Improving Environmental Risk Assessment with Data Literacy and the Bayesian Regression Toolbox

From a data analysis perspective, environmental risk assessments (ERAs) are highly dependent on their domain context and data availability. The collection of appropriate data for an ERA can be challenging, as a variety of experimental and synthetic data are available, often with varying degrees of reliability. Identification of appropriate data sources is important, especially because expert curation of reliable data can be time and resource consuming.

For high quality ERAs, the understanding of the collected data is therefore crucial, especially the data-generating processes. Other important considerations are variations between experiments, incomplete data sets, or measurement limits. The evaluation of this information is important for choosing an appropriate modelling strategy, and, in the case of a Bayesian data analysis, the construction of meaningful priors.

Bayesian regression models (BRMs) offer great flexibility to address these challenges and can quantify uncertainties. Outside the Gaussian default, more appropriate probability distributions is available for count, survival, or concentration data. Information such as weights (reliability scores), censoring (values below measurement limits), or truncation (known lower or upper limits of a response variable) can also be included. This ensures that the data can be used "as is", without subjective data manipulation. More complex variations are possible, for example by accounting for non-linearity or by adding a hierarchical component ("random effects").

A good understanding of BRM results is necessary to evaluate the quality of the model, as well as to draw the right conclusions of the posterior model results for an ERA. This includes both the understanding, but also the visualization and communication of the results and their uncertainties.

The above points can be collectively described as data literacy. A stronger focus on this concept, together with BRMs, has the potential to greatly improve the quality of ERAs under regulatory frameworks, as well as improve a shared understanding of the involved stakeholders.