Catchment-scale thawing and greening decreases long-term nitrogen export in NE Greenland

Climate change is expected to alter nitrogen (N) export from Arctic rivers, with potential implications for fragile coastal ecosystems and fisheries. Yet, the directionality of change is poorly understood, as increased mobilization of N in a ‘thawing’ Arctic is countered by higher rates of vegetative uptake in a ‘greening’ Arctic, particularly in the understudied region of Greenland. We use an unprecedented dataset of long-term (n = 18 years) river chemistry, streamflow, and catchment-scale changes in snow and vegetation to document changing riverine N loss in Greenland. We documented decreasing inorganic and organic N loads, linked to decreasing snow stores, warming soils, and enhanced plant uptake. Higher variability in N export across years also points to the increasing role of high flow events in driving downstream N loss. This alteration in N cycling may significantly reduce both inorganic and organic N transport across the terrestrial-aquatic boundary during the open water season in a rapidly warming Greenland.


Introduction
Climate change is driving extensive landscape alterations across the Arctic.With temperatures increasing six-fold faster than the global average [1,2], warming-induced changes in vegetation and thawing of permanently frozen ground cover (e.g.glaciers, permafrost) will have significant implications for carbon and nutrient transport across the terrestrialaquatic continuum.Numerous studies have documented drastic changes to carbon cycling and transport in Arctic rivers as a result of permafrost thaw [3][4][5] and Arctic 'greening' [6,7], which is defined as the widespread positive trends in satellite-derived vegetation indices across the Arctic.However, fewer studies have explored changing nitrogen (N) transport dynamics [8][9][10].In particular, studies often lack comparisons across inorganic and organic forms of N or focus on dissolved inorganic N (DIN) alone [11][12][13][14].
Understanding changes in N cycling in this rapidly changing landscape is essential, as N limits production across the Arctic [15,16].Predominant sources of N to the Arctic include biological N fixation [17,18], atmospheric deposition [19,20], N bound to soil organic matter [21,22], and geologic sources [23].Moreover, permafrost thaw and enhanced N mineralization can provide Arctic ecosystems with novel sources of N available for plant uptake and microbial turnover [16,[24][25][26].Changes in transport from terrestrial to riverine habitats have important implications for the severely N-limited coastal regions [27,28].Coastal regions play both important economic and cultural roles, serving as critical fisheries for local indigenous communities [29,30].Should N loads to coastal regions decrease, this could further exacerbate N limitation and devastate local fisheries.Alternatively, increasing loads and warming temperatures could drive harmful algal blooms with similarly damaging effects to coastal areas.Yet, we currently lack an understanding of how N export from large Arctic rivers is altered by climate change, limiting our ability to understand and predict the directionality and magnitude of impacts to coastal ecosystems.
