Influence of Wildfire on Downstream Transport of Dissolved Carbon, Nutrients, and Mercury in the Permafrost Zone of Boreal Western Canada
This article was corrected on 16 OCT 2023. See the end of the full text for details.
Abstract
Northern regions are undergoing rapid change with wildfires increasing in frequency and severity alongside thawing permafrost and altered water balance. These disturbances could cause significant change in the export of carbon, nutrients, and metals to aquatic systems, with implications for food webs and ecosystem processes. Here, we examine chemical data from a series of 52 streams and rivers that were sampled across a 250,000 km2 expanse of the Taiga Plains and Taiga Shield ecozones of the Northwest Territories, Canada. Samples were collected immediately after and for 3 years following a “megafire” that occurred in this region in 2014, and included wildfire-affected and non-affected catchments. While wildfire has been observed to cause significant impacts on water quality in other regions, we here report weak relationships with percent watershed burn with minor to moderate effect sizes, the greatest being a reduction in dissolved organic carbon (−32% concentration). Watershed-specific properties were a strong driver of large spatial variability in stream water chemistry, which may overwhelm or obscure lesser wildfire effects. The watershed chemical yield-specific response to wildfire was weaker than the response for concentrations, due to substantial variation and uncertainty in runoff among sites and years. This suggests that watershed chemical yields in this region are more sensitive to changes in water balance due to climate than to altered wildfire regimes.
Plain Language Summary
Our study examines how wildfires affect the chemistry of water in northern regions. We collected data from 52 streams and rivers in the Taiga Plains and Taiga Shield ecozones of Canada. The results show that the effects of wildfires on water quality are relatively small. Other factors, such as climate and the presence of wetlands, have a much stronger influence on water chemistry in each ecozone. Understanding these changes in water chemistry is crucial for managing aquatic ecosystems. However, we found that the impact of wildfires on water quality in this region differs from that in other locations. The variability in runoff among different sites and years also adds uncertainty to our findings. Looking ahead, climate change and altered hydrology may have more substantial impacts on fluvial networks compared to wildfires in this rapidly changing region. Our research provides valuable insights into the relative importance of various landscape factors in regulating water quality, including the effects of wildfires. By examining the chemical changes occurring due to wildfires and considering the broader environmental context, we contribute to a better understanding of how these factors influence water quality in northern regions.
Key Points
-
Wildfires in the Taiga regions of Canada had modest effects on water chemistry, with the largest impact being a reduction in DOC (−32%)
-
Variability in stream water chemistry was driven by watershed-specific properties overshadowing the influence of wildfires
-
Future impacts on streams are likely more substantially driven by climate and altered hydrology than increasing wildfire frequency
1 Introduction
The Boreal biome is currently experiencing pronounced changes and is warming faster than the global average (Stocker et al., 2013). As a consequence, wildfires in this region are increasing in frequency and severity (Coogan et al., 2019; Coops et al., 2018). In addition, rising temperatures are causing permafrost to warm and thaw (Biskaborn et al., 2019) which is further enhanced by wildfire (Gibson et al., 2018). Together, wildfire and permafrost thaw change hydrologic flow paths (Connon et al., 2018; Walvoord & Kurylyk, 2016; Haynes et al., 2018) influencing chemical fluxes to aquatic systems (Betts & Jones, 2009; Spence et al., 2015; Toohey et al., 2016). These changes in the delivery of carbon (Abbott et al., 2014; Larouche et al., 2015; Wauthy et al., 2018), nutrients (Smith et al., 2011; Spencer et al., 2015), and metals (Garcia & Carignan, 2011; St. Pierre et al., 2018) to aquatic systems have implications for food webs and ecosystem processes (Bixby et al., 2015). Thus, understanding how wildfire influences watershed chemical export in permafrost regions is critical to predicting future impacts on biogeochemical cycling and water quality.
The permafrost zone of Boreal western Canada (Taiga Plains and Taiga Shield ecozones) is an exemplar of the rapid changes occurring in the Boreal biome, with warming occurring at three times the global average (Stocker et al., 2013). Further, increasing wildfire frequency and associated permafrost thaw are particularly pronounced in this region (Coops et al., 2018; Gibson et al., 2018). This region is underlain by sporadic to discontinuous permafrost, but recent thaw has caused drastic changes to the landscape, including the conversion of forests to wetlands, and increased wetland connectivity that has both drained bogs and increased drainage network connectivity and discharge (Carpino et al., 2018; Haynes et al., 2018; Connon et al., 2014). Change across this landscape has been further exacerbated by a “megafire” in 2014, which burned 28,500 km2 of land, accelerating thaw by removing insulative surface soils drastically altering vegetation (Walker et al., 2018) and ground surface spectral properties (Ackley et al., 2021) across burned areas.
