Bayesian multivariate normal modelling

Abstract

Water pollution is an important global problem. Mostly, it is contaminated by excessive artificial chemicals from human activities. As a result, it will be harm to people and ecosystems. In order to prevention, we need to estimate the quantity of these pollutants which have been circulated along water cycle and end up at water resources.
We use the data provided by Christchurch Harbour Macronutrients Project in the UK and and the Hubbard Brook Ecosystem Study in US to construct a statistical model describing the amount of chemicals in rivers and precipitation. Moreover, we have to deal with uncertainties which affect to chemical concentration such as seasonal, storm, etc. In addition, examining the water sample is expensive then the most collected data are limited and ecology data itself has many specific problem such as extreme outlier, missing value, or truncated data. This uncertainty may be dealt with the Bayesian hierarchical model.
We are focusing on constructing Bayesian multivariate Normal model to use the most information based on geography of catchment area. As a result of fitting Bayesian multivariate normal model, the model has better performance comparing with individual models. Moreover, it shows a better performance if more sources are included into the same Bayesian Multivariate normal model. In addition, the annual chemical loads are calculated from the selected model including with credible intervals of total loads which can not be obtained from individual models.

More information

Identifiers

ORCID for Sujit Sahu: orcid.org/0000-0003-2315-3598

Catalogue record

Export record