Despite the importance of Arctic climate change impacts, long-term records of riverine water chemistry in the Arctic are relatively rare [31,32].Documenting long-term changes is critical to unlocking linkages between climate-driven changes in terrestrial ecosystems and changing watershed N export.Most studies on riverine nutrient export in the Arctic are short-term, limited to a single summer [9,10,13,33,34] or a few years [35][36][37][38].Critically, Greenland accounts for approximately 27% of the Arctic landmass, yet research on changing N dynamics, especially at the basin scale, is lacking from this region.Several short-term studies from Northeast Greenland have explored nutrient uptake [14,39], linkages between hydrochemistry and vegetation [34], and the role of snow cover in controlling stream chemistry [12].However, long-term patterns in riverine N export from Greenland, a region in which temperatures are increasing by 0.4 • C per decade [40], have not yet been documented.
Over the past 40 years, annual discharge has increased in Arctic rivers [41].However, previous studies excluded the Greenland ice sheet from analysis; thus, the directionality of runoff changes from Greenland and implications for N export are largely unknown.Researchers have postulated increasing discharge may result in increasing N delivery to coastal areas, particularly as nitrate (NO 3 − -N) export generally increases with increasing runoff [42].Additionally, permafrost thaw and increased soil temperatures release organic and inorganic forms of N [43], which are available for transport to adjacent streams and rivers.Yet, few studies have explored how Arctic 'greening' may offset expected increases in runoff-driven N export.For example, given terrestrial production is highly N-limited [44] and the growing season is short, assimilatory uptake of N by vegetation could represent a sink of nutrients on the landscape, limiting soluble N that may runoff into waterways.Consistent with these hypothesized processes, decreasing trends in N loads have been very recently reported in large riverine systems in Siberia and North America [45], yet the extent of these systems makes it challenging to directly link trends to changes in catchment-scale land surface processes.In NE Greenland, three previous studies on small, headwater streams have linked landscape characteristics with nutrient biogeochemistry [12,34,39]; however, studies linking watershed-scale land cover to riverine chemistry in large rivers are lacking for this region and the eastern coast of Greenland.Thus, we lack information on the interaction between landscape changes and N export for large rivers in Greenland, particularly due to the limited availability of long-term data for such systems.Therefore, understanding the balance between the counteracting effects of changing runoff and vegetative cover is critical in determining how N export to coastal areas may change in the future.
We address these gaps using a unique dataset combining basin-wide remote sensing data on changing snow and vegetation patterns with a longterm (n = 18 years) record of river discharge and chemistry data from the Zackenberg River Basin (ZRB; figure 1(A)), a catchment in the high Arctic of Northeast Greenland.We note this dataset constitutes the longest and most comprehensive suite of time series available from the high Arctic [46].Over the record, we document a marked decrease in both inorganic and organic N export from the basin to the coast (figure 2), along with decreased snow cover and increased vegetation across the catchment (figures 1(B)-(D)).Notably, we do not report substantial changes in runoff during that period.Based on these findings, we hypothesize changes in N coastal export in Northeast Greenland are dominated by basin-scale changes in vegetation (Arctic 'greening'), which offset the influence of changing hydrology (Arctic 'thawing'; i.e. decreasing snow volume, changing precipitation patterns, and more frequent Glacier Lake Outburst Flood, or GLOF, events).These interacting processes may have serious implications for the N-limited coastal systems of Greenland.