Despite the accelerated changes occurring in the western Canadian Taiga, investigations of impacts on water chemistry are rare (Burd et al., 2018; Coleman et al., 2015; Spence et al., 2015, 2020), especially at large scales. Changes in lateral carbon export in this region are highly relevant, given that the western Canadian Taiga hosts some of the highest concentrations of soil organic carbon per unit area on Earth (Tifafi et al., 2018), while aquatic systems function as critical reaction sites within the global carbon cycle (Cole et al., 2007), particularly in the face of thaw (Plaza et al., 2019). Additionally, the changing export of nutrients can have important impacts on aquatic food webs (Guigue et al., 2016) and broader biogeochemical cycles (Elberling et al., 2010). Furthermore, permafrost stores large amounts of mercury (Schuster et al., 2018) and there is a risk that mercury will be mobilized by thaw or fire (Coleman et al., 2015; Fraser et al., 2018; Gordon et al., 2016). Mercury released into the aquatic system can biomagnify through food webs (Selin, 2009), with subsequent consequences for northern residents who consume upper trophic level organisms (Laird et al., 2013). Yet, little is known about water chemistry in the changing, vast (area of 1.3 million km2) and remote western Canadian Taiga. This lack of knowledge is compounded by the fact that the region is divided into two contrasting ecozones (the Taiga Shield and Taiga Plains), which are defined by different quaternary geologies, bedrock types, landforms, and vegetation, which may in turn regulate the chemical response to wildfire and thaw. For example, the Taiga Shield's abundant lakes and thin soils may lead to different hydrological and biogeochemical processes than the Taiga Plains' widespread peatland complexes and thick sedimentary deposits. These differences could influence nutrient cycling, export, and uptake, as well as the fate of contaminants like mercury. Additionally, the varying connectivity of streams and lakes between the ecozones could further influence how chemicals are transported and transformed across the landscape. Overall, a better understanding of the potential ecozone-specific effects of wildfire is needed to improve our ability to predict and manage their impacts on aquatic ecosystems in this region.
In this study, we present aquatic chemical data from 52 streams and rivers over an area of 250,000 km2 across the Taiga Shield and Taiga Plains of the Northwest Territories (Canada). Data collection for this work occurred over a 3-year period, beginning during the initial runoff pulse following the 2014 megafires. We investigate patterns in dissolved organic carbon (DOC), nitrogen, phosphorous, and mercury concentrations across ecozones and fire-disturbed and undisturbed watersheds. We then estimate watershed chemical yields (area-specific export) for these chemical species, to provide a baseline of current conditions, allow extrapolation across this broad and understudied region, and explore more broadly the potential impacts of wildfire within this rapidly thawing landscape.
2 Materials and Methods
2.1 Sampling and Site Description
A total of 52 streams and rivers were sampled multiple times from 2015 to 2017 across the Northwest Territories, Canada (Figure 1; Hutchins et al., 2020; Tank et al., 2019). During each of summer 2016 and spring 2017, we sampled 50 sites in large-scale field campaigns. From spring 2015 to fall 2017, six sites (Table S1 in Supporting Information S1, four in common with the large-scale sites) were sampled more frequently (bi-weekly to monthly during the open water season) by the Government of Northwest Territories Department of Environment and Natural Resources. The study area was distributed over the Taiga Shield and Taiga Plains ecozones (Marshall et al., 1999; Figure 1), where 23 and 29 sites were sampled, respectively. A total of 24 sites had their watersheds burned (2% or more) in the 2014 wildfires. There were 212 samplings, with 106 in each ecozone, 74 in burned watersheds, and 138 in unburned.
The study region has a subarctic climate with snow cover beginning in October and lasting until the end of April or beginning of May and ∼40% of precipitation falling as snow. The bedrock geology of the Taiga Shield is characterized by mainly Archean/Precambrian intrusive and metamorphic rocks. Minimal surficial deposits are comprised of thin tills, with bedrock mainly overlain by thin soils extending to the bedrock. In contrast, bedrock of the Taiga Plains is dominated by Phanerozoic sedimentary deposits containing carbonate minerals. Bedrock on the Plains is overlain by thick glaciolacustrine and lacustrine sediments (Wheeler et al., 1996). The hydrology and streamflow generation in both the Taiga Shield and Plains is highly variable and complex but with different underlying processes. In the Taiga Shield, the aquatic landscape is dominated by chains of lakes that are subject to fill-and-spill hydrology, making flow highly variable as all upstream segments must fill to generate streamflow (Spence et al., 2014; Spence & Woo, 2003). In contrast, the streamflow network on the Taiga Plains contains interconnected bog complexes with highly dynamic streamflow transmission, whereby water is generated on peat plateaus and upland regions, and either captured in bogs or channeled to throughflow fens. The complex groundwater recharge and discharge processes in the Taiga Plains region, which involve both local and regional flow paths, can be influenced by factors such as vegetation cover, soil type, and climate (Winter 1999). Additionally, permafrost thaw in the Taiga Plains enables “bog capture” that causes partial drainage of bogs and increased discharge through downstream watershed components (Connon et al., 2015, 2014; Quinton et al., 2003).
2.2 Field and Laboratory Analyses
We measured pH in situ at each sampling occasion, and water samples were 0.45 μm filtered representing the dissolved fraction. Water samples for Hg analysis were collected into certified pre-cleaned glass amber bottles (Environmental Sampling Supply, Sam Leandro, CA, USA) using the clean hands/dirty hands sampling protocol and filtered through HCl-washed disposable Nalgene 0.45 μm cellulose nitrate filter towers. Filtrate was poured into new glass amber bottles, acidified to 0.2% with trace-metal grade HCl and kept cool until analysis. All filtered samples were analyzed in accredited laboratories (Canadian Association for Laboratory Accreditation) for concentrations of: DOC, total dissolved nitrogen (TDN), total dissolved phosphorous (TDP), dissolved total mercury (DHg), dissolved sodium (Na), dissolved chloride (Cl), dissolved calcium (Ca), dissolved magnesium (Mg), dissolved potassium (K), and alkalinity. Method specifics (instrumentation, Quality Assurance/Quality Control, and method detection limits) are included in the supplement.
