Sampling Methods A total of 50 ponds were sampled during three summer field campaigns between 2014-2016. The ponds were spread across three polar ecosystems (all above 68 degrees latitude). Specifically, we sampled ponds in tundra and larch forest ecosystems in northeastern Siberia as well as ponds located in a mire in northern Sweden (Fig. 4). Sampled ponds at each location were chosen at random. Standard sampling techniques were used across all locations to measure dissolved GHG gas concentrations in addition to physical and environmental variables including: pond depth, surface area, water temperature, dissolved oxygen (DO), pH, conductivity, dissolved organic carbon (DOC), total dissolved nitrogen (TN), and sediment organic carbon content (SOC). Importantly, while the research campaigns took place over different years, only samples from July were compared when doing cross-ecosystem type analyses.
Fig 4. The ponds sampled in this study were located across the circumpolar region including sites in northern Sweden and northeastern Siberia (marked in red). Unique ecosystem types within those locations are also marked.
Fig 5. To estimate percent cover for plant functional group, we used 1x1m quadrants (triplicate reps) within the pond and outside of the pond. Quadrants were placed randomly in both locations. White squares represent an example of the locations where we made estimates.
Additionally, each sampling location supported unique environmental variables that are also assessed in this study. In the tundra site, the ponds were located along an elevation gradient going down a hill-slope towards the Kolyma River (Fig 6-left). In the larch forest site, three sub-sites were identified based on the intensity of recent wildfire activity in those locations (Fig 6-right). Sub-sites identified include a low-severity burn, high-severity burn, and an unburned site. Lastly, in the mire, vegetation communities within and surrounding the ponds were also accounted for using a 1x1 meter quadrant and percent cover analyses (Fig. 5). Additionally, the ponds in the mire were measured over the entire summer, not just July, so in mire-specific analyses, initial data exploration was done to look at the effects of time on pond characteristics.
Fig 6. Left) A map of the tundra sampling site. Green dots represent ponds. The ponds were sampled going down a hillslope towards the Kolyma River (upper left corner). Center) A map of the greater northeast Siberia sampling region which encompasses both the tundra and larch forest sites. RIght) Diagrams representing the 3 sub-sites within the larch forest site and examples of typical ponds found in those sites.
Multivariate Analyses For the statistical analysis, I used Principle Component Analyses (PCA), permutational multivariate analysis of variance (perManova), and clustering. PCA involves the combination of variables, transformed orthogonally and ranked by highest variance, in order to explain the data in fewer “principle components”. The direction of vectors, describes how related those variables are. The proximity of individual data points indicates how related data points are. I used PCA to look at the relationships between different physical and chemical variables, with an emphasis on specific relationships with methane and carbon dioxide. I used the location of individual data points, which represent ponds, to examine how individual ponds form into groups.
PerManova tests multiple variables and their interactions within groups at the same time in order to test for differences between groups. The technique utilizes distance-based measures and randomly shuffles the variables within the data set many times to estimate a p-value. I used perManova to test for differences in pond characteristics between the burned and unburned larch stands. Clustering uses distance matrices to group alike observations based on similarities and can be used to identify groups within a data set. Using the Bray-Curtis distance, ponds were grouped in the mire based on vegetation communities (percent cover) both within and surrounding each pond. Finally, while not a multivariate technique, I then used repeated measures ANOVA to test for differences in methane concentrations between the selected groups since measurements were taken many times over a summer.