain a substantial proportion of the spatial variation in the observed arsenic concentrations. Thus in prediction problem of arsenic concentrations it is able to obtain predictive maps or other properties of predictive maps for different levels of depth.

In the results of this study we obtained the predictive map for different levels of depth and it reveals that the arsenic concentration is much lower in deeper tubewells. As this study provide the predictive map of arsenic concentrations it will be more useful for understanding of the arsenic concentrations in groundwater of Bangladesh in different levels of depth.