2.3 Hydrometric Data and Runoff
For the 50 sites in large-scale field campaigns (2016–2017), velocity was measured at 60% depth at 0.5 m intervals across the streams and rivers using a handheld FlowTracker2 (SonTek, San Diego, CA, USA). Discharge was estimated using the velocity-area method, where velocity was multiplied by the cross-sectional area along each transect with the FlowTracker2 software. Instantaneous runoff for each site was calculated using the estimated discharge and divided by the estimated watershed area (determined as described in the section below).
Daily discharge for 17 hydrometric stations was obtained for the 2016–2017 period from the Water Survey of Canada (https://wateroffice.ec.gc.ca/). Eight stations were located in the Taiga Shield and nine in the Taiga Plains (Table S2 in Supporting Information S1). Annual runoff for each hydrometric station was calculated using annual discharge from summed daily measurements obtained from the Water Survey of Canada and divided by watershed area.
2.4 Geospatial Analyses
Watersheds were delineated with the hydrology toolbox in ArcGIS 10.6.1 using digital elevation models from ArcticDEM v2.0 (Porter et al., 2018) with a 5 m resolution. The resulting watershed polygons were used to calculate land cover (% exposed bedrock, water [lakes and ponds], forest, wetlands), peatlands, net primary production (NPP), watershed slope, elevation, and percent watershed burn area using the Zonal Statistics toolbox in ArcGIS. Landsat-derived land cover was from Natural Resources Canada (http://geogratis.cgdi.gc.ca/) and the extent of the 2014 wildfires was derived from the National Fire Database (NFDB) polygons obtained from Canadian Forest Service (https://cwfis.cfs.nrcan.gc.ca/). Peatland coverage was from the Peatlands of Canada map from the Geological Survey of Canada (Tarnocai et al., 2011). Average annual NPP (2000–2014) was from the National Aeronautics and Space Administration's Moderate Resolution Imaging Spectroradiometer (NASA MODIS) satellite product (Running & Mu, 2015). The ArcticDEM was used for elevation and to calculate slope in the Surface Statistics toolbox in ArcGIS.
2.5 Statistical Analyses and Chemical Yield Estimates
All statistics were performed using R version 3.5 (R Core Team, 2014). All variables were square root or natural log transformed to improve normality. Multilevel models (lme4 package; Bates et al., 2014) were used to examine the variance explained (variancePartition package; Hoffman & Schadt, 2016) by a series of levels (site, ecozone, month and year) and percent watershed burned, for each chemical species concentration and instantaneous runoff. Multilevel or hierarchical models are appropriate for this data set given its observational nature with the 212 observations nested in geographical areas and where responses are correlated over time and not independent (Hutchins et al., 2021). We report variance explained by each level in the multilevel models, which provides an estimate of effect size and relative importance.
We used constrained ordinations to further explore the multivariate responses of the chemical species to landscape properties. Canonical analysis of principal coordinates (CAP), a distance-based redundancy analysis using Bray-Curtis dissimilarity (vegan package; Oksanen et al., 2020), was used to explore the relationships between watershed-specific landscape properties, percent watershed burned, ecozone, and water chemistry. Input variables for CAP were site averages for each of the 52 watersheds. CAP has the advantage over other constrained ordinations of accounting for the correlation structure among the chemical species (Anderson & Willis, 2003).
Annual runoff was constrained using 100,000 iterations of Markov chain Monte Carlo (MCMC) from the 2016–2017 hydrometric station data (Data Set S1). Chemical yield estimates, the interquartile range, and 95% confidence intervals (from 2.5 to 97.5 percentiles) for each chemical species were calculated from 100,000 modeled concentrations and annual runoff in each group. Chemical yield estimates are reported for burned and unburned watersheds within either the Taiga Shield or Plains ecozones or pooled if these divisions did not explain any variance in the multilevel models (singular fits).
3 Results
Our sampling sites spanned 250,000 km2, and showed substantial variation in chemical species concentration between sites (Table 1). In the frequently sampled sites, chemical species' concentrations varied predictably with discharge, but did not show clear differentiation by burn status (Figures S1–S4 in Supporting Information S1). Overall, DHg and TDP site-averaged concentrations were higher in the Taiga Shield than in the Taiga Plains (Figure S5 in Supporting Information S1, Wilcoxon rank sum test, p < 0.01), while TDN was lower in the Taiga Shield (Wilcoxon rank sum test, p < 0.01), and DOC concentrations were similar between ecozones (Wilcoxon rank sum test, p > 0.05, Figure S6 in Supporting Information S1). In the Taiga Plains, all chemical species were similar between burned and unburned watersheds (Wilcoxon rank sum test, p > 0.5). In contrast, in the Taiga Shield, TDN and TDP concentrations were similar in burned and unburned watersheds (Wilcoxon rank sum test, p > 0.05), but DHg and DOC were higher in unburned watersheds (Wilcoxon rank sum test, p < 0.01). However, directly comparing concentration differences between ecozone and burn status has the limitation of not capturing confounding factors and multivariate relationships. As a result, we used multilevel models to estimate mean concentrations, which showed different patterns than the Wilcoxon rank sum tests (Table S3 in Supporting Information S1).
