Differential impact of thermal and physical permafrost disturbances on High Arctic dissolved and particulate fluvial fluxes

Climate warming and changing precipitation patterns have thermally (active layer deepening) and physically (permafrost-thaw related mass movements) disturbed permafrost-underlain watersheds across much of the Arctic, increasing the transfer of dissolved and particulate material from terrestrial to aquatic ecosystems. We examined the multiyear (2006–2017) impact of thermal and physical permafrost disturbances on all of the major components of fluvial flux. Thermal disturbances increased the flux of dissolved organic carbon (DOC), but localized physical disturbances decreased multiyear DOC flux. Physical disturbances increased major ion and suspended sediment flux, which remained elevated a decade after disturbance, and changed carbon export from a DOC to a particulate organic carbon (POC) dominated system. As the magnitude and frequency of physical permafrost disturbance intensifies in response to Arctic climate change, disturbances will become an increasingly important mechanism to deliver POC from terrestrial to aquatic ecosystems. Although nival runoff remained the primary hydrological driver, the importance of pluvial runoff as driver of fluvial flux increased following both thermal and physical permafrost disturbance. We conclude the transition from a nival-dominated fluvial regime to a regime where rainfall runoff is proportionately more important will be a likely tipping point to accelerated High Arctic change.

. (a) Annotated-oblique aerial photograph of a localized active layer detachment (ALD) at the CBAWO. (b) Photograph of the newly formed, well-connected channels within the scar/track zone of ALDs in PT (photo location indicated in (c)). Note that surface runoff has been redirected around the intact-active layer soils in the toe/compression zone. (c) Geomorphic evolution of localized ALDs in the PT watershed. Mapping was carried out annually with a handheld Garmin GPS. Hydrological connectivity of the internal channel configuration is indicated as mapped with handheld GPS in 2010. Photographs (a/b): S.F. Lamoureux.

Particulate organic carbon (POC) method 2.1 Blank corrections
It is important to note a key difference in our method of POC determination. Here, we opportunistically use the non-combusted, 1-µm glass fiber (GF) filters from suspended sediment concentration ([SS]) determination, which differs from standard methods that use combusted 0.7-µm GF filters. Although pre-combustion of filters is recommended to eliminate any initial C and to reduce blank values, combusted 0.7-µm GF filters were not available for POC analysis in this study. All combusted filters used for DOC sampling were discarded immediately following filtration in the field. 1-µm GF filters used in this study were manufactured by Whatman®, which are free of organic binders and typically have low C blank values 2 . We determined the OC content of 25 new 1-µm GF filters to determine a blank correction value to account for the lack of combustion. The average OC content of the blank filters was 0.001 ± 0.0004 % C (Fig. S2).
Process blanks and LECO certified reference materials (LOT 1000: 10.8 ± 0.26 % C, 0.86 ± 0.03 % N) were run at the beginning and throughout every run to ensure consistency and to determine instrument accuracy and stability. Blank values for LECO foil cups (n = 200) had an average OC content of 0.033 ± 0.006 % C (Fig. S2). All samples were corrected for these trace amounts of OC in blank filters and foil cups.  S2. Boxplot of OC content (wt%) in new 1-μm glass fiber filters and LECO foil cups. All blank filters were acid fumed following standard methods prior to analysis. All samples were corrected for these trace amounts of OC.

Duplicate Sampling
Duplicate water samples (n = 46) collected from four river systems of varying watershed size (10-100 km 2 ) at the CBAWO, during different hydrological periods (nival, baseflow, stormflow) in 2017 were used to compare differences between the [POC] measured from non-combusted 1-µm GF filters and combusted 0.7-µm GF filters. Studies of [POC] in the Mackenzie River noted that using 1.2-µm pore size GF filters versus filters with smaller pore sizes showed negligible differences in [POC] [3][4] . Although the relationship is not 1:1, measured [POC] is strongly correlated (r 2 = 0.93) and within 1-standard deviation (± 1s) using 1-µm and 0.7-µm GF filters (Fig. S3). Further, small root-mean-square-error (RMSE) relative to the linear correlation suggests little variation in [POC] between filters. Together, these findings suggest that the use of non-combusted, 1-µm GF filters provide an accurate measure of [POC] in this setting. We also compare the correlation between [SS] and [POC] for the different filters ( Fig. S4). One-way ANCOVA indicates that the slopes are significantly different from each other (1-µm vs. 0.7-µm), but are within ± 1σ, further validating our use of filters in this study.

Fig. S3. Comparison of [POC]
determined from duplicate water samples filtered through non-combusted 1-μm GF filters and combusted 0.7-μm GF filter. Duplicate samples were collected across the different hydrological periods (nival, baseflow, stormflow) from four river systems of differing catchment size (10-100 km 2 ) at the CBAWO. Although the relationship is not 1:1, measured [POC] are strongly correlated and within 1 standard deviation (± 1σ) of each other, with little variation around the 1:1 line (RMSE = 0.19).

Fig. S4. Comparison of the relationship between [SS] and [POC]
for duplicate water samples filtered through non-combusted 1-μm GF filters and combusted 0.7-μm GF filters. One-way ANCOVA indicates that the slopes are significantly different from each other, but within 1 standard deviation (± 1σ).

Replicate Samples
Due to subsampling of the filters into quarters for analysis (maximum size that for the LECO), randomly selected duplicate samples from both streams (n = 32 each for PT and GS) were analyzed (   Fig. 2 as a single color denoting the dominance (>50 % of the total pluvial flux) of either DOCflux (orange), major ionflux (blue) or SSflux (black). Data available in Supplementary Table S1.  Tables   Table S1. Rainfall events with measurable pluvial responses for Ptarmigan (PT, n = 11) and Goose (GS, n = 15). Rainfall duration, intensity and estimated recurrence interval (years) are shown, as is the total pluvial runoff for each rainfall event, SSflux (POCflux is 1.4 % of SS), major ionflux, and DOCflux (kg). Seasonal timing, early (Jun), Mid (Jul), Late (Aug) indicated. Labels refer to the events displayed graphically in Fig. 2.

Site
Year Label (Fig. 2 Table S6. Statistically significant (p <0.05) slope changes in the linear relationship between cumulative DOCyield, major ionyield and SSyield (POC is 1.4 % of SS in both watersheds) for Ptarmigan (PT) and Goose (GS). Significant changes were identified using the double-mass curve approach in combination with one-way analysis of covariance (ANCOVA) using Matlab® (version 2020a; aoctool). No hydrological monitoring in 2011No hydrological monitoring in , 2013No hydrological monitoring in , and 2015. Displayed graphically in Fig. 5.