Study site
The ZRB is a 514 km 2 high Arctic catchment located in Northeast Greenland flowing into the fjord Young Sound (figure 1(A)).The ZRB is representative of the predominant river type along the coast of Eastern Greenland.The ZRB belongs to the bioclimate subzone C of the Circumpolar Arctic Vegetation Map, which represents 23% of the total non-glacierized Arctic area [34].In the ZRB, altitude ranges from 0-1469 m above sea level.Vegetation is primarily found below 300 m a.s.l.[47], while barren rock and loose sediments are characteristic of higher elevations [48].The A.P. Olsen Glacier is the major source of streamflow.Additionally, approximately halfway along the river channel, there is a drainage lake that experiences GLOF events in some years [49].There is also a small fen in the valley, comprising approximately 0.4% of the catchment area.

Remote sensing
We obtained 30-meter resolution multispectral imagery from the Landsat 7 and 8 missions (Collection 2 Tier 1) from the United States Geological Survey (USGS) using the Google Earth Engine platform [50].The acquired images represent surface reflectance with calibration and atmospheric correction previously carried out by USGS as described in Chander et al [51] and Masek et al [52].In particular, this includes inter-calibration to enable time series analysis across sensors (ETM+ and OLI) [53,54].The acquired images overlap spatially with the ZRB and cover the 2000-2022 period.A detailed description of quality assessment can be found in the supplemental information.Vegetation was detected by computing the normalized difference vegetation index (NDVI) using the reflectance values captured by the red and near infrared bands of each sensor (i.e.Bands 4 and 3 for the ETM+ sensor of Landsat 7, and Bands 5 and 4 for the OLI sensor of Landsat 8).The NDVI is a widely used indicator of photosynthetic activity, and a value above a certain threshold for any given pixel is indicative of vegetative cover [55].We used a method similar to Mullen et al [56] to determine that threshold.We first constructed annual composite images formed of the 90% percentile of NDVI values that year for each pixel.We then used an unsupervised classification algorithm (Weka K-means implemented on Google Earth Engine) to segment the pixels of each mosaic into two clusters based on their NDVI value, each year recording the NDVI threshold separating the two clusters.Annual NDVI values (displayed in table S4) were then used to classify all images for the corresponding year with vegetation-covered pixels determined as having an NDVI value equal to or larger than the relevant annual threshold.The overall process classified each image into 4 classes: cloud/cloud shadows (masked), snow, vegetation, and a fourth class containing all other land cover types.
Once the classification process was completed, we created annual frequency images for snow and vegetation.These images show, for each pixel, the proportion of clear sky overpasses covered by snow or vegetation, respectively.We used annual frequency images to construct figures 1(B)-(D) by retaining the pixels with snow cover frequencies exceeding 90% of available images each year, or with vegetation coverage frequencies exceeding 10% and 40% of available annual images.Finally, we partitioned results for each 300-meter elevation band, which were obtained from the Greenland Ice Mapping Project (GIMP) [57] acquired on Google Earth Engine.

Long-term weather data
We obtained weather data from 2003-2020 (n = 18 years) for the ZRB (74 • 18´N, 20 • 18´W) from the ClimateBasis Monitoring Program, which is part of the long-term Greenland ecosystem monitoring (GEM) Program run by Asiaq (www.asiaqgreenlandsurvey.gl).For consistency across the 18 year period, we considered June 15 to August 24 as the open water season for data analysis; this represented the period during which we could consistently obtain river chemistry data for each study year.For each summer season, we obtained air temperature measurements ( • C; at 60 min intervals) from a climate station located north of the Zackenberg Research Station; we then calculated a daily mean air temperature.We also obtained summer soil temperature data ( • C; at 60 min intervals) at 10 cm depth; we calculated a daily mean soil temperature for the season.We obtained river temperature data ( • C; measured 2× daily) for the catchment outlet; we averaged measurements to determine daily average water temperature for the season.
We calculated total precipitation during the open water season, as well as the maximum snow depth measured in the winter prior to each season (1 October to 31 May) using data collected at the same climate station as air temperature.Finally, we obtained summer measurements of seasonal soil thaw depth (weekly or biweekly) from two sites within the catchment to determine the active layer thickness (ALT; in cm), or the soil depth down to permafrost.

River chemistry data
To examine riverine N export over time, we analyzed all N data available to date from the outlet of the ZRB (2003-2020; 18 years) collected by the GeoBasis Monitoring Program, as part of the GEM Program.As noted above, we analyzed the open water season (June 15 to August 24) for interannual comparisons.To determine daily discharge (Q; L s −1 ), we averaged 15-minute Q data to develop a record of continuous daily Q for each year, which we converted to daily specific runoff (RO; in mm d −1 ) by normalizing discharge by catchment area.From 2003-2011, we obtained daily grab samples for NO 3 − -N (µg l −1 ), ammonium (NH 4 + -N; µg l −1 ), and total dissolved N (TDN; µg l −1 ).From 2012-2020, grab samples for N chemistry were collected 3x per week and analyzed similarly.All solutes were analyzed on Fiastar 5000 Flow Analyzer; however, from 2015-2020 TDN was analyzed on Shimadzu total organic carbon analyzer TOC/TN.We then calculated DIN (NO 3 − -N + NH 4 + -N) and dissolved organic N (DON; TDN-DIN) concentrations.