Unit | Taiga Plains | Taiga Shield | Combined | |
---|---|---|---|---|
Latitude | °N | 60.8–64.1 | 62.5–64.2 | 60.8–64.2 |
Longitude | °W | 116.2–121.5 | 113.8–117.3 | 113.8–121.5 |
Mean annual temperature (weather stations) | °C | −3 to −5 | 2 | 4 |
Mean annual precipitation (weather stations) | mm | 275–350 | 243–283 | 243–350 |
Number of rivers/streams | 29 | 23 | 52 | |
Strahler orders | 1–5 | 1–5 | 1–5 | |
Annual runoff (hydrometric stations 2016–2017) | mm yr−1 | 105 (77.7–137) | 109 (78.9–125) | 102 (65.8–183) |
Sampling time runoff (sites sampled 2016) | mm yr−1 | 84 (34–158) | 88 (24–156) | 84 (29–156) |
Sampling time runoff (sites sampled 2017) | mm yr−1 | 323 (123–486) | 201 (103–303) | 255 (105–285) |
Net primary production | g m−2 yr−1 | 351 (229–413) | 212 (194–247) | 249 (198–385) |
Watershed area | km−2 | 97 (26–326) | 4 (2–42) | 32 (4–180) |
Mean watershed elevation | m | 278 (257–287) | 286 (225–385) | 278 (248–310) |
Mean watershed slope | ° | 0.5 (0.4–0.7) | 4 (3–4) | 1 (0.4–3) |
Forest cover (from Landstat) | % | 50 (36–60) | 43 (19–61) | 49 (29–61) |
Shrub land cover (from Landstat) | % | 3 (1–9) | 17 (3–34) | 4 (2–16) |
Exposed rock (from Landstat) | % | 4 (2–10) | 10 (3–26) | 5 (2–14) |
Water (from Landstat) | % | 6 (2–8) | 11 (5–17) | 7 (3–14) |
Wetlands (from Landstat) | % | 24 (11–43) | 4 (1–9) | 11 (4–27) |
Peatland (from Tarnocai2011) | % | 49 (37–54) | 6 (2–10) | 25 (8–51) |
Percent watershed burned (from NFDB) | % | 0 (0–83) | 23 (0–94) | 6 (0–88) |
pH | 7.9 (7.5–8.2) | 7.0 (6.9–7.2) | 7.5 (7.0–7.9) | |
Alkalinity | 162 (116–183) | 43 (19–38) | 76 (26–164) | |
Mg | mg L−1 | 13.9 (7.9–24.7) | 2.5 (1.7–4.1) | 6.9 (2.7–15.4) |
Ca | mg L−1 | 49.4 (34.5–68.8) | 6.2 (5.5–9.5) | 26.7 (7.2–51.7) |
Na | mg L−1 | 2.5 (1.5–3.3) | 2.2 (1.7–4.1) | 2.4 (1.6–3.6) |
K | mg L−1 | 1.1 (0.6–1.7) | 1.3 (0.9–2.3) | 1.1 (0.8–1.9) |
Cl | mg L−1 | 0.8 (0.4–2.0) | 0.9 (0.5–4.2) | 0.9 (0.4–2.3) |
DHg | ng L−1 | 1.3 (1.1–1.7) | 2.1 (1.5–3.1) | 1.5 (1.1–2.1) |
DOC | mg L−1 | 16.6 (12.9–20.8) | 16.2 (10.0–23.5) | 16.5 (11.0–22.9) |
TDN | µg L−1 | 673 (558–783) | 511 (352–900) | 630 (829–283) |
TDP | µg L−1 | 5 (4–7) | 9 (5–13) | 6 (4–10) |
- Note. Median values of percentages may not add up to 100% and peatlands can overlap wetlands, forests, and shrub lands.
3.1 Partitioning Spatial and Temporal Controls on Water Chemistry
Using multilevel models, the observed variance in runoff, and chemical species' concentration was partitioned between temporal (month and year) factors, site, ecozone, and watershed burn status (Figure 2). Relatively little variance was explained by the temporal levels: the month or year of sample collection. Month explained less than 10% of variance for DOC (0% of variance explained), TDP (6.6%), Mg (3.8%), Ca (2.8%), alkalinity (5.1%), Cl (1.6%), and Na (10.0%), but explained greater variation in TDN (15.4%), DHg (14.3%), K (16.9%), and runoff (22.2%). Year of collection explained even less variance than month, with less than 1% of variance explained except for TDP (30.4%) showing TDP lower in the first year and higher in the latter. Site-specific variance was greater than 50% for DOC, TDN, Cl, and Na. In contrast, TDP (7.7%), DHg (18.3%), Mg (24.1%), Ca (15.5%), K (46.7%), alkalinity (15.1%), and runoff (40.5%) showed less variance partitioned by site. For Mg, Ca, and alkalinity, ecozone rather than site explained most of the variance, accounting for 67.8%, 79.0%, and 75.8% of total variance, respectively. The Taiga Plains with sedimentary deposits had greater Mg, Ca, and alkalinity than the Taiga Shield. Percent watershed area burned explained 5.5%, 1.4%, 0.1%, 0.2%, 1.1%, 12.3%, and 0.6% of the variance in DOC, TDN, TDP, DHg, Mg, K, and Na, respectively. Overall, the multilevel models explained more than 75% of the variance in most of the chemical species except for TDP, DHg, and runoff which had residuals of 46.1%, 44.1%, and 36.3%, respectively. In general, spatial factors (site and ecozone) were most important for explaining the variance observed, except for TDP, where temporal variance (year) was most important. Within the spatial factors examined, wildfire appeared to have a relatively modest effect.
