The Athabasca River basin makes up approximately 24 per cent of the province’s landmass, with the river flowing beside oil sands’ tailings ponds and open-pit mining operations. As a result, there is growing concern about whether and how the oil sand industrial development will impact the local and downstream aquatic ecosystems—home to 31 of Alberta’s 59 fish species. Alberta’s fish species are of great importance as an economic resource and essential to the aquatic ecosystem.

The trout-perch (Percopsis omiscomaycus) is a key indicator species in the Athabasca River. Understanding how environmental changes affect trout-perch’s health provides insights into how the oil sands may impact the Athabasca River ecosystem. Previous approaches to modelling fish health had moderate success when location and year information were used in the models. However, their predictive power was limited. The ability to make accurate predictions regarding the health of fish in this region and identifying features driving the variability is instrumental in the development of a longer-term environmental monitoring program.

Improving aquatic health monitoring models using statistical techniques

University of Guelph statistics professors Drs. Zeny Feng [1] and Lorna Deeth [2], along with Dr. Tim Arciszewski (Alberta Environment and Parks) and their former MSc student Patrick McMillan, are improving environmental monitoring models and techniques to predict trout-perch health in the Athabasca River. In contrast to classical approaches, the team focused on developing associations between key physiological measurements: growth (body weight) and organ development (liver and gonad weight); and Athabasca River water quality measurements: water nutrients, pH and metals.

With these measurements, the team used two statistical methods, least absolute shrinkage and selection operator (Lasso) and elastic net (EN), to systematically select water variables (e.g., pH) to see if they substantially improve or worsen the model’s ability to predict trout-perch health and then ensure that the model does not over- or undergeneralize the data. They showed that using Lasso and EN were able to select the available variables that were the best predictors of fish health.

Water quality measurements create better models for fish health

Through their analysis, Feng, Deeth and team identified that the presence and potentially the accumulation of nonessential metals (e.g uranium, tellurium, molybdenum, and antimony) in the Athabasca River may influence the growth and development of the resident fish. While the origins of these metals were not identified, the use of statistical methods such as Lasso and EN helped to improve the prediction of trout-perch health, specifically growth and organ development. These results have established a new standard for monitoring the health of aquatic species and set the stage for future research.

“River acidification by contamination is believed to pose a toxic risk to the fish species found in the Athabasca River,” says Feng. “It is interesting that pH was identified as a variable related to liver weight, which may imply that liver development is sensitive to the pH level of the water. Additional research is required to fully understand how the various factors we identified as predictors of fish health are entering the watershed and impacting Athabasca River fish development.”

Dr. Zeny Feng is an Associate Professor in the Department of Mathematics and Statistics.

Dr. Lorna Deeth is an Associate Professor in the Department of Mathematics and Statistics.

This work was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant and a grant from the Oil Sands Monitoring program via Alberta Environment and Parks. The information contained in this article does not represent an official position of the Oil Sands Monitoring program or its participants.

McMillian P, Feng Z, Deeth L, Arciszewski, T. Improving monitoring of fish health in the oil sands region using regularization techniques and water quality variables. [3] Sci of the Total Environ. 2021 Dec 10. doi: 10.1016/j.scitotenv.2021.152301