Data analyses
We estimated cumulative seasonal N loads (concentration x Q) at the ZRB outlet using the loadflex package in R [58], which uses a composite model based on continuous daily Q and grab sample concentration data to estimate daily solute export for NO To compare NO 3 − -N vs. NH 4 + -N export, we analyzed flow duration curves to understand the proportion of DIN loss occurring across different flow conditions, which provides a composite signature of streamflow variability and its role in N export [59].Using the 18 year dataset, we split the flow duration curve into three intervals: Baseflow (0-60th percentile), Moist Flows (60-90th percentile), and High Flows (90th percentile = top 10% of flows) [60].To determine the relative contribution of each flow interval to seasonal NO 3 − -N and NH 4 + -N loads, we calculated the total NO 3 − -N and NH 4 + -N load, respectively, for all days in each flow interval category per year.

Statistical analyses
We used a one-way analysis of variance (ANOVA) for differences in both total and mean runoff, ammonium, nitrate, and DON, respectively, across the early, middle, and later periods.We treated period as our fixed effect.If a significant effect of period was detected, we used Tukey's HSD to determine which periods were significantly different from each other.
We also explored changes in climatic variables from 2003-2020 using simple linear regression (SLR), regressing air temperature, soil temperature, water temperature, total precipitation, and mean ALT, respectively, over time.We noted increases in water temperature and ALT in the final few years of the study; thus, we conducted a breakpoint analysis using the segmented package in R (version 1.4.0)[61] to statistically generate two distinct relationships for both water temperature and ALT over time.
For our flow duration analysis results, we used a two-way ANOVA to determine if the effect of baseflow, moist flow, and high flows on total NO 3 − -N and NH 4 + -N export changed across the three periods.We treated flow interval and period as our fixed effects.If a significant effect of period was detected, we used Tukey's HSD to determine which periods were significantly different from each other.We also conducted the same test to determine the proportion of NO 3 − -N and NH 4 + -N exported at each flow interval changed over the three periods.
To determine drivers of variation on N loads, we regressed loads versus soil and water temperatures, precipitation, snow, ALT and MEI, (as a holistic metric of ENSO intensity) [62].We considered p < 0.05 to be significant for statistical analyses and conducted all analyses in R (version 4.1.0)[63].

Decreasing nitrogen export to coastal areas was not explained by changing runoff
Despite predictions of both increased runoff and solute export to coastal areas as the Arctic rapidly warms [16,[64][65][66], we did not document significant increases in discharge (Q) or N loss from the ZRB.In fact, we observed a substantial decrease in cumulative loads of nitrate (NO 3 − -N) and dissolved organic nitrogen (DON) loads, but no change in ammonium (NH 4 + -N) loads over the record (figure 2).We partitioned the record into early (2003)(2004)(2005)(2006)(2007)(2008), middle (2009-2014), and late (2015-2020) periods.For the open water season, NO 3 − -N export during the late period (3230 ± 561 kg season −1 , mean ± stand-ard error) was 37% lower than in the early period (4677 ± 783 kg season −1 ; table S1).Similarly, while cumulative DON loads were comparable between early and middle periods (figure 2, Row 4), we saw a 22% decrease in DON export in the late period.However, it is important to note that high interannual variability prevented us from detecting statistically significant trends in total runoff and N loads across the three periods (one-way ANOVA; p > 0.05 for all).
This persistent decrease in N loads over time was surprising given the 5 cm increase in maximum ALT (figure 3, table S3) and the small increase in runoff (figure 2, table 1) between the early and late periods.Maximum ALT is influenced by seasonal permafrost thaw and increased steadily by approximately 0.5 cm per year during the study period (with notable outliers in 2015 and 2018, see figure 3(D)).Mean daily runoff increased by only 8% from 323 ± 10 mm to 351 ± 37 mm in the early and late periods, respectively (figure 2, Row 1; table 1).The relatively small change in Q may be linked to substantial decreases in snow cover and increases vegetative cover throughout the catchment documented via remote sensing (figures 1(B)-(D)).It also contrasts with findings from other Arctic rivers, which have documented increased discharge [41,67,68].
The long-term record from the ZRB also revealed significant interannual variability in the climatic drivers of runoff (e.g.air and soil temperatures, precipitation; figure 3).Interannual variability in Q was high and increased by 114% from the early to late period (table S3).High interannual variability may mask long-term signals of warming at local scales, though we observed sharp increases in water temperatures (figure 3(C)) and maximum ALT (figure 3(F)) after 2016.In fact, we documented the highest water temperature (5.9 • C) and deepest ALT (85 cm) in 2020, which were 0.6 • C warmer and 4.5 cm deeper than the next highest measurements for water temperature and ALT, respectively (table S2).Moreover, breakpoint analysis reveals a significant shift in water temperature and mean ALT patterns beginning in 2018 (figure S1), where both water temperatures and mean ALT increase by 1 • C per year and 6 cm per year, respectively.
As rivers are well-known integrators of spatially distributed catchment characteristics [69], rapidly warming water temperatures documented at the watershed outlet in the later years may reflect broad scale patterns in warming across the entire ZRB.Together, these results suggest climate change may be accelerating in the ZRB, reducing the stability of Greenland ecosystems over time [70].Notably, the marked decrease in NO 3 − -N and DON export to coastal areas we documented cannot be explained by corresponding changes in runoff or its main climatic drivers, which was surprising given NO 3 − -N loss, in particular, is typically tightly coupled to runoff [42].Table 1.Summer means (± standard error) in runoff and N loads for each six-year period: early (2003)(2004)(2005)(2006)(2007)(2008), middle (2009-2014), and late (2015-2020).We also calculated the percent difference (%Dif) in mean loads between the early and middle periods, as well as the early and late periods, to show changes over the study period.