To assess how stream chemical species responded to variation in landscape properties across watersheds, we applied constrained ordinations with watershed-specific landscape properties as the constraining matrix. In line with the high degree of variability in Mg, Ca, and alkalinity explained by ecozone in multilevel models, these chemical species loaded heavily in ordinations (Figure S7 in Supporting Information S1). Excluding Mg, Ca, and alkalinity from further ordinations allowed for an exploration of the association between the other chemical species and landscape properties unhindered by heavy constrained loading on the ordination axes. A CAP ordination of remaining dissolved chemical species concentrations (DOC, TDN, TDP, DHg, K, Cl, and Na) constrained to percent watershed burn area and ecozone explained 7.3% between CAP1 and CAP2 axes (Figure 3a). TDN scores were the furthest from the origin aligning in the same quadrant as the centroid of the Taiga Plains ecozone. The other chemical species (DOC, TDP, Hg, K, Cl, and Na) had scores that loaded weakly on both the CAP1 and CAP2 axes. Scores for DOC most closely aligned with the percent watershed burn area vector in the opposite direction. Hg and TDP scores were closest to the Taiga Shield ecozone centroid, while K, Cl, and Na were closer to the origin. Overall, watershed burn status and ecozone explained little variation in DOC, TDN, TDP, DHg, K, Cl, and Na concentrations in either multilevel models or CAP ordination.
Since the most variance in chemical species concentration was explained by site in our multilevel models, we performed a CAP ordination constrained to remotely sensed watershed-specific properties: mean slope, mean elevation, Landsat-derived land cover (% forest, % water, % wetlands, % exposed bedrock), proportion peat, and NPP. This approach resulted in constrained ordination axes CAP1 and CAP2 that explained 36.3% of the variation in chemical species concentrations across the landscape (Figure 3b). Ecozone and watershed burn status did not fit significantly to this CAP ordination using the factorfit function (R2 < 0.04, p value > 0.1, 9999 permutations); showing those factors occupy a different dimensional space than the watershed-specific properties included here, validating our use of these factors as different levels in the multilevel models. Axis CAP1 explained 31.6% of the variation in chemical species and was mainly constrained by slope and elevation, which was negatively associated with NPP, wetlands, and peat coverage. Slope and elevation were negatively associated with most chemical species (TDN, TDP, DOC, Cl, and Na) but close to neutral for K and Hg. TDN loaded most heavily on the CAP1 axis, followed by DOC; both were associated with higher NPP and greater proportion of wetlands and peat. Constrained axis CAP2 explained much less variation (4.8%) than CAP1. Further, CAP2 was most strongly constrained by watershed area, proportion exposed land, and proportion water and to a lesser extent mean slope in the positive half-axis; and proportion wetlands, peat and forest cover, NPP, and elevation in the negative half-axis. TDN, Cl, Na, and K loaded on the positive half of the CAP2 axis, whereas DOC and Hg loaded on the negative half with TDP close to neutral. Overall, lower elevation/sloped watersheds with greater NPP and wetland/peat coverage had greater nutrients (TDN and TDP) and DOC.
Using outputs from multilevel models, the relative magnitude of variation in constituent concentration from burn status alone was expressed as percent difference (Figure 4). K, Mg, and TDP showed estimated increases in completely burned watersheds but with very large uncertainty. Ca and Cl showed no difference with burn since this parameter offered no explanation of variance in the multilevel models. DOC (median modeled decline of 32%) showed the greatest potential decrease in burned watersheds, whereas TDN (18%), DHg (11%), and Na (18%) showed more modest median and interquartile range (IQR) declines, but with the 95% confidence interval (CI) of modeled outputs overlapping zero change. Overall, our models indicate a change in stream concentrations of most chemical species in the recently burned watersheds, but with broad uncertainty around estimates.
3.2 Assessing Runoff and Constituent Yields Across the Taiga Plains and Shield
Runoff is a key driver of the delivery of chemical species to the aquatic environment. Above, we included instantaneous runoff values from the sampling sites in the multilevel models to assess how runoff varied with spatial and temporal controls. However, the remote nature of our numerous sampling sites meant that continuous annual measurements were not available for most locations, and thus we obtained discharge data from surrounding hydrometric stations (Table S2 in Supporting Information S1). Annual runoff was highly variable with an IQR of 117 mm yr−1 (median 102) at the hydrometric stations, compared to 156 mm yr−1 (median 105) calculated from instantaneous measurements at the sampling sites (Table 1). There were no significant differences in annual runoff between the Taiga Shield and the Taiga Plains, using either 2016–2017 hydrometric data, or calculations using instantaneous runoff from sampling sites (Wilcoxon rank sum tests, p > 0.2). The variance in instantaneous runoff measured at sampling sites was mostly explained by month of sampling in the multilevel modeling; since the instantaneous variation would skew annual runoff estimates, the hydrometric station data was used in the chemical yield calculations as an analogous data set. These MCMC-estimated annual runoff values from the hydrometric stations (111, 95% CI: 87–137 mm yr−1, Figure S8 in Supporting Information S1) are similar to runoff for the last 30 years (median 118, IQR: 79–183 mm y−1 from 1987 to 2017, Figure S9 in Supporting Information S1) for all the hydrometric stations.
