Results & Discussion
Landscape factors affect different water biogeochemistry parameters
Multivariate regression trees identified different landscape factors as drivers of each group of lake biogeochemistry. Thermokarst lake type, vegetation cover, and ecodistrict were considered explanatory landscape variables, while ground ice and permafrost zone did not appear to have a strong influence on lake biogeochemistry.
For ease of interpretation, landscape variables were shortened to 1-3 letter codes (tables 5-7).
Table 5. Codes for thermokarst lake types and their corresponding description.
Table 6. Codes for vegetation types and their corresponding description.
Table 7. Codes for ecodistrict and their corresponding description.
Thermokarst Type & Vegetation drive organic matter composition in northern lakes
Figure 10. A multivariate regression tree depicting the landscape drivers of dissolved organic matter composition in northern lakes.
Thermokarst landform is the strongest driver of dissolved organic matter composition, with the first split explaining ~20% of variation observed across lakes (figure 10). Vegetation cover was also an important driver of dissolved organic matter composition. Ground ice and permafrost zone did not appear to influence dissolved organic matter composition.
As humic-like compounds are typically terrestrially-sourced, lakes in Group 1, 2, and 3 (thermokarst types alluvial plain lakes, bedrock lakes, quarry lakes, and terrain-aligned lakes) likely have high land-water connectivity and/or surface runoff into lakes.
The first two principal components explained 81% of variation in dissolved organic matter composition across lakes (figure 11).
Variation along the first component describes a continuum from more labile to more humic organic matter. Variation along the second component is strongly associated with dissolved organic carbon concentration.
This PCA shows relatively low within-group variation of clusters produced from the MRT.
Figure 11. A principal component analysis showing variation in dissolved organic matter composition across northern lakes. Each point represents one lake. Colours of points and ellipses (75% confidence) represent the group derived from the multivariate regression tree (figure 10).
Thermokarst Type & Ecodistrict drive physico-chemical lake properties
Figure 12. A multivariate regression tree depicting the landscape drivers of physico-chemical properties of northern lakes.
Thermokarst landform is the strongest driver of physico-chemical properties, with the first split explaining ~31% of variation observed across lakes (figure 12). Ecodistrict is also an important driver of physico-chemical properties, with the first two splits explaining nearly half (47%) of the variation observed in lake physico-chemical properties.
Ground ice and permafrost zone did not appear to influence physico-chemical properties.
Interestingly, vegetation cover was not a strong influence over physico-chemical properties, which includes nutrient content. Vegetation cover did however distinguish groups 4 and 5, which varied in their nutrient content.
Variation in physico-chemical properties is likely linked to groundwater input. Quarry and terrain-aligned lakes (group 6) are characterized by high cation concentrations, which are typically associated with high groundwater input.
The first two principal components explained 65% of variation in lake physico-chemical properties (figure 13).
Variation along the first component is related to cation concentration. Variation along the second component is strongly associated with nutrient (nitrogen and phosphorous) content.
This PCA shows relatively low within-group variation of clusters produced from the MRT.
Figure 13. A principal component analysis showing variation in physico-chemical properties across northern lakes. Each point represents one lake. Colours of points and ellipses (75% confidence) represent the Group derived from the multivariate regression tree (figure 12).
Ecodistrict is the main driver of trace metal concentrations in northern lakes
Figure 14. A multivariate regression tree depicting the landscape drivers of trace metal concentrations in northern lakes.
Ecodistrict is the strongest driver of trace metal content, with the first two splits explaining 32% of variation observed across lakes (figure 14). Thermokarst landform was the second strongest driver than ecodistrict, although to a much lesser extent, appearing in the third split.
Ground ice and permafrost zone did not appear to influence trace metal concentrations.
The relatively high levels of heavy metals seen in the Great Slave area (ecodistricts Great Slave Plains (GSP) and Great Slave Uplands (GSU)) are likely tied to mining legacies around the city of Yellowknife. High levels of arsenic and other heavy metal contaminants have been reported in the area over the past 20 years. These ecoregions are also in the discontinuous permafrost zone and high trace metal concentrations may be reflective of deeper water flow pathways.
The first two components explained 57% of the variation in trace metal concentrations across lakes (figure 15).
Variation along the first component is primarily associated with aluminum and titanium concentrations. Variation along the second component is primarily associated with rubidium and arsenic concentrations.
Figure 15. A principal component analysis showing variation in trace metal concentrations across northern lakes. Each point represents one lake. Colours of points and ellipses (75% confidence) represent the Group derived from the multivariate regression tree (figure 14).