Climate-related changes in land cover explained decreasing N export
Our results point to changes in land cover (i.e.snow and vegetation) as key drivers of the observed decrease in N export in the ZRB.Using biweekly satellite imagery from Landsat 7 and Landsat 8 for the 1999-2021 period, we documented a slight decrease in summer snow cover in the upper ZRB (see Methods), particularly at altitudes <1200 m above sea level (a.s.l.), which generally correspond to glacier snouts (dark blue; figure 1(D)).While vegetative cover varied substantially from year to year, we also documented an increase in persistent vegetation cover (figure 1(D), dark green), measured as the number of pixels covered with vegetation during the full length of the snow-free season (⩾40% of yearly images), predominantly at lower altitudes (figure 1(C)).This increase in persistent vegetative cover is consistent with the increase in maximum seasonal ALT measured in the lower reaches of the basin (figure 3(F)).Together, our observations in the ZRB lead us to hypothesize three distinct mechanisms through which these climate-related landscape alterations might have affected N export to the coast.First, decreasing summer snow cover in the basin can be linked to a decrease in riverine NO 3 − -N loads at the beginning of the season (June 15; figure 4, row 2), as the snow in Greenland contains notable NO 3 − -N stores derived from atmospheric deposition that can be released during snowmelt [71,72].The decrease in early season loads is particularly noticeable between the early and middle periods (figure 4), which corresponds to the period of decreasing snow cover (figure 1).Together, this suggests reduced landscape stores of snow in the ZRB decreased the amount of deposition-derived NO 3 − -N available to be transported to the river during the 'first flush' event over the 18 year study period.
Second, landscape 'greening' may also decrease N transport across the terrestrial-aquatic interface, which drives riverine N export.As a result of N limitation of Arctic plant growth [44], the increase in seasonal vegetative cover we documented in the lower ZRB basin (figure 1(C)) may contribute to decreasing riverine N export (figure 2).In addition, DON in the soil rhizosphere has long been known to be an important N source for tundra plants [73][74][75].However, recent work has documented up to onethird of the bulk N used by tundra plants is derived from soil NO 3 − -N [76], representing a much larger contribution to plant assimilation than previously thought.Thus, warmer Arctic temperatures may increase plant productivity [77,78] and could, in turn, drive increased NO 3 − -N and DON removal from the soil, resulting in less available N to be transported to adjacent rivers.In fact, we observed a negative relationship between seasonal NO 3 − -N export and ALT (figure S2, R 2 = 0.41, p < 0.01); this pattern may be driven by higher NO 3 − -N uptake rates from the soil, which is likely linked to increased vegetative cover and a relatively shallow permafrost layer in this basin.
Third, warming soil temperatures and increasing soil ALT in the late period (figure 3(F)) have likely increased DON mineralization rates in the soil [16], allowing N to cycle more rapidly in the terrestrial landscape and limiting N runoff to the river.Additionally, water column breakdown of DON may be enhanced given warmer water temperatures (figure 3(C)).While this increased mineralization can reduce DON export from the catchment, it may also help sustain NH 4 + -N loads in the river.In turn, this may explain why we documented decreases in NO 3 − -N and DON loss but observed little change in NH 4 + -N export over the 18 year period (figures 2 and 4).Warmer soil temperatures could also enhance denitrification rates.However, NO 3 − -N concentrations observed in the river were low.As NO 3 − -N concentration is a key driver of denitrification rates [79], it is unlikely that N losses to the atmosphere (as nitrous oxide or dinitrogen gas) via denitrification played a significant role in the decrease in NO 3 − -N export to the coast documented here.In fact, previous work did not detect measurable denitrification rates in Arctic soils in Sweden [80].