Using MCMC-estimated annual runoff (Figure S8 in Supporting Information S1) and chemical species concentrations from the multilevel models (Table S3 in Supporting Information S1), yields for DOC, TDN, TDP, DHg, K, Cl, and Na were estimated for the Taiga Plains and Shield ecozones (Table 2). Compared to the percent difference with burn status alone, incorporating runoff and its uncertainty resulted in a wide 95% CI of the yield estimates, and therefore significant overlap between burn status and ecozone groups for CI-based comparisons. Mg and Ca had the least overlap in 95% CI between ecozones, and—as assessed by their IQR—were both higher in the Taiga Plains than in the Taiga Shield. In contrast, the IQR for DHg indicated higher concentrations in the Taiga Shield than in the Taiga Plains. Yields for all other chemical species (DOC, TDN, TDP, K, Cl, and Na) showed broad IQR overlap between ecozones. When comparing burn status, the greatest difference in yields was observed in K with the burned watersheds 0.1 g m−2 L−1 (100%) higher than unburned watersheds in both ecozones, with no overlap in IQR. DOC showed a similarly large difference with the burned group 0.7 g m−2 L−1 (33%) lower than the unburned with no overlap in IQR. All other comparisons showed substantial IQR overlap. Notably, the Wilcoxon-associated differences in nutrients (TDN and TDP) across ecoregions were not evident when runoff and variance partitioning of chemistry were taken into account. Similarly, incorporating runoff into our calculations caused the IQR-assessed differences in nutrients, DHg, K, and Ca concentrations from the multilevel models to disappear in the resulting watershed chemical yields.
Chemical species | Ecozone | Burn status | Units | Estimate | Q1 | Q3 | CI 2.5% | CI 97.5% |
---|---|---|---|---|---|---|---|---|
DOC | Both | Burned | g m−2 yr−1 | 1.3 | 1.2 | 1.6 | 0.9 | 2.1 |
DOC | Both | Unburned | g m−2 yr−1 | 2.0 | 1.8 | 2.2 | 1.5 | 2.7 |
TDN | Plains | Burned | mg m−2 yr−1 | 78 | 66 | 93 | 47 | 129 |
TDN | Plains | Unburned | mg m−2 yr−1 | 95 | 83 | 109 | 63 | 141 |
TDN | Shield | Burned | mg m−2 yr−1 | 67 | 57 | 80 | 41 | 111 |
TDN | Shield | Unburned | mg m−2 yr−1 | 82 | 71 | 94 | 54 | 122 |
TDP | Plains | Burned | mg m−2 yr−1 | 0.49 | 0.3 | 0.70 | 0.18 | 1.4 |
TDP | Plains | Unburned | mg m−2 yr−1 | 0.46 | 0.3 | 0.64 | 0.18 | 1.2 |
TDP | Shield | Burned | mg m−2 yr−1 | 0.73 | 0.5 | 1.04 | 0.26 | 2.0 |
TDP | Shield | Unburned | mg m−2 yr−1 | 0.69 | 0.5 | 1.0 | 0.26 | 1.8 |
DHg | Plains | Burned | µg m−2 yr−1 | 0.12 | 0.09 | 0.15 | 0.05 | 0.26 |
DHg | Plains | Unburned | µg m−2 yr−1 | 0.13 | 0.10 | 0.17 | 0.06 | 0.27 |
DHg | Shield | Burned | µg m−2 yr−1 | 0.20 | 0.15 | 0.27 | 0.09 | 0.46 |
DHg | Shield | Unburned | µg m−2 yr−1 | 0.23 | 0.18 | 0.30 | 0.11 | 0.48 |
Mg | Shield | Both | g m−2 yr−1 | 2.0 | 1.1 | 3.5 | 0.38 | 10 |
Mg | Plains | Both | g m−2 yr−1 | 0.40 | 0.2 | 0.71 | 0.08 | 2.1 |
Ca | Plains | Burned | g m−2 yr−1 | 7.0 | 3.5 | 14 | 0.93 | 53 |
Ca | Plains | Unburned | g m−2 yr−1 | 7.2 | 3.7 | 14 | 1.00 | 52 |
Ca | Shield | Burned | g m−2 yr−1 | 0.99 | 0.49 | 2.0 | 0.13 | 7.5 |
Ca | Shield | Unburned | g m−2 yr−1 | 1.0 | 0.52 | 2.0 | 0.14 | 7.4 |
K | Plains | Burned | g m−2 yr−1 | 0.20 | 0.16 | 0.25 | 0.10 | 0.37 |
K | Plains | Unburned | g m−2 yr−1 | 0.10 | 0.09 | 0.13 | 0.06 | 0.18 |
K | Shield | Burned | g m−2 yr−1 | 0.20 | 0.16 | 0.25 | 0.11 | 0.39 |
K | Shield | Unburned | g m−2 yr−1 | 0.11 | 0.09 | 0.13 | 0.06 | 0.19 |
Cl | Both | Both | g m−2 yr−1 | 0.13 | 0.11 | 0.16 | 0.08 | 0.21 |
Na | Both | Burned | g m−2 yr−1 | 0.32 | 0.25 | 0.40 | 0.16 | 0.63 |
Na | Both | Unburned | g m−2 yr−1 | 0.39 | 0.34 | 0.45 | 0.26 | 0.59 |
- Note. Yield estimates are divided into burned and unburned landscapes within the Taiga Shield and Taiga Plains ecozones. Estimates, interquartile ranges (first quartile—Q1, third quartile—Q3), and 95% confidence intervals were calculated from multilevel models using Monte Carlo simulations with 100,000 iterations. Yields are grouped by ecozone and/or burn status if these parameters had singular fits in multilevel models (explained zero variance).
