Consultancy Corner Sketch

For this month’s Consultancy Corner workshop, Mike Dunbar (Environment Agency), mike.dunbar@environment-agency.gov.uk) presented some challenges of fitting statistical models of pollutant concentrations using spot sample water quality monitoring data from rivers:

  • General Additive Models (GAMs) are a robust approach for modelling seasonality, multi-year trends (arising e.g. changing effluent treatment regulations) and datasets with concentrations below analytical limits of detection. River flow can be added as a covariate – allowing bivariate smoothing with time for longer-term dynamics, such as regulation changes. However, this approach doesn’t explicitly distinguish between contributions from different sources of pollution e.g. point or diffuse sources

    • The Load Apportionment Model (LAM) does explicitly model the contributions of point and diffuse sources as power law functions of flow. It could benefit from more robust statistical formulation for limits of detection, uncertain/time-varying parameters and explicit residual distributions, and comparison with the above GAM approach.

  • We are planning a masters project to investigate these topics further.