Increasing influence of high flows on N export
Climate warming has been shown to drive changes in precipitation patterns in the Arctic [81,82], shifting the magnitude and timing of streamflow and nutrient export.In the ZRB, this may have driven a relative increase in the contribution of high flows to N export.Flow duration analysis suggests baseflow (<60th percentile) accounts for ∼40% of annual DIN (NO 3 − -N + NH 4 + -N) export to the fjord, which was relatively consistent over the observation period (figure 5); this is a significantly greater proportion of DIN export than export at high flows (two-way ANOVA; p < 0.001).In contrast, we documented an increase in the influence of high flows in driving DIN export from the basin (figure 5); high flow contributions to total export increased from 19% to 26% for NO 3 − -N and 17% to 25% for NH 4 + -N from early to late period, respectively.Similarly, the relative contribution of moist flows (60-90th percentiles) decreased by a comparable amount (figure 5).This increased relative contribution of high flows may be partially attributed to the occurrence of more late season precipitation events, which has been shown to increase terrestrial-aquatic linkages in Arctic catchments and influence dissolved and particulate material fluxes [83].Moreover, the ZRB experiences periodic glacier lake outburst flood (GLOF) events, which have increased in frequency over the study period (11 out of 18 years), resulting in short periods (1-2 weeks) of elevated flow and material fluxes to the fjord [49].We documented two GLOFs in the early period (2003)(2004)(2005)(2006)(2007)(2008) and five and four GLOFs in the middle (2009-2014) and late (2015-2020) periods, respectively.Given the potential influence of GLOFs on summer hydrology [84], GLOF frequency in the ZRB may play a significant role in driving the increased influence of high flows to DIN export during the open water season.

ENSO was linked to interannual variability in N export
Our results also highlight the importance of interannual variability in driving N export patterns in the ZRB in Northeast Greenland.Variability in annual NO 3 − -N loads (as standard error) has decreased by 33% between the early and late period, with a concurrent 64% increase in variability in annual DON loads (table 1).Numerous studies have documented the impacts of the El Niño-Southern Oscillation (ENSO) on weather in the North Atlantic and Europe [85][86][87][88] and sea ice fluctuations [89], which points to its potential to influence interannual variability in riverine N export in Arctic regions.We investigated this relationship using the multivariate ENSO index (MEI) as a metric of ENSO intensity, which is calculated as a composite index of sea surface temperature, surface air temperature, sea-level pressure, zonal (i.e.east-west) surface wind, meridional (i.e.northsouth) surface wind, and total cloudiness [62].We found mean annual MEI varied substantially over While one study also found ENSO influenced the timing of runoff in the Mackenzie River Basin in northern Canada [90], other examples of linkages to N export are lacking.El Niño years are associated with decreased precipitation and streamflow, while overall precipitation and streamflow increase during La Niña years [91]; therefore, the fluctuations in climate driven by ENSO cycles can potentially influence runoff, the strength of terrestrial-aquatic linkages, and, in turn, N export to coastal areas.This association between large scale climate variability and N export from freshwater to the coast is poorly documented in Arctic systems and new for Greenland.The correlation between ENSO and N export is worthy of exploration in other Arctic river basins.