4 Discussion
We found modest influences of recent fires on stream water chemistry and watershed solute exports in Boreal western Canada. Our results show that the main driver of the variability among streams was watershed-specific properties. Our analysis used three complementary approaches—variance partitioning of multilevel models, constrained ordinations, and modeled chemical yields—to investigate patterns in carbon, nutrients, ions, and mercury across ecozones and fire-disturbed and undisturbed watersheds. While burned watersheds exhibited lower concentrations of DOC, TDN, and DHg and higher concentrations of TDP, Mg, and K (with a lesser impact on yields), the contribution of watershed burn to the overall between-site variation observed was limited. Instead, ecozone and site-specific properties were the main drivers of the variation in constituent concentration observed between sites, with substantial variation in runoff among sites and years causing the chemical yield-specific response to be weaker than for concentration. The substantial variability in runoff observed among our study sites and between years can be attributed to the unique climate and watershed characteristics of the Boreal region, in particular, highly variable water storage thresholds—with small lake chains in the Taiga Shield ecozone and bog cascades in the Taiga Plain. This variability highlights the importance of considering both concentration and chemical yield when assessing the impacts of wildfires on water quality in this region. Below, we discuss in more detail the drivers of DOC, DHg, TDN, and TDP across this subarctic landscape, and discuss the importance and implications of landscape chemical yields under altered fire regimes.
Little work has been done to study the effects of wildfire on surface water chemistry and watershed solute exports within Boreal permafrost regions. However, two recent studies shed some light on this issue. Pretty et al. (2021) found that while few water quality variables were affected by wildfire in the catchments of lakes in the Sahtú Settlement Area of the Northwest Territories, remote sensing and field observations suggested that macrophyte biomass was higher in lakes affected by burns. Ackley et al. (2021) examined the impacts of a low-severity wildfire on a permafrost plateau in the wetland-dominated landscape of the watershed of Scotty Creek—a stream in this study. Their findings showed an increase in solutes and DOC but also longer pore water residence times.
While wildfire effects on surface water chemistry and watershed solute exports in Boreal permafrost regions have not been well-studied, several studies from Boreal non-permafrost regions have reported variable effects on DOC concentrations. Wildfire has been shown to have variable effects on DOC concentrations, ranging from no (Lamontagne et al., 2000; Marchand et al., 2009; Olefeldt et al., 2013) to increased (Carignan et al., 2000; Emmerton et al., 2020; McEachern et al., 2000) or decreased concentrations (Betts & Jones, 2009; Rodríguez-Cardona et al., 2020). Whereas, nutrients (both N and P) have typically been shown to increase following wildfire (Abbott et al., 2021; Betts & Jones, 2009; Burd et al., 2018; Carignan et al., 2000; Kelly et al., 2006; Lamontagne et al., 2000; Marchand et al., 2009; McEachern et al., 2000; Rodríguez-Cardona et al., 2020), with the exception of one study in Minnesota lakes which showed no effect (McColl & Grigal, 1977). The potential impacts of wildfires on Hg concentrations and yields in the Arctic and subarctic have been largely unexplored, although Boreal studies have found either no effect (Garcia et al., 2007) or increased tissue accumulation caused by food web restructuring from increased nutrient loads (Garcia et al., 2007). However, in many of these studies, the lack of pre-wildfire measurements may have impeded the ability to observe a wildfire-associated effect.
Similar to our study, most studies investigating the effect of wildfires on surface water chemistry and watershed solute exports are observational, comparing burned and unburned watersheds in the same region (e.g., Betts & Jones, 2009; Burd et al., 2018; Carignan et al., 2000; Emmerton et al., 2020; Rodríguez-Cardona et al., 2020). In observational studies, establishing a causal relationship between watershed burn status and stream concentrations is difficult. Often, other intrinsic watershed-specific landscape properties that control the delivery of chemical species to the aquatic system are overlooked. Here, we show that watershed landscape differences represent confounding factors explaining most of the variation in concentration between watersheds. For example, carbon and nutrient concentrations are lower in high slope and elevation watersheds and higher in more productive (NPP) and peat/wetland-rich watersheds (Figure 3). By including site-specific variation in multilevel models, we were able to account for other landscape factors and more effectively understand the possible effects of wildfires. Therefore, when it is not possible to monitor sites pre-burn and post-burn, all intrinsic factors need to be taken into account to understand how streams are affected by wildfires and climate change.
Our results indicated that wildfires had a minor effect on summertime solute concentrations. The largest potential decreases due to wildfires found here were in DOC, TDN, DHg, and Na. Possible causes of these decreases could include the large amount of soil carbon combusted and lost from the watersheds (Walker et al., 2018) may decrease the amount available for aquatic export. Further, dilution may play a large role in the decreases observed. In the first few years after a fire, the reduced vegetation could lead to decreased transpiration and increased surface runoff causing dilution. Alternatively, accelerated permafrost thaw from wildfires (Gibson et al., 2018) causing greater connectivity with deeper groundwater (Walvoord & Kurylyk, 2016) could cause dilution. In the Taiga Shield, where “fill-and-spill” runoff dominates (Spence & Woo, 2003), deeper frost tables may cause runoff to be transported through deeper soil horizons that may have low DOC and solute availability (Spence et al., 2015). Alternatively, grow back of vegetation following fires may increase nutrient uptake and thus lower runoff concentrations (McColl & Grigal, 1977). Additionally, biomass burning during wildfires causes widespread Hg deposition (Fraser et al., 2018) impacting both burned and adjacent unburned watersheds. While multilevel models indicated small decreases in DHg concentrations due to wildfire, these differences were not always reflected in watershed chemical yields.
