The relationship between groundwater and its receiving environment is of particular interest
in the Alberta oil sands region (OSR) where industrial operations have the potential to affect
both the quality and quantity of groundwater resources via e.g., landscape disturbance,
groundwater withdrawals, and tailings pond seepage. Despite groundwater’s importance to
natural environments and species communities, monitoring of these interactions has been
limited. Groundwater dependent ecosystems (GDEs) are ecosystems that are maintained by
direct or indirect access to groundwater, and rely on the flow or chemical characteristics of
groundwater for some or all of their water requirements (Rohde et al., 2017). Here, we present
the results from the first year of a literature review and modeling effort to map aquatic GDEs
within the OSR.
Our first year literature review included three components: (1) groundwater indicators of GDEs; (2) biological indicators of GDEs to support mapping, with a focus on aquatic environments; and (3) empirical methods for mapping GDEs.
We identified and collated available geographic, geologic, hydrologic and landcover data for
mapping GDEs. Over 50 datasets were identified, with over 40 datasets compiled. From the
available data, we selected appropriate data to serve as training & validation data and
explanatory variables in MLMapper model and identified data gaps. The key data gaps are
access to the McKay River Integrated Surface Water-Groundwater Model, hydraulic head data,
and higher resolution thermal data, among others
Based on the results of the three literature reviews and data compilation, we undertook a
machine-learning based modeling approach in the McKay and Steepbank River watersheds
using a variety of topographic, hydrogeologic, and wetland/vegetation predictor data, with
indicators from the literature review informing variable selection. Final variables included in
the modeling were aquifer hosting sediment, bedrock, depth to water, elevation, flow
accumulation, normalized difference vegetation index (NDVI), permeability, wetness index, slope, soil drainage, and wetland class. Model fit of the top-performing models, as assessed by internal cross-validation, was very high. Outputs from the top five models were averaged into a final ensemble model of GDE probability.
The GDE maps identify lower river reaches, riparian areas, and wetlands (e.g., fens) as GDEs, but do not capture lakes, likely due to the lack of training data in the modeling pipeline. Upland areas are mostly categorized as non-GDEs. We conclude with suggestions for next steps in model development and application, as well as for potential improved or additional datasets that could be integrated going forward.