High level of microbial activity likely in subset of lakes
To identify lakes that are likely to have a high level of microbial activity, lakes with high dissolved organic carbon concentrations, high nutrient concentrations, and more labile (protein-like) dissolved organic matter were targeted.
Lakes that fall under Groups 4, 5, and 7 in the organic matter regression tree (figure 10) and Group 5 in the physico-chemical properties regression tree (figure 12) therefore are likely to have high microbial activity.
Of 68 sampled lakes, 16 lakes meet this criteria. These lakes form two distinct clusters: one in the northern mainland Northwest Territories and one in the south (figure 16).
Figure 16. Lakes across the Northwest Territories, Canada (n=68) expected to have greater levels of microbial activity (red) and lower levels of microbial activity (white) based on water biogeochemistry and landscape variables. Lakes expected to have high microbial activity formed two distinct clusters: one in the north (panel B) and one in the south (panel C).
Interestingly, the two clusters of lakes identified as likely to have high microbial activity did not share any of the measured landscape characteristics (table 8), suggesting water biogeochemistry is driven by multiple landscape factors rather than one dominant factor. However, both clusters of lakes have landscape features that suggest high land-water connectivity. Discontinuous permafrost in the south cluster would allow for more movement of surface and subsurface water in the surrounding landscape, similar to the wetlands observed in the north cluster. As northern climates continue to drive permafrost thaw, land-water connectivity may increase in lakes that currently have low movement of surface and subsurface waters. This could increase the number of lakes with water biogeochemistry that can support high microbial activity.
Table 8. Landscape descriptors of two clusters of lakes identified as likely to have a high degree of microbial activity based on water biogeochemistry.
Figure 17. Satellite imagery of lakes in the Tuktoyaktuk Peninsula Coastal Plains (A) and the Great Slave Lowlands (B) ecodistricts in Northwest Territories, Canada. Lakes that are expected to have greater levels of microbial activity are identified with a red point and lakes expected to have lower levels of microbial activity are identified with a white point, based on water biogeochemistry and landscape variables.
Satellite imagery was used to further explore distinguishing features between lakes geographically related, but identified as having different potential for microbial activity. In the northern cluster of high activity lakes (panel A of figure 17), lakes expected to have high microbial activity appear to be smaller than those expected to have lower activity. This suggests that lake shoreline and surface area may be important drivers of lake water biogeochemistry; both variables could be determined with spatial techniques that would not require expensive on-the-ground fieldwork.
In the southern cluster (panel B of figure 17), lakes expected to have high microbial activity appear to be larger and have less distinguished shoreline boundaries than other lakes in the area. These shorelines could allow for greater land-water connectivity, resulting in the deposition of more nutrients and organic matter into the lake. In contrast, many lakes in this area that aren't expected to have high microbial activity have distinct borders likely reflective of bedrock depressions. These lakes likely have little surrounding vegetation and less interaction with the shoreline terrestrial material.
Our approach to predicting microbial activity in lakes is supported by evidence of algal activity (bright green around lake shores) in several of the lakes identified as likely to have higher microbial activity (red points in panel B in figure 17). Further sampling of microbial communities present in these lakes will be required to confirm the efficacy of these predictions.
Additional landscape factors required
While the landscape factors included in this study did explain much of the variation in lake water biogeochemistry, additional landscape variables should be included for a more robust approach. Surficial geology, past glacial histories, and permafrost depth would provide more information on the geological setting of lakes. Lake surface area and shoreline length may also be important landscape factors, as lake size appears to be a distinguishing factor in lake biogeochemistry between geographically related lakes (figure 17). Distinguishing lakes that are experiencing substantial thaw, such as landslides (figure 18), from those with more stable permafrost conditions may strengthen the classification system by providing more information on land-water connectivity of northern lakes.
Figure 18. A lake in the Northwest Territories with an adjoining permafrost thaw-driven landslide (outlined in red).
Take-Home Message
Northern lakes are complex, highly variable, and closely related to their landscape.
Grouping lakes by their landscape factors, particularly thermokarst feature, ecodistrict, and vegetation cover, can be an effective way to understand patterns in water biogeochemistry. However, additional landscape variables need to be explored to optimize this approach. The use of landscape features to predict lakes with high potential for microbial activity will reduce the need for resource-intensive sampling and help to identify areas to expect greater greenhouse gas outputs.
As permafrost thaw changes land-water connectivity across northern regions, landscape features will be an important tool for monitoring the resulting changes to lake biogeochemistry.