In contrast to our findings here, DOC concentrations have also been observed to increase as peatlands become more hydrologically connected to streams with permafrost thaw (Olefeldt et al., 2014; Olefeldt & Roulet, 2012). Fire-induced permafrost thaw in peatlands is more pronounced after 10 years (Gibson et al., 2018) and increases in DOC concentrations may instead be observed over longer timescales than this study. Additionally, previous studies have shown that wildfires often increase nutrient concentrations in streams due to the release of nutrients from the burned landscape, with P concentrations often increasing after wildfire, as P is not lost during combustion and nutrient demands from regenerating vegetation are initially low (Ackley et al., 2021; Burd et al., 2018). In our study, we only observed a very small potential increase in P concentrations, which may be due to factors such as flushing during freshet (Burd et al., 2018) or other confounding factors not captured with our data. Further, our study did not explicitly examine whether the observed changes were associated with different forms of nutrients (inorganic or organic), DOC composition, or methyl-Hg. It is possible that terrestrial inorganic P export could have increased, but high in-stream biological uptake may have removed it before it could be measured downstream at our sites. Additionally, the Monte Carlo uncertainty analysis indicated a range of possible outcomes, including the possibility of much more elevated P concentrations, although these were not statistically significant in our model. Thus, while our results suggest a minor effect of wildfire on P concentrations, further research is needed to fully understand the complex interactions among fire, permafrost thaw, and nutrient export in this region.
Runoff varied greatly between sites, but there was no detectable difference in runoff with watershed burn. While several other studies have recorded increased runoff in burned watersheds (Burke et al., 2005; Pelster et al., 2008), here, complex hydrological processes seem to dampen any wildfire effect. In both ecozones, bog cascade “element threshold” hydrology in the Taiga Plains (Connon et al., 2015) and lake “fill-and-spill” hydrology in the Taiga Shield (Spence, 2006) likely substantially mitigate the impact of wildfires. A detailed study of hydrological wildfire effects in Taiga Shield watersheds found no impact on open water stream flow and hydrological resilience to fire, but indications of increased winter flow and icings were observed (Spence et al., 2020). However, the study lacked sufficient winter samples to examine the possible impacts of wildfires on winter chemical yields. It is worth noting that chemical species concentrations showed much less temporal variation than stream flow and runoff. This finding may reflect the unique hydrological characteristics of the region and emphasizes the disproportionate influence of runoff on chemical yields. Future research is required to better understand the potential effects of wildfire on hydrological processes in this region, especially in light of changing climate conditions. However, it seems that the impact of wildfires on hydrology is minor in the short term, while future changes in runoff due to climate change may have a disproportionate effect on watershed chemical yields.
5 Implications
We assessed current conditions across the broad and understudied Taiga Plans and Taiga Shield ecoregions, and explored the potential impacts of wildfire within this rapidly thawing region. Overall, there was a detectable decrease in DOC, TDN, DHg, and Na concentrations and an increase in TDP, Mg, and K concentrations due to wildfires, although the magnitudes were mostly modest and uncertain. Instead, other watershed-specific properties, including climate, wetlands/peatlands, and slope, drove the majority of variation in concentrations that we observed. There was considerable uncertainty in watershed chemical yield estimates due to the overriding influence of variation in runoff. Despite the unprecedented increases in wildfires, it is likely climate and altered hydrology in this rapidly changing region that will have the most substantial future impacts on fluvial networks.
Acknowledgments
The authors extend our deepest appreciation to the Tłįcho directors and local guides in Whatì, Gamètì, and Wekweètì including April Alexis, Shirley Dokum, Gloria Ekendia-Gon, Adeline Football, Lloyd Bishop, Alfred Arrowmaker, and William Quitte, whose invaluable assistance was essential to the success of this research. The authors would also like to acknowledge Erin MacDonald and Luke Gjini for their logistical and field support, as well as the funding provided by the Northwest Territories Cumulative Impacts Monitoring Program (CIMP; grant CIMP180), Polar Knowledge Canada (Grant 1617-0009), and the Campus Alberta Innovates Program. The authors would also like to thank the Polar Geospatial Center for providing the DEMs under NSF-OPP Awards 1043681, 1559691, and 1542736.
Conflict of Interest
The authors declare no conflicts of interest relevant to this study.
Open Research
Data Availability Statement
Map of the study area (Figure 1) was made using ArcGIS (version 10.6.1). Figures 2-4 were made using open-source R statistical programming (Version 3.5) (R Core Team, 2014). All data presented in this study have been included in the Supporting Information and have been deposited in the Figshare Digital Repository (https://doi.org/10.6084/m9.figshare.23812095).
References
Erratum
The originally published version of this article contained errors in Figure 1. The symbols representing 2016/2017 Snapshot Sites, 2015-2017 Frequent Sites, and Towns should have been represented with red circles, green triangles, and black squares, respectively. The errors have been corrected, and this may be considered the authoritative version of record.