Mercury biomonitoring in freshwater fish is foundational for environmental science, public health, and Indigenous community well-being, given mercury’s persistence and toxicity. Accurate characterization of length–mercury relationships is central to environmental monitoring, risk assessment, and the development of fish consumption guidelines. However, most monitoring programs rely on a single default model, typically a log–log or power regression, which may misrepresent true patterns across heterogeneous lake–species combinations. This study introduces a decision-based regression framework that evaluates multiple candidate models using predefined statistical criteria and incorporates sensitivity analyses to assess model stability. Applying this framework to community-based monitoring data from northern Ontario revealed substantial variability in the form and strength of length–mercury relationships. No model type was universally optimal; several lake–species groups exhibited weak or absent relationships, indicating that automatic regression-based approaches can produce misleading estimates. Sensitivity analyses (leave-one-out cross-validation and outlier diagnostics) identified model fragility in data-limited or biologically heterogeneous groups, highlighting the need for explicit uncertainty evaluation. This flexible and transparent approach improves methodological rigor, supports defensible ecological interpretation, and strengthens mercury exposure estimates. This framework reduces the risk of biased exposure estimates, strengthens the scientific defensibility of consumption guidelines, and provides a reproducible, adaptable modeling workflow that can be adopted across environmental monitoring programs, particularly those working in northern, remote, or community-led contexts.
Fleming et al. (Mon,) studied this question.