Conclusions
Our results have important implications for N loss dynamics in Greenland.Given the combination of decreasing snow cover at lower altitudes (figure 1(B)), and the link between higher DIN loads and greater snow cover, it is likely snow acts as a key source of DIN to the basin that is shrinking over time.Moreover, we documented an increase in vegetative cover at low altitudes (figure 1(C)) and decreased NO 3 − -N and DON loads in years with higher vegetative cover.This indicates increased vegetative cover in an N-limited environment is potentially decreasing riverine N export via increased biological activity across the basin, which has been suggested as a key driver of decreasing NO 3 − -N loads in particular [45].Together, this suggests sources of DIN within the basin are decreasing in summer, and available N may be cycling more quickly via warming soils and plant growth responses.
More broadly, this sheds light on the relative effect of hydrologic processes (the 'thawing' of the Arctic) versus biological and landscape processes (the 'greening' of the Arctic) in Arctic river basins.Combined, these have the potential to drastically alter N transport across the terrestrial-aquatic boundary and decrease N loss from the basin to the coast.Decreasing N delivery to already N-limited coastal areas may have significant cultural and economic implications for the fisheries in Greenland, on which many indigenous communities rely [29,30].Overall, our results highlight critical linkages between climate-driven changes in landscape cover and riverine N loss in Greenland, an understudied, yet substantial, region of the Arctic.

Figure 1 .
Figure 1.(A) Map of the Zackenberg River Basin (ZRB) with colors representing elevation bands.Inset shows the location of the ZRB in Northeast Greenland.(B) Time series of persistent snow and ice cover by altitude band.(C) Time series of seasonal vegetation cover by altitude band.(D) Maps of basin-wide snow and vegetation cover in summer for 2000, 2008 and 2016.Regions covered by snow or ice are represented in light and dark blue for high (>1200 m a.s.l) and low (<1200 m a.s.l) elevations, respectively.Regions covered by vegetation on >40% and on 10%-40% of annual images are represented in dark and light green, respectively.

3 −
-N, NH 4 + -N, and DON.We examined annual patterns in runoff, as well as NO 3 − -N, NH 4 + -N, and DON loads, across three six-year intervals, designated as the early (2003-2008), middle (2009-2014), and late (2015-2020) periods.By partitioning the data into three periods, we were able to detect longterm changes in runoff and N loads despite significant interannual variation.

Figure 3 .
Figure 3. (A) Mean air temperature, (B) mean soil temperature (at 10 cm depth), (C) mean water temperature at the watershed outlet, and (D) total precipitation during summer (June 15 to August 24) in the Zackenberg River Basin (ZRB) from 2003-2021.(E) Maximum active layer thickness (ALT) for the summer in the ZRB from 2003-2021.Black lines represent the three-year rolling mean, while colors indicate a shift from early (cool colors) to later years (warm colors).Note that data in panels (A)-(B), (D) are derived from a single weather station near the mouth of the river and may not reflect trends across the entire ZRB.

Figure 5 .
Figure 5. (A) NO3 − -N and (B) NH4 + -N loads over time divided by flow category: top 10% of flows (high flows), the 60%-90% of flows (mid-range flow), and the bottom 60% of flows (baseflow).Insets depict the percentage of NO3 − -N and NH4 + -N loads, respectively, that are lost at a given flow category during each of the six-year periods.

Figure 6 .
Figure 6.(A) Patterns in the annual mean multivariate ENSO index (MEI) over the study period.Both the relationships between (A) runoff and MEI and (B) DIN loads and MEI were statistically significant.Colors indicate a shift from early (cool colors) to